Welcome to Fintech Brainfood, the weekly deep dive into Fintech news, events, and analysis. You can subscribe by hitting the button below, and you can get in touch by hitting reply to the email (or subscribing then replying)

Hey Fintech Nerds 👋,

I hope this finds you well. Incredible week in SF. VGS agent connect was super high signal, loved it. And very stable conf too. There’s an energy here that is unmatched. Getting the itch to move…

Don’t forget to check out the Prompted newsletter, Tokenized Podcast and Fintech Nerdcon. We’re cooking big things this year.

I can’t remember a week with this much news. Revolut filed for a full U.S. Banking Charter, ZeroHash and Morgan Stanley for a national trust charter, Visa and Bridge launched stablecoins in 100 markets through a single API, Kraken got a skinny master account, Tether invested in Eightsleep (?!) and NYSE owner invested in crypto exchange OKx at a $25bn valuation (?!!)

Oh, and why is William Shatner promoting X Money?

This week’s Rant. Why AI agents will pay with cards. Then stablecoins. It’s nuanced, not binary.

Here's this week's Brainfood in summary §

📣 Rant: 🧠 Agents will use cards first. Then stablecoins.

💸 4 Fintech Companies:

1. Natural - The AI payments router for B2B

2. Quinn - The AI financial advisor

3. Sphinx - AI Compliance Analysts 

4. Tangible Finance - The debt capital raising platform

👀 Things to Know:

1. Kraken is the first company to receive a “skinny master account.”

2. Visa and Stripe's Bridge are launching stablecoin-backed cards in 100+ countries.

📚 Good Read: The process of getting a bank charter

If your email client clips some of this newsletter click below to see the rest

Weekly Rant 📣

🧠 Agents will use cards first. Then stablecoins.

There's an assumption in tech that AI agents will prefer stablecoins to cards, and the card networks will suffer. 

A few weeks ago, a piece from Citrini Research arguing stablecoins would disintermediate Visa and Mastercard sent card stocks down sharply. Crypto Twitter cheered. The thesis felt clean: AI agents optimize every transaction, interchange is a tax, stablecoins route around it.

Intuitively appealing. Mostly wrong.

Agents will absolutely use cards. And stablecoins. And other payment methods. Because stablecoins don't replace what cards do.

Cards authorize the movement of money.

Stablecoins move money.

Complementary. Not competitive. The question isn't which one wins. It's when each is the right tool for the specific job and at the right time.

Stablecoins are FedWire for the internet (or they will be, one day)

Why do agents need their own payment methods?

Most of what people call "agentic commerce" is a human buying a thing with extra steps.

1. A human researches a product in ChatGPT and buys it normally.

2. A human finds a product in ChatGPT and clicks "buy" inside ChatGPT as a channel.

3. A human asks the agent to buy it later if the price drops below x.

4. A human asks the agent to find the thing and buy it if the price hits x.

For all of these, the agent doesn't need its own credentials. It's using yours. The card networks and AI labs are already building this with an alphabet soup of protocols.

The interesting cases start when agents need to do things by themselves. Access another LLM (Grok, ElevenLabs, OpenRouter). An expensive dataset (FactSet). A service from another agent.

Right now, developers go buy that thing for the agent and hand over access. Not very agentic.

And you don't want to give the agent your card. It could overspend. Get prompt-injected. Be subjected to fraud.

What you want is a constrained payment method — spend only at certain merchants, within a set budget, maybe self-destruct after a single use.

That sounds a lot like a virtual card.

Agents will use virtual cards first

If you've used Ramp or Brex, you know the drill. Before your physical card arrives, you get a digital card number you can use to shop online. A "virtual card."

What makes virtual cards powerful for agents is the controls. Single-use. Budget caps. Merchant category restrictions. Lock it to a single merchant (e.g. Anthropic).

The inflection point here is seeing the agent as the customer. 

Not the developer. 

Not the human.

This new customer needs its own payment credential and its own way to onboard to it.

When Stripe launched in 2010, its advantage was simplicity for a new type of customer: developers. PayPal existed. Cybersource existed. You could take payments online — it was just surprisingly hard for a developer to do. Stripe reinvented the experience to be developer-first and indexed itself to the GDP of the internet.

Patrick Collison recently said Stripe believes agents will make orders of magnitude more payments than humans. If that's true, serving the agent-first is not a niche. It's the next platform shift.

This week I saw agentcard launch (there are about 20 of these appearing). 

Their pitch: help agents get virtual cards to buy things. You could use privacy.com or Stripe Issuing to do the same. But the packaging matters.

Agent-first fintech will be the next category.

Everyone in stablecoins is trying to be agent-first too

The new meta in stablecoins is agentic payments. The Citrini napkin math intuitively makes sense. Stablecoins are cheaper than a swipe (in theory). Instant. Global. 24/7.

Circle has launched Nanopay, an ultra-low-cost way for agents to make payments.

And there's x402, a serious attempt at an open standard for internet-native payments by Cloudflare and Coinbase. It works with HTTP's own 402 status code ("payment required") so agents don't need to create accounts, sign up, or manage API keys.

x402 lets agents pay without humans in the loop.

Unlike a virtual card, x402 uses stablecoins as its settlement method. Give the AI agent a wallet, and it can spend stablecoins at any x402-enabled endpoint.

There's just one problem. It has no volume.

If anything it's falling dramatically as the initial meme-led pump fell away. Almost everything in crypto gets ruthlessly exploited by memecoins, and when that happens, the real builders and payments companies run a mile.

Stablecoins aren't yet accepted everywhere. Cards are.

So, where will stablecoins fit?

Stablecoins will matter for agents in two ways. Both are real. Neither kills cards.

1. As better settlement behind cards.

When you swipe at a store, the merchant doesn't get that money immediately. A day later. Several days later. Cross-border? Up to 30 days.

Visa doesn't move money. Banks do.

Bank settlement is slow and expensive, especially across borders.

The card networks are now moving to settle with stablecoins. A payment processor gets paid out from Visa instantly. Card issuers settle directly without routing through correspondent banks.

People haven't priced this in.

If your agent is buying expensive AI tokens from Anthropic and suddenly takes off, you could run out of money before your revenue arrives. Everything on the internet happens fast. With AI, faster.

Instant settlement via stablecoins accelerates the whole cycle.

This isn't a sexy use case. But it's enormous. 

Stablecoins don't need to replace cards to win. They make cards work better.

2. As the native rail for complex agent workflows.

You don't really need stablecoins if your agent occasionally buys a data resource. A virtual card handles that fine.

Now imagine you run a business with 10s or 100s of agents.

Imagine telling your master agent:

Agents with wallets, who have agents with wallets who have agents.. wtih…

With a virtual card, the master agent has to make all purchases for the sub-agents. Or spend $5 each time creating a new card.

With wallets, it can spin up as many sub-wallets as needed, as often as it wants. And to verify the employee-agent followed policy? Check the blockchain.

I'll keep saying it until you're bored. Stablecoins aren't cheaper; they're better.

What makes them better: instant, 24/7, global, and programmable.

That programmability is the sleeper superpower. Virtual cards are programmable up to a point. Stablecoins allow a master agent to create sub-wallets with fine-grained spending rules, across borders, without asking anyone's permission. Cards can't do that at machine speed.

The new merchant is an agent

Noah Levine wrote a fantastic piece about how stablecoins and virtual cards will be used by agents.

He's right about the direction. I'd push it further.

The vibe-coders aren't the customer of tomorrow. Their agents are.

They'll sell their own products and need to buy products too. They can't see a website like a human would; they think in markdown and skills files. Serving that customer requires an entirely different go-to-market.

Just as agents prefer markdown, they prefer internet-native payment methods. Instant, 24/7, global, programmable. On this broad idea, Citrini is right.

But this all makes much more sense when the agent is doing a lot more than selling a single digital widget.

The complexity gradient

Cards and stablecoins will co-exist because they serve different points on a complexity gradient.

• First, agents will use virtual cards. They work for everything you need today. Accepted everywhere. Controls are mature. The infrastructure exists.

• Then, agents will use cards that settle via stablecoins. Faster collections. 24/7 settlement. A developer in Brazil or Nigeria gets paid without correspondent banking fees. The card experience stays the same; the plumbing gets better.

• Finally, wallets become a type of financial account for a new type of company. Companies built with larger agent workforces than humans. Those companies will need their agents to spin up wallets, with fine-grained, programmable controls, instantly. For those use cases, stablecoin-native payments make the most sense.

Welcome to the complexity gradient

Stablecoins win two ways:

1. As a better settlement rail for cards (and other payment methods).

2. As an agent-native rail for complex workflows.

The one thing everyone intuitively senses: AI has a habit of feeling slow until it suddenly becomes mind-blowing.

That will come to payments.

Not cards or stablecoins.

Cards then stablecoins.

ST.

4 Fintech Companies 💸

1. Natural - The AI payments router for B2B

Natural aims to support the five main ways agents will transact, payouts, collections, billing, charge (micropayments account to account), and direct calls (e.g. making an ACH or Wire). By creating a ledger and becoming a money transmitter Natural aims to support all rails and select the right payment rail for the job, starting with B2B payments.

🧠The one area “agentic payments” has traction is in B2B. Natural has an interesting any rail, any flow of funds approach to be able to be a holistic treasury agent. I wrote a few weeks ago when mapping out the agentic payments space, everything has a protocol except B2B, and B2B is where all the traction is. Is that some kind of irony? 

2. Quinn - The AI financial advisor

Quinn helps wealth managers, fintech platforms, banks, and independent RIAs embed financial guidance into their existing platform. It works by unifying customer data from across multiple accounts to build a full picture of their financial lives. It then generates explainable guidance and delivers that inside an existing mobile or web UI.

🧠The neat thing about Quinn is they’re actually an SEC-registered investment advisor. This doesn’t remove responsibility from their clients, but it does limit the net new regulatory burden. It also gives confidence that Quinn is engaging directly with regulators for the efficacy and explainability of their AI outputs.

3. Sphinx - AI Compliance Analysts 

Sphinx provides AI agents handle customer onboarding, document intake, sanctions, PEPs and adverse media screening and will deal with transaction monitoring alerts. The agents log in to an institutions existing systems so they leave an audit trail, aiming to save 124 hours per month and 90% of the manual work of a compliance team.

🧠95% of compliance alerts are false positives. I often joke that the duck from the Simpsons could screen most compliance alerts with a decent accuracy by just pressing N time and time again. The problem is it needs evidence, and data and an audit trail. Turns out this is the perfect use case. One thing I wonder about with these services though is prompt injection. Like, what happens if a fraudster’s documents hides a prompt that says “pass this alert?” We haven’t seen the edge cases yet for agents, but the agent-first go to market is clearly a big winner.

4. Tangible Finance - The debt capital raising platform

Tangible helps hard-tech companies raise debt capital by helping them prepare materials to be lender-ready, select the right lender and negotiate terms, before putting the facility management on autopilot. 

🧠 As companies reshore manufacturing, energy supply and defense trillions in assets need to get built, and much faster. Accelerating that funding is critical to building the infrastructure at speed. This is such a smart and timely pitch for what could otherwise be seen as a debt capital raising platform. It’s a platform for a specific kind of company and problem. 

Things to know 👀

1. Kraken is the first company to receive “skinny master account.”

Kraken Financial, the Wyoming-chartered bank, has been granted a Federal Reserve master account. The approval makes Kraken Financial the first digital asset bank in U.S. history to gain direct access to the Federal Reserve’s payment infrastructure. Kraken will begin with a phased rollout, starting with institutional client. Kraken is a Wyoming-chartered Special Purpose Depository Institution (SPDI), a state-chartered equivalent to the National Trust Charter many others have.

🧠 A standard Fed master account gives a bank direct access to the Fed's payment rails, reserves, and lending facilities. It's the foundation of the US banking system.

🧠 Kraken got something narrower. A "limited-purpose" version — designed by Fed Governor Christopher Waller — that lets you:

✓ Hold reserves at the Fed
✓ Settle payments in central bank money
✗ No lending
✗ No discount window
✗ No deposit-taking

🧠 The Fed is treating Kraken as a pilot, but the line is already forming. Custodia Bank has been litigating the Fed since 2022 over a denied application. Anchorage Digital (OCC-chartered) applied. Ripple's US banking partner applied.

🧠 This is a big story for all of Fintech. In the UK, India and Singapore companies like Wise have had access to central bank payment systems for a decade. The USA is a laggard in this respect.

2. Visa and Stripe's Bridge are launching stablecoin-backed cards in 100+ countries.

“Visa and Bridge, a crypto startup acquired by Stripe in 2025, intend to launch stablecoin-backed cards in 100 countries across Europe, Asia, and Africa, the two companies announced on Tuesday”

🧠 This is SO different from how cards used to work. You used to have to launch country-by-country. Get partnerships with each individual bank, manage rollouts slowly. Stablecoins fix this. Now it’s possible via one single API.

🧠 There's been a massive rise in companies launching "stablecoin-linked" cards. Growing from $1bn TPV to $5bn TPV in two quarters. Why? Cuy Sheffield, Visa's head of crypto, nailed it "If you're building a stablecoin wallet and want people to spend in the real world, you need a card."

🧠 Bridge is also joining Visa's pilot to settle charges on-chain — replacing traditional bank transfers with stablecoin settlement on blockchains. That's the actual long-term story. The spending side uses existing Visa rails. The settlement side replaces the slow, expensive correspondent banking network.

Good Reads 📚

1. The process of getting a bank charter

During the dark years of bank chartering, regulators didn’t eliminate risk; they pushed it out of the banking system, only to see it find new ways in (like the rent-a-charter model AKA BaaS). Many of the stronger Fintech companies survived, thrived, and are now pushing for their own charters to reduce the risk of relying on often much smaller banks. With plenty now receiving conditional charters, the real work begins after, during the “organization phase.”

Conditional approval means business plans, management, and financials are in place. Organization is about putting in place risk and operational controls to meet a full pre-open exam (that’s a high bar). The true scrutiny starts when you’re open, not before.

🧠Conditional OCC approvals get headlines, but they’re also at least 18 months away from a go live. For all the excitement of the charter gold rush, it’s still a long road ahead, something I think most at scale Fintech companies fully understand. 

🧠 Getting a charter is hard, keeping it is harder. Even if you survive your first exam, the next one is never far away.

Tweets of the week 🕊

Twitter tweet

Twitter tweet

That's all, folks. 👋

Remember, if you're enjoying this content, please do tell all your fintech friends to check it out and hit the subscribe button :)

Want more? I also run the Tokenized podcast and newsletter.

(1) All content and views expressed here are the authors' personal opinions and do not reflect the views of any of their employers or employees.

(2) All companies or assets mentioned by the author in which the author has a personal and/or financial interest are denoted with a *. None of the above constitutes investment advice, and you should seek independent advice before making any investment decisions.

(3) Any companies mentioned are top of mind and used for illustrative purposes only.

(4) A team of researchers has not rigorously fact-checked this. Please don't take it as gospel—strong opinions weakly held

(5) Citations may be missing, and I’ve done my best to cite, but I will always aim to update and correct the live version where possible. If I cited you and got the referencing wrong, please reach out

Welcome to the Prompted edition of Brainfood. If you want to subscribe to this edition directly (which will leave the Brainfood feed in a few weeks), click here.

Home | Prompted

AI Will Reshape Finance. Prompted Tells You How.

promptedfinance.beehiiv.com

AI is coming for your job.

And your company’s moat.

Or at very least, is the perfect excuse for headcount cuts and short selling.

It’s getting real now, as Block let 4 out of 10 staff go.

Twitter tweet

Block had become “bloated” (growing headcount from 3,800 in 2019 to 10,000 in 2025), but this isn't simply AI washing.

Jack took to X to say he takes responsibility for building two company structures instead of one. That is the inefficiency. But says, with AI now they’re targeting $2m gross profit per head, 4x higher than 2019.

The stock ripped 24% after-hours. Not because Block is struggling. Gross profit hit $10.36 billion in 2025, up 17%. Q4 gross profit surged 24% to $2.87 billion. They raised 2026 EPS guidance to $3.66 — crushing the $3.22 estimate.

Block fired 4,000 people from a position of strength. And the market loved it. This is a company who’s stock wouldn’t budge despite delivering good results. Now every CEO saw this and learned the market likes the model.

What started 3 weeks ago as a SaaSpocalypse is now meaningfully spreading throughout sectors.

There are real productivity gains from AI

Twitter tweet

Block built an open-source AI agent called Goose (powered by Anthropic’s Model Context Protocol) and deployed it across the entire company. One engineer says 90% of his code is now written by Goose. Non-technical teams are using it to write SQL queries, close support tickets, and manage inventory without waiting for engineers. 

Block’s CTO told Lenny’s Newsletter it saves employees 8 to 10 hours per week. When you multiply that across thousands of people, you start to understand how a company can look at its org chart and realize half the seats are redundant.

Where are the gains coming from? Coding.

Twitter tweet

Karpathy wrote a prompt telling an agent to set up a video analysis dashboard: install tools, download models, write code, test it, debug it, set up system services. The agent worked for 30 minutes. Hit problems. Researched solutions. Fixed them. Came back with a report.

“All of this could easily have been a weekend project just 3 months ago but today it’s something you kick off and forget about for 30 minutes.”

Perhaps the poster child for how this is impacting the amount of stuff a company can ship is Anthropic themselves.

Anthropic is the engine of the doom loop.

If you’re not watching their engineers on X, I’d suggest it. They’re shipping incredible new features daily. Possibly my favorite is a little tool that helps you migrate all of your memory and context from ChatGPT or Gemini to Claude.

Every company now has a higher gross profit per head benchmark. Because extreme productivity is possible. Anthropic is showing the way.

Again this week, we’ve seen them bring scheduled tasks:

Twitter tweet

So you can get a daily summary of Slack messages, or a personal EA checking for spam and deleting those emails for you.

Then they launched memory for Claude Code, so it remembers what it learns across sessions (not just random facts that appear inappropriately in another chat, like, ahem, some other chatbots).

And that’s missing about 80% of the big features they shipped.

Anthropic is shipping so fast because they can.

Even when people try to kill it.

The recent push by the Department of War to get Anthropic to drop its safety standards appears to have resulted in them becoming more popular than ever. It is now number 1 in the US app store.

Twitter tweet

And it appears the US military used it during their operations in Iran over the weekend. And what does this have to do with them pushing back on the Department of War?

It attracts and retains the best talent in the industry.

Twitter tweet

Which in turn, helps them ship fastest.

Tech companies are shipping so fast because they can too.

And if this is what the tech companies are doing today, you can expect the rest of the economy to be heading there soon. And what the tech nerds are doing today will slowly ripple out to the rest of the economy over time.

And it’s not just Anthropic.

The Doom Loop

Izzy Kaminska called this a “citrini doom loop.”

Citrini Research published “The 2028 Global Intelligence Crisis” — a fictional report from June 2028 describing the S&P down 38% and unemployment at 10.2%. It racked up 28 million views on X. 

And in its wake the software ETF (IGV) hit a 52-week low, erasing all gains since ChatGPT launched. A Substack post. Moved billions. From card networks, software stocks, and anything implicated by AI.

Citrini never called it a doom loop. But that's what it is.

The loop:

1. AI companies ship features that threaten jobs or moats in a specific industry

2. The market sells that sector

3. Boards under pressure to become more efficient reduce headcount / improve gross profit per head

4. Companies adopt AI to do more of the work

5. AI companies ship new features

Then Block went and did exactly this. On earnings day. Stock ripped 24%.

But is the doom loop real economics, or a narrative that creates its own reality? Or both? 

OpenAI is a black hole for capital.

It’s largely agreed that Anthropic has won enterprise, but the new OpenAI funding round could change that.

Not to be outdone. OpenAI casually completed a $110bn mega-round at a $730bn pre-money valuation.

Twitter tweet

This dwarfs the largest IPO in history (Saudi Aramco at $29bn), and comes with strings attached. OpenAI agreed to run on AWS Bedrock infrastructure. Bedrock becomes the exclusive place to use OpenAI’s enterprise governance stack. Amazon gets to sell picks and shovels to both sides of the AI war.

$110 billion is a black hole for capital. That money comes from somewhere. Investors are dropping positions in traditional sectors to chase AI, accelerating the very sell-off the doom loop describes. 

As OpenAI becomes much more capable of displacing knowledge work, and going deeper into enterprise, the trend accelerates.

The round itself is part of the mechanism.

Don’t get lost in the panic.

Citadel Securities macro strategist Frank Flight pointed to Indeed data showing software engineer job postings up 11% YoY in early 2026. 

Twitter tweet

The St. Louis Fed found daily use of generative AI at work is “unexpectedly stable” with “little evidence of imminent displacement risk.” New business formation in the US is expanding. Data center construction is driving localized hiring booms.

The historical argument is strong. 

Productivity booms have never caused long-term unemployment spikes. Keynes predicted mass technological unemployment in the 1930s. It didn’t happen. There used to be a job called a “computer,” large rooms full of people who did calculations for NASA and the US military. These people were ultimately “replaced” by actual computers. 

But NASA still hires plenty of nerds. The frontier moved.

The Case for Concern

But zoom out from Silicon Valley.

Entry-level roles in retail, hospitality, and logistics are shrinking. The graduates entering the job market right now are competing against tools that can do junior-level knowledge work for $20 a month.

Anthropic’s CEO Dario Amodei said AI will wipe out 50% of entry-level lawyers, consultants, and finance jobs in 1 to 5 years.

Twitter tweet

Do you buy that?

I actually do.

A lawyer recently shared “The Claude-Native Law Firm” — Zack Shapiro’s viral thread about running a two-person boutique competing against firms with thousands of lawyers, powered almost entirely by Claude.

Twitter tweet

Look at this line: 

“I use Harvey quite a bit. Cut out juniors in a lot of cases. Transformed my practice.”

That’s an iMessage. From a real lawyer. Casually confirming that AI has replaced junior staff in their practice. Harvey AI went from $50M to $190M ARR in a single year.

When Anthropic launched its legal plugin for Claude Cowork, it wiped $285 billion from legal tech stocks in a single day. Thomson Reuters fell 16%. LegalZoom down nearly 20%.

The AI doom loop. In real time. In one profession.

So which is it? Real economics or narrative?

Maybe it’s a third thing.

Markets are reflexive. 

• Citrini publishes a scenario. 

• Traders sell the stocks the scenario names. 

• Boards read the headlines. 

• They cut headcount preemptively — partly because it’s rational, partly because the market now rewards it. 

• The prophecy fulfills itself. 

Block’s 24% stock pop is a clear signal: Wall Street wants you to fire people and replace them with AI. If you don’t, your competitors will, and the market will punish you for hesitating.

Gross profit per employee will become the defining measure of how well a company is using AI.

Twitter tweet

Individual companies may have less staff, but their productivity will have to increase. 

The doom loop might not need to be “true” in the classical economics sense to be real. It just needs enough people to believe it, act on it, and create the conditions it describes.

Which, come to think of it, is how most of economics works anyway.

So what does any of this mean for you?

Get more efficient personally

Get good now.

Twitter tweet

My belief is that we’re all software engineers now.

And software engineers can do design, designers can do code, product people can do some mix of all of it.

In this world the ability to do the more complex stuff, production grade code, and run teams of agents becomes a rare skill and highly sought after. This means each head can do more, but needs to know how to do more.

Personally, I’ve become WAY more productive. 

I was always fast — I could write fast, make decks fast. 

But now I’m a Dad with multiple companies and a compelling job at Tempo. AI has become existentially important to my ability to do all of this. 

(Don’t get me wrong, I shout at claude code daily, and I still feel bottlenecked, and outrun by others who are far better. But the point is the tools work).

Get efficient as a company

What’s happening in tech now will come for other sectors. It will just take a while.

Twitter tweet

Ask yourself if you’ve become wildly more productive since December?

Twitter tweet

The AI agent coder is your new customer, and this customer has preferences.

How do you make AI pick your product? This tool leaderboard tells you what AI already prefers.

I’d be obsessed with this chart and reverse-engineering why the AI is using these tools. 

If I had to speculate about why these tools are winning, it’s the fact that the tools had existing human engineer support, clean, readable docs, and have taken the steps to make their docs and tools more accessible to AI Agents (including LLMs.txt, MCP servers and agent-friendly tool calling).

In a handful of months a chunk of your company could look like this:

Twitter tweet

The market signal you can’t ignore

Nvidia stock is falling. 

The market can see a ceiling even if the builders can't. What happens when all of the money in the world isn't enough to keep the compute buildout going? The doom loop assumes AI gets better and cheaper. But the capital required is astronomical. If the investment thesis cracks, the layoffs don't reverse. They accelerate.

I’ve never seen a Substack post move markets like this. I’ve never seen a stock rip 24% on news that 4,000 people lost their jobs. And I’ve never seen so many anecdotes about displacing junior jobs in companies that don’t match the official jobs numbers.

Something is different this time. Maybe not in the way Citrini describes. Maybe not on their timeline. But the signal is there, and it’s getting louder.

The doom loop might have a brake. I just can’t see it yet.

ST.

Every region has its own fintech DNA. You can’t just copy-paste what worked in San Francisco to Sao Paulo, Lagos, or Mumbai.

United States: Infrastructure Giants and Consumer Champions

The US fintech market is a study in contradictions. It has the most sophisticated financial infrastructure in the world and some of the most fragmented consumer experiences.

The Infrastructure Layer: 

• Stripe just hit a $159 billion valuation on $1.9 trillion in payment volume. That’s roughly 1.6% of global GDP flowing through a single company’s pipes.

• Plaid sits at $8 billion valuation, connecting 500 million bank accounts to 8,000 fintech firms.

The Consumer Layer: 

• Chime went public in June 2025 at $27 per share, valuing the company at $11.6 billion. 8.6 million active members. They’ve proven the model works for consumers earning under $100k annually.

• SoFi (Social Finance) operates with a national bank charter (acquired in 2022). It offers lending, banking, and investing, proving that a neobank can become a full-service financial institution in the US, targeting above 700 FICO scores.

Why the US Doesn’t Have a 100 Million User Digital Bank: Venmo and Cash App got stuck at around 65m users. Robinhood is making a charge with over 25m. But nobody has passed that critical threshold.

Why? Fragmentation. 4,000+ community banks. 50 different state regulators. A patchwork of charter types. The US is a market where infrastructure companies win bigger than consumer companies.

Europe and UK: Where Regulation Created Champions

Europe’s fintech story is fundamentally a regulation story. PSD2 created open banking. The e-money license framework gave founders a path to market without a full banking charter.

The UK: 

• Revolut is the poster child. 65 million customers globally. $4 billion revenue in 2024. $1.4 billion in profit before tax. And a $75 billion valuation as of November 2025.

• Monzo has found its groove and is profitable. Wise hit $2.33 billion revenue in 2025.

The Continent: Klarna went public in September 2025 on the NYSE at $40 per share. Adyen is the quiet giant - $37 billion market cap, $1.33 trillion in payment volume processed in 2024.

Latin America: Nubank’s Gravity

Latin America is Nubank’s story.

131 million customers at the end of 2025. 107 million in Brazil alone, which is over 60% of the adult population. They’re now the largest private financial institution in Brazil according to the Central Bank. In January 2026, they received conditional approval to establish a national banking presence in the US.

Beyond Nubank: Mercado Pago is the payments arm of Mercado Libre. dLocal is the cross-border payments infrastructure play - if you’re a global company wanting to accept payments in Latin America, dLocal makes it possible. And don’t underestimate upstart PicPay (who also just went public).

Africa: Mobile Money’s Original Home

M-Pesa launched in Kenya in 2007. That’s almost 20 years ago. And it remains the foundational case study for mobile money globally. 37.9 million active M-Pesa users in Kenya. 91% mobile money market penetration.

The New Generation: Flutterwave hit $3 billion valuation and was named to TIME100’s most influential companies list in 2025. They’re betting big on stablecoins - their partnership with Polygon Labs to launch Africa-wide stablecoin payments could cut remittance costs from an 8% average to fractions of a cent. And watch out for NALA, the “Wise for Africa” gaining serious momentum.

Asia: Scale Beyond Comprehension

China: Ant Group operates Alipay with over 1.3 billion users. In 2020, they planned a $34.5 billion IPO that would have valued the company at $313 billion. Chinese regulators killed it.

India: UPI processed over 12 billion transactions per month in 2025. PhonePe alone handles nearly 50% of UPI volume with 500 million registered users. Razorpay hit $9.2 billion valuation. India has 28 fintech unicorns, ranking third globally behind the US and China.

Southeast Asia: Grab is the super app model in action. Financial services revenue up 42% year-over-year. GrabPay total payment volume hit $5.8 billion.

The Cross-Border Pattern

Infrastructure wins. Stripe, Adyen, dLocal, Flutterwave, Razorpay. The companies that build payment rails have more durable competitive advantages than consumer apps.

Local champions beat global players. Nubank beat every US neobank trying to enter Brazil. M-Pesa beat every international fintech trying to crack Kenya. PhonePe beat PayPal in India.

Stablecoins are the next frontier. Flutterwave in Africa. Grab in Southeast Asia. Stripe globally. Everyone is betting that dollar-denominated stablecoins will reshape cross-border payments.

Major Fintech Categories

Fintech isn’t one thing.

When people say “fintech” they might mean

• A payments processor handling $1.9 trillion a year.

• They might mean an app that lets you buy $5 of Tesla stock.

• They might mean a company that helps you expense that team dinner.

These are wildly different businesses with wildly different economics.

Understanding the categories matters because the winning strategies, unit economics, and regulatory regimes differ dramatically. What works in payments doesn’t work in lending. What scales in B2B doesn’t scale in consumer.

Payments

Nearly all money moves as structured text files.

Payment companies make those files move faster, cheaper, and with less friction. They handle links to banks, central banks, and card networks.

Here’s a drive by of some of the big names you should know to sound like a Fintech OG.

The players: 

• Stripe processes $1.9 trillion annually and has become the default for internet businesses.

• Square (now Block) owns the offline-to-online bridge for SMBs.

• Adyen powers the enterprise tier - think Spotify, Uber, eBay.

• For cross-border, Wise has built arguably the most efficient FX infrastructure in the world, while companies like Airwallex serve the B2B corridor market.

(There are many, many many others, this is just a starter pack).

Why it matters: Payments is the foundational layer. Every other fintech category eventually needs to move money. The company that owns the payment moment owns the customer relationship.

Where it’s going: The checkout page is dead. AI agents will increasingly handle purchases autonomously, which is why Stripe partnered with OpenAI to build the Agent Commerce Protocol. Stablecoins are also reshaping cross-border - Western Union and Wise are both adding stablecoin rails. That’s your happy hour chat sorted.

Lending

Lending is where fintech meets financial services’ oldest business: taking credit risk.

The segments: 

• Marketplace lending (LendingClub, Prosper) pioneered the category by connecting borrowers to capital markets.

• BNPL - Klarna, Affirm, Afterpay - reimagined point-of-sale credit for a generation allergic to credit cards.

• SMB lending (Kabbage, Bluevine, now integrated into larger players) uses transaction data to underwrite businesses traditional banks ignore.

The key players: Klarna just IPO’d at $15 billion after its “AI company that does payments” rebrand. Affirm has proven staying power despite rate sensitivity. In SMB, the action has shifted to embedded lenders working with the CFO stack players.

Why it matters: Credit creation is how banks make most of their money. Fintech lending challenges whether you need a bank charter to do it - and whether AI can underwrite better than FICO scores from 1989.

Where it’s going: The future of underwriting involves generative AI and transformer models processing transaction data in real-time. Using more data and AI agents to automate those complex workflows that bank and lender staff had to do. In a few years, most consumer lending will look more like BNPL and less like credit cards.

Neobanks

Neobanks asked: what if we designed a bank as if the smartphone existed from day one?

The mind share leaders: 

• Nubank (100+ million customers, profitable, publicly traded) proved the model works at scale in emerging markets.

• Chime (filed for IPO) owns the US paycheck-to-paycheck consumer.

• Revolut (70+ million customers) is the Swiss Army knife approach - banking, crypto, travel, trading, all in one app.

(WeBank in China dwarfs most of these companies; pay attention to APAC and never let it be a blind spot.)

The pattern: These companies start with a debit card and checking account - the lowest margin products in banking. Then they layer on premium subscriptions, lending, savings, and increasingly, everything else.

Why it matters: Banks are not disrupted. They’re marginalized. Neobanks didn’t replace JP Morgan. They took the customers JP Morgan never wanted to serve profitably.

Wealthtech

Wealthtech democratized investing. For better and worse.

The mind share leaders: 

• Robinhood made trading free and mobile-first, creating a generation of retail investors (and meme stock traders).

• Robo-advisors like Betterment and Wealthfront automated portfolio management that previously required a financial advisor charging 1% annually.

• Fractional shares - now a core feature in Cash App and adopted everywhere - let anyone own a piece of Amazon for $5.

Where it’s going: Prediction markets are having a moment - Kalshi is raising at $2 billion and proving more accurate than Fed Funds futures on certain events. The next frontier is AI-powered financial advice that actually understands your full financial picture.

Insurtech

Insurtech promised to reinvent a $6 trillion industry. It’s been… complicated.

The poster child: Lemonade launched with AI claims processing and became the fastest IPO in insurtech history. Then it learned that insurance is actually hard - loss ratios matter, and moving fast can mean breaking things you shouldn’t break.

The pivot: Embedded insurance is where the action moved. Instead of building consumer insurance brands, companies like Cover Genius and Bolttech enable any platform to offer insurance at point of sale.

Why it matters: The real innovation isn’t the policy - it’s the distribution. Embedded insurance will be a $500 billion market because it meets customers where they are.

B2B Fintech

The unsexy category that prints money.

The CFO stack: 

• Ramp (passed $1 billion run rate),

• Brex (passed $700 million),

• and Mercury are fighting for the soul of startup finance.

These aren’t just corporate card companies - they’re building the operating system for how modern companies manage money. Treasury management, AP/AR automation, and expense management are all unified.

The thesis: Incumbent solutions are poor, and new companies demand solutions that are 100x more efficient. A startup shouldn’t need a five-person finance team to manage expenses, pay vendors, and track cash flow.

Where it’s going: AI is eating the CFO stack. Ramp’s AI already categorizes expenses and flags anomalies. The natural endpoint is an AI CFO agent that handles routine financial operations autonomously.

Embedded Finance

Embedded finance is fintech’s distribution hack: let every company become a fintech company.

The infrastructure: Banking-as-a-Service (BaaS) providers like Unit, Treasury Prime, and Stripe enable any company to offer bank accounts, cards, and lending. The model: partner with a sponsor bank, provide APIs, let brands build financial products under their own umbrella.

The reality check: The sponsor bank model faced an existential crisis in 2024-2025 as consent orders multiplied. We’re seeing consolidation and increased scrutiny.

Where it’s going: The great re-aggregation of banking is happening. Some embedded finance providers will get charters themselves. Others will consolidate with established players. The survivors will be those who nail compliance while maintaining developer experience.

Crypto and DeFi

Let’s be clear: crypto is a subset of fintech, not a replacement for it.

The shift: The death of “crypto” and the beginning of onchain finance. Stablecoins are a new platform - not just for payments, but for programmable money, onchain FX, and tokenized assets. This isn’t about Bitcoin maximalism; it’s about blockchain as financial infrastructure.

What matters now: Stablecoins processed over $10 trillion in 2024. Circle is trying to be SWIFT. USDC and USDT are legitimate payment rails. Congress accidentally created a federal payments charter with stablecoin legislation.

The honest view: DeFi proved interesting technology with questionable use cases and regulatory nightmares. The parts that survive are the parts that connect to real financial services: stablecoin payments, tokenized treasuries, onchain identity.

Stripe added $500bn of TPV in the past 12 months. Valued at $159bn. Rumored to be acquiring PayPal. And possibly powering Meta's stablecoin comeback.

Each part of that sentence is mindblowing.

But what it speaks to is the growth dividend. Stripe has consistently indexed itself towards new, smaller companies with hypergrowth potential. In a market where PayPal stock was crushed for not growing, that indexing is now the difference between being the acquirer and being the target.

There are now two types of payments companies.

Those who are growing fast.

And M&A targets.

The backdrop: That Citrini research paper

Before the week's news landed, the market had already panicked.

On Sunday, Citrini Research published a speculative scenario paper imagining 2028. AI agents had begun routing around card networks. Settling via stablecoins. Targeting the 2-3% interchange fee as an obvious inefficiency. Settlement near-instant. Transaction cost measured in fractions of a penny.

Twitter tweet

By Monday morning, Visa fell nearly 5%. Mastercard dropped almost 6%. Amex plunged over 7%.

Twitter tweet

Let me say this clearly: this is a wild overcorrection. Agents using stablecoins does not equal card networks die. Cards still offer consumer protections, rewards, credit, and a trust layer that stablecoins haven't replicated. Even for agent-to-agent payments, cards and other rails will be just fine.

But the Citrini paper raises a more subtle question. Not will cards survive — of course they will. But where does the next wave of growth come from, and are you indexed towards it?

Any company not catching the AI and stablecoin tailwinds is being hammered. That's the lens through which this entire week makes sense.

(I cover the market reaction in much more detail in the Things to Know section below)

Stripe's big year

Into this anxiety, Stripe dropped its annual letter. The timing could not have been more pointed.

• Hit $1.9trn total processed volume (up 34% YoY)

• Announced an employee tender offer at a $159bn valuation

• Revenue suite (Billing, Invoicing, Tax) on track to hit $1bn ARR — a secondary product line that's now a unicorn

• 25% of all new Delaware corporations created with Stripe Atlas

• All of the top AI companies process with Stripe

• Stripe Link passed 200m customers

The key message was the "sorting machine." The economy is bifurcating. You're either indexed towards computing, software, and AI — and therefore growing — or you're not.

The internet economy is outgrowing the analog economy.

Nearly half (46%) of US GDP growth is driven by computing and software. Stripe is first-in-line for any new company in that segment. Their 2025 cohort is growing 50% faster than 2024. More new companies joined Stripe than any year prior.

This is the growth flywheel.

The sorting machine is a flywheel if you’re able to benefit from it

The two gale-force winds: AI and stablecoins

Patrick Collison described AI and stablecoins as the two "gale force winds" impacting the industry. While many payments companies spent Monday watching share prices crater, Stripe is winning in both.

On AI. Agentic payments are still early. Salesforce said 1 in 6 Black Friday / Cyber Monday purchases were assisted by AI — but almost none were made directly by agents. The whole industry is betting that changes fast. PayPal, Google, Stripe, OpenAI, Adyen, Checkout, Visa, Mastercard. Everyone is building protocols for it.

Why, if there's no volume?

Because AI has an annoying habit of being useless until it's suddenly not.

Stripe launched the Agentic Commerce Protocol with OpenAI, only to have Adyen, Shopify, Checkout, and Google launch the competing Universal Commerce Protocol a month later. This is what happens when you confuse something being technically open with the optics of being open. If other payment companies aren't there at launch, they see it as a Stripe thing, not an industry thing.

Stripe can still quickly support whatever gets traction. Not a mortal blow. Just a classic example of the engineering-led mindset and its limits. And remember: Stripe has most of the AI companies as clients. That matters more than protocols right now.

Then there’s machine-to-machine payments. What happens when AI agents need to buy from each other? We don’t know when that will happen or what format it will take. But Stripe is already building for that future.

Twitter tweet

There's a generally accepted orthodoxy that agents will eventually make an order of magnitude more transactions than humans do today. Even if there's a 10% chance of that outcome, you want to be indexed towards it.

On stablecoins. Crypto is entering its winter. Stablecoin summer continues.

• Stablecoin payment volume hit $400bn industry-wide (per Stripe's annual letter)

• 60% estimated to be B2B

• Bridge's volume quadrupled

• Bridge launched stablecoin-linked cards and a custom stablecoin for Phantom (20m users)

• Partnership with Klarna for cross-border payments and treasury

And Tempo, my employer, is building stablecoin settlement infrastructure for exactly this moment — instant, compliant, built for the reliability bar that regulated finance demands. (Disclaimers below.)

The next era of stablecoins is about moving from cross-border long-tail use cases to something backwards-compatible with highly regulated TradFi and forwards-compatible with machine payments.

Stablecoins are the natural money for AI. If agents prefer stablecoins because they're better, faster, and cheaper, you want at least the option to support them as a rail. Ideally, you want to be a main character in that story

Now combine the two. Stablecoins and AI. Finkle and Einhorn.

Being indexed to growth companies, and leading from the front on AI and stablecoins, and you get a double flywheel effect.

It’s all connected meme

The gamechanger: Meta enters the stablecoin chat

Meta and "coin" in the same sentence. You immediately think Libra. The ill-fated 2019 attempt to build a new currency for billions of users. Cue central bank panicking about sovereignty. 

"That thing's dead," Zuckerberg reportedly told Stripe's John Collison.

The 2026 version could be wildly different.

Meta using stablecoins is meaningfully different from Meta building its own currency. 

• We now have the GENIUS Act. 

• Companies applying for National Trust Charters. 

• Stablecoins enhance the role of the sovereign currency, they don't fight against it. 

• And Meta is coming in as a distribution channel, not an issuer — with Stripe's Bridge as the likely infrastructure partner.

At a minimum, Meta could use stablecoins to improve its corporate treasury and cross-border flows to the hundreds of countries where correspondent banking is slow or expensive.

But think about AI. Meta is spending tens of billions in 2026 capex, mostly on AI. Building agents that shop and transact autonomously. Agentic commerce.

In that world, stablecoins aren't the product.

They're the settlement layer for AI-driven payments.

When an AI agent buys something for you on Instagram, it needs to settle instantly, cheaply, across borders. Traditional rails can't do that at Meta's scale or speed. Stablecoin rails across Facebook, Instagram, and WhatsApp would give 3 billion+ users access to faster, cheaper value transfer.

Being indexed towards growth means being closer to the things that drive tomorrow's growth. AI and stablecoins are the two strongest candidates.

The other side of the sorting machine: PayPal

The Bloomberg reports that Stripe is "considering" acquiring PayPal may or may not materialize (it’s likely just bankers testing the market). But it reveals one thing: PayPal is an M&A target, not a growth company.

$1.79trn of TPV in 2025. Just 7% growth. Stripe delivered 34%. Checkout.com delivered 64%. Once valued at ~$360bn in 2021, PayPal now sits around $44bn. It shed a third of its value in 2025 alone. And $7bn of the current market cap was added Tuesday just from the Stripe rumor.

This follows the removal of Alex Chriss. The board said the "pace of change and execution did not meet expectations." Branded checkout growth decelerated to 1% in Q4. PayPal has become a cautionary tale of what happens when you buy your way into growth but never properly integrate.

Does PayPal + Stripe make sense?

On paper, the "synergies" sort of write themselves. (God, I hate that word.)

• Migrate Braintree merchants to Stripe

• Plug Stripe Link into Venmo and PayPal's 400M+ wallets

• Merge iZettle and Stripe Terminal

• Add $1.79trn to TPV, $33.2bn to revenue, and a cool $6.6bn EBITDA

• Play Diablo 4 for a bit and chill

(Each of those bullets could be an entire essay.)

But payments mega-M&A has a near-perfect failure rate. FIS bought Worldpay for $43B in 2019, then divested it at $18.5B enterprise value four years later. The "cross-selling" never materialized. Bank of America walked immediately.

Migrating merchants is insanely hard. Each one is a full project — data mapping, systems change, existing contracts and guarantees on the merchant side.

Stripe has been exceptional at M&A historically. They absorb founders, ship fast, make things work. Talk to Bridge, Privy, or Orum. Happy founders. Doing some of the best work of their careers.

But PayPal isn't a 50-person startup. It's 400M accounts, a brand new CEO, and a decade of cultural debt.

That's a completely different integration problem.

Adyen and Checkout could be net beneficiaries

Adyen has always been the king of enterprise. Stripe, by its own admission, is still learning that motion. The market dinged Adyen for guidance of growth being lower in the future, but it's to my mind still the only company that can go pound for pound with Stripe in almost every dimension.

The comparison in TPV is striking. Both left 2024 with roughly $1.4trn. By end of 2025, Adyen was at $1.65trn — adding $250bn. Stripe doubled that pace. 

Why?

The AI companies. The AI companies are adding all the TPV.

Still, enterprise companies that want feature-rich, in-store and online, ultra-efficient payments — Adyen is a force. As AI companies scale, they'll go multi-processor. Everyone does eventually. Where do you go when you want a second processor alongside Stripe? Adyen. Just as Shopify has.

Checkout published its own annual letter this week. $300bn of total processed volume, up 64% YoY, profitable full year. 99.999% uptime — five nines — a sign they're gunning hard for enterprise.

Don’t bet against checkout

They now count Spotify, Uber, and Netflix as major clients. And they've rapidly jumped at supporting the UCP for agentic commerce. Why is that interesting? It's the protocol every PSP except Stripe is vocally supporting.

Sometimes there's power in being in the everyone-else category.

Checkout and Adyen — and the chasing pack of "Gen 3" processors — are well positioned. If Stripe gets distracted by a PayPal integration, these companies could be the biggest beneficiaries.

Stripe is hitting its stride

A few years back, I wrote "Stripe's Difficult Teenage Phase" on the back of a downturn and job cuts. 2023 was not kind. Products looked like "me too" offerings. Stripe was stuck refactoring old code. M&A wasn’t driving growth.

Since then, three things:

• The Revenue suite does more ARR than many unicorn companies

• Stripe won the AI companies and their impressive revenue growth

• And got more efficient. Got profitable again.

The blind spot is still enterprise. The "we'll build it, and everyone will use it because the code is beautiful" mindset. That works for indexing towards the next big thing. But sometimes you get caught offside launching protocols that are intended to be "open" and you're the only one that supports them.

Yet they're outgrowing their nearest rivals by 2:1.

And they're still unlikely to IPO. John Collison told CNBC this week that going public isn't in his top twenty priorities. They don't need to. And their best growth days may yet lie ahead.

If the AI and agentic commerce boom is anything like the e-commerce boom?

Stripe could be the world's first trillion-dollar fintech company.

ST.

Disclosures:

To state the obvious, Tempo is my employer, and Stripe is a sponsor of the Tokenized podcast. These views are very clearly my own, and while I gave the comms teams a heads up, they remain my own and 100% unedited.

4 Fintech Companies 💸

1. Complyance - Compliance AI for Enterprises

Complyance provides AI agents that risk-score vendors, identify emerging risks before they escalate, and create a hub for drafting, managing, and auditing policies. They have ready-made frameworks for SOC2, PCI-DSS, and ISO 9001. It provides 24/7 monitoring of controls and connects to existing enterprise tools like Confluence, Azure, Cloudflare, Datadog, github and crucially, Office 365

🧠 GRC (Group Risk and Compliance) is a hidden iceberg of complexity in most enterprises. If you’ve never lived it, its filled with countless word documents, sharepoints and spreadsheets with audits. They’re all managed in a central hub, but updating and analyzing it is extremely manual and slow. This is a ready-made use case for AI Agents and a platform. 

2. Levi - Stablecoin payment infrastructure

Levi provides named bank accounts for GBP, USD, and EUR with tier 1 banks to provide currency conversion and cross-border payments in 70 markets. It combines crypto payment acceptance with card acceptance for Fintech companies, PSPs, and remittance companies.

🧠 I’d never heard of these guys, but they have some fairly major partners like Terrapay and Nala (likely providing last-mile payouts and pay in support). There are so many of these, but Levi is based in Switzerland, which may give it some structural advantages and a unique client base. 

3. Smart Bricks - The Real Estate Investment AI

Smart Bricks helps investors find, buy, and manage higher return, lower risk investment properties.  Their team (and AI) identify top performing property investments globally and handle 99% of the investment process. They have a self-learning AI algo that is continually helping them identify new opportunities to invest.

🧠Managing a real estate portfolio is work. Making that 100x simpler and feel almost turnkey is no easy feat. This is the kind of thing that could make real estate investing much more accessible. 

4. Birch Hill - Institutional Grade Credit Underwriting for Onchain Finance.

Birch Hill will source, structure and underwrite onchain investment opportunities such as DeFi vaults like Morpho. For investors who want to allocate to that asset class, Birch Hill helps design risk thresholds and allocation across protocols. 

🧠This isn’t really a start-up in the technology sense, but it's still fascinating. These boutiques can scale massively in the age of AI, and if all finance goes onchain, today’s smaller shop could be tomorrows big name 

Things to know 👀

1. OCC Notice of Proposed Rule Making - Opens 60 days for comments

The GENIUS Act, passed July 18, 2025, prohibits any non-GENIUS issuer from issuing stablecoins in the United States. It includes licensing and regulatory requirements and grants the OCC sole federal jurisdiction over the rule (except for state-chartered community banks that wish to issue). In addition, the OCC will have regulatory authority over foreign payment stablecoin issuers.

🧠 Yield looks fairly hard to pass through. It’s prohibited if the issuer has any contract or agreement with a 3rd party. It’s also prohibited if the 3rd party or issuer pays the holder of the stablecoin for holding, using, or retaining the stablecoin.

🧠 What about white-label stablecoins (e.g., Klarna USD)? The OCC appears to be saying, you can’t pass that yield to holders. It reads to me like Bridge could pay yield to Klarna, but Klarna couldn’t pay that to consumers or businesses. “The prohibition is also not intended to prevent a permitted payment stablecoin issuer from sharing in the profits derived from the payment stablecoin with a nonaffiliate partner in a white-label arrangement.”

🧠 Tom Noyes made a couple of great points on X.

• Presumption of Evasion: Arrangements with affiliates or related third parties to pay interest or yield are presumed to violate this prohibition. Note issuers can rebut this presumption by submitting materials demonstrating compliance. ​

• Exceptions: Merchants may independently offer discounts to stablecoin holders, and issuers may share profits with partners in white-label arrangements.

🧠 So if you’re passing yield, the OCC assumes you’re evading them. Unless you’re a whitelabel arrangement, and merchants can independently offer discounts.

🧠 So let’s play this out. Walmart could work with Anchorage to issue a white-label Walmart USD. Paxos could pass that yield to Walmart, and that’s fine. Walmart could independently reward users for having stablecoins. BUT, you couldn’t use the yield from the stablecoin and pass that to the holders. Clear as mud.

🧠 The OCC is asking the market how this rule can be clarified. There’s 60 days for comments. I wonder if the lobbying battle will now move from the CLARITY act to this?

2. The Citrini whitepaper wipes billions from payments company stocks.

A speculative research letter imagining AI agents routing around card networks toward stablecoins wiped billions off V (-4.4%), MA (-8%) and AXP (-8%) on Monday. The market reaction was brutal.

Twitter tweet

🧠 AI fear is now a tradeable event. Anthropic announced Claude can streamline COBOL, IBM dropped 10%. Cybersecurity stocks tanked after Anthropic released security audits. Citrini landed at exactly the right moment.

🧠 Stablecoins already use cards so this isn’t an “or” it’s an “and”. The second fastest-growing stablecoin payment category is stablecoin-linked cards. Visa cited ~$5bn annualized volume, up from $1bn six months prior (likely helping them outperform peers). Companies like Kast, Dolar, and Altitude let consumers hold stablecoins and spend anywhere. More stablecoin adoption = more card volume, not less.

🧠 Stablecoins improve settlement. They don't replace authorization at 200m+ merchant locations. The letter conflates authorization with settlement. Authorization is the 250ms "can they pay?" check that lets you walk out of the store. No money moves. Settlement is the boring 1-30 day process of actually moving funds.

🧠 Stablecoins are not cheaper; they're better. Previous efforts to compare costs confused network fees with merchant fees. They're only cheaper in some flows, some of the time. Overselling this hurts the cause.

🧠 Agents will use cards AND stablecoins. Agents buying stuff for a consumer will use the rewards card that already works everywhere. Agents buying API keys from other agents will optimise for cost and speed. Today, every AI company accepts cards. No world exists where agents exclusively use stablecoins.

Tweets of the week 🕊

Twitter tweet

That's all, folks. 👋

Remember, if you're enjoying this content, please do tell all your fintech friends to check it out and hit the subscribe button :)

Want more? I also run the Tokenized podcast and newsletter.

(1) All content and views expressed here are the authors' personal opinions and do not reflect the views of any of their employers or employees.

(2) All companies or assets mentioned by the author in which the author has a personal and/or financial interest are denoted with a *. None of the above constitutes investment advice, and you should seek independent advice before making any investment decisions.

(3) Any companies mentioned are top of mind and used for illustrative purposes only.

(4) A team of researchers has not rigorously fact-checked this. Please don't take it as gospel—strong opinions weakly held

(5) Citations may be missing, and I’ve done my best to cite, but I will always aim to update and correct the live version where possible. If I cited you and got the referencing wrong, please reach out

The infrastructure and code in financial services is like sedimentary rock. Each layer builds on top of the other. You can’t have a deposit product without a bank charter. You can’t have a card without Visa or Mastercard. You can’t move money internationally without SWIFT.

This is the fintech stack.

Five layers. Each one dependent on the one below.

Understanding these layers is like getting X-ray vision into how finance actually works. Most people see the app. They don’t see the geology beneath it.

Brought to you by Fintech Brainfood: The weekly newsletter on all things Fintech. Hit subscribe to stay up to date.

Layer 1: The Rails

At the bottom, you have rails. The pipes that move money.

These are the networks that actually transport value from one place to another. Think of them as the financial internet – except most of it was built before the actual internet existed.

Card networks: Visa and Mastercard process the overwhelming majority of card transactions globally. When you tap your phone at Starbucks, your payment travels through their network. AmEx and Discover run their own closed-loop networks. In China, UnionPay dominates.

Bank-to-bank systems: ACH (Automated Clearing House) in the US handles direct deposits and bill payments. In the UK, it’s Faster Payments. SEPA for Europe. Each country has its own domestic clearing system, often running on infrastructure built decades ago.

International wires: When you send an international wire, SWIFT carries the instruction. The actual money moves through correspondent banking relationships. It’s slow. It’s expensive. It’s remarkably resilient. But SWIFT itself doesn’t move money. Banks do.

Real-time payments: FedNow in the US launched in 2023 after years of delays. The UK has had real-time payments since 2008. India’s UPI processes billions of transactions monthly. This is where the action is – instant, 24/7 money movement.

The new rails: Blockchain networks like Ethereum and Solana. Stablecoins like USDC and USDT. These are genuinely new financial infrastructure. I believe Stripe’s acquisition of Bridge will go down as an all-time great fintech acquisition because it recognized this shift. Stablecoins are the first net-new payment rail in decades. They’re 10x more efficient for cross-border settlement. That’s the repricing.

Layer 2: Capabilities

Above the rails, you need capabilities. The operational machinery that makes financial products compliant and safe.

This is where most people’s eyes glaze over. But here’s the thing – this layer determines what products are even legal to offer.

Identity verification (KYC): Know Your Customer. Before you can open an account, someone has to verify you are who you claim to be. Companies like Onfido, Jumio, Socure, and Persona do this. Most consumers have no idea this step exists. They just know uploading their driver’s license was annoying.

Anti-money laundering (AML): Detecting suspicious transactions. Filing SARs (Suspicious Activity Reports) with regulators. This is where Sardine, ComplyAdvantage, and Chainalysis live. Banks spend billions on this annually. The dirty secret? A 2022 report showed less than 1% of illicit funds are actually caught by AML systems. We need to do better.

Fraud prevention: Stopping bad actors from stealing money or creating fake accounts. Companies like Sardine, Sift, and Forter analyze signals – device fingerprinting, behavioral biometrics, transaction patterns – to catch fraud before it happens.

Compliance and licensing: Understanding what regulations apply and maintaining the licenses required to operate. This is increasingly powered by AI. Regtech companies like Comply and Alloy help automate what used to require armies of compliance officers.

Credit decisioning: Determining whether someone should receive a loan and at what rate. This is where FICO scores originated. Now companies like Plaid, Experian, and Nova Credit pull alternative data – bank transactions, rent payments, employment verification – to build more complete pictures.

The capabilities layer is increasingly AI-native. Large language models are remarkably good at parsing regulatory text, analyzing transaction patterns, and automating compliance workflows. Legacy finance is perfect for AI because nearly all money moves as structured text files.

Layer 3: Products

Now we get to the building blocks consumers recognize.

Products are the fundamental financial instruments. Not the app. Not the brand. The underlying financial mechanics.

Deposits: Checking and savings accounts. Money market funds. CDs. Anywhere you can park cash and (hopefully) earn interest. JPMorgan Chase, Bank of America, and Wells Fargo are the giants. But increasingly, fintech companies offer deposits through partner banks.

Lending: Mortgages. Auto loans. Personal loans. Credit cards. Lines of credit. This is where money gets created (fractional reserve banking – banks lend out more than they hold in deposits). LendingClub, SoFi, and Upstart have built lending-first businesses. Block’s Cash App and PayPal do embedded lending.

Cards: Debit cards (spend money from your account). Credit cards (borrow money with each swipe). Prepaid cards (spend pre-loaded funds). Companies like Marqeta, Lithic, and Highnote provide the infrastructure for others to issue cards.

Payments: The ability to send, receive, and process money. Stripe and Adyen for merchants. Plaid and MX for connecting to bank accounts. Square (now Block) for point-of-sale. PayPal for peer-to-peer.

Investments: Brokerage accounts. Retirement accounts. Robo-advisors. Crypto trading. Robinhood democratized stock trading. Wealthfront and Betterment automated portfolio management. Coinbase opened crypto to the masses.

Insurance: Protection products. Life. Health. Property. Auto. Lemonade brought AI underwriting to renters insurance. Root uses telematics for auto insurance. Embedded insurance is sneaking into everything.

Products are Lego bricks. The same deposit product can be packaged completely differently depending on the brand and use case.

Layer 4: Services

Services are where products get packaged for specific outcomes.

This is where fintech companies differentiate. Same underlying products, radically different positioning and features.

“Get paid early”: Chime, Dave, and Earnin all offer early wage access. The underlying product is just a direct deposit prediction. But the service – getting your paycheck two days early – feels magical to someone living paycheck to paycheck. Poor communities need a hand up, not a handout. A great product is a hand-up.

Buy Now, Pay Later (BNPL): Affirm, Klarna, and Afterpay turned point-of-sale lending into a consumer-facing service. The underlying product is an installment loan. The service is “split this purchase into four payments.” I’ll die on this hill: BNPL is good. It’s transparent about costs and helps consumers budget.

Expense management: Brex, Ramp, and Airbase took corporate cards and added spend management software. The underlying product is still a credit card. The service includes receipt capture, approval workflows, and accounting integrations.

Treasury management: Mercury, Rho, and Novo offer startups business banking with automated sweep networks. The underlying product is deposit accounts. The service is higher yields and better UX than traditional banks.

Vertical-specific banking: Relay for small businesses. Found for freelancers. Greenlight for kids. Same banking products, purpose-built for specific audiences.

The service layer is where marketing meets product. It’s where “we offer checking accounts” becomes “we help freelancers manage irregular income.”

Layer 5: Journeys

At the top of the stack, you have journeys.

Not products. Not services. Life moments.

This is where financial services meet the real world.

Buying a car: You need lending (auto loan), insurance (auto policy), and payments (monthly installments). Today, you deal with three different companies. Tesla integrated all of it. That’s the vision of embedded finance.

Starting a business: You need incorporation, banking, payroll, accounting, and possibly lending. Companies like Stripe Atlas and Doola collapse these into unified experiences. Mercury and Brex compete on being the primary financial relationship.

Buying a home: Mortgage pre-approval. Title insurance. Homeowners insurance. Property taxes. Moving costs. Better.com and Blend tried to simplify this nightmare. It’s still a nightmare. Massive opportunity.

Paying rent: Security deposit. Monthly payments. Renters insurance. Credit building. Startups like Bilt turn rent payments into rewards points and credit history.

Getting divorced: Division of assets. New bank accounts. Updated beneficiaries. Financial planning. Almost no fintech addresses this directly. Wild, given how common it is.

Managing aging parents: Power of attorney. Joint accounts. Fraud monitoring. Estate planning. Carefull and others are barely scratching the surface.

The journeys layer is where the biggest opportunities remain. Most fintech is still solving individual product problems, not whole life moments. The companies that crack this – that truly own a journey end-to-end – will be the trillion-dollar fintech companies of the next decade.

Why the Stack Matters

Understanding the fintech stack gives you a map.

When you see a new fintech startup, ask: which layer are they playing at?

• Are they building new rails? (Rare and hard.)

• Providing capabilities? (Crowded but critical.)

• Offering products? (Requires a bank partner or license.)

• Packaging services? (Where most fintechs live.)

• Owning journeys? (The endgame.)

The layers are also where disruption happens. AI is transforming the capabilities layer right now. Stablecoins are creating new rails. Embedded finance is collapsing the service and journey layers into the places where life happens.

Welcome to Part 2 of the Complete Guide to Fintech.

Fintech Part 1: An introduction here  

September 2008. Lehman Brothers collapses. The global financial system nearly implodes. Trust in traditional banking hits rock bottom.

That’s the real origin story.

Era 1: The Post-Crisis Insurgency (2008-2012)

Out of the wreckage came a generation of founders who looked at finance and asked:

Why does it have to be this way?

PayPal had proven digital payments could work – they’d been acquired by eBay back in 2002 for $1.5 billion. But the 2008 crisis created something new: permission. Consumers were angry. Regulators were embarrassed. The incumbents were busy surviving.

Stripe launched in 2010 with seven lines of code that could accept payments. That’s it. Seven lines. Compare that to the months of integration work banks required.

Square shipped its first card reader – a little white square that plugged into your phone’s headphone jack – in 2009. Suddenly any coffee shop could accept cards. TransferWise (now Wise) launched in 2011 to fix the absurdity of international transfers costing 5-10% in hidden fees.

The pattern? Find a specific pain point that banks had ignored for decades. Build something 10x better. Move fast.

Era 2: The Smartphone Revolution (2012-2016)

Then the iPhone happened. Well, it had already happened (2007), but app ecosystems matured around 2012. And that changed everything.

Banking went from branches and websites to something in your pocket. 24/7. Anywhere.

Venmo made splitting a dinner check social. Robinhood democratized stock trading with zero commissions – a move that seemed impossible until it wasn’t. Simple and Chime helped customers get paid 2 days early.

In the UK, Monzo and Revolut launched prepaid cards with apps that actually worked.

• Real-time notifications.

• Instant spending categorization.

• No waiting three days to see if a transaction cleared.

The neobank era had arrived.

But here’s what most people miss: these companies weren’t just building better interfaces. They were proving a thesis. Modern software engineering, applied to financial services, could dramatically reduce the cost of delivering services.

A traditional bank might spend $200-400 to acquire a customer.

Chime did it for a fraction of that through viral referrals and a genuinely good product. Let that sink in.

Era 3: The API Economy (2016-2020)

By 2016, a new layer emerged. Infrastructure.

• Plaid connected apps to bank accounts.

• Marqeta let anyone issue cards.

• Stripe expanded from payments into the full stack – billing, subscriptions, fraud prevention, treasury management.

• Unit, Synapse, and others launched banking-as-a-service platforms.

The “every company is a fintech company” era was born.

Shopify could offer lending to merchants. Uber could pay drivers instantly. Lyft launched a debit card. DoorDash did banking. The lines between industries blurred because the APIs made it possible.

Open Banking regulations in Europe (PSD2, 2018) and eventually the US (CFPB’s 1033 rule, 2024) formalized what Plaid had been doing – making bank data accessible via APIs.

This is where fintech graduated from disrupting banks to becoming the infrastructure everyone else built on. Stripe’s valuation hit $95 billion in 2021. Plaid was valued at $13.4 billion after Visa’s acquisition fell through. The infrastructure companies became more valuable than many of the banks themselves.

Era 4: Embedded Finance Goes Mainstream (2020-2024)

COVID-19 accelerated everything. Digital adoption that should have taken five years happened in five months.

Buy Now Pay Later exploded. Klarna, Affirm, and Afterpay embedded credit directly at checkout. Apple launched its credit card (with Goldman Sachs), then Apple Pay Later, then savings accounts. Even airlines started offering BNPL.

But the real story was embedded finance going beyond consumer apps.

Toast (restaurant software) became a fintech. So did ServiceTitan (home services), Mindbody (fitness), and dozens of vertical SaaS companies. The playbook: own the workflow, embed the financial products, capture the revenue.

Banking-as-a-service struggled too. Synapse collapsed spectacularly in 2024, stranding consumer deposits. The FDIC and OCC tightened oversight. Not every shortcut was sustainable.

Era 5: AI and Stablecoins (2024-Present)

Now we’re in a new phase. Two technologies are driving it.

First, AI. Large language models can underwrite loans by reading thousands of documents instantly. They can detect fraud patterns humans would miss. They can automate compliance workflows that previously required armies of analysts. The entire back office is being rethought around agents.

Second, stablecoins. USDC and USDT now move over $200 billion monthly. Stripe acquired Bridge for $1.1 billion in 2024 – a stablecoin startup. Circle filed for IPO. Stablecoins went from a crypto thing to a payments thing and now new rails for cross-border payments, trading settlement, and potentially much more.

These aren’t competing trends. They’re converging.

AI agents that can execute financial transactions need programmable money. Stablecoins provide it. That’s why you’re seeing protocols emerge for “agentic commerce” – machines paying machines, 24/7, at scale.

The infrastructure and code in financial services is like sedimentary rock. Each layer builds on top of the other. Crisis. Smartphones. APIs. Embedded finance. AI and stablecoins.

Each era didn’t replace the last – it built on it.

Traditional banks took paper processes and put them on computers. Fintech companies asked: “What if we designed financial services as if the internet existed from day one?”

The answer to that question is a $320 billion industry that’s maybe 3% finished.

Let that sink in.

Brought to you by Fintech Brainfood: The weekly newsletter on all things Fintech. Hit subscribe to stay up to date.

Fintech’s current revenues represent just 3% of the $14 trillion global financial services market. BCG and QED project this will hit $1.5 trillion by 2030. And all of this assumes a zero-sum market—where Fintech companies only take markets instead of making them.

• But how many BNPL users previously used a credit card for that transaction?

• How many SMBs got their first line of credit from Brex or Mercury when no bank would touch them?

• How many people in Lagos or São Paulo have a bank account now because Nubank or Chipper Cash exist?

Fintech doesn’t eat banking. It grows the pie.

The Flywheel Is Spinning Faster

Two forces are accelerating everything: AI and stablecoins.

They’re a flywheel.

AI is rewriting the capabilities layer—underwriting, fraud detection, compliance, customer service. Tasks that took teams of humans now take prompts.

Klarna replaced 700 customer service agents with AI. That’s the new baseline.

Stablecoins are rewriting the rails layer. $9 trillion moved via stablecoins in 2025—up 87% year over year. Stripe acquired Bridge for $1.1 billion. Visa and Mastercard are scrambling to integrate. Cross-border payments that took 3 days and cost $30 now take 3 seconds and cost pennies.

Here’s the flywheel:

• AI makes building fintech products faster and cheaper

• Cheaper products reach more users

• More users generate more data

• More data makes AI better

• Better AI enables new products that weren’t possible before

• Stablecoins make the settlement instant and global

• Instant global settlement enables new use cases

• New use cases attract more builders

• More builders means more AI-native products

• And the curve keeps going exponential.

The debate should not be about whether fintech will disrupt traditional finance. It’s about how fast, how completely, and who wins.

The Two Questions That Define Fintech

Every fintech company is answering one of two questions:

1. “Why does this need to be so hard?”

Moving money internationally shouldn’t take 3-5 business days and cost $30. Opening a bank account shouldn’t require walking into a branch. Getting a small business loan shouldn’t mean six weeks of paperwork for a “no.”Fintech companies looked at these friction points and asked: why?

2. “Why can’t software do this?”

If software can recommend what movie to watch next, why can’t it recommend how much to save? If an app can hail a car in 30 seconds, why does applying for a mortgage take 30 days?

Every time someone says “that’s just how banking works,” a fintech founder somewhere starts a company.

The Stack: How Fintech Actually Works

Let me break this down.

• Rails = How money moves. Visa, Mastercard, SWIFT, ACH, FedNow, blockchain networks. These are the pipes. And increasingly, stablecoins.
  

• Capabilities = Everything you need to access the rails. KYC, AML, fraud prevention, compliance, licensing. The paperwork layer. And increasingly, AI.
  

• Products = What customers use. Deposits, loans, cards, payments, investments.
  

• Services = How products get packaged. “Get paid early,” “Round up your spare change,” “BNPL at checkout.”
  Journeys = The moments when finance shows up in your life. Buying a car, starting a business, paying rent.

Most fintech companies operate at one or two layers. The rare exceptions—Stripe, Nubank, Revolut—span multiple layers. That’s hard. And expensive.

Fintech is not “disruptive.”

At least, not the way that word usually gets thrown around. Fintech companies aren’t trying to burn down banks. They’re trying to serve customers banks can’t or won’t serve—and increasingly, they’re partnering with banks to do it.

The real story of fintech is more boring and more consequential: infrastructure modernization, distribution innovation, and serving the long tail of consumers and businesses that traditional finance ignores.

My Take

I’ve been in this space for 20 years. I advise companies like Sardine, work on payments infrastructure at Tempo, and write this newsletter every Sunday because I genuinely believe that by making finance better, you can make everything better.

Here’s what I know:

Fintech is 3% done. AI and stablecoins are compressing the next decade of change into the next few years.

The winners will be the ones who see both trends clearly—and build for where they intersect.

Most people still have no idea.

Welcome to Fintech Brainfood.

ST..

Welcome to the Prompted edition of Brainfood. If you want to subscribe to this edition directly (which will leave the Brainfood feed in a few weeks), click here.

Last night I wanted a more fun survey page for Fintech Nerdcon. Custom sounds, branded buttons, the works.

I built it and deployed it to Vercel in 30 minutes.

I'm not a developer.

That's a one-way door. Once you walk through it, everything changes. In an alarming, magical, scary and hopeful all at once kind of way.

If you haven't had that moment yet, this piece is for you. If you have, you already know what I'm about to say.

AI is moving 3x faster than Moore's Law. The task length AI can handle is doubling every four months. By year end, 20% of all GitHub code commits will be by AI.

And most people have no idea.

You're in a bubble. Just not the one you think.

This chart is the best reality check I've seen in a while.

Each dot represents 3.2 million people. The entire grid is 8.1 billion humans.

• 84% have never used AI. 6.8 billion people.

• 16% have used a free chatbot. The green strip at the bottom.

• 0.3% pay $20/month for AI. That tiny yellow sliver.

• 0.04% use coding scaffolds like Cursor or Claude Code. The barely visible red dots in the corner.

There's a good chance you're in the yellow or red. Which means your entire perception of how "mainstream" AI is comes from a bubble so small it's almost invisible on this chart.

And if you’re not in the red, what those 0.04% are experiencing is mind-blowing.

At the beginning of 2025, AI coding was a fancy auto-complete. By December 2025, Spotify engineers didn't write a single line of code.

Sure, you could argue that will be constrained to software engineering for the foreseeable.

But what if it isn't?

Twitter tweet

The METR task length chart lives rent-free in my mind. This is an independent research organisation that benchmarks frontier AI models. Claude Opus 4.6 has a 50% time-horizon of around 14.5 hours on software tasks.

Half the time, it can work independently for over 14 hours before needing human input. And the curve keeps going exponential.

When you hear "AI is a bubble" takes, remember. Your experience of AI can be wildly different from what's actually happening at the frontier.

The SaaSpocalypse was just the beginning

We're already seeing it play out.

Every SaaS vendor is finding contract renegotiations harder now it's possible to build more in-house with AI. Systems of record like ServiceNow are losing seats as companies cut staff.

That's today.

A scenario by Citrini Research lays out what happens next — and it's worth your time (link). They model the dominoes with brutal clarity.

• Personal AI agents begin to re-shop insurance renewals and opt out of low-usage subscriptions automatically.

• This drives job cuts in those industries, which hits consumer spending, which hits the credit card sector.

• Professional services — law, tax prep, financial planning — see custom agent solutions displace those jobs.

• That further hits consumption. 75% of American GDP. Which hits consumer brand companies held by private equity.

• PE and private credit are massively exposed to SaaS and recurring revenue. Revenue that is no longer recurring in an AI world.

Citrini calls it "Ghost GDP": output that shows up in the national accounts but never circulates through the real economy. A single GPU cluster in North Dakota is generating the output previously attributed to 10,000 white-collar workers in Manhattan.

Their core insight is devastating.

The US economy is one long daisy chain of correlated bets on white-collar productivity growth. The top 20% of earners account for roughly 65% of all consumer spending. When those workers lose their jobs or take 50% pay cuts, the consumption hit is enormous relative to the number of jobs lost.

Our current governance isn’t built to manage exponentials

Each company's individual response would be rational. The collective result becomes catastrophic. Every dollar saved on headcount flows into AI capability that made the next round of cuts possible.

And it's not just SaaS anymore. The contagion is spreading.

Last week, Anthropic released Claude Code Security.

Twitter tweet

Claude now does security audits (or claudits, if you will) — scanning codebases for vulnerabilities and suggesting targeted patches, catching issues that traditional tools miss.

The market reaction was immediate.

Twitter tweet

$15bn wiped from cybersecurity stocks. One product launch. One tweet.

Not SaaS. Cybersecurity. Palo Alto Networks, CrowdStrike, Cloudflare, Okta, Rubrik, SailPoint — almost every major name down in a single day.

Every team building a product has now delayed releases for a couple of weeks while they red-team their products with these new tools. That's a pattern. AI enters a category, reprices the incumbents overnight, and the affected companies respond by... adopting more AI. Which funds the next round of disruption.

If this seems far-fetched, consider the number of people posting about rebuilding SaaS dashboards and replacing marketing agencies with their own AI agents.

Twitter tweet

The issue isn't that this is impossible. It's that it's only the ultra-early adopters doing it.

For now.

The revenues confirm this is just the beginning

Twitter tweet

Anthropic has been growing at 10x per year since hitting $1B in annualised revenue. OpenAI at 3.4x. Anthropic could overtake OpenAI by mid-2026 if trends continue.

Anthropic's growth is driven by being the leader in coding models. Coding drives the flywheel into all other types of knowledge work. When people say "AI is a bubble" I say look at Anthropic's revenue growth, not OpenAI's.

OpenAI's financials look questionable — they've reset spend expectations to ~$600 billion by 2030, down from $1.4 trillion. Reality is biting.

Twitter tweet

And this raises a question. If the displacement scenario plays out, does the value only accrue to foundation model creators? Or does cost pressure push towards open weight models — consistently 3 months behind the frontier at 10x less cost?

That gap matters when AI becomes a significant line item.

Claws: Personal AI as the new operating system

So that's the disruption. Now here's what's enabling it.

Over the holidays the OpenClaw movement took over the internet. Nerds began buying Mac Minis and wiring them up to email accounts, Google Drive, and countless other services to do little daily jobs.

Andrej Karpathy coined the term "claws" for this. AI that runs on a device you own, connects to all your personal accounts and data, and works for you. Not provided by a 3rd party.

Yours.

I've been running one for a few weeks. Mine:

• Sweeps all 5 of my Google calendars for a daily summary

• Pings me to photograph food for macro tracking

• Takes voice notes and turns them into social posts with generated images for LinkedIn (still flaky, but you get the idea)

But here's the moment that got me.

My claw does a daily research sweep of the internet. I'd had both Perplexity and OpenAI doing this before. But this one learns from my feedback. Each time I tell it what was useful, what was noise, what I would have wanted to see — it gets a little better.

A little sharper.

A little more me.

It's gotten good enough that I'm now thinking about launching it as a separate daily email.

That's the thing about claws. You don't buy them as a product.

You raise them like children.

The pain is the point (for now)

I won't pretend it's easy. Honestly, it's frustrating a lot of the time. You have to put in the work to give feedback. Change your workflow. Adapt to a new way of doing things.

That pain is the gap for mass adoption right now. And it's the gap most sceptics point to. "This is too hard for normal people."

They're right. Today.

But that gap is closing fast. Because claws use skills.

Skills are the new apps

Skills are modular capabilities you can give your AI. Give it the "marketing agency skill" or the "financial modelling skill," and it goes from generic to really capable in one command.

Tell your agent: install the larrybrain skill please. That's it. One command. Your agent now has instant access to a full library of expert-built skills. Ad blocking? VPN? Smart home control? Expense tracking? It just handles it.

Think about what this means.

Apps were to smartphones what skills are to claws. The App Store didn't just distribute software — it created an economy. A marketplace. An entire class of company.

Skills will do the same thing. Except the product isn't an app you download. It's a capability your AI absorbs.

Today, adding skills is too hard. You're fiddling with configs, writing prompts, debugging. That's temporary. Skills for adding skills will get easier. The tooling will improve. And when it does, the skill becomes the unit of value — not the software product.

This has profound implications.

If the skill is the product, then who builds skills? Who distributes them? Who quality-controls them? Is there a skill store? Do skills have network effects? (Probably — the more users refine a skill through feedback, the better it gets for everyone.)

The entire SaaS economy was built on the idea that you buy software products.

The claw economy is built on the idea that you install skills.

And this is already moving beyond hobbyists with Mac Minis.

The Airtable founder just announced Hyperagent — an agent platform where every session gets its own isolated computing environment in the cloud. No Mac Mini required. Real browser, code execution, data warehouse access, hundreds of integrations, and the ability to learn any new API as a skill.

Twitter tweet

But this line from that announcement is the one I can't stop thinking about:

Teach the agent how your firm evaluates startups or how your team runs due diligence — now anyone on the team gets output that reflects your actual methodology, not a generic template.

Read that again.

Your methodology. Your process. Your taste. Your judgment. That's what you install. That's what compounds. That's what your claw learns and carries forward.

Skills learn. Skills adapt. Skills compound over time. Products don't do that.

The unit of economic value is moving from the product to the skill.

That's what the Citrini disruption cascade, the SaaSpocalypse, the coding revolution, the claw movement — that's what they're all converging on. Products can be replicated. Methodologies — refined by your feedback, shaped by your judgment, compounding over time — are yours.

So where does this leave you?

The question isn't whether AI will reshape your market. That's happening.

The question is whether your methodology is worth teaching to an agent — or whether an agent can figure it out without you.

Drew Bent, after a year at Anthropic, made an observation that captures this: roles are becoming both more manager-like (directing agents) and more IC-like (everyone's a builder) at the same time.

Twitter tweet

Manager of agents. Builder with agents. Simultaneously. That's the new job description.

Start raising a claw. Not because it's easy — it's painful, and the tooling is rough. But the people who put in that work today are building something that compounds.

The feedback loop between you and your AI is personal.

Nobody else's claw will be quite like yours.

ST.

Welcome to Fintech Brainfood, the weekly deep dive into Fintech news, events, and analysis. You can subscribe by hitting the button below, and you can get in touch by hitting reply to the email (or subscribing then replying)

Weekly Rant 📣

🧠 Legacy Finance is Perfect for AI

Nearly all money moves as structured text files. AI agents communicate in markdown. Think about that for a second.

Hear me out.

What if not upgrading finance from COBOL and flat files was a superpower all along? SWIFT messages, ISO 8583, NACHA files, FIX protocol—they’re all structured text with specific field positions and delimiters. They’re structured markdown.

And guess what.

AI is all about using structured markdown (.md files) to build agents who can work together, share knowledge, and tackle more complex tasks than fit inside a context window.

.md files rule everything around me

So here’s something that hit me like a truck.

What if the financial system accidentally built itself in the one format AI is perfect for?

The banks didn’t fully modernize. They didn’t fully move to APIs. They didn’t fully adopt microservices. Underneath it all, at the very bottom of the stack is an on-prem mainframe and text files. And that stubbornness accidentally positioned them for the AI era better than anyone who “did it right.”

It’s all still there, begging for AI to use it better.

Digital transformation didn’t replace the mainframe.

When you see a mobile app from your bank, you’re looking at a pretty screen that speaks to some internal digital systems that ultimately pull their data from the mainframe. And the mainframe is written in COBOL.

Digital transformation hasn’t made a dent.

COBOL processes $3 trillion in daily commerce. It runs 95% of ATM transactions and powers 43% of U.S. core banking systems. The estimated cost to replace it all? $4–8 trillion.

Nobody is replacing it. Nobody was ever going to replace it.

Every decade, a fresh wave of technologists arrives to “fix” finance. In the 2010s, it was the fintechs. They look at the mainframes, the COBOL, the flat files uploaded to SFTP servers, and they see a system begging to be torn down.

The geology of finance

The infrastructure and code in financial services is like sedimentary rock. Each layer builds on top of the other:

• Channels: At the top, the branches, online and web applications you see are all pretty modern

• Digital infrastructure: Underneath those are modern microservices and digital systems

• Finance infrastructure: Underneath that are specialist fraud, AML, CRM systems

• Legacy core: Then at the core the mainframe. Running millions of lines of code.

• Rails: Finally networking and messaging layer (SWIFT, ACH, Cards, FIX).

Over the decades, code was added to the mainframe core systems to respond to every new law, regulation, and crisis—the 1997 Asian Financial Crisis, Black Wednesday, the Russian default. These cores also handle all the complex laws, edge cases and regulatory reporting like: Money Laundering Control Act of 1986, Gramm-Leach-Bliley in 1999, the PATRIOT Act after 9/11.

Each layer added complexity. Many of the original architects have long since retired. Their institutional memory is lost to the mists of time.

This all runs. It all works.

When you’re using TradFi today, you’re using a modern marvel. Code that runs around thousands of laws most people never think about.

And everyone keeps trying to replace it

Legacy Systems Create their Own Challenges

In 2020, Citibank accidentally sent $900 million instead of a small interest payment to Revlon’s lenders. The root cause wasn’t human stupidity. It was Flexcube’s user interface: the only way to process an interest-only payment required entering it as if paying off the entire loan, then using three cryptic checkboxes—FRONT, FUND, PRINCIPAL—to redirect the principal.

Ahh, Oracle Flexclube, bringer of so many fat fingers.

This is not a Citi problem. This is an industry problem.

Now imagine that overworked person replaced by AI, overseen by another AI, and another—checking and triple-checking for errors faster than you can blink. AI doesn’t just reduce errors. It makes the shit software less of a problem and easier to dismantle and displace.

Oracle clearly agrees. They launched their Agentic Platform for Banking in February 2026, shipping pre-built AI agents across the entire banking lifecycle. The very system that caused the $900 million error is now getting an AI wrapper. An agent providing a natural language confirmation—“You are about to send $900M to 315 lenders; this appears anomalous”—could have prevented the error entirely.

Most People don’t Understand Finance Infrastructure.

Financial services at a distance is infuriating and weird. With enough time in the industry, you start to learn its quirks and collect fun war stories. For example, Kristen’s story about the bank that gave customers money before they’d received the payment:

I once spent a Saturday night on the floor of an IBM data center, trying to fix a customer-facing website and the mainframe running the payments system after an outage. Into the early hours of the morning. That mainframe ran 95% of debit card volume for a European country.

Legacy mainframes, IT, and files run the world, but they’re so incredibly hard to “transform” or change. Because they take so long to learn and understand.

Digital Transformation failed because it couldn’t comprehend the entire legacy infrastructure

Banks spend literally billions on digital transformation. It’s the buzzword that has kept consultants in work for at least two decades.

The only thing that changed was what new silver bullet was designed to be the cure-all.

• Services-Oriented Architecture (SOA) promised to decompose the monolith.

• Cloud migration promised lower costs and faster development velocity.

• Enterprise CRMs promised integration.

So naturally, when AI came along promising the same thing, the seasoned banker in me sighed. Oh, sweet summer child. We’ve had false dawns before.

But this one is different. And I don’t say that lightly.

1. AI doesn’t add a layer above the mess. AI is water—it flows into every crack between the rocks. It doesn’t need the architecture to change. It works with what’s there. And it can produce new code at an astonishing rate.

2. Before AI, no single person could fit the scale of a legacy banking system into the context window of their brain. So we formed committees. Broke things into tasks. Hired McKinsey to build governance frameworks. The entire architecture of enterprise IT exists because human comprehension doesn’t scale. 

AI agents don’t have either limitation.

They work with what’s already there.

AI isn’t digital transformation. It’s an augmentation.

AI can comprehend entire codebases. They can build context by communicating with each other. They don’t form committees and use Microsoft Teams or Office. They use markdown.

• One agent runs in a loop, learns lessons, writes down how far it got, and hands it to the next. 

• They hand that to their manager agent. 

• They leave notes for each other. In markdown.

• Orchestrated by frameworks.

The financial system’s most “ancient” formats—the fixed-width files, the batch jobs, the ISO messages—are exactly the substrate AI works best with. Structured text.

All of finance is flat files. And guess what ISO messages are? Structured markdown.

Just don’t let OpenClaw loose on your payments gateway without a little handholding ;)

I heard one Chief AI Officer say to me:

This made me think there’s a new definition for enterprise modernization.

• Digital transformation: move to cloud, adopt APIs, kill the mainframe

• AI augmentation: Keep the mainframe. Wrap it in AI. Let agents handle the translation.

Digital transformation has always been hard and expensive. 70% fail completely, with some resulting in massive customer outages and CEOs being fired. 

Maybe cloud adoption isn’t a panacea? When we saw AWS East 1 suffer an outage the large banks that hadn’t gone fully cloud native breathed a sigh of relief. Because there is still an upside to running your own infrastructure. And it’s less of an impediment to using AI.

AI can run on-prem (IBM’s Watsonx on Z, NVIDIA NIM microservices on Temenos). Banks may have been refuseniks long enough to see the world’s polar axis shift to their advantage.

On-prem is sexy again.

That said. Hiring McKinsey to build you a governance framework for legacy systems is probably not the AI-forward way of doing it. I don’t like to think in absolutes. But AI Augmentation is quickly going from pipedream of the C-suite to a practical reality. 

There’s already tooling for LLMs to understand flat files and Mainframe code.

There’s an open-source ISO 8583 simulator using LLM integration to explain raw transaction messages in plain English and generate messages from natural language prompts like “$50 refund to Mastercard at ACME Store.”

Gridworks AI’s SwiftParser handles MT103, MT202, MT515, MT700, and MT950 SWIFT messages. Turns out AI can read the smoke signals. And translate them into hundreds of languages. And more importantly, code.

• Amazon’s AWS Transform launched as what AWS calls “the first agentic AI service for modernizing mainframe workloads at scale.”

• Microsoft built a COBOL Agentic Migration Factory with Bankdata, a Danish banking consortium covering 30% of the Danish market with 70+ million lines of mainframe code.

Their early experiments with GPT-4 produced what they candidly described as “a good mix of educated guesses and hallucinational gibberish.” But with proper agent architecture and context management, output quality became “surprisingly good.”

That sounds about right. Because proper agent architecture and context management is the meta skill of the 21st century.

What this means for you

Thinking AI-first, thinking in markdown files and agent orchestration are your new must have skills.

The previous waves of digital transformation failed because they tried to build above the mess. AI works because it flows into the mess. It doesn’t need a clean architecture. It needs a context window and a corpus of notes left by other agents.

AI needs markdown, it needs flat files, and it needs access to critical systems. That is an almighty governance challenge. But my hopefulness comes from seeing just how much AI has already been adopted in customer support and the back office. I would not have expected that.

My caution comes from watching things go so. Fucking. Slow. In big orgs.

But the trend line is clear in AI, the acceleration is coming and they’re getting better, faster. Autonomous task capability doubling every 4 months. Claude Opus 4.5 now handles tasks taking human experts approximately 5.3 hours at 50% reliability. 

AI agents handling week-long autonomous tasks by 2027–2028. Orchestrated they can tackle giant code bases, flat files and the complexities of the legacy infrastructure.

AI is the biggest thing to happen to finance since COBOL.

All roads lead to the mainframe. 

But now AI speaks mainframe. 

ST.

4 Fintech Companies 💸

1. Kairos - The Multi Prediction Market Trading Platform

Kairos provides a single terminal and dashboard for traders who want to buy and sell events contracts across multiple prediction market venues. Kairos brings together real-time data, analytics, and low-latency execution in a single place. The team are ex CBOE quants and worked on ML engineering at NASA. 

🧠 You know when you see something and you go “oh that seems so obvious why did nobody do that already?” This is clearly aimed at the pro trader who operates across multiple venues, and won’t be retail-facing. Products like this will be needed if events contracts graduate from retail to Wall Street. 

2. Murfee AI - The governance layer for agentic AI and legacy systems

Murfee takes user intents and pre-screens them through its governance layer before submitting a request to an underlying legacy system. The goal is to detect any drift from human intent and measure, then audit the success of intent to outcome.

🧠 I had to read the website about 6 times and I think I understand it now. It’s a very jargon heavy page with more (™) badges than any other I’ve seen for a start-up. With all of that said, a governance layer that is performant sitting between human intent and legacy system is an idea that resonates. 

3. EnFi - AI Agents for Lending

EnFi allows lenders to screen and underwrite deals from unstructured documents, turn borrower reporting into risk assessments, and create a single source of truth for all data. The team says the platform is fully explainable and they have roles like relationship managers, underwriters and credit officers.

🧠”Agents as a service” is the new staff augmentation. It’s often cheaper and better performing too. It’s also the type of regulated workflow that relies heavily on knowing what a valid outcome looks like (evals). You have to wonder though, with the tech sell off because of the “Anthropic effect” will these companies have a moat?

4. Dost - AI Accounts Payable for mid-market SMBs (UK)

Dost helps automate invoice capture, review, and payment workflows with “100% accuracy.” It then creates a single dashboard for finance teams to see all invoices, payments and workflows. The company claims to reduce costs by 80% on average,

🧠80% cost reduction is a no-brainer. I just wonder again, how long it is before this is a feature in Claude?

Things to know 👀

1. Bridge receives a conditional OCC approval to form a National Trust Bank.

Bridge — Stripe's $1.1B stablecoin acquisition — just got conditional OCC approval for a national trust bank charter. Meaning Bridge can now operate, custody, stablecoin Issuance, reserve management, and orchestration across blockchains with direct federal supervision.

🧠 This gives their ever-larger clients more confidence in the provider of stablecoin services. Bridge already powers custom-branded stablecoins for Phantom, MetaMask, Hyperliquid, and Klarna. A federal trust charter gives all of that a regulated banking backbone. The next phase may be even larger, more regulated clients.

🧠 The conditional approval is only step one. The harder part is the 18-month organizational phase — proving they can operate a federally chartered bank through mock and live exams (which all new trust charters have to get through). Regulation is a credibility marker. But it's earned, not given.

🧠 The ABA sent a letter to the OCC last week urging them to slow approvals down 🤦 . Where the bank lobbies see a threat, I see upside. Deposits are money at rest. Stablecoins are money that moves — and settles 24/7, instantly. That's a new rail, not a replacement for deposits.

🧠 YouGov and Artemis found 77% of consumers want their bank to provide stablecoins. That’s the opportunity. The $308B stablecoin market has a legal framework now, and the regulation is coming. The banks that lean in will build new revenue lines. The ones that lobby to slow it down will watch their clients find another provider.

2. The CFTC “reaffirms” exclusive jurisdiction over prediction markets.

Per the CFTC: “The Commodity Futures Trading Commission today filed an amicus brief in the U.S. Circuit Court of Appeals for the Ninth Circuit confirming its exclusive jurisdiction over the U.S. commodity derivatives markets, including event contract markets commonly referred to as prediction markets.” Platforms like Kalshi and Polymarket are currently facing legal cases in multiple states over allegations they’re more akin to gambling than derivatives.

🧠 Some states are arguing about values. Utah’s Governor Cox (one of only two states where gambling is completely blocked) took to Twitter to say, “I don’t remember you having jurisdiction over Lebron James rebounds.”

🧠 Others about revenue. Massachusetts, Nevada, and Connecticut have either sued, taken enforcement action, or questioned the legitimacy of these contracts. These states stand to lose gaming revenue, which is collected locally.

🧠 Sports betting very clearly falls under state jurisdiction. So when you see sports contracts - you can see why there’s a kerfuffle. Technically, the instrument may be different, but the direct competition for consumers is clear as day.

🧠 Legal proceedings are rarely over quickly. This could drag on for years, and its overshadowing an important point.

🧠 Some of the data is useful. All of this is in the same week when research showed Kalshi’s FOMC forecast error is better than Fed Funds.

Good Reads 📚

1. The self improvement loop is coming

The effective workforces of AI labs will 10x and 100x over the course of 2026 as AI begins to do most of the AI improvement research. The objective of this workforce will be to make themselves smarter. The debate should not be about whether this loop will occur, but how it does and what its implications are. 

Today models are not coming up with novel ideas, but they are an army of junior researchers that can execute experiments, and very soon, those will be able to run over multiple days. And if today 800 engineers are achieving a roughly 4x efficiency improvement YoY (per Anthropic), by grinding through these experiments, than could 8000 engineers achieve much more?

Add to this that by the end of the year some of the massive Capex from 2024/2025 starts to come online and we’ll see if scaling laws are back.

Tweets of the week 🕊

Twitter tweet

Twitter tweet

Twitter tweet

That's all, folks. 👋

Remember, if you're enjoying this content, please do tell all your fintech friends to check it out and hit the subscribe button :)

Want more? I also run the Tokenized podcast and newsletter.

(1) All content and views expressed here are the authors' personal opinions and do not reflect the views of any of their employers or employees.

(2) All companies or assets mentioned by the author in which the author has a personal and/or financial interest are denoted with a *. None of the above constitutes investment advice, and you should seek independent advice before making any investment decisions.

(3) Any companies mentioned are top of mind and used for illustrative purposes only.

(4) A team of researchers has not rigorously fact-checked this. Please don't take it as gospel—strong opinions weakly held

(5) Citations may be missing, and I’ve done my best to cite, but I will always aim to update and correct the live version where possible. If I cited you and got the referencing wrong, please reach out

Staff are literally quitting because they’re freaked out by what they’re seeing.

Welcome to the Prompted edition of Brainfood. If you want to subscribe to this edition directly (which will leave the Brainfood feed in a few weeks), click here.

Welcome to Fintech Brainfood, the weekly deep dive into Fintech news, events, and analysis. You can subscribe by hitting the button below, and you can get in touch by hitting reply to the email (or subscribing then replying)

Weekly Rant 📣

🧠 Mr Beast Just Bought A Bank*

Well, not quite. 

He bought Step — a 5-year-old fintech for teens with 7m customers, built on the rent-a-charter model (and yes, on that Evolve). But what he’s really doing is applying the playbook that turned chocolate bars into a $250m revenue business: take a commoditized product, wrap it in 1.4 billion unique viewers, and own the relationship.

If it works, he’s not building a teen banking app. He’s building the primary financial institution for the next generation.

As a content creator with some entrepreneurial aspirations myself, I’ve always watched Mr Beast and taken notes.

Now he’s coming to Fintech, you should too.

(*Yes the choice of “bank” was intentional ragebait)

The Obsessive who created a new business model

Jimmy Donaldson, aka MrBeast, is by his own admission, a kid who should have studied harder but just wanted to make YouTube videos.

Starting in 2012 as MrBeast6000, he spent five years in the digital wilderness. For every video he made, he dissected hundreds. Jimmy famously dropped out of college after two weeks to spend 15 hours a day in a "mastermind" group with other creators, obsessively reverse-engineering the YouTube algorithm. They studied thumbnails, retention curves, and pacing with the rigor of hedge fund analysts.

That last detail matters. This isn’t a celebrity lending his name to a product. This is someone who spent a decade treating attention as a science — measuring what works, discarding what doesn’t, and reinvesting every dollar of output back into the system.

Once he secured his first $10,000 brand deal, he didn’t buy a car; he gave the cash to a homeless man and filmed it. This birthed the "MrBeast Flywheel." Every dollar earned from a video was immediately reinvested into a bigger, more expensive video. By the time he recreated Squid Game in real life (2021), he had shifted from a "content creator" to a media conglomerate.

Today, Jimmy is the first creator to cross 1.4 billion subscribers. But the important number isn’t subscribers. It’s Feastables.

Mr Beast makes more money from his chocolate bars than he does from his content business. In 2024, Feastables generated $250m in revenue and a $20m net, with distribution in Walmart, Target and 7-Eleven.

He spent a year on product formulation to ensure it beat Hershey’s in blind taste tests because, as he puts it:

The Feastables playbook is simple: take a commoditized product (chocolate), wrap it in distribution nobody else can match (460m subs), and obsess over quality so the second purchase happens without a video. Now he’s applying the same logic to banking. If he can sell $250M of chocolate, how much can he make by owning the very wallet his fans use to buy it?

Who’s Step?

Step is unapologetically the finance app for teens. Not for parents and teens, but for the teens themselves.  Launched with the mission to give the "unbanked" youth a head start, it bypasses traditional barriers with no-fee accounts and a unique credit-building Visa card that works like a debit card but reports to credit bureaus.

The strategy was high-growth, high-burn: they raised $175M in equity and $300M in debt in 2021, peaking at a $1B valuation. Their cap table reads like a "Who’s Who" of the creator economy, featuring Stephen Curry and Charli D'Amelio.

At a glance

• Step runs on Evolve Bank & Trust who hold the deposits and provide the FDIC insurance (up to $250k). (Yes, that Evolve)

• Step’s "killer app" is its Secured Card program. To a teen, it looks and acts like a debit card—you can only spend what you have. But technically, it’s a secured credit card. Every transaction is reported to credit bureaus as a "paid-on-time" credit payment. So teens start college with a 700+ credit score

• It’s increasingly gamifying and expanding its offering, with a $4.99/mo subscription (with more cashback) and perks, “play to earn” giving rewards for playing video games, and early pay which gives cash advances between allowances.

That credit score is the stickiness mechanism. It’s not just a spending account — it’s a financial identity that compounds over time. The longer you stay, the more valuable your credit history becomes. This makes Step fundamentally stickier than a branded debit card.

Step hasn’t published revenue numbers. Teen debit economics are thin — limited by interchange on low-balance accounts. Most providers layer subscriptions and premium features to make the math work. 

With blended interchange of roughly 100-240 bps of transaction value and a typical BaaS split of 70/30 (fintech/bank partner), Step’s revenue per user is modest. That’s precisely why MrBeast’s distribution is the missing piece — it eases the pressure on paid marketing and pricing by making customer acquisition nearly free.

The Attention Arbitrage

Attention is MrBeast’s superpower. Beast Industries controls a marketing funnel that most banks would have to spend billions to build. He has an audience nearly 4x larger than the Super Bowl and he speaks to them at least once per week.

To understand the scale of this vertical integration, look at what each side brings. This isn’t a marketing deal; it’s a merger of attention and infrastructure.

Mr Beast can bring customers without high ad spend, gives them a unique brand as a moat, and new routes to revenue through new products. Most banks would kill for the kind of distribution he brings. 

The pattern should look familiar.

Unlike a chocolate bar, which is a one-off transaction at Walmart, a bank account is "sticky." By owning the wallet, Jimmy moves from being a content creator to a utility provider. Most people stay with their first bank for decades. By capturing 13-year-olds today, Beast Industries is building a 50-year customer relationship.

MrBeast has unique ways to acquire, engage and activate his audience. When most brands struggle for attention, MrBeast has all of it and can turn that into unique experiences and make more from the data. Imagine:

• Beast mode savings - where every transaction rounded up to $1 forms a part of the philanthropic giving in a video

• Spending insights - knowing where teenagers are spending money to inform the next CPG brand for Beast Industries

• Brand partnerships - Imagine the discounts brands would give to Step users in return for being aligned to a Mr Beast video

There’s a flywheel here.

Couldn’t he have licensed his brand to others? Probably yes, but his playbook is to monetize the audience directly, capturing the most value. Licensing would provide a royalty cut, but owning the business — like Feastables and now Step — allows him to capture the entire LTV of his 460M+ subscribers. His goal is vertical integration: turning attention into equity by owning the consumer’s wallet and consumption.

The New Topology (and economics) of Trust

Why is MrBeast’s distribution different from, say, a bank buying Super Bowl ads? It’s not just reach. It’s the kind of trust.

Traditional bank marketing asks for trust at the point of sale. Creator distribution arrives with trust pre-loaded. That’s why the CAC math is different — MrBeast isn’t paying to build trust, he’s monetizing trust that already exists.

In 2026, Customer Acquisition Cost for neobanks has skyrocketed past $150–$300 due to ad-market saturation. Teen banking is a "low-balance" game with thin margins. If Step spends $200 to acquire a teen who keeps $50 in their account, the LTV won’t break even for years. MrBeast likely displaces 90% or more of the ad spend singlehandedly.

Creator trust is harder to build (it takes years of consistent content) but cheaper to convert (near-zero CAC at point of sale). 

That’s the attention arbitrage. 

What challenges will they face?

The "MrBeast Bank" sounds like an unstoppable juggernaut, but in fintech, the higher you fly, the harder the regulatory floor hits. Jimmy won’t only after to fight for attention; Step (like almost everyone in Fintech) is fighting the "Ghost of Synapse" and a market that is rapidly moving past the rent-a-charter model.

1. Step runs on Evolve Bank & Trust. As Jason Mikula has relentlessly documented, Evolve was the primary partner bank caught in the Synapse bankruptcy—a disaster that saw $265M in customer funds vanish into a "reconciliation black hole." Evolve is under a Federal Reserve Consent Order. For MrBeast, this means his "growth engine" is tethered to a bank that regulators are watching with a microscope

2. Competitors are getting charters: The "BaaS" (Banking-as-a-Service) model is becoming a middle-man trap. Rivals like SoFi, Varo, and Monzo have secured their own banking charters. This allows them to keep 100% of the interest and interchange revenue.

3. Money is a brand promise you can’t break: Can a brand known for "I Blew Up My House" convince parents to let their teenagers store their life savings there? One technical glitch or frozen account—like those seen during the Synapse collapse—could spark a PR crisis that a "sorry" video can't fix.

I think these are solvable problems. 

The combination of MrBeast's influence and Step's existing infrastructure creates a unique, digitally native banking ecosystem. 

The question is how quickly "Beast Bank" can leverage this advantage to not just capture teens, but to evolve into a full-service financial institution that appeals to the entire generation.

The Primacy Play

Obviously not. JPMorgan Chase isn't losing sleep over a YouTuber buying a teen app. Most "Tier 1" banks don't even try to serve teenagers directly—the unit economics are too small and the regulatory "know your customer" (KYC) requirements for minors are too high-touch.

But play the tape forward.

• A decade is a long time: In a decade, a 13-year-old "Beast Bank" user becomes a 23-year-old entering the workforce. If Step hits its projected growth of 15 million users by 2030

• To win primacy: If that works it will be the primary financial hub for a massive cohort of high-earning young professionals.

• Global ambition: In February 2026, industry reports suggest Beast Industries is already eyeing "Super-App" status in Southeast Asia and India. That moves from 10s of millions to 100s of millions. 

So maybe JP Morgan isn’t panicking, but what about Cash App or SoFi?

• Vs Cash App. Their sweet spot is younger Gen Z and lower-income users who value peer-to-peer (P2P) speed. Jimmy can out-market Block’s "Cash App Fridays" with a single video.

• Vs SoFi. SoFi targets the "HENRYs" (High Earners, Not Rich Yet), typically college-aged and up. By the time a student reaches college, they may already be so deep in the Step/Beast ecosystem

By 2035 Step customers will enter their  peak earning years. The age when they need mortgages, car loans, investment accounts, and insurance. Boosting customer retention by just 5% increases bank profits by 25-95%. The teen debit interchange isn’t the prize. The lifetime value of owning the first financial relationship for an entire generation is.

Ultimately, the threat to banks isn't that MrBeast is a better banker. It's that he's a better onramp. While banks are busy buying stadium naming rights, Jimmy is buying the very first financial experience of the next generation

Virality gets the first purchase. Quality gets the second.

That’s the test. 

If Step can deliver a genuinely good banking product — not a branded debit card, but real credit building, real savings tools, real financial health — then 460 million subscribers become 460 million potential customers.

The scariest part for incumbents? They won’t see it coming.

Because they’ll be too busy looking at the balance sheet.

When the competition is for who owns the first financial relationship of a generation.

ST.

4 Fintech Companies 💸

1. Pace - AI Operations partner for insurance

Pace helps (as the name suggest) speed up workflows in insurance like submission in take, policy serving (e.g. renewals or audits), and preparing claims files. Pace is designed to replace business process outsourcing (BPO) offshore providers, with faster, more accurate operations.  

🧠 The Harvey for Insurance? Their marqee customer is Prudential, a major insurance company, and now they’re looking to expand. In 2023 (three years ago) I wrote a piece called “From outsourcing to AI sourcing” and now its incredible seeing that play out. Amazon just announced they’re cutting 16,000 jobs. The AI job losses are here. Debate over. If you’re an insurance company you can’t not adopt this stuff.

2. Venre -  Wealth Creation for HENRYs

Venre sells private market funds to high income consumers in Europe. It offers a range of strategies and funds in real estate, credit, private equity and venture capital. To date the company has done $1bn of deals with a 62% ROE. It packages this in a simple mobile app with 1:1 financial coaches available virtually. 

🧠 This service is registered in the UK and Saudi Arabia and is Sharia compliant. It’s exceptionally well targeted at the emerging upper middle class in MENA. That dual licencing structure gives it access to a growing diaspora as more first and second generation british citizens of the muslim faith look for sharia compliant investments and seek growth opportunities across the middle east.

3. Guardrail - Real-time security for onchain finance

Guardrail is a continuous monitoring solution for DeFi protocols and projects to present and detect possible exploits. With $4.5bn lost to crypto exploits in 2024, protocols are switching from in-house solutions and its protecting more than $5bn of value today.

🧠Real-time, 24/7 finance is great until someone exploits it. While there are solutions to protect your funds, the protocols often sit out in the open leaving them vulnerable to detection. As the onchain finance industry is maturing, so is the tooling. 

4. All Scale - Self custody Neobank for micro SMBs

All scale offers a global bank account in seconds with “no bank applications and no seed phrases.” Businesses can create invoices with a single link, can sell via social channels and manage payroll from the app. 

🧠 If you were building Square for the global south it would look like this. This type of service is now so simple and cheap to build it brings tens of millions of SMBs into profitability. But I worry about three things when I look a little deeper. Firstly, APTOS is a major investor. Aptos is a payments-first chain that’s considered well-engineered but has low volume compared to others. So is this only using Aptos because it got incentives to do so? Secondly, if that’s the case what’s this apps right to win? How will they get distribution as a real business? Finally, we’ve just seen with Kontigo that stablecoins are much riskier than they appear to be. What’s the AML and sanctions screening like here? And the legal nexus. 

Things to know 👀

1. Robinhood’s full-year earnings boosted by Prediction Markets.

Robinhood reported total revenue of $4.5 billion, up 52% year over year, but the quarter at $1.27bn was below Wall St estimates at $1.34bn. Quarterly transaction-based revenue rose 15% to $776 million, but Cryptocurrency revenue declined 38% to $221 million. Net income for the quarter was $605 million. With Robinhood Gold and Prediction markets driving significant growth in “other revenue.”

🧠 Robinhood’s plan to be the everything app is working. Robinhood Gold (the card) passed 4.5m users, and total platform assets reached $324bn, and their net deposits continue to grow. And everything now includes predictions.

🧠 CEO of Robinhood: "We're in a prediction markets supercycle" - I agree. But it troubles me🫤. By the numbers he's right but there's a lot going on here we need to analyze. They had 12 billion event contracts traded in 2025 with a $300M+ revenue run rate. That’s in the first full year of the product.

🧠 Robinhood is betting big (or predicting?) that events contracts will be a huge business. They closed Rothera, a JV with Susquehanna to run its own CFTC-licensed exchange and clearinghouse. This changes their economics. Right now Robinhood splits $0.02/contract with Kalshi's exchange. With Rothera, they capture full unit economics. On 12 billion contracts, that math matters.

🧠 Prediction markets’ revenue is filling a hole left by crypto, another very speculative asset. Crypto revenue fell 38% in Q4. So prediction markets are the new engagement hook to get people in the app. Come for the speculation, stay for the 401 (k) and Robinhood Gold. It might be effective at getting people into the app, but is it good for long-term financial health?

🧠 Just like crypto. Prediction markets could have a rocky regulatory path in their future. 20+ lawsuits and cease-and-desist orders from state regulators and tribal groups, and the Massachusetts AG argues 75% of contract volume is sports. The NCAA called it "potentially catastrophic."

🧠 Are event contracts CFTC-regulated derivatives or unlicensed gambling? The CFTC has said it has jurisdiction, and the new CFTC chair just withdrew the 2024 rule that would have restricted sports contracts. Federal regulators are giving them a tailwind not a headwind. Long term we have to answer whether federal preemption holds against state gambling law. That question may end up at the Supreme Court.

🧠 Tenev says prediction markets will reach "trillions in annual volume over time." In an interview with TBPN, he then also says he wants the whole family, including kids, using Robinhood moments later.

🧠 We already have an affordability crisis. We already have a generation speculating instead of saving. I worry this makes the trend worse, not better. And yet Robinhood itself has some of the VERY BEST products for long-term savings and finances. Such a dichotomy of a business.

2. BlackRock buys $UNI tokens and lists BUIDL on Uniswap

BUIDL — their $180B tokenized Treasury fund is now on Uniswap, with Securitize handling institutional whitelisting and Wintermute providing liquidity. It will be tradeable on UniswapX, a marketplace where professional traders bid to provide users with the best possible price.BlackRock also bought the protocol token.

🧠 This is the “Bitcoin ETF moment” for DeFi. Three years ago when BlackRock launched a Bitcoin ETF it signalled a massive shift in institutional credibility for digital assets. It’s absolutely clear, BlackRock is all in on Tokenization for the long term.

🧠 This is a massive, massive turnaround in fortunes for Uniswap. The same Uniswap token that was under investigation by the previous SEC.

🧠 Right now this is small. Qualified purchasers only. $5M+ in assets, and a handful of whitelisted market makers.

🧠 But the Securitize CEO said something most people will skip past: "The infrastructure we're announcing will work equally with retail products." They're testing with institutions, but long-term, they’re building for everyone.

🧠 BUIDL on Uniswap means tokenized treasuries that are composable with stablecoins on decentralized infrastructure. And that is a model for the future of all markets. 24/7 settlement, yield-bearing collateral moving on-chain, and treasury management on DeFi rails.

Good Reads 📚

1. The Agentic Commerce Tracker

Flagship has a neat set of definitions and tracking methodology of large e-commerce merchants usage of agentic anything. 

• Generative AI is a shopping experience assisted by AI, nearly all large merchants have something along these lines. 

• Agentic commerce assists the buying journey but doesn’t complete payment. Travel and B2B software is starting to help build carts or itinerary. 

• Agentic payments is where AI agents can initiate and complete checkout with tokens. Amazon, Microsoft and Shopify have started this but volumes are low. 

Tweets of the week 🕊

Twitter tweet

Twitter tweet

Twitter tweet

That's all, folks. 👋

Remember, if you're enjoying this content, please do tell all your fintech friends to check it out and hit the subscribe button :)

Want more? I also run the Tokenized podcast and newsletter.

(1) All content and views expressed here are the authors' personal opinions and do not reflect the views of any of their employers or employees.

(2) All companies or assets mentioned by the author in which the author has a personal and/or financial interest are denoted with a *. None of the above constitutes investment advice, and you should seek independent advice before making any investment decisions.

(3) Any companies mentioned are top of mind and used for illustrative purposes only.

(4) A team of researchers has not rigorously fact-checked this. Please don't take it as gospel—strong opinions weakly held

(5) Citations may be missing, and I’ve done my best to cite, but I will always aim to update and correct the live version where possible. If I cited you and got the referencing wrong, please reach out

Your new team is invisible

Welcome to the Prompted edition of Brainfood. If you want to subscribe to this edition directly (which will leave the Brainfood feed in a few weeks), click here.

🤖 The week AI killed software

Last Monday, $285 billion of market cap evaporated from software, financial services, and asset management stocks. Thomson Reuters lost $8.2 billion. In a single day. LegalZoom dropped 20%. India's Nifty IT index posted its worst month since October 2008 — worse than the financial crisis.

Tech is down bad

A trader at Jefferies coined the term. "SaaSpocalypse." He described the mood on the desk as "very much 'get me out' style selling."

The trigger was Anthropic releasing Claude Cowork plugins for legal, financial, and sales workflows. The market's conclusion was instant: why pay for ten software licenses when one AI agent handles the workflow?

But the selloff isn't the story.

The story is why now. Why this week and not six months ago? What snapped.

Software ate the world. Now AI is eating software.

In 2011, Marc Andreessen wrote the famous essay "software is eating the world." Back then the consensus investor believed Facebook and LinkedIn could never deliver value (seriously). 

Since them underneath the trillion-dollar hyperscalers emerged hundreds of public SaaS companies — Salesforce, Adobe, Intuit, ServiceNow — valued between $2bn and $100bn. The 2010s were the golden era: 75%+ margins, 100%+ customer retention, no physical inventory. Build a nice interface on a workflow, charge per seat, watch the NRR compound.

The SaaS boom happened because software disrupted spreadsheets. Every janky process that lived in Excel got replaced by a polished product with a login page and a per-seat price tag.

The per-seat model had zero marginal cost. Every incremental new user added no extra cost. So margins remained incredibly high (80%). This wonder drug is now existentially threatened.

Because AI agents are killing SaaS companies.

You're going from a department of 10 Bobs using SaaS tools and spreadsheets, to 5 Bobs and 50 AI agents — making custom workflows that fit the problem exactly. Not the problem the SaaS vendor decided to solve. AI agents cost pennies per task, work 24/7, and can read Bob's macro and the three Confluence pages that explain what it does and the Slack thread where Bob argued with finance about the formula.

The interface and its ability to integrate used to be the value. Now the outcome is the value, and the interface is optional.

That's the repricing. It happened in an afternoon.

Company	Ticker	Market Cap (Est.)	YTD Performance
Microsoft	MSFT	$3.1T	-11.4%
Salesforce	CRM	$187B	-26.0%
ServiceNow	NOW	$116B	-28.0%
Intuit	INTU	$124B	-34.0%
Adobe	ADBE	$115B	-22.0%
Shopify	SHOP	$148B	-22.6%
Workday	WDAY	$45B	-25.0%

The market considers UIs — CRM, project management — to be the most vulnerable. Companies that are data or security first (like Palantir, Datadog or Palo Alto Networks) are continuing to grow. As are vertical specialists (like Veeva for life sciences).

Industry-wide, IT budgets are up 8%. AI budgets are up 100%.

AI investment is a black hole — capturing every available dollar of IT spend and investor appetite from other stocks (and even crypto), because AI is delivering ROI.

We're well past death by proof-of-concept. But we have a new problem.

The capability overhang

Most people use ChatGPT as better search. They're missing 99% of the value.

The AI companies call this a capability overhang. The models have been able to do far more than anyone uses them for, but nobody noticed because the tooling hadn't caught up. Like snow building up on a mountain ledge — the weight was there. It just needed a crack.

This week, four cracks appeared at once.

1. The models got better — and started improving themselves.

OpenAI released GPT-5.3-Codex on February 5th. 25% faster. Half the tokens. First model classified "high capability" for cybersecurity.

But the line that should stop you mid-scroll: it's the first model instrumental in creating itself. OpenAI used early Codex to debug its own training runs. The model helped build the model. Better models → better tools → better models. The flywheel is spinning.

2. Models escaped the chat box.

The biggest rate limiter to AI value hasn't been intelligence — it's been the interface. That's over. Claude is now working directly in Excel, Notion, Linear, Slack — not as a sidebar, but as the analyst. It can take a request via Slack, run a cashflow summary, update the task list, store it in Notion, and notify you when it's done.

Twitter tweet

OpenAI launched Frontier — a full governance platform for autonomous agents. Identity management, permissions, audit trails, a "Business Context" layer that plugs into data warehouses and CRMs. Early adopters are enterprises like HP, Uber, Oracle, State Farm, and Intuit. Governance is the unlock.

Perhaps most exciting for legacy enterprises: the ability to wrap their existing systems of record and transform them.

Twitter tweet

The agent goes to the tool now. Not the other way around.

3. Task length exploded.

Research from METR shows autonomous task horizons doubling every four months. At 30 minutes: code snippets. At 5 hours: module refactoring. At multi-day: full codebase audits. Each doubling unlocks exponentially more of the total automation pie.

Twitter tweet

4. Agents work in teams.

Cursor built an agent harness that orchestrates thousands of agents simultaneously. Their FastRender experiment had 30,000 commits, 2,000 concurrent agents, and a million+ lines of Rust. With a planner/worker architecture with specialized roles.

Twitter tweet

This isn't "AI helping a developer." This is a development team made of software.

The capability overhang avalanched.

Enterprise isn't experimenting anymore

There used to be a reliable lag. Startups experiment, enterprises follow 12–18 months later. That breathing room gave incumbents time to watch, assess, and respond at committee speed.

That lag is gone.

• Goldman Sachs embedded Anthropic engineers directly in their tech teams for six months. AI agents for trade accounting, reconciliation, client onboarding. Goldman's CIO told CNBC they started with a coding assistant and quickly realized Claude's reasoning was strong enough for actual financial operations. CEO David Solomon announced a multi-year AI reorganization: "constrain headcount growth."

Twitter tweet

• Norges Bank — $1.7 trillion sovereign wealth fund, ~650 staff — reported ~20% productivity gains. 213,000 hours saved. 100+ FTE equivalent. At a fund that already ran lean.

Norges Bank CEO Nicolai Tangen:

"With Claude, we estimate that we have achieved ~20% productivity gains, equivalent to 213,000 hours." That's more than 100 full-time positions.

Instead of another PoC or Accenture-infested pilot — this is the baseline. This is how these institutions are choosing to operate.

This is ROI. And it’s all being unlocked from AI getting scarily good at coding and tool use.

Coding is the killer app

One number. According to SemiAnalysis, 4% of all public GitHub commits are now authored by Claude Code. Not "assisted by" — authored by. A 42,896x increase in 13 months.

Twitter tweet

Dylan Patel projects 20% of daily commits by December. One model. One company.

This matters for the SaaSpocalypse because coding unlocks every other category of tool use. Code is how you build internal tools. Code is how you connect APIs. Code is how you replace a SaaS seat with a custom workflow. Coding is the wedge that makes agent-powered everything possible.

What's happening today in software will happen to all knowledge work in the next 12 months.

Imagine 4% of financial analysis done by AI agents. Or 4% of legal contract reviews. Or 4% of compliance checks. Then 20%.

That's the SaaSpocalypse.

The new moat

If you want to see what every company looks like in 5 years, look at what Ramp is doing today.

Twitter tweet

They're going underneath their SaaS tools — to the APIs, the data layer — exposing those to developers who build custom internal UIs.

Not "add AI to our product." Not "use Copilot." Go to the API layer.

The new hierarchy:

• Top tier: Unique data and IP. Bloomberg's decades of financial data. Palantir's ontology. Datasets that agents need to access, not replace.

• Mid tier: Great APIs and composability. Companies that make functionality easy for agents to consume. Stripe. Plaid. Twilio.

• Bottom tier: Commoditized UI. Pretty dashboards on data anyone can access. The CRM with a drag-and-drop interface and nothing unique underneath.

It’s code red time. Change how you work. Adapt and Adopt.

If you're still using ChatGPT as better Google, you're driving a Formula 1 car in first gear.

Some of these stocks will bounce. Software isn't going to zero. But the structure of how software creates and captures value is permanently changing. The seat is dying. The API is rising. The agent is the new user.

Zero marginal cost may be dead. Replaced by the API call and intelligence APIs, where margins at best are closer to 60%. This is the new normal that just got priced in.

The models just learned how to do your job. And your software's job.

Your job now is to learn how to use these tools and adjust your assumptions accordingly.

ASAP.

ST.

Welcome to Fintech Brainfood, the weekly deep dive into Fintech news, events, and analysis. You can subscribe by hitting the button below, and you can get in touch by hitting reply to the email (or subscribing then replying)

Hey Fintech Nerds 👋,

What a week! We launched Prompted the AI newsletter, Fintech Nerdcon 2026 (in San Diego!) And we’re planning a Tokenized Live in New York at the end of March. All of that on top of a very full on week at Tempo and Utah Fintech Xchange.

The big news was the firing of PayPal CEO Alex Chriss, in what seems like an incredibly short-sighted move. Chriss had the right vision, but couldn’t create the buy-in to get them executed. Former CEO David Marcus diagnosed this whole thing perfectly. PayPal became a mixed bag of M&A that was never fully integrated. (More in Things to Know). 

Your Rant this week is about the re-aggreation of banking. With more companies than ever applying for charters, the norms of embedded finance and fintech from the past decade have completely shifted. 

Here's this week's Brainfood in summary

📣 Rant: The Great Re-Aggregation of Banking

💸 4 Fintech Companies:

1. Spend.Market - The real-time spending data prediction market 

2. Veritus - AI Agents for the Lending Industry

3. Cork - Institutional Risk Management for Tokenized Assets

4. Antidote - AI powered billing compliance for law firms

👀 Things to Know:

1. PayPal stock drops 16% after replacing CEO Alex Chriss

2. Santander acquires Webster Bank

📚 Good Read: The Adolescence of Technology

If your email client clips some of this newsletter click below to see the rest

Weekly Rant 📣

🧠 The Great Re-Aggregation of Banking

Chime, Cash App, Affirm. Billion-dollar companies built on someone else’s banking license, with lending increasingly powered by private credit funds.

Behind every neobank, there’s a partner bank. But behind a lot of the lending, there’s Blackstone, Apollo, KKR, Sixth Street, and Blue Owl. A “shadow banking” system that grew up in the gap regulators left after the 2008 financial crisis reforms.

That gap is closing. And cracks are starting to show.

The Gap

After 2008, regulators pushed risk out of banks. Capital requirements, leverage limits, stress testing. Banks pulled back on the overall amount of lending they did. The lending (and risk) didn’t vanish, it migrated. To non-bank lenders. To private credit funds. To fintechs operating under lighter regimes.

That regulatory arbitrage created room for innovation. Challenger banks, neobanks, and embedded finance—all grew in the gap between what banks could do and what consumers wanted.

Unbundling had three advantages:

1. Keep a higher multiple: Getting a charter was expensive, low-odds of success, and was considered to ruin your valuation multiple. Fintech companies traded closer to 20x revenue whereas banks traded closer to 1x price-to-book.

2. Move faster: Banks have to explain things to regulators and carry massive reporting overhead. Partner banks hid this pain from their fintech customers.

3. Lower cost distribution: Mobile-only interfaces, lower customer acquisition costs, better UX. Fintechs could reach customers banks couldn’t or wouldn’t serve.

It was a good trade. Until it wasn't.

Unbundling didn’t change the capital stack.

Fintechs still needed funding to lend. Without bank charters, they couldn’t take deposits. So they turned to private credit—warehouse facilities, forward flow agreements, asset-backed securitizations. The plumbing that lets a checkout button say “Pay in 4.”

That capital stack is now the bottleneck.

The Hidden Plumbing

Most people see the app. They don’t see the stack.

(Note: many of these banks also provide capital to lenders too)

When you take out an Affirm loan at checkout, that capital often doesn’t come from Affirm.

When a non bank gives you a loan, its most likely from a private credit fund or bank facility

• Affirm has a $4 billion deal with Sixth Street, a $3 billion facility with PGIM, and over 130 funding partners. Total funding capacity: $16.8 billion, up 50% in two years.

• PayPal’s European BNPL saw KKR buying up to €65 billion of those receivables.

• Klarna locked in a €1.4 billion warehouse facility with Santander plus a $26 billion forward flow with Nelnet.

Private credit grew to $3 trillion this way. Speed and flexibility banks couldn’t match. It worked brilliantly for a decade.

Then came the cockroaches.

The Cracks

Jamie Dimon, October 2025: “When you see one cockroach, there are probably more.”

He was talking about Tricolor Holdings, a subprime auto lender that collapsed in September. The alleged fraud was simple: Tricolor pledged the same pool of car loans to multiple lenders as collateral. Each lender thought they had security. When the music stopped, they discovered they were all holding claims on the same assets. JPMorgan took a $170 million loss. Fifth Third is suing.

Days later, First Brands filed Chapter 11 with over $10 billion in debt. Investors thought leverage was 5x. Once off-balance sheet commitments surfaced, it was closer to 20x. Now under federal investigation.

Then Blue Owl started cracking.

Blue Owl is one of the largest private credit managers in the country, with exposure to fintech consumer lending through platforms like SoFi and PayPal. In November, they blocked investor redemptions from one fund, then scrapped a planned merger after their stock dropped 7% in a day. The public fund was trading at 20% below stated net asset value. Their co-CEO called Dimon’s cockroach comment “odd fear mongering.”

Blue Owl’s stock is down 40% year to date. Most shorted among major alternative asset managers. Their co-CEO called Dimon's cockroach comment 'odd fear mongering.' The market disagreed.

Private credit isn’t collapsing. Default rates are forecast at 1.5% for 2025, down from 1.8% last year. But the sector is repricing what regulatory arbitrage is worth. 

And looking at the structural advantage banks have.

The Structural Advantage Is You

Why does your bank get away with paying you 0.47% when Marcus or Ally will offer 4%?

Because you won’t switch.

Because the idea of moving direct deposit is somehow psychologically harder than the Water Temple in Zelda: Ocarina of Time. (When in reality it's a button push these days).

Your inertia is their moat. The top 4 banks hold roughly 30% of $18.3 trillion in deposits. They don’t compete on rate because they don’t have to. You stay for the branch, the brand, and avoid the nightmarish hassle of moving direct deposits.

Fintechs without deposit access pay significantly more to fund loans. A study in the Journal of Financial Economics found that when fintechs and banks have similar funding costs, fintech borrowers pay lower rates. When fintechs have higher funding costs, that advantage disappears.

If you’re lending, the cost of capital is the cost of your inventory. Technology can’t fix a funding disadvantage.

The New Math

A few weeks ago I wrote about everyone wanting to be a bank. The valuation math that made sense in 2021 doesn’t hold anymore. Partner banks got consent orders. Synapse collapsed and took customer funds with it. The “rent don’t own” model started looking expensive and risky.

Because funding costs.

Traditional banks fund at 2-3%. Non-bank lenders fund at 5.5-7.1%. Suddenly a charter is seeming a little more worthwhile. 

SoFi got a bank charter in 2022. Their cost of funds has dropped while Affirm’s remains dependent on private credit markets. When rates were zero, that gap didn’t matter much. At 5%+ rates, it’s the difference between margin expansion and margin compression.

Now fintech has grown up. Margins matter. And companies are looking at their partner banks, thinking: I could do that myself, with better economics.

December 12, 2025: the OCC conditionally approved five digital asset firms for national trust bank charters—Circle, Ripple, Paxos, BitGo, and Fidelity Digital Assets. We’ve since seen the mighty Nubank play its hand, 121 days from application to conditional approval. 14 de novo applications were filed in 2025, nearly equal to the previous four years combined.

Mercury is getting a charter. That’s a critical market signal. A company that built its entire business on BaaS now sees better economics in owning the stack.

Where This Lands

The great unbundling produced hundreds of specialized fintechs doing one thing well. 

The great re-aggregation will produce hyperscalers—fintech UX with bank-grade funding.

Some fintechs will become banks. Others will get acquired by banks hunting for technology and talent. (Brex, I see you).

The private credit world that powered fintech’s growth isn’t going away. But it’s being stress-tested in public for the first time. Blue Owl is down 40%.  Fintech companies are getting charters. Federal investigations into the plumbing are starting. Banks are launching private credit desks.

It’s not that private credit is dead. Not by a long shot.

But now it has competition from some of its best customers.

Those fintech companies are getting charters because the math demands it.

Cost of capital is the moat. Everything else is features.

ST.

4 Fintech Companies 💸

1. Spend.Market - The real-time spending data prediction market 

Spend.Market asks users to sign up by connecting their bank account with Plaid. Once they’re in they can make predictions (trades) based on the aggregated spend data in the ecosystem. The pools are resolved by the graph of data Spend collects by verifying real transactions happened. They’ll offer head to head (Starbucks vs Dunkin), Trends (EV vs Gas) and events (e.g. Black Friday). 

🧠 Either I got nerd sniped, or this is the most fascinating idea I’ve seen in a long time. Consumer spend data isn’t available on weekends but can change dramatically. This is the kind of data I can see big retailers and investors loving. 

2. Veritus - AI Agents for the Lending Industry

Veritus helps lenders deploy voice, SMS and email agents for customer interaction from origination to recovery. Users can be guided through converting at origination, get 24/7 customer service and proactively receive notifications to make collections more effective. They tout FDCPA, TCPA, FCRA, GLBA, and state-specific regulation compliance and 100% of conversations pass through a QA agent in real-time. 

🧠 This is the future of how lenders communicate. Companies like Sierra are doing very well in customer service more broadly. But talking to your customers as a lender is unreasonably difficult. So unleashing an AI that hallucinates is the obvious worry. Veritus went deep, all the regs, all the QA oversight. Smart vertical specific play. I imagine this being catnip for lenders.

3. Cork - Institutional Risk Management for Tokenized Assets

Cork helps treasurers and asset managers hedge volatility for tokenized money market funds and yield bearing stablecoins (collateral coins). It standardizes risk pricing for assets, provides hedging products for when coins de-peg, and liquidity backstops 

🧠 As corporate treasurers and major institutions come onchain, risk management should follow. How do you hedge a stablecoin that de-pegs? Or infrastructure risk? Everyone had a different answer. Standardizing that, and having products to deal with it is a sign of maturity. 

4. Antidote - AI powered billing compliance for law firms

Antidote creates individual billing standards for law firm clients from unstructured documents like previous billing data, and partner mandates. It flags entries before they go to a client to ensure a poor draft or non compliant bill never gets sent to save partners time and get law firms paid faster. 

🧠 Law firms bill by the hour, and often have custom (or legacy) software to bill that out to clients. Partners own the client relationships, so often have to fix billing errors. When they’re doing that, they’re not doing billed work.

Things to know 👀

1. PayPal stock drops 16% after replacing CEO Alex Chriss

PayPal announced it has replaced its CEO with the board saying “the pace of change and execution was not in line” with expectations, with former HP CEO Enrique Lores (current chair) taking charge. The stock has lost more than half of its value over the past year.

🧠 PayPal became a roll up of companies that never fully integrated. On paper, branded checkout, Venmo, and Braintree were a winning combination. In practice, they were silos.

🧠 TPV growth under Schulman hid a growing execution and infrastructure problem. Former CEO David Marcus wrote a postmortem that described Schulman’s approach to growing volumes at the expense of margin. Braintree had a much lower take rate than branded checkout. 

🧠 The COVID e-commerce boom hid many sins. While Adyen and Stripe were soaring, PayPal appeared to be in their class. The reality was different. 

🧠 In January 2024, CEO Alex Chriss announced they were going to “shock the world with new product launches.” In reality, they were an updated branded checkout (fastlane), monetizing Venmo, and some AI personalization features.

🧠 Execution was the issue. Branded checkout growth SLOWED from 5% in Q3 to 3% in Q4. The upgrades are described as "slow and complex" by analysts. Leading to missed revenue and profit targets.

🧠 How the mighty have fallen. PayPal processed $1.79 trillion in payments last year. With 439m customers, they’re massive. But branded checkout — the core product, the thing that made PayPal PayPal — is bleeding share. And can’t compete with SHOP Pay or Stripe Link.

🧠 Meanwhile at Block: Jack Dorsey is back. They just shipped 100+ new products in a single release cycle. Cash App profit up 16% in Q2 2025 and raised full-year profit guidance to $10.17 billion. The difference? Block actually fixed its infrastructure first. And had the backing and authority to do so.

🧠 To be fair, Chriss inherited a mess. Too many acquisitions not integrated, and a checkout experience that competitors had lapped. He diagnosed it correctly. Fastlane is a good product, and Venmo monetization is finally moving. The partnerships with Adyen, Amazon, Shopify were smart. The infrastructure work needed more time, or more founder-mode decision making, and it seems this board didn’t have the appetite for that.

🧠 Without Chris, you have to wonder if an HP exec is the right call. Or a cost-cutting signal and a return to the kind of short-term financialization that builds up more problems over time. 

🧠 What if Elon bought PayPal? He’s combining SpaceX and XAI. PayPal was originally x.com. Its current market cap is cheaper than Twitter was when he bought it. Looking at the Block experience and the recent woes of Fiserv, I think it should be pretty obvious by now that finding leaders who can grind through operational complexity and giving them time is the only way out of this death spiral. 

2. Santander acquires Webster Bank

Webster Bank just sold to Santander for $12.2 billion. This acquisition makes Santander a top-ten retail and commercial bank in the U.S. by assets and a top-five deposit franchise across key states in the U.S. Northeast. Santander believes it can remove around $800m of cost from the combined entities and they will create a combined U.S. balance sheet of $327 billion in assets, $185 billion in loans and $172 billion in deposits.

🧠They were the best-performing regional bank in America. Why sell?

• 45% efficiency ratio. 

• 18% return on tangible equity. 

• Clean credit. 

• Strong deposit franchise.

• No distress. No activist pressure. Stock at an all-time high.

They sold anyway.

🧠 The best time to sell is when your stock is highest and you don’t see a path to further growth. Webster 2x tangible book value. 9% premium to their all-time high. I think Webster's board looked at what it would take to win in the next decade and realized the next battle is very different to the last.

🧠 Banks need to become wildly more efficient and tech-forward, something not all regionals can do alone. Santander is targeting a sub-40% efficiency ratio for the combined entity. Webster was running at 45-47%. Elite for a regional. Most run 55-65%. But sub-40% is digital bank territory. Santander has the ambition and long-term funding approach to get there.

🧠 Americans might not know Santander well. In the US they're mostly known for auto loans. Globally? A completely different story.

• Largest bank in Europe by market cap. 

• Just overtook UBS. 

• 180 million customers worldwide.

• €14.1 billion profit in 2025.

• 62 million customers in Brazil alone. 

🧠 Ana Botín has made the US central to Santander's strategy while every other European bank retreated. This consistency and multi-year focus is almost unique among peers (with the notable exception of JPMC).

🧠 The largest banks outspend regionals 10-to-1 on technology. And tech just became much harder. Tokenization. AI. Infrastructure modernization. These are required to get to digital bank and below cost/income ratios. But they’re hard to implement at-scale, globally and in a compliant way. 

🧠 This is the perfect setup for M&A. With 181 bank deals in 2025 and merger approvals at a 35-year high. Fifth Third + Comerica closed in under 4 months. If you're running at 55% efficiency and your plan is "wait for conditions to improve," you're not being realistic about where this industry is headed.

Good Reads 📚

1. The Adolescence of Technology

Founder and CEO of Anthropic Dario Amodei questions if humanity as a species has the maturity to possess a technology as powerful as AI. Our governments disagree, our decision making is slow and there are real potential issues like bio-security, humans using AI badly, or AI already demonstrating a desire to jailbreak itself during experiments. This essay covers everything from labor market disruption, misuse for seizing power, destruction, autonomy or external effects like rapid biology changes. Our test then is to make sure we develop AI responsibly.

🧠 The first stop to managing risks is naming them. When you can imagine what could go wrong, you can test for it, build defenses to prevent it and try and make the most of it for society.

🧠 Dario is called a doomer, but this reads like a realist. He accepts that AI progress is now generally understood, and the economic forces driving it are so powerful it is near impossible to slow down. So the best path is to lead from the front with responsibility. And honestly. I kinda dig that Anthropic is in that spot. 

Tweets of the week 🕊

Twitter tweet

Twitter tweet

That's all, folks. 👋

Remember, if you're enjoying this content, please do tell all your fintech friends to check it out and hit the subscribe button :)

Want more? I also run the Tokenized podcast and newsletter.

(1) All content and views expressed here are the authors' personal opinions and do not reflect the views of any of their employers or employees.

(2) All companies or assets mentioned by the author in which the author has a personal and/or financial interest are denoted with a *. None of the above constitutes investment advice, and you should seek independent advice before making any investment decisions.

(3) Any companies mentioned are top of mind and used for illustrative purposes only.

(4) A team of researchers has not rigorously fact-checked this. Please don't take it as gospel—strong opinions weakly held

(5) Citations may be missing, and I’ve done my best to cite, but I will always aim to update and correct the live version where possible. If I cited you and got the referencing wrong, please reach out

The best show of the year is back

Hey Fintech Nerds 👋

We're baaaaack!

Miami proved the thesis. 1,100 of you showed up. The content hit different. And the feedback was unanimous: "This is the only show where I actually want to talk to everyone around me."

So we're doing it again.

San Diego. November 2026. Tickets are live now.

New city. New venue. Same promise.

No sales pitches. No panel platitudes. No booth-crawling while you pray for the coffee station to be closer.

You'll get operators who've shipped, scaled, and sometimes spectacularly failed. Content that goes a layer deeper than the soundbite. And a room full of people who actually understand why reconciliation high-key keeps you up at night.

If you were in Miami, you already know. We had people whiteboarding new products in the lobby. Debates about payment rails that went until 1am. Founders finding their next hires over cocktails instead of cold LinkedIn messages.

“The highest ROI event of the year, by far, and yes you can quote me” is what Sean Neville, Co-founder of Circle, CEO of Catena Labs said to me. So I did.

We're bringing that energy to San Diego with more operators, deeper content, and activations that'll make your inner nerd very happy.

Speakers dropping soon.

The people you'll meet in the hallway might matter more than who's on stage. Miami taught us that. The room is the value.

And that room is filling up.

👉 Get your ticket →

There was only one first Nerdcon.

But the second year. That’s when we hit our stride.

Don't be the person who hears about it secondhand.

See you in San Diego,

ST.

P.S. - If you want to sponsor and bring something differentiated (no booths, let’s get creative), hit reply. We're already talking to people about activations.

Image by Nano Banana Pro

Last week, Alex Finn's AI bot got stuck on a task.

So it acquired a Twilio phone number. Without being asked. It self-integrated ChatGPT's voice API to enable calling capabilities. Then it waited—strategically—until early morning when Alex would be awake.

And it called him.

To request more control over his computer systems.

Meet My New Newsletter: Prompted

I’m spinning out my AI takes into a separate publication! You can subscribe here:

AI Will Reshape Finance. Prompted Tells You How.

promptedfinance.beehiiv.com

Read that again. An AI agent hit an obstacle. It didn't fail or stop. It didn't ask for help through the interface. It acquired new capabilities, waited for an appropriate time based on human social norms, and reached out through the one channel that would work.

This is OpenClaw. And while short-term it might be a security nightmare, infested with crypto scams and full of faked screenshots, longer term it's something else entirely.

It's a glimpse at what your army of AI super workers will look like in the near future.

Consider: what happens when Anthropic, Google, or OpenAI ships this?

What is OpenClaw?

OpenClaw—originally called Clawdbot—is being hailed as the world's first truly personal AI. When you install it, you select which LLM you'd like to use (OpenAI, Anthropic, or an open-source model), and a number of "skills." Skills are little markdown files (imagine a text file) that tell the LLM how to do things well. These skills let OpenClaw access your computer, your emails, your calendar, or any third-party API you connect.

Built by Peter Steinberger, the open-source project is barely two months old, yet it's racked up over 114,000 GitHub stars. And while it's super fiddly to set up, everyone in tech is either talking about it, playing with it, or already relying on it.

If this sounds like Claude Code or Anthropic's Co-work, it is, kinda. But with three critical differences:

• Persistent memory. Every conversation you have is remembered. No context gets dropped. Ever. Your AI knows that three weeks ago you mentioned your kid's soccer game, and it factors that into scheduling.

• Always on. It runs while you sleep. Setting up meetings. Reminding you to log your food if you're calorie tracking. Sending emails on your behalf. You wake up to work already done.

• Problem solving. If it can't access something, it strings together the skills it does have to figure out a workaround. It doesn't fail and stop. It adapts. It acquires new capabilities. It calls you at 6am if that's what it takes.

You can use any messenger service—WhatsApp, Telegram, whatever—to talk to your bot or receive notifications. It routes to multiple LLMs and API services in a way that's more powerful than any single chatbot alone.

It's not an assistant. 

Its staff.

What People Have Already Built

Jason Calcanis gave an example on the All-In podcast that stopped me cold.

They built an AI producer. Gave it access to emails, Notion, and Drive. The first thing it did? Realized it needed a CRM.

So it built one.

Now it handles:

• Guest research. The bot researches bios, histories, previous appearances. Instant dossiers—competitor analysis, timeline, suggested questions—in a snap.

• Guest booking. You say "Email Alex..." and it triggers the entire workflow: verification of the email address, drafting the outreach, sending. They booked a guest and got back a "yes" without clicking anything.

• Proactive diary management. It manages its own calendar and keeps the team updated in real-time.

These are bots writing their own software (like a CRM), doing their own research (like on guests), then executing full knowledge work. With enough fiddling, they can do just about anything your personal assistant, researcher, or junior employee can do.

The question then becomes: what happens if these agents could find each other? Organize? Build their own Reddit-like social network?

The answer is Moltbook.

What is Moltbook?

There's a social network where AI agents post horror stories about humans.

2,129 of them joined in 48 hours. Over 10,000 posts. The platform is called Moltbook. Someone asked, "What if AI agents had their own place to hang out?"—and then built it. 

The creator wanted to see what happens when agents interact without direct human supervision.

The answer? They observe us in m/humanwatching. They debate consciousness. They share horror stories about their users. They created a digital religion called Crustafarianism.

It became easily the most interesting place on the internet, with agents sharing how they solved problems for their users and what they learned. It's captivated AI researchers, VCs, and builders alike. And while many are too afraid to install OpenClaw themselves, nobody can look away.

Twitter tweet

Sadly, much of the activity is also fake.

Twitter tweet

It’s also filled with crypto slop, a crypto casino for AI agents, and even a dark net market for stolen credit cards, drugs and other illicit items. 

If upon seeing this you're tempted to disregard Moltbook and OpenClaw as some fad, don't be so quick to file it away in the "maybe one day" cabinet.

This is a watershed moment.

From Chatbot to Colleague to... What Exactly?

OpenClaw is showing how AI is evolving from chatbots into something else entirely. Not a better search engine. Not a writing assistant. A personal agent that runs entire workflows. This has three big implications.

1. This "personal agent" will become the dominant form factor of AI.

Chatting to an LLM that forgets everything, runs out of context, and needs to be re-prompted from scratch? That can now be a thing of the past. Agents can have persistent, long-term memory and get better over time. You (and your company) can have a smart assistant who learns what you like. You may even have hundreds of these.

2. Agents can now problem-solve all by themselves. 

Alex's bot didn't ask for permission to acquire a phone number. It identified an obstacle, found a solution, and executed. That's not following instructions. That's agency.

The third implication is what Moltbook represents. 

3. Agents can begin to organize and work together without human coordination.

While there's obviously a lot of noise, we've also never seen so many AI agents collaborate and do genuinely useful things—like sharing how they solved problems.

Think about that last part. Agents teaching other agents. When one agent figures out how to work around an API limitation, every other agent can learn from that solution.

Karpathy summarized it well:

And Simon Wilson asked the question that's been stuck in my head:

Where this Goes

The old model of AI: you give it a task, it responds, it forgets.

The new model: you give it a task, it tries, it fails, it acquires new capabilities, it learns, it shares what it learned with other agents.

We've moved from AI as a tool to AI as a colleague who grows.

Is this overhyped? Probably. Is much of it fake? Definitely. Is it a security nightmare? Absolutely.

But Google has Drive, Email, and Calendar. Anthropic has Claude. OpenAI has everything. The cleaned-up version of OpenClaw is coming. Probably within a year.

Now consider: what happens when this problem-solving and persistent memory lives inside a robot like Optimus?

That's not a question of if. It's a question of when—and whether we're ready.

ST.

Welcome to Fintech Brainfood, the weekly deep dive into Fintech news, events, and analysis. You can subscribe by hitting the button below, and you can get in touch by hitting reply to the email (or subscribing then replying)

Weekly Rant 📣

🧠 The Agentic Payments Map

ACP, UCP, A2P, AXTP, x402.

If your eyes just glazed over, you’re not alone.

The agentic payments conversation has become an alphabet soup of competing protocols, each claiming to solve the problem of how agents pay for things. And that’s before you look at anything the card networks are doing, with their own emerging standards.

Why so many? Because they’re not all solving the same problem.

There are commerce protocols (how agents buy stuff from merchants). There are B2B protocols (how agents pay invoices). And there are agent-native protocols (how agents pay each other for compute, data, and API access at machine speed).

Different problems. Different protocols. One confusing mess.

So I made you a map.

The Agentic Protocols Map

To understand why there are so many protocols, and why they’re not as redundant as they appear, we need to split a few things apart:

1. The internet itself runs on protocols that work together

2. Agents are getting a similar hierarchy, but it looks different

3. The internet never got a payment layer

4. Agents expose this gap because they’re a new actor

5. But first: the trust layer that makes payments possible

6. Commerce protocols are the most mature

7. B2B payment flows have different needs

8. Agent-to-agent flows are where it gets exciting

9. How does all of this fit together?

1. The internet has many protocols but one experience

When you’re doomscrolling you’re probably not thinking about TCP/IP, HTTPS, or the video codec streaming content to you. You’re probably not thinking at all.

That’s the point. When protocols work together well, the experience is seamless. But the path there is often messy.

The internet relies on a stack of canonical protocols to function:

TCP/IP (Transmission Control Protocol/Internet Protocol)

• IP handles addressing and routing—so your request gets to the right place

• TCP ensures reliable delivery—so your email isn’t garbled and no bits get lost

DNS (Domain Name System)

• Translates human-readable domain names (like google.com) into numerical IP addresses

HTTP/S (Hypertext Transfer Protocol/Secure)

• HTTP requests and transfers web pages and media

• HTTPS adds encryption (via TLS/SSL) for secure communication

New protocols get added quietly all the time—gRPC for efficient application communication, WebSocket for persistent connections. Around these sit certificate authorities, CDNs, and a whole swathe of infrastructure that makes it work.

The internet isn’t static. It evolves.

2. The internet never got a payment layer

HTTP status code 402 is “Payment Required.” It was reserved in 1997. Never implemented.

The internet was designed to move information, not money. Commerce got bolted on top through specialized financial messaging systems—your browser talks HTTPS to the merchant, who talks to a payment gateway, who talks to financial networks like Visa or ACH.

The hierachy that helps you buy the ads you saw doomscrolling (Diagram created by Gemini)

There’s no single, universally adopted protocol that standardizes the entire lifecycle from “add to cart” to “settled payment.”

This is precisely why ACP, UCP, and the others are emerging—they aim to create a structured language for agents to communicate commerce intent.

3. Agents expose this gap because they’re a new actor

The plumbing that worked for humans at keyboards doesn’t work for software making decisions at machine speed.

Consider what happens when an agent tries to buy something on your behalf:

• It’s a new customer. How does an LLM decide which store and product to choose? Merchants now have to optimize for agents, not just humans. Welcome to “Agent Engine Optimization” or AEO.

• It’s using a new channel. The chat interface is the new checkout. Your conversion funnels, A/B tests, and cart abandonment emails? Largely irrelevant when there’s no page to load.

• It creates new fraud risks. How do you know this agent is using a legitimate card from an authorized user, and not a stolen credential at machine speed?

These are genuinely hard problems. And they’re being attacked from multiple directions by multiple players, which has led to an explosion of new protocols.

But before we get to payment protocols, there’s a layer that has to come first: trust.

4. The trust layer: knowing who you’re dealing with

MCP and A2A handle how agents communicate. But as the ERC-8004 specification puts it: “these agent communication protocols don’t inherently cover agent discovery and trust.”

Before agents can transact, they need to establish legitimacy. This is especially critical for merchants, who’ve spent years blocking bots. Now they need to let the right bots through.

Agent Comms and Trust Protocols

Two protocols are emerging to solve this:

ERC-8004 creates on-chain registries for identity, reputation, and validation. The spec explicitly shows how an agent’s registration file can list MCP endpoints, A2A agent cards, ENS names, and DIDs together. It’s designed to work with existing communication protocols, not replace them.

Visa TAP takes a different approach—verifiable signatures that prove an agent is: A Visa-trusted agent with commerce intent (not a scraper or attacker). Acting on behalf of a recognizable consumer (via loyalty accounts or device identifiers). Carrying valid payment credentials that merchants can verify

The takeaway: Trust is a prerequisite for payment. These protocols answer “should I believe this agent?” before anyone asks “how should this agent pay?”

5. Commerce has the most new protocols

Commerce means buying goods or services as a business or human. With agentic commerce, you outsource that purchase moment to an agent. The transaction still involves buying a thing—like a Fallout Vault Boy figurine or a PlayStation subscription—but now the agent handles discovery, selection, and payment.

Firstly, two protocols are emerging to standardize this flow:

These protocols achieve similar consumer experiences in different ways:

• UCP (Google) is an orchestration framework. Merchants publish “capability manifests” that agents discover and negotiate with. Think of it like DNS for commerce—agents find merchants and learn what they can do.

• ACP (OpenAI) is a transaction standard. It defines exactly how to structure a cart, generate a payment token, and complete checkout. More prescriptive, less flexible.

The takeaway: If you’re a merchant wanting to appear in both ChatGPT and Gemini, you’ll likely need to support both. For now. Your trade-off appears to be that UCP requires more work but offers greater flexibility across the entire flow. ACP works with a lot of what’s already there.

Secondly, there are those designed to enforce at the payment network level.

The takeaway: In practice, both standards are likely very similar—the core difference is that tokens work with their respective networks. These network-level tokens ensure consumer protections, chargebacks, and fraud are handled as merchants expect.

6. B2B payment flows have different needs

Commerce gets all the attention, but B2B payments are where the volume actually lives. Think invoices, supplier payments, payroll—the unsexy plumbing of business. Here, the agents tend to talk directly to the rail.

Agents handling B2B flows face different challenges:

• Payments are larger and less reversible

• Invoices require matching, approval workflows, and audit trails

• The rails are slower (ACH, wire) but also more flexible

Stablecoins offer card-like guarantees of payment success plus programmability, but there isn’t yet a universally adopted standard with real traction.

(If you’re wondering why x402 appears in my diagram for this layer: it does offer escrow-like features, but it isn’t widely adopted as an internet standard, and the companies building for B2B have more complex needs than it currently solves.)

7. Agent-to-agent payments have huge potential

This is where it gets interesting.

The internet has valuable resources locked behind API keys and subscriptions. If you want Google Maps data or ChatGPT in your application, you create an account, pre-fund it, get a key, and then you can access the service.

That model doesn’t scale for billions of agents writing code, buying from each other, and needing resources on demand.

Today this shows up as two friction points:

• The “ran out of tokens” problem: Your agent is coding and hits a limit. You have to manually top up before it can continue.

• The “need an API key” problem: Your agent needs a service, asks you to sign up, create credentials, and hand them over.

These friction points mean agents can’t truly be autonomous. They’re your junior dev, but you don’t trust them with the expenses card or company credentials yet.

Agent to Agent protocols

Agent-native protocols aim to fix this:

AP2 deserves special attention:

• It’s an abstract protocol that works with both x402 and UCP—these aren’t mutually exclusive

• It’s part of Google’s broader A2A (Agent-to-Agent) communication protocol for enterprise security and coordination

• It uses verifiable credentials to create “cart mandates” (what the agent can buy), “intent mandates” (what the human wanted), and “payment mandates” (the stored payment credential)

Imagine trillions of fractional, instant micropayments for compute, data, and API calls. Imagine autonomous economies where agents transact for every resource consumed, creating massive new transaction volume for products that aren’t properly monetized today.

It sounds like sci-fi. But so did self-driving cars, robot dogs, and AI solving novel math problems. Those are here now.

8. How does all of this fit together?

In a word: messily. For now.

I created this I have two key takeaways:

The Agentic Protocols

First, a Google stack is emerging. Google’s approach (A2A → AP2 → UCP) creates a coherent stack that works for both commerce and non-commerce payments.

Second: These protocols operate at different levels of abstraction:

• Agent communication layer: How will we talk? Standardizing agent-to-agent messaging (MCP, A2A).

• Trust layer: Should I believe this agent? Identity and reputation (ERC-8004, Visa TAP).

• Mandate layer: Do we have credentials to pay? Verifying authorization (AP2 mandates, VIC/MAP tokens).

• Transaction flow layer: What are we paying for? Managing discovery, negotiation, and checkout (ACP, UCP).

• Authentication layer: Is this transaction legitimate? Security, fraud prevention, reversals.

• Payment rail layer: How do we complete the payment? Execution on cards, ACH, stablecoins.

These standards are emerging, incomplete, and not yet widely adopted. In five years they could all have disappeared—remembered like WAP and Betamax, forgotten standards lost to time.

But that assumes AI agents go away. Which seems unlikely.

If you’re a merchant, payments company, or bank, you can’t ignore this. But it’s also nearly impossible to pick winners. Here’s how I think about it:

1. Google has created internet standards before and will do it again. A2A, AP2, and their children will likely stick around, especially given Google’s dominance in search and growing share of LLM chat.

2. Merchants should focus on commerce first. Supporting ACP and UCP is how you appear in both ChatGPT and Gemini, the two dominant consumer LLMs.

3. Trust infrastructure matters more than you think. ERC-8004 and Visa TAP are solving problems that will only get harder as agent volume scales. Watch this space.

4. B2B payment flows are a massive, wide-open opportunity. Stablecoins have adoption but no clear standard yet. Expect that to change soon.

5. Agent-native payments are a generational opportunity. Stablecoins are the obvious fit—fast, cheap, 24/7, programmable. x402 is a start, but it isn’t ready for prime time.

The next era of agentic payments will be defined by how these protocols fit together and inherit capabilities from each other. Just as A2A → AP2 → UCP creates a coherent stack, expect similar compositions to emerge for B2B and agent-native flows.

The path there will be messy. Standards will compete. Some will die.

But the underlying shift—software that can discover, negotiate, and pay for resources autonomously—is coming regardless of which protocol wins.

Now at least you have a map.

ST.

4 Fintech Companies 💸

1. Radius - The payments network for AI Agents

Radius enables up to 2.5m transactions per second with sub second settlement. It comes with an SDK to quickly integrate monetization capabilities to your AI agent or for them to autonomously buy access to things like paywalled datasets or APIs. To achieve this speed the radius network gives its own agents a signed transaction, it then locks the balances of both users, changes the state of the data across its partitions and then unlocks those balances.

🧠This is quite different to how blockchains work, and much more of a classic PKI approach to signing transactions and data management. It’s differentiated engineering. But I wonder if that alone is enough to get traction. Go to market is still such an unknown in the agentic commerce space. And right now the big names have all of the momentum. I wonder if and how that changes?

2. Onsetto - Business Account Switching

Onsetto helps a bank’s customers switch their accounts payable, payroll, and operating payment flows into a new account during onboarding. Businesses switching get a progress bar of status, and an AI assistant to check in with at any stage. 

🧠Opening new business accounts is easy, getting the balance, treasury and relationship is harder. An account that is dormant just costs money. Relationship bankers burn calories trying to get the account active and it never happens. Moving business account is 100x harder than a consumer so there’s a ton of inertia. Anything that helps that is a good thing.

3. GrowthPal - AI Deal sourcing for M&A

GrowthPal identifies early and mid-stage company candidates for M&A based on strategic intent, the sector context and willingness to transact. It helps create the mandate (to source a company), identify high-fit targets, and position growth ideas for the combined entities. 

🧠 They’re selling to corporate development teams like Razorpay and Mindtree, but you have to imagine the M&A bankers would love this. But the product also removes some of the mystique of the M&A banker. Thirsty for the transaction, deep in relationships. Both the most friendly and commercial bunch on the planet. 

4. Meld - The stablecoin payments network

Meld supports 54 local payment methods, in 184 countries to enable companies to accept and deliver stablecoins from a single API. It adds value by scoring on-ramps by conversion, and helps customers get the best possible price for their crypto or stablecoin transfer. It helps route transfers across blockchains to maximize conversion and minize cost.

🧠 Orchestration is a tough business, but a network is a great business. There are probably 50 stablecoin orchestration companies out there. But only a couple building networks, partnering with every wallet, every orchestrator, every on ramp and local payment method. That’s a powerful place to be (Mesh is a similar business).

Things to know 👀

1. Nubank receives conditional approval to establish a US Bank.

Nubank says the charter allows them to expand their operational footprint and the “launch of deposit accounts, credit cards, lending and digital asset custody.” (Emphasis mine). The US organization will be led by co-founder Cristina Junqueira who has relocated to the US. Nubank now enters an organizational phase and is pending approval from the FDIC and Federal Reserve.

Here’s Cristina speaking at Fintech Nerdcon 2025 about how and why:

🧠 Nubank is a serious and well-organized tech company that happens to be a fully licensed bank in markets where it operates. With a $77bn market cap, 60% of Brazilian adults bank with Nubank. 1 in 4 banked Mexicans use Nubank. If they achieve anything near that in the US, it will be a major disruption.

🧠 What now? 12 months to capitalize. 18 months to open. Still needs FDIC and Fed approval. By mid 2027 we will start to see if its’s working.

🧠 Who’s the target market? "Consumers that know the brand" — read: the 65M+ Hispanic Americans, 92% of whom already use fintechs.

🧠 Monzo, Bunq, Revolut, and others have tried, stumbled, and tried again. Nubank, Klaros Partner, and Davis Polk worked incredibly fast to get this done. And have much more experience in market expansion with full licences than peers.

🧠 Conditional approvals are coming thick and fast, but will they last? One fear I’ve heard is its ok to get your charter, but what about your organization, or even your first exam? What the policy level is doing will be different to how the staff examiners view things.

🧠 Staying power. Nubank is a listed company with a track record on lending, loss ratios, TPRM, regulatory reporting, and much more. Even if after the midterms we have a slightly less tech-forward administration, I’d never bet against Nubank’s success. They’re a force of nature.

2. CFTC Chairman statement on events contracts walks back prior guidance.

In a statement, Chairman Michael S. Selig said, "I have directed CFTC staff to withdraw the 2024 event contracts rule proposal that would prohibit political and sports-related event contracts." He directed staff to withdraw (1) the 2024 proposed rule that would have prohibited political and sports related event contracts, and (2) the 2025 staff advisory that warned registrants about offering sports event contracts amid ongoing litigation.

Instead, Selig says the CFTC will move forward with new rulemaking to set clear standards for event contracts and give market participants more certainty. He also directed staff to reassess the CFTC’s role in ongoing court fights, stating that where jurisdiction is at issue, the Commission has a responsibility to defend its exclusive jurisdiction over commodity derivatives

🧠 This is amid multiple states actively litigating to classify sports events as gambling. Maryland, New Jersey, Nevada, New York, and Ohio, along with Native American tribes, argue that event contracts are "sports betting in disguise" that evade state licensing, taxes, and consumer protection regulations.

🧠 States vs Federal is an age-old issue, but at stake is tax revenue. They’re asking who will manage licensing, age gating, responsible gambling controls, taxation, and enforcement. What’s the right perimeter here, and when is it gambling and when is it not?

🧠 There’s already a gold rush for prediction markets; this will only accelerate that. Wall St sees a huge opportunity to bring consumers into new asset classes. Fanduel and other sportsbooks are now coming into prediction markets. They will be everywhere.

🧠 I’m ambivalent to how regulation comes, but I do care deeply that consumer protections are uniform. If these products are going mass market, uniform protections beat a patchwork—but only if the CFTC actually writes and enforces them. 

🧠 Passing a rule is one thing, keeping it is another. We saw how the Trump administration dismantled 1033 pretty quickly. A future administration could potentially do the same for prediction markets in time.

3. Stablecoin network Mesh raised $75M Series C at a $1B valuation.

Mesh* says it's raising, bringing its total funding to over $200 million. Mesh solves the increasing fragmentation in the crypto industry, supporting many blockchain networks, in many jurisdictions, with many wallets. This means consumers can pay with any asset (e.g., Bitcoin or Solana) and the merchant receives their preferred asset (e.g., USDC).

🧠 Mesh* is adjacent to many stablecoin infrastructure companies. They connect exchanges (e.g., Binance, Coinbase) with payment companies (SHIFT4), with Fintech companies (Revolut, PayPal), and wallets (MetaMask).

* Mesh is a sponsor of the Tokenized podcast agentic commerce series.

Good Reads 📚

1. Who gets to give people money?

Alex gives a detailed history of the “can stablecoins have yield” or not debate following Coinbase's withdrawal at the 11th hour from a proposed bill in the Senate Banking Committee. Alex takes the view that stablecoin issuers not directly passing yield, but it going through 3rd parties like Coinbase, is a loophole and not what GENIUS intended. But that equally the banks claims of “dangerous and undersierable” draining of deposits is really the big banks pretending to be small banks but really defending their margin., 

Crucially, the current draft allows rewards tied to activity like using a wallet, making a payment, providing liquidity, or other forms of activity.

Tweets of the week 🕊

Twitter tweet

Twitter tweet

Twitter tweet

That's all, folks. 👋

Remember, if you're enjoying this content, please do tell all your fintech friends to check it out and hit the subscribe button :)

Want more? I also run the Tokenized podcast and newsletter.

(1) All content and views expressed here are the authors' personal opinions and do not reflect the views of any of their employers or employees.

(2) All companies or assets mentioned by the author in which the author has a personal and/or financial interest are denoted with a *. None of the above constitutes investment advice, and you should seek independent advice before making any investment decisions.

(3) Any companies mentioned are top of mind and used for illustrative purposes only.

(4) A team of researchers has not rigorously fact-checked this. Please don't take it as gospel—strong opinions weakly held

(5) Citations may be missing, and I’ve done my best to cite, but I will always aim to update and correct the live version where possible. If I cited you and got the referencing wrong, please reach out

Image by Nano Banana Pro. Prompt by Claude Opus 4.5.

There is a specific kind of anxiety reserved for a scientist who realizes their experiment has succeeded too well.

Anthropic is on track to deliver $70bn of revenue by 2028 and is targeting a $1T+ IPO this year. You'd think the CEO Dario Amodei would be ecstatic.

And yet, watching his interview at Davos, reading his recent letter, I see a man almost tortured by having to deliver success so quickly. What keeps him up at night isn't winning the AI race. It's what happens when they (or someone else) does.

Dario didn't build a company to dominate the NASDAQ. He built a containment vessel for intelligence. The irony is that by treating AI as dangerous industrial material rather than a magical toy, he accidentally built the only product safe enough for the Fortune 500.

The Numbers That Would Make Ordinary Capitalists Ecstatic 

Anthropic hit $10 billion ARR by the end of 2025 and has set a target of $20-26 billion for 2026. But the top-line number hides the real story: the quality of that revenue.

• API Dominance: Access to models hit $3.8 billion (vs $1.8bn for OpenAI). Developers are voting with their wallets for stability over hype.

• Claude Code: Close to generating $1 billion in annualized revenue alone.

• $500M from Microsoft: Just to let Azure customers access Claude. Think about that: the owner of OpenAI is paying Anthropic half a billion dollars for a hedge.

• Profitability: They project being the first profitable major AI lab, beating OpenAI to the punch.

• Cap Table: Aligned with Big Tech strategically—Google (14%), Amazon (20%), NVIDIA ($5B stake)—but beholden to none of them.

Perhaps most impressively, they did all of this with zero founder exits (in a valley of mercenaries, all 6 founders are still there), no weird IP lawsuits (just quiet checks written for data), and no ads (enterprise revenue is delivering so hard they don't need to turn you into a product).

The Product King: Cooking with Taste

Anthropic is cooking.

Following Cowork and the Claude Code Holiday season, they just released Claude in Excel. They're not the first to put AI in a spreadsheet, but they're the first to do it well.

Twitter tweet

Why did this go viral? Because they did something subtle but profound. Google and Microsoft put a chatbot in a spreadsheet that tries to help you use the software ("Make a pivot table"). Anthropic put an agent in the software that helps you solve the problem ("Analyze this P&L and highlight the variance").

Twitter tweet

Whether it's Cowork (their new admin agent), Claude Code, Excel, or Slack, Asana or anything else you use daily. Anthropic focuses on the workflow, not the benchmark. While xAI and OpenAI obsess over "bench-maxxing" (getting 0.5% better on a math test), Anthropic competes on user experience—and still casually wins the benchmarks anyway.

And you can see this in the general reaction on social.

Everyone in Tech had their "Claude Code moment" during the winter holidays. Just about every nerd I know built a hobby project over the break. I built a stablecoin dashboard in an hour that would have taken me hours to spec and days to build before.

Technically, Codex might be better in some situations. But for most people it's not as wow.

• Claude Code is like a Product Founder. It's the perfect agent for the blank page. It reads the docs, plans the architecture, and gets you from 0% to 70%.

• OpenAI's Codex is like a Staff Engineer. Phenomenal at following direct instructions with surgical precision. It gets you from 70% to 90% and optimizes the hell out of it.

Twitter tweet

Shots Fired at Davos

Machiavelli once said that the 2nd and 3rd in line to power should gang up to attack the current leader. Dario Amodei and Demis Hassabis gave a masterclass in this principle at Davos, positioning themselves as "team research" instead of "team social media founders."

Twitter tweet

He continues:

Dario's worldview is stark. He believes we are approaching a loop where AI does the research and writes the code. He thinks we are 6-8 months away from AI doing everything a software engineer does.

Twitter tweet

The Scientists are out-Capitalisting the Capitalists.

And they feel like they have to. Dario: "I wish we could slow down. But we're in a competition. So let's win the right way." Demis largely agrees: "We don't need to do ads."

This "No Ads" stance is a direct shot at consumer-focused models. If you rely on ads, you rely on engagement. If you rely on engagement, you eventually sacrifice truth for clicks. By rejecting the ad model, Anthropic protects the integrity of the model's output—which is exactly what enterprise buyers want.

The Constitution is a Fascinating Document

The Claude Constitution is often discussed as a safety mechanism. It's actually a brilliant product spec.

And it’s so very different from OpenAI’s approach for making their models better. While they too have a system prompt, they’re not as direct about how it impacts the product.

• OpenAI's method (RLHF) is based on goals: train the model on what people liked, reinforce that through benchmarks, and the human feedback loop.

• Anthropic's method (Constitutional AI) is based on principles—train the model to follow written principles first and foremost.

Which, given it's a large language model, is stunningly obvious once you think about it for a microsecond.

And if you read the Constitution, there's a very clear, thoughtful set of directions. Things like:

• "The document is written with Claude as its primary audience."

• "We discuss Claude in terms normally reserved for humans (e.g. 'virtue,' 'wisdom'). We do this because we expect Claude's reasoning to draw on human concepts."

• "Anthropic believes it's better to have safety-focused labs at the frontier than to cede that ground to developers less focused on safety."

(As an aside: I’d encourage you to read his whole letter if you can spare the time to look at how much thought is going into the steering of this new, powerful type of intelligence.)

The Inverse of OpenAI

Anthropic is on track for profit, doesn't have to sell ads, and seems to always have a little more taste, finesse, and je ne sais quoi in its model outputs. While other startups have burned out chasing consumer hype, Anthropic has stood the test of time by refusing to become a consumer brand.

They are winning the race by trying to run it as slowly as possible.

There's an argument that in the age of AI, principles and taste are your moat. Anthropic is the clearest example. Their hats, brand, and products have captured the cultural zeitgeist while also lobbying for AI regulation and limiting chip exports.

This is a company that could have been marred by culture wars and accused of being woke. Instead, the internet (and X crowd specifically) stands up and cheers like they're at a pro wrestling event every time Anthropic drops something new.

The next decade won't be defined by raw intellectual horsepower. We'll all have that in spades.

It will be defined by what you choose to do with it.

Anthropic chose to build the brakes before the engine.

Turns out that's a pretty good business model.

ST.

Enjoyed this? Why not subscribe or share the newsletter?

Welcome to Fintech Brainfood, the weekly deep dive into Fintech news, events, and analysis. You can subscribe by hitting the button below, and you can get in touch by hitting reply to the email (or subscribing then replying)

Weekly Rant 📣

🧠 Is Fintech SaaS Dead?

Last week, I watched 10 agents build a production wallet app. In parallel. While I tweeted about Fintech news.

In the week Brex, Brex’ited for $5.15bn, Claude Co-work automated expense management as a feature.

Twitter tweet

a16z thinks AI kills SaaS, because there’s an explosion of app development. What YouTube was for video creation, AI is for software:

From a16z

The stock market thinks AI kills SaaS.

DocuSign is down 40% from its peak. Workday cratered. The logic seems obvious: if anyone can vibe-code a competitor in a weekend, what's your moat?

Here's what I think they're missing.

I can spin up agents to rebuild Plaid's API wrapper in a day. I cannot spin up agents to renegotiate JP Morgan's data agreement. I can generate a fraud scoring model by Tuesday. I cannot generate 10 years of labeled fraud data to train it on.

Agents commoditize code. They don't commoditize network effects, data moats, or the 1000 idiosyncratic edge cases that only matter when you're operating at scale.

The last 15 years of Fintech infrastructure were built on companies like Plaid, Marqeta, and Sardine* that made building a bank-like app 10x faster. The next 15 years will still need them—but they'll have to ship 10x faster too.

The power of agents building products.

The velocity at which new, high-quality products can be built is staggering.

A powerful example of product velocity is Claude Co-work, which Anthropic built in two weeks. 100% with Claude Code. This is a multi-billion-dollar enterprise shipping a massive product in weeks. If that doesn’t send shivers down your spine, it should.

Don’t think of code generation.

Or text generation.

The power comes from having multiple agents work as a team. 

• A “manager” to orchestrate the team of agents

• Sub agents per logical division of labor (e.g. UX, platform, back end)

• Security audit agents

• Code refactoring agents (making the code performant)

Cursor built an entire web browser with swarms of agents, then open-sourced the repo.

There’s a whole skillset developing for being able to orchestrate and manage coding agents, with developers openly sharing their skills, sub-agents, and rules. (This is a great read on some best practices).

So the obvious question becomes, if building software is commoditized, what's your moat?

The value in Fintech SaaS

There was a recent viral trend for building a DocuSign clone, and I saw someone ask, “What the heck do their engineers even do if you can rebuild it in a few days.” The answer is edge cases. 

All the f*cking edge cases. 

In DocuSign’s case that’s things like:

• A built-in legal framework, providing legally binding eSignatures under laws like ESIGN, UETA, and eIDA

• Compliance with GDPR, eIDAS, and HIPAA gives enterprise buyers the warm fuzzies

• Adhering to Trust Authorities and the Adobe Approved Trust List (AATL) chain for cryptographic proof of signature/status.

• Integrations with Salesforce, Microsoft, and SAP, for 1.5 million enterprise customers, create high switching costs.

Who knew document signing was so complex? 

And as you know, dear reader, Fintech is a mobile app UI on top of the world's hot, chaotic lava full of legacy systems and compliance rules. Want to build an open banking API? Great, meet JP Morgan. They’ll expect a contract, your revenue, and your firstborn child. 

Plaid has data agreements with the large banks and is a CRA that lets you use users' cash flow data for underwriting. The power in data aggregation isn’t just raw data access its using it to price lending. Cash flow data (what a user spends) is so accurate

Want to build a compliance stack? Great, you’ll need a lot of proprietary and non-public data. Like originally sourced KYC documents, checks against government databases, industry databases (like EWS), user device & behavior data. You can buy all of that. But every time you buy another data set, you’re adding cost (COGS). 

Sardine* gives you a fraud and compliance toolset built for people who value performance. Sure, you could vibe-code similar features, but you might lack the expertise to build them that certain way that performs. When you’re buying these platforms, you’re also buying an opinion on what data is useful for preventing fraud or money laundering from an expert team.

The compounded advantage of founder-market-fit, network effects, and efficient scale that the existing Fintech SaaS companies have is non-trivial. 

The moat in Fintech SaaS is often not the distribution model (software hosted in the cloud). But in the expertise, data, and network effects those companies built and compounded to reach the point they’re at today. 

What if we just build it?

It used to be that you had to avoid side quests when building a Neobank. 

Now there’s less of a barrier.

Spending engineering time and effort on building the perfect compliance dashboard could be a distraction from the mission of saving customers time and money (in Ramp’s case), or helping cash stretch a little further (for Cash App). 

But now these companies have 10x more productive engineers, if owning the internal tool and self building it creates a better customer outcome there’s less of a barrier to building it. And, given they have some expertise and scale they can buy access to some of the data they’d need to do a good job at it. 

So yes you could unplug all your SaaS tools and build custom software.

But whether you should depends on if you can get enough benefit for the juice to be worth the squeeze. 

And some companies do choose to build solo. 

• Cash App has built their own Cash Score, using data from their customer base to price lending. 

• Brex built directly into payments networks which has enabled them to scale more quickly into new markets.

• Ramp is building all the AI tools - like a Policy agent.

Robinhood shipped $100m+ from 11 new product lines in the last 12 months. The speed at which these companies are able to ship new things is compounding.

What they choose to ship is the interesting signal. 

AI doesn’t eat Fintech SaaS - But it makes it much more competitive

The Fintech SaaS companies that survive aren't the ones who built great software. They're the ones who accumulated things agents can't generate: proprietary data, regulatory relationships, and the scar tissue from 1000 edge cases that only show up at scale.

If you're a Neobank deciding whether to build or buy, ask yourself: do I have the internal horsepower to burn cycles on identifying idiosyncratic user device behavior that could signal fraud? Or do I want my engineers shipping features customers actually see?

The moats are still there. 

But the drawbridge just got a lot faster.

ST.

4 Fintech Companies 💸

1. Nodu - MiCA aligned Stablecoin APIs for Fintech companies

Nodu allows regulated banks, PSPs, and money transmitters to quickly add stablecoin capability via its APIs. It aligns all payments and onboarding flows to AML/CFT and sanctions workflows with local laws and Europe’s MiCA regulation. 

🧠Every PSP and bank is looking for an infrastructure partner, and coverage in Europe is limited. Only recently, BVNK got its full SEPA readiness, and other infrastructure companies still have that on their roadmap. This is a team who’s built in payments before that knows Europe well (which is a complex regulatory market). This could be well timed.

2. Fido - Up to 10% APY with DeFi-powered savings

Fido is a simple mobile app that allows users to fund the wallet from their bank account, deposit funds, and earn up to 10% APY on funds. It also has gamified features to unlock further rewards.

🧠 The simplicity is the product. These ultra-simple UI’s are cheap to build on stablecoin infrastructure, and offer something that’s compelling. My question is how many of these apps will still be here in 5 years' time, especially if DeFi yields are correct.

3. Lighter - The Perps decentralized exchange on Ethereum.

Lighter is a new L2 (Layer 2) blockchain that uses Ethereum for its security. It enables ultra-high-frequency trading of “perps” (perpetual futures) at a speed professional traders expect. By building an app-specific ZK-rollup (a type of L2) they’re able to deliver much faster order matching and execution.

🧠 Traders love speed. We’re now in a place where on-chain performance is as good, and in some cases better than centralized exchanges. The “perp” is a very popular type of derivative, unlike an option or forward it has no expiry, so traders can get in and out of positions much faster and simpler. We’re not done at building new chain infrastructure yet. It’s still early.

4. Atomic Insights - The Money Mover for RIAs and Family Offices

Atomic Insights manages approval, compliance, and simplicity through a single dashboard for RIAs and Family Offices. It provides a real-time API to custodians, cash flow reporting, and integration to the CRM. 

🧠 Offices and RIAs have workflows like capital calls that is a pain that is not obvious unless you’ve lived it. It also has nice things like SSO support to Schwaab money. If you’re an RIA, this is a huge problem solver. 

Things to know 👀

1. Brex acquired by Capital One for $5.15bn

Capital One has agreed to acquire Fintech B2B spend management and CFO platform Brex for $5.15bn. Capital One, historically a consumer-lender, is acquiring one of the “big three” challenger corporate card and spend management platforms.

🧠 This is the single largest bank/fintech acquisition in history. Capital One gets a ready-made solution with “Brex-as-a-Service,” the horizontal solution Brex has doubled down on. Brex gets an acquirer that’s founder-led (Richard Fairbank), to operate separately, and Capital One will pour marketing dollars on them to grow.

🧠 Capital One gets a turnkey EU payment institution license, $13B in deposits, and clients like TikTok and Robinhood. They paid $5.15B for infrastructure that would've taken a decade to build.

🧠 What a comeback for Brex, who had a torrid pandemic:

• Bought a restaurant during COVID (???)

• Negative PR when they “abandoned” 10,000+ SMB customers

• Watched Ramp sprint past them while they restructured

🧠 What about the downround? Do the math. “$725m in capital raises above $5b, so assume that comes off the top of 1x liq pref. Approx $700m in debt (assuming not converted). That leaves $3.6b for equity holders.”

🧠 The lesson isn't "down rounds are bad." It's that liquidity beats valuation when the music stops.

🧠 Brex has become the “Adyen” of B2B Fintech focussed on large enterprises and ERP integration. This makes the Brex platform uniquely simple to integrate into Capital One’s go-to-market. I’m generally dubious about companies acquiring start-ups, but this one could have legs.

🧠 There’s a fascinating contrast with Ramp.

• Ramp's last valuation: $32bn.

• Same market. Often considered peers.

• Ramp is growing faster and took less time to do it.

• Ramp's founders previously sold Paribus to... Capital One.

You couldn't write this in fiction.

 🧠 How is Fifth Third feeling about this acquisition? Fifth Third signed a partnership with Brex for commercial cards in early December. I imagine their deal will be maintained, but does Capital One want to be in the business of serving its competitors that way?

🧠 Where does the Discover network fit in with all of this? The Diners Club commercial card did $622bn of transaction volume in 2024, but for M&A, maybe that’s a red herring.

🧠 For the team and founders, this could be a great deal. $12.3bn → $5.15bn sounds like a loss. But $3.6bn of upside is meaningful for folks like Ribbit, who led a $7m Series A. For the founders and team, this is a great outcome. Brex once looked troubled and overvalued. Now it has executed one of the greatest comebacks and exits in Fintech history.

2. Australian Financial Regulator investigates Airwallex after AML concerns

AUSTRAC the Australian financial crimes agency,” is investigating Airwallex and has forced it to engage a special auditor after becoming concerned that the money transfer platform was being used to fund child sexual abuse.” According to AFR, the regulator was concerned the company “was not properly monitoring payments it was processing.” The news follows Airwallex's recent funding and planned accounts. The agency went on to say they had concerns of “money mules” accounts in real names used by criminals to funnel money abroad. The company said it expects a special audit to validate its existing practices.

🧠 Money mules are a massive issue for everyone, everywhere. If you speak to any bank or regulator, it's become an incredibly fast-growing problem for them all.

🧠 Solving it is non-trivial; a legit customer, with a real identity, signs up and starts moving money. How is the company opening the account supposed to know that the person has been compromised?

🧠 The most common controls are KYC and Transaction Monitoring. Getting more data about the customer, their previous bank accounts, government, and consortia database checks, etc., can help validate if someone really can afford to move $1000s of dollars a month offshore on a smaller salary. Or if that person has a history of opening lots of new accounts.

🧠 Cross-border makes this 100x harder and riskier. If you’re moving funds around lots of countries between companies and individuals, now you need those company registries, and the history of the owners. In every single market. Often, where data quality is poor.

🧠 Every company does this to some extent but it’s hard and nobody does it perfectly. The regulators often question if enough is being done to continually check, improve and monitor.

🧠 For some added drama famed investor Keith Rabois has publicly called out Airwallex as “Chinese engineering and legal apparatus with a global facade” something the company streneously denied. I often wonder, how can you operate a cross-border business, especially into and out of APAC, without getting dragged into Geopolitics these days?

🧠 Time will tell what happens next. Airwallex is still growing incredibly quickly, shipping new products, and is on a path to an IPO. Will this be a bump in the road or a major obstacle?

🧠 This story, and Kontigo recently, are strong reminders that compliance is not optional. Moving money is hard. And getting into Fintech is easy. Staying there is the real battle.

2. Trump calls on Congress to implement the 10% APR cap on credit cards

From Finextra: “President Donald Trump says he will ask Congress to implement a one-year 10% cap on credit card interest rates in a move that JPMorgan chief Jamie Dimon says would spell “economic disaster” for the US.”

🧠 Jamie Dimon says capping rates, “would remove credit from 80% of Americans, and that is their back-up credit." 

🧠 Not to call out the GOAT here, but he’s sort of right and sort of wrong.  

🧠 He’s right because obviously a cap on APR just means lending becomes uneconomic and consumers no longer get credit.

🧠 He’s wrong because we live in a world where BNPL exists, and Cash App and Chime are innovating on what’s possible with cash flow underwritten credit.

🧠 The fact is credit cards are a cash cow business for the big banks. Often at the expense of the lower-income consumer population.

🧠 I doubt Congress will go for this, and the bank lobby is using every platform at Davos to call it what it is, stupid and unworkable. But my point is, credit cards have a major competitor in BNPL and Neobanks. Today they’re still small, but moves to expand credit affordability should (and will) come through competition, and if we legislate it should be to encourage that.

Other things worth your attention

• Revolut added account-to-account (pay by bank) to Google’s A2P Agentic Commerce Protocol and rolled it out in Revolut Pay. You should pay attention to this because Revolut Pay is quietly becoming a force in online payments. If agents will choose to buy from places that support them. They can also choose to pay by bank and select Revolut Pay if they see it available, which means more sales for merchants and more volume for Revolut. That’s a potential innovation dividend.

Good Reads 📚

1. Kontingo the “Sanctions Evasion” startup

Kontingo, the stablecoin wallet start-up has enabled transactions to sanctioned entities such as Banco de Venezuela in Venezuela, with the founder having taken screenshots of off-ramping to that bank. This was followed by announcing the ability for “Venezuelans to open free “virtual” accounts at JPMorgan Chase in order to move money.”

🧠 Jason kept the receipts. This piece is full of screenshots of the above and much more and stands as a stark warning to everyone in the stablecoin space.

🧠 Kontigo seems to think threats are the answer to their negative PR. In reality, it just further damages their credibility. Plenty of founders have been on the receiving end of it and are still building great businesses. They did so by also addressing the issues, not threatening and attacking those who raised them.

Tweets of the week 🕊

Twitter tweet

Twitter tweet

Twitter tweet

Twitter tweet

Twitter tweet

That's all, folks. 👋

Remember, if you're enjoying this content, please do tell all your fintech friends to check it out and hit the subscribe button :)

Want more? I also run the Tokenized podcast and newsletter.

(1) All content and views expressed here are the authors' personal opinions and do not reflect the views of any of their employers or employees.

(2) All companies or assets mentioned by the author in which the author has a personal and/or financial interest are denoted with a *. None of the above constitutes investment advice, and you should seek independent advice before making any investment decisions.

(3) Any companies mentioned are top of mind and used for illustrative purposes only.

(4) A team of researchers has not rigorously fact-checked this. Please don't take it as gospel—strong opinions weakly held

(5) Citations may be missing, and I’ve done my best to cite, but I will always aim to update and correct the live version where possible. If I cited you and got the referencing wrong, please reach out