Failory

Fintech on Fumes

Nicolás Cerdeira — Thu, 11 Jun 2026 15:01:59 +0000

Hey - It’s Nico.

Welcome to another Failory edition. This issue takes 5 minutes to read.

If you only have one, here are the 3 most important things:

Chimoney, a startup building a global payout API, shut down — learn why below
LLMs are picking winners. Here’s how to become one
There’s a new Claude model: Fable 5 — learn why you should try it below

A big thanks to today’s sponsor, SparrowDesk. Let their AI agents handle 60%+ of your support tickets automatically.

What if 60% of your support tickets never reached you? ^AD

As a startup, most of what lands in your inbox is repetitive. The same questions, the same responses, resolved manually by someone who should be closing deals or building your product.

SparrowDesk fixes this. Our AI agents auto-resolve 60%+ of queries before it reaches your team. One unified inbox for chat, email, and more, so nothing goes missed, and every customer gets a reply.

Early-stage teams get 90% off the first year through our startup program.

→ Try SparrowDesk for free

This Week In Startups

🔗 Resources

How AI changes the software value chain

Why you should focus on the Return on Tokens (ROT)

LLMs are picking winners. Here’s how to become one

Your startup didn't work out. Now what? *

📰 News

Decart’s new world model can simulate hours of photorealistic driving

Google just fired a warning shot in the AI subscription price wars

OpenAI files confidentially for IPO, following Anthropic

Apple’s long-awaited AI Siri overhaul is finally here

💸 Fundraising

Arpio, an AI-native cloud recovery and resilience platform, raises $15M Series A

Forage, a financial infrastructure platform for government benefits payments, raises $40M Series B

Terra AI, an AI platform for mineral and energy exploration, raises $20M Series A

ZeroDrift, an AI compliance firewall for enterprise communications, raises $10M seed round

* sponsored

Fail(St)ory

Money Out, Not Money In

Chimoney tried to make messy global payouts feel simple.

It wanted to build a payout layer that could sit between companies, recipients, banks, wallets, local rails, and regulators across multiple markets.

It raised less than $1M to do it, which is roughly “nice seed round for a calendar app” money, not “move money across continents without breaking anything” money.

What Was Chimoney:

Chimoney was a payout orchestration layer for businesses sending money into fragmented markets.

It lived after the moment a company already owed someone money:

A marketplace needed to pay sellers.
A remote-work company needed to pay contractors.
A creator or affiliate platform needed to pay users across countries where the preferred payout method changed by market.

That was the job Chimoney tried to handle. Integrate once, fund the payout, and let recipients receive money through the method that actually worked where they lived.

The product supported 41 currencies across North America, Africa, and Latin America. Customers could send bank transfers, mobile money, airtime, gift cards, and even stablecoins.

It was not trying to be a new Stripe. It did not help businesses accept card payments, run checkout, manage subscriptions, or process merchant payments. It focused on the other side of the flow: getting money out to recipients who might not have the same banking access, cash-out options, or local rails.

The messy part sat underneath that simple integration. If one recipient wanted mobile money, another needed a bank transfer, and another could only use airtime or a gift card, Chimoney had to make those options work behind the scenes.

That meant routing payouts, managing local partners, handling failed transactions, reconciling balances, processing refunds, answering support tickets, and keeping enough liquidity in the right places. The customer saw one API. Chimoney carried the market-by-market mess underneath it.

Near the end, Chimoney tried to reposition part of this infrastructure around AI wallets and programmable payments. It was a cleaner story for the future, but the company still needed the payout business to work in the present.

By the time Chimoney stopped accepting new transactions in April 2026, the shape of the failure was clear: the product solved a real last-mile payout problem, but the company had to carry too much infrastructure before enough volume arrived.

The Numbers:

💸 Funding: Under $1M raised over the company’s life
🌍 Coverage: 41 currencies supported
🏢 Customers: Hundreds of businesses served
🛑 Shutdown: Stopped accepting new transactions on April 30, 2026

Reasons for Failure:

Distribution never became a real engine. The founder’s own explanation was direct: Chimoney had a technically solid, licensed platform, but no customer acquisition engine. In payout infrastructure, buyers need proof that money lands, failures get handled, compliance is covered, and switching risk is worth taking.
The last mile was expensive to operate. Chimoney simplified payouts for customers by absorbing the operational mess itself. Every local rail, mobile money option, gift card route, off-ramp, refund path, and partner relationship added work behind the API.
Fintech costs arrived before fintech scale. Under $1M is thin for a company moving money across multiple jurisdictions. Licenses, audits, banking partners, compliance, support, and liquidity management all cost money before transaction volume is large enough to carry them.
Volume density never caught up. Payout infrastructure gets healthier when repeat transactions flow through the same rails often enough. Without strong distribution, Chimoney still had to maintain the platform, partners, support, refunds, and compliance systems on too small a revenue base.

Why It Matters:

Payout startups need dense, repeat transaction volume before the operational load becomes manageable.
Developer experience helps, but payout buyers mostly care about whether money arrives and who fixes it when it does not.
Future-market repositioning is hard when the current infrastructure business is already running out of room.

Trend

Claude Fable 5

There’s a new Claude model in town, so naturally, everyone is talking about it.

Anthropic says it’s SOTA, many developers claim it’s the new coding model to use, and a lot of people seem angry that it won’t help them make a bomb.

Here’s everything you should know about it: the good, the bad, and the biohazard disclaimers.

Why it Matters

The best model is no longer always the model you get. Fable 5 is public. Mythos 5 is for vetted partners. Same family, different permissions, which means AI access is becoming a status layer.
The coding hype seems earned. Fable 5 looks built for the annoying work developers actually hate: messy repos, migrations, broken tests, weird dependencies, and tasks that take more than one clean prompt.
The guardrails are part of the product now. Sometimes Fable answers. Sometimes it refuses. Sometimes it quietly sends you to Opus 4.8.

The Coding Hype Has Substance

The strongest case for Fable 5 is coding.

Anthropic says it is better at long-running agentic work, large migrations, complex implementations, test writing, design fidelity, and reasoning across big codebases.

Early developer reactions mostly point in the same direction: it stays useful for longer, handles more context, and needs less babysitting once the task gets messy.

The specs help explain why. It reportedly has a 1 million token context window and 128,000 max output tokens. That gives it more room to read, plan, edit, test, and produce larger changes in one run.

Anthropic’s biggest example is Stripe using it to cut a 50-million-line Ruby migration from months to a day. My favorite example so far has to be this one:

The problem: it is very very expensive. Fable 5 costs twice as much as Opus 4.8, so letting it loose on a repo has the energy of giving a very smart intern your credit card.

The Guardrails Are Everywhere

The weirdest part of Fable 5 is that it is not just a model. It is a model with a bouncer.

Anthropic says Fable 5 uses the same underlying model as Mythos 5, but with stronger safeguards for public use. Mythos 5 is the more restricted version, available only to vetted partners, because it can help in areas that could be dangerous (mostly cybersecurity and biology).

So Fable 5 sometimes does something strange. If a cyber or bio prompt gets flagged, it may route the request to Opus 4.8 instead.

The biology filters seem especially jumpy. Reports say Fable 5 refused or rerouted questions about mitochondria, mRNA vaccines, hay fever, asthma medication and antibiotic resistance. It's like if a high school worksheet got stopped at airport security.

Mythos Already Changed the Security Workload

Anthropic’s caution makes more sense when you look at what Mythos has already done.

In Project Glasswing, around 50 partners used Mythos Preview for defensive cybersecurity. Anthropic says they found more than 10,000 critical severity vulnerabilities.

Then Mythos scanned more than 1,000 open-source projects, initially estimated 23,019 vulnerabilities, and produced thousands of critical severity findings. Anthropic says independently reviewed findings had a 90.6% true-positive rate.

That’s the fear: the same model that can find vulnerabilities for defenders can also help attackers find them first. So Anthropic is keeping the public version on a much shorter leash.

That is the tradeoff with Fable 5. The model looks genuinely useful, especially for coding, but the launch is also a reminder that the best AI tools may increasingly come with routing, permissions, and a bouncer at the door.

Help Me Improve Failory

That's all for today’s edition.

Cheers,

Nico

The Pizza Robot Graveyard

Nicolás Cerdeira — Thu, 04 Jun 2026 17:33:41 +0000

Hey - It’s Nico.

Welcome to another Failory edition. This issue takes 5 minutes to read.

If you only have one, here are the 3 most important things:

Picnic, a startup building robots that made pizza, has shut down — learn why below
24 tips for giving S-tier demos
Microsoft Build 2026 was this week — all important announcements below

A huge thanks to today’s sponsor, HubSpot. Download their free AI guide to grow your business today.

Demystify AI for Your Business in Just 4 Steps ^AD

Overwhelmed by AI? HubSpot's free guide cuts through the noise. Get the ultimate crash course for non-technical entrepreneurs who want to harness AI's power—without getting lost in the jargon.

"AI for Business Builders" delivers:

A 4-part roadmap to AI mastery
Jargon-free explanations of large language models
Practical prompt engineering tips you can use today
Real-world examples of AI boosting businesses like yours

Arm yourself with the knowledge to make informed AI investments and skyrocket your startup's growth.

Download Your Free Guide Now

This Week In Startups

🔗 Resources

Startups become what they hunt

24 tips for giving S-tier demos

Small Businesses are the Next Frontier for AI

Your startup didn't work out. Now what? *

📰 News

OpenAI launches new Codex tools for white-collar work

Anthropic files to go public

Claude Opus 4.8 is out and is SOTA

Meta rolls out a new AI creator assistant on Facebook

💸 Fundraising

OpenRouter, an AI model exchange and inference routing platform, raises $113M Series B funding

Reactor, a real-time generative video platform for developers, raises $59M Series A funding

Catena Labs, an AI-native banking infrastructure startup, raises $30M Series A funding

Saris, an AI workflow automation platform for banks and credit unions, raises $28.8M Series A funding

* sponsored

Fail(St)ory

Robot Aquarium

Picnic built a robot that could sauce, cheese, and toppings at assembly-line speed.

It could sit inside a commercial kitchen and help one worker turn out 100+ pizzas an hour. That made it easy to like and easy to demo.

The problem was turning a pizza-making robot into something enough operators needed every month.

What Was Picnic:

Picnic built the Picnic Pizza Station, a robotic pizza assembly system for commercial kitchens. It applied sauce, cheese, pepperoni, and other toppings onto dough, then handed the pizza off to the rest of the kitchen workflow.

Pizza was a smart first target. The work is repetitive, the ingredients are standardized, and rush-hour demand can be painful to staff.

The company claimed one employee could produce roughly 100+ customized 12-inch pizzas per hour with the system, depending on setup and size.

That was the pitch: fewer hands on the line, more predictable output, less waste from inconsistent topping.

Picnic sold the machine through a leasing model. When it reached the market in 2021, pricing started around $3,500 per month.

That made the product feel closer to an operating expense than a large capital purchase, but the customer still had to believe the machine would earn its place every month.

The strongest use cases were dense-volume environments: stadiums, campuses, military sites, large venues, and high-throughput foodservice operators. Those are the kinds of places where pizza behaves more like production.

A normal restaurant is messier: some hours are slow, menu changes matter, and kitchen space is tight. Staff still need to load ingredients, monitor the machine, clean it, recover from jams, and trust it during the exact moments when failure is most expensive.

Picnic had real customer interest. It partnered with Ethan Stowell Restaurants, tested with Domino’s Pizza Enterprises in Berlin, listed customers like Aramark, Chartwells, and Compass Group, and pushed into venues and institutional foodservice.

It also carried the cost structure of a hardware company. Machines had to be manufactured, shipped, installed, maintained, monitored, and supported. In 2024, Picnic raised $5M to scale production and meet demand across North America.

By May 2026, the company had become insolvent. Its assets and IP were sold to an undisclosed buyer through liquidation.

One customer described being left with a $250,000 “robot aquarium,” which says a lot about the risk operators take when their automation vendor disappears.

The Numbers:

🍕 Founded: 2016 in Seattle
⚙️ Throughput claim: roughly 100+ pizzas per hour
💸 Lease pricing: started around $3,500/month in 2021
📈 Funding: roughly $50M-$53M raised
👥 Peak team: about 100 employees by 2023
🏟️ Customers and pilots: MOTO Pizza, Domino’s Pizza Enterprises, Aramark, Chartwells, Compass Group, military and venue installs

Reasons for Failure:

The ROI depended on dense, predictable pizza volume: Picnic made the most sense when pizza production was concentrated: stadium rushes, campuses, bases, large venues, and very busy shops. In lower-volume restaurants, a $3,500/month lease had to compete with slow hours, uneven staffing needs, and limited kitchen space.
The robot automated only one visible part of the job: Topping is important, but it is only one step in a pizza operation. People still had to prep ingredients, manage dough, bake, cut, box, serve, clean, and fix problems when the workflow broke. For operators, the buying decision was really about total kitchen reliability, not demo throughput. A robot that needs monitoring and support can still save labor, but the savings need to survive real shifts.
Hardware burn met a colder funding market. Former CEO Clayton Wood said Picnic was “caught in the squeeze” between the free-money period and the harsher post-2022 market. That matters for a company selling physical systems. Picnic needed capital for manufacturing, deployment, maintenance, and sales cycles that move slower than software.
Food robotics had become a scarred category. Picnic was selling into a market where Zume, Basil Street, Piestro, and other restaurant automation startups had already failed. That gave buyers a fair reason to ask what would happen if the vendor disappeared and the robot became expensive furniture.

Why It Matters:

Sell the customer the full outcome, because automating the most visible task may still leave them with the annoying parts.
Your best use case is only a market if there are enough buyers with the same pain, budget, and workflow.
A buyer’s fear of being stuck with dead equipment can matter as much as your ROI math.

Trend

Microsoft Build 2026

Microsoft used Build 2026 to show how much of the enterprise agent market it can absorb by default.

It already controls the work context, identity layer, data platform, dev workflow, runtime, and endpoint. Now it is wiring those pieces together so agents can use company files, calendars, permissions, business data, and approval rules without customers stitching the stack themselves.

For founders, generic workplace agents just got harder to sell.

Why it Matters

Enterprise context becomes distribution. Microsoft IQ gives agents company context from files, people, workflows, data, and the web. Startups have to rebuild that through integrations; Microsoft already sits where the work happens.
Generic coordination agents get squeezed. Scout handles meetings, prep, stalled work, calendars, and deliverables across Microsoft 365. Startups need a narrower workflow, unique data, or a buyer Microsoft will not serve well.
Microsoft wants model control. Copilot leaned heavily on OpenAI. The new MAI models give Microsoft more control over cost, latency, routing, and product behavior.

Microsoft IQ makes company context the product

Microsoft IQ is Microsoft’s name for the context layer underneath its agents.

Think of it as the system that tells an agent what it needs to know before it acts: who works on what, which files matter, how the business defines key terms, what happened in meetings, and what outside information is current.

It has four pieces:

Work IQ: context from Microsoft 365: files, chats, meetings, people, calendars, and workflows.
Foundry IQ: internal company knowledge for agents built in Microsoft Foundry.
Fabric IQ: business meaning for company data, like how customers, products, invoices, and metrics connect.
Web IQ: fresh external information from web pages, news, images, and video.

The idea is that companies define this context once, then reuse it across Copilot, Foundry, Copilot Studio, and Microsoft 365. That makes agents less dependent on long prompts and one-off integrations.

Scout is Microsoft’s always-on work agent

Scout is Microsoft’s first Autopilot: an agent that can run in the background instead of waiting for a prompt every time.

It connects to Microsoft 365 tools like Teams, Outlook, OneDrive, SharePoint, calendar, email, contacts, and browser context. The job is coordination work: scheduling meetings, preparing users, flagging stalled decisions, blocking focus time, and noticing upcoming deliverables.

Under the hood, Scout is built on OpenClaw, Microsoft’s open-source agent runtime for autonomous workflows.

The important part is control: Scout gets its own governed Entra identity, so its actions are visible and attributable. Companies can limit what it can access, require approval for sensitive actions, and apply existing Purview policies when it handles data.

GitHub Copilot is getting a new app

The new GitHub Copilot app is a desktop app for running coding agents as separate work sessions.

Instead of asking Copilot for code in the editor, a developer can spin up agent sessions from GitHub Issues, Pull Requests, or previous sessions. Each session gets its own branch, files, conversation history, and task state through git worktrees.

The practical change is parallel work. One agent can investigate a bug, another can draft a feature, and another can respond to review comments, without mixing files or branches.

The developer still decides what gets merged. Copilot is taking on more of the execution loop, while the human keeps the review, testing, and approval role.

Microsoft is reducing its OpenAI dependency

Microsoft also announced seven in-house MAI models for reasoning, coding, image generation, transcription, and voice.

The main one is MAI-Thinking-1, a 35B-active-parameter reasoning model for multi-step tasks, long-context work, and code generation. It is still limited to early partners.

This is important because Copilot and Microsoft’s AI products have relied heavily on OpenAI models. The MAI models give Microsoft more control over cost, latency, routing, and product packaging. Simple tasks can run on cheaper Microsoft models, while harder tasks can still use frontier models when needed.

Help Me Improve Failory

That's all for today’s edition.

Cheers,

Nico

Turned Into A Feature

Nicolás Cerdeira — Thu, 28 May 2026 18:54:45 +0000

Hey - It’s Nico.

Welcome to another Failory edition. This issue takes 5 minutes to read.

If you only have one, here are the 3 most important things:

Huxe, an AI audio app that turned your inbox into a personal daily briefing, has shut down — learn why below
How to Avoid AI Code Slop
Tokenmaxxing is the new obsession with measuring AI productivity — learn why it matters below

A huge thanks to today’s sponsor, SimpleClosure. When startups hit the wall, close cleanly and move forward with expert-guided support.

When It’s Time to Close, Do It Properly ^AD

Across the startup landscape, founders are facing tougher markets, crowded categories, and harder paths to sustainable growth.

When shutdown becomes the right next step, the work is not over.

SimpleClosure helps founders bring structure to the process, from state filings and investor communications to final distributions, compliance, and selling remaining assets. So you can close cleanly, responsibly, and with a clearer path through.

👉 Simplify Your Shutdown

This Week In Startups

🔗 Resources

How to Avoid AI Code Slop

In B2B customer support, AI is a Copilot, Not a Replacement

The Founder's Black Box

Anthropic and OpenAI have finally found PMF

📰 News

Meta launches Instagram, Facebook, and WhatsApp subscriptions

YouTube adds new podcast features, including an AI recommendation tool

Pope Leo uses first major papal text to warn about dangers of AI

Robinhood now lets your AI agents trade stocks

💸 Fundraising

AI coding startup Cognition raises $1B

Mercury, a fintech banking platform for startups, raises $200M Series D funding

Moment, an AI operating system for investment management, raises $78M Series C funding

Modal, a serverless cloud platform for AI infrastructure, raises $355M Series C funding

Fail(St)ory

Personalized Podcasts

Huxe was an AI audio app that turned your inbox, calendar, and interests into a personal daily briefing.

You opened it in the morning, pressed play, and got something closer to a podcast about your day.

It came from the same world as NotebookLM’s viral Audio Overviews. Ex-Googlers built it. Good investors backed it.

Then, on May 21, 2026, Huxe announced it was winding down, in the same month Spotify and Amazon pushed deeper into AI-generated personal audio.

What Was Huxe:

Huxe was a consumer AI startup in the Bay Area built around one bet: people do not want to read everything.

They want the important stuff spoken to them while they drive, walk, cook, work out, or avoid another screen.

The team came from Google’s NotebookLM project, where they had already seen AI-generated audio take off.

At first, Huxe looked more like a work tool. The early idea was to connect systems like Salesforce and let teams chat with company data.

Then the team shifted toward consumers. The bigger opportunity seemed to be audio.

The core product was a personalized daily briefing. Users connected email, calendar, and interests, then got a morning audio summary covering meetings, important emails, news, and topics they cared about.

The app also had personalized audio feeds, AI-generated DeepCasts built from prompts, live AI radio-style stations around topics like tech or sports, and interactive hosts users could interrupt to ask questions or change direction mid-conversation.

It was pretty ambitious. Huxe touched productivity, news, learning, search, and voice assistants all at once.

However, at the same time, Google, Spotify, and Amazon were pushing into the same category with products tied directly into platforms people already used every day.

Huxe raised $4.6 million in September 2025. Eight months later, it shut down.

The Numbers:

🗓️ Founded: late 2024, after the team left Google
💰 Funding: $4.6M
📱 Public launch: June 5, 2025
🛑 Shutdown announced: May 21, 2026
📉 Public launch to shutdown: about 11.5 months

Reasons for Failure:

The product was turning into a platform feature: A personalized AI briefing fits naturally inside Google, Spotify, Amazon, Apple, or Microsoft. Google had NotebookLM and Daily Listen. Spotify announced Personal Podcasts the same week Huxe wound down. Amazon had Alexa Podcasts. Huxe had to convince users to download a new app, while the platforms could place similar features inside products people already used every day.
The trust barrier was high: Huxe worked best when users connected email, calendar, and personal context. That is sensitive territory for a young startup. Many users hesitate before giving inbox access to a product they discovered a few weeks ago. Big platforms already had the accounts, the data, and years of user familiarity behind them
The product needed near-perfect reliability: The whole pitch was saving users time. That breaks fast if people still feel the need to double-check their inbox afterward. Some users liked the summaries but still worried important emails could get missed. Once that doubt appears, the product shifts from essential tool to optional layer.
Audio has very little margin for error: Users tolerate mediocre text more than mediocre audio. You can skim weak writing in seconds. Audio forces you to sit through every awkward pause, robotic sentence, and bad transition. Small quality issues feel much bigger in a product people are supposed to use every morning.

Why It Matters:

In consumer AI, the biggest risk is often becoming a feature inside a larger platform.
Audio products have a brutal quality bar because users feel every second of bad output.

Trend

Burning Tokens

This week, Uber’s COO said AI token spending is getting harder to justify. A few days earlier, OpenAI offered YC startups up to $2 million in API credits and encouraged founders to go “tokenmaxxing.”

That’s where the market is right now. One side is subsidizing massive AI consumption. The other is asking whether all this compute actually produces better products.

The interesting thing is how quickly tokenmaxxing escaped startup Twitter and became corporate behavior. Internal leaderboards. Usage quotas. Employees running agents for no reason just to look AI-active.

A productivity tool quietly turned into a status game.

Why it Matters

AI spend is becoming a hiring decision. Founders increasingly compare token costs against salaries, not software budgets. Uber even said it’s hiring fewer people while trying to balance AI spend against headcount.
AI adoption and AI productivity aren’t the same thing. Companies like Meta and Amazon reportedly tracked token usage aggressively. Others, like Indeed, focused on outcomes instead.
Tokens are becoming a real cost center. AI usage scales unpredictably through agents and automation loops. Companies are starting to treat tokens more like cloud infrastructure than SaaS seats.

Tokenmaxxing

Tokenmaxxing is what happens when companies start treating AI usage as proof of productivity.

Executives want employees using AI more aggressively. But AI adoption is hard to measure. Most workers say they use ChatGPT or Claude occasionally. Few actually change how they work around it.

So companies started looking for a measurable signal.

Tokens became the obvious choice. Every prompt, response, workflow, or agent loop consumes them. More tokens usually means more AI usage. That made token consumption an easy internal metric.

Then the dashboards appeared.

Meta built an internal leaderboard called “Claudeonomics” tracking AI usage across roughly 85,000 employees. Microsoft created internal token leaderboards as well. Amazon managers monitored AI adoption rates and usage dashboards internally.

YC encouraged founders to think about “tokenmaxxing over headcountmaxxing.” OpenAI reinforced the trend by offering YC startups up to $2 million in API credits. Small startups started viewing tokens as a substitute for hiring.

Nectir reportedly pushed engineers to spend hundreds or even thousands of dollars per month on AI tools because leadership believed aggressive usage would force employees to discover better workflows. One founder described AI agents as becoming “an army of coders.”

Why it Broke

The problem with tokenmaxxing is simple: activity is easy to measure. Value isn’t.

Once token usage became visible inside companies, employees started optimizing for the metric itself.

According to reports, some Meta employees left AI agents running for hours just to inflate usage numbers. One user consumed 281 billion tokens in a month.

Amazon employees used internal AI tools for unnecessary tasks because they believed managers were paying attention to adoption dashboards

Then the financial reality showed up.

Uber burned through its annual Claude Code budget just four months into the year. Leadership started asking the question: does more token consumption actually create proportional business value?

Some people still believe tokenmaxxing is a useful signal, at least before employees start gaming it. Their argument: strong engineers using AI seriously naturally burn more tokens because they’re testing more ideas, running more workflows, and shipping faster.

Others think the metric corrupts itself the moment companies start tracking it publicly. Employees stop optimizing for output and start optimizing for visibility instead.

The real lesson is probably that token usage works as an early adoption push, but breaks once consumption becomes the goal rather than the byproduct of useful work.

Help Me Improve Failory

That's all for today’s edition.

Cheers,

Nico

23M Users. Only 75 Paid.

Nicolás Cerdeira — Thu, 21 May 2026 18:36:49 +0000

Hey - It’s Nico.

Welcome to another Failory edition. This issue takes 5 minutes to read.

If you only have one, here are the 3 most important things:

StreamElements, a startup building creator tools, is about to shut down — learn why below
The Dark Marketplace: why Vertical AI is coming for commerce
Google’s I/O 2026 event just finished — learn every important announcement below

A huge thanks to today’s sponsor, SimpleClosure. When startups like StreamElements face an uncertain future, close cleanly and move forward with structured, expert-guided support.

Closing Well in Uncertain Times ^AD

StreamElements’ potential shutdown is a familiar startup story: a company can build for years, raise significant funding, support millions of users, and still reach a point where the next step is uncertain.

Across the creator economy and AI tooling space, founders are running into similar pressure: crowded markets, changing customer behavior, and harder paths to sustainable growth.

When shutdown becomes the right next step, the work is not over.

SimpleClosure helps founders bring structure to the process, from state filings and investor communications to final distributions, compliance, and asset sales. So you can close cleanly, responsibly, and with a clearer path through.

👉 Learn more

This Week In Startups

🔗 Resources

YC’s Request for Startups - Summer 2026

The Dark Marketplace: why Vertical AI is coming for commerce

After Automation: AI progress creates more work for humans, not less

Stuck between US LLC and EU? Honest comparison of 12 countries, real tax data *

📰 News

OpenAI co-founder Andrej Karpathy joins Anthropic’s pre-training team

Elon Musk has lost his lawsuit against Sam Altman and OpenAI

OpenAI launches ChatGPT for personal finance, will let you connect bank accounts

Figma adds an AI assistant to its collaborative canvas

💸 Fundraising

Hark raises $700M Series A at a $6B valuation

Mercury raises $200M Series D funding

GridCARE raises $64M to squeeze more power out of grids

Commure, an AI platform for Healtchare, raises $70M in funding

* sponsored

Fail(St)ory

The Cost of Free

StreamElements was a creator-tools company built for livestreamers.

It had millions of creators, deep Twitch and YouTube roots, $111 million in funding, and a product people actually used.

Then, in May 2026, the company entered shutdown mode while trying to find a buyer, with creators reportedly told to save their assets during a 30-day window.

What Was StreamElements:

Launched in 2017 the company started with a simple pitch: give streamers the tools they need to run a professional live channel, for free.

That meant overlays, alerts, chatbots, tipping pages, loyalty tools, merch stores, analytics, moderation, and sponsorship infrastructure. The plumbing behind the show.

A streamer does not just press “go live.” They need alerts when someone subscribes, commands for chat, widgets for donations, sponsor graphics, and a setup that does not collapse the second the audience shows up.

StreamElements became one of the default options in that stack.

By 2019, more than 200,000 Twitch and YouTube channels were using it. Then the pandemic hit and livestreaming exploded. By 2021 more than 1.1 million creators were using the tool.

That same year, SoftBank led a $100 million round, bringing total funding to roughly $111 million.

So, they had the users, but how were they making money?

StreamElements kept the tools free, then made money through brand sponsorships. Brands paid to run campaigns with creators, creators got paid for the work, and StreamElements took a cut in the middle.

In theory, it was a good idea. Creators loved free tools. Brands wanted creator distribution. StreamElements sat in between.

However, free software still has bills.

By 2026, StreamElements said it was spending $300,000 per month just to keep its tools running. That did not include the heavier stuff around payroll, sales, support, and product.

So in January 2026, it launched Keep It Live, a crowdfunding-style campaign asking creators and viewers to help keep StreamElements free.

The campaign accidentally told the real story.

StreamElements said 23 million-plus creators had used its tools over the past decade, and 2 million-plus had gone live that month. The campaign had raised $1,700 from 75 supporters in the research snapshot.

Millions used it. Almost nobody paid for it.

By May 2026, GamesBeat reported that StreamElements was preparing to close its website and sunset its creator tools. Creators were reportedly told they had 30 days to log in and save their assets.

The company later said it was in “positive discussions with potential acquirers.”

The Numbers:

💰 Funding: $111M
👥 Users: 23M+ creators served.
🧾 Costs: $300K monthly tool costs
📉 3 major layoff rounds between June 2022 and August 2023.
🧑‍💻 200+ employees at peak, down to about 72 listed on LinkedIn by May 2026.

Reasons for Failure:

The free-tools bargain broke. StreamElements built enormous reach, but most of that reach sat outside the paywall because there was no real paywall. That only works when sponsorship revenue is growing fast enough to subsidize the product.
Sponsorships were too cyclical for the cost base. The company depended heavily on brand campaigns, and those budgets were the first thing to wobble when the ad market slowed. StreamElements said as much during its 2023 layoffs, pointing to weaker advertising demand and reduced budgets from existing clients. That is a bad place to be when your product costs are fixed and your revenue depends on marketers staying brave.
The 2021 round sized the company for a boom that did not last. StreamElements raised $100 million when livestreaming, gaming, and creator monetization all looked like they would keep climbing forever. Then growth cooled, ad spend tightened, and the company had to cut staff three times in 14 months
The creator product and the buyer product were not the same thing. Creators came for alerts, overlays, bots, and tipping tools. Brands came for measurable sponsorship performance. Those are two different products, two different sales motions, and two different definitions of success. StreamElements had creator adoption, but the business depended on advertiser conversion.

Why It Matters:

Creator love does not matter if the buyer is the advertiser.
Boom-era funding can lock a startup into boom-era assumptions.

Trend

Google I/O 2026

Google just wrapped its I/O 2026 event with a ton of AI announcements.

Most of them point in the same direction: Google wants AI embedded across everything it already owns. Search, Gmail, Chrome, YouTube, Shopping, Android.

Gemini 3.5 Flash became the foundation for most of the launches. Spark handles long-running tasks in the background. Omni pushes AI deeper into editing workflows. Search gets more agentic. Shopping gets automated.

Here’s what matters and where this is going.

Why it Matters

Google launched the operational model first: Gemini 3.5 Flash shipped before Pro because agents care more about reliability and speed across repeated actions than perfect reasoning on one big query.
Google is quietly turning Search into a monitoring product: The shift is from “search this” to “keep watching this for me.” That changes how people discover products, content, and information online.
Omni is a bigger threat to creative software than Veo was: Generating clips is one thing. Editing existing assets through conversation is much closer to actual creative work.

Gemini 3.5 Flash

The biggest launch from I/O was Gemini 3.5 Flash.

Google says it outperforms Gemini 3.1 Pro across several coding and agent benchmarks while running much faster. More importantly, Google immediately made it the default model behind AI Mode in Search.

Google clearly optimized Flash for constant execution across tools and workflows. Search queries, coding agents, background tasks, enterprise agents, shopping flows.

These systems constantly make API calls, retrieve information, compare outputs, and retry actions. Speed starts mattering a lot once that stack grows.

Spark

Spark feels much closer to OpenClaw-style agents than to a traditional assistant. Google is basically building a persistent cloud worker connected to Gmail, Calendar, Search, and eventually payments.

The examples were practical: reviewing inboxes, tracking internship openings, turning emails into task lists, scheduling calendar blocks, preparing recurring workflows, the kind of thing you would expect OpenClaw to be able to handle.

Spark runs on dedicated virtual machines inside Google Cloud, so it keeps running even when your devices are offline.

Google also said Spark will support third-party tools through MCP integrations later on. The rollout starts with trusted testers and AI Ultra subscribers in the U.S.

Daily Brief

Daily Brief organizes your morning inside Gemini: inbox updates, schedule context, tasks, reminders, priorities.

It’s nothing groundbreaking technically, but the packaging is probably the important part. Google is wrapping agent behavior into an existing habit instead of asking users to build workflows manually. Open app in the morning, get context, move on.

That approach showed up across a lot of the I/O announcements. Less “build your own AI system,” more lightweight automation built directly into products people already use.

Omni

Google also introduced Gemini Omni, a multimodal model focused on video generation and editing.

The important part is that Omni works with existing assets, not just prompts. Users can combine text, images, audio, and video inputs, then edit outputs conversationally inside tools like Flow and YouTube Shorts.

Search and Commerce

Google also upgraded AI Mode in Search with Gemini 3.5 Flash and introduced information agents that continuously monitor topics for users.

The shift here is pretty straightforward: instead of repeatedly searching for the same things manually, users hand monitoring tasks to the system.

Google connected this directly into shopping too.

Universal Cart and the new Agent Payments Protocol let users define spending limits, product constraints, and purchase conditions while agents handle transactions automatically once those conditions are met.

Overall, most of the I/O announcements pointed in the same direction: AI systems that stay connected to ongoing workflows instead of isolated prompts.

Help Me Improve Failory

That's all for today’s edition.

Cheers,

Nico

Sudden Bankruptcy

Nicolás Cerdeira — Thu, 14 May 2026 19:28:24 +0000

Hey - It’s Nico.

Welcome to another Failory edition. This issue takes 5 minutes to read.

If you only have one, here are the 3 most important things:

Parker, a startup making corporate cards for ecommerce, shut down abruptly — learn why below
Is Software Losing Its Head?
OpenAI’s new Voice Models and Thinking Machines new “Interaction Models” — learn everything about them below

A huge thanks to today’s sponsor, Warmy. Fix your email deliverability so your outreach reaches inboxes and generates leads.

Sending Emails but Getting Zero Replies? ^AD

It might not be your copy.

Many startups unknowingly have broken email deliverability, causing messages to land straight in spam folders.

That means lost leads and wasted outreach. Warmy.io automatically warms up your email accounts, protects your sender reputation, and keeps your emails landing in real inboxes.

If email is part of your growth strategy, this matters.

Fix your deliverability before your next campaign

This Week In Startups

🔗 Resources

How to Build Services-as-Software Business

The state of B2B monetization in 2026

Is Software Losing Its Head?

Your startup didn't work out. Now what? *

📰 News

Notion just turned its workspace into a hub for AI agents

Amazon launches an AI shopping assistant for the search bar

Anthropic now has more business customers than OpenAI

Google unveils Googlebook, a new line of AI-native laptops

💸 Fundraising

Vapi raises $50M Series B powering the next generation of enterprise voice AI

AI-powered wholesale broker Novella raises $21M

Rivian spinoff Mind Robotics raises another $400M

Exaforce Raises $125M Series B to Combat AI-Powered Attacks

* sponsored

Fail(St)ory

Parker

Parker built corporate cards and financing tools for ecommerce brands that needed cash before revenue showed up.

Then, almost overnight, it disappeared.

Three days after shutting down, the company filed for Chapter 7 bankruptcy. Reports suggest Parker had been trying to sell itself shortly before the collapse, and whatever was supposed to happen behind the scenes clearly did not happen.

What Was Parker:

Parker launched in 2019 and went through YC’s Winter 2019 batch.

Its focus was entirely on ecommerce brands, especially businesses doing a few million to tens of millions in annual sales.

Most startup card companies are built around SaaS startups. Those companies have predictable margins, recurring revenue, and relatively clean cash flow.

Ecommerce businesses are different. They constantly spend money before revenue shows up. Inventory gets paid upfront. Ads get paid upfront. Freight, packaging, suppliers, warehouses, agencies, manufacturers, all upfront. The money comes back later, if everything goes well.

Parker built its product around that gap.

The company’s pitch was that traditional banks and corporate cards underwrite ecommerce brands badly because they look at outdated financials and generic risk models. Parker wanted to underwrite using live business data instead.

Customers connected systems like Shopify, Amazon, and QuickBooks. Parker used that data to decide credit limits and repayment terms.

That let the company make a much more aggressive offer than normal business cards. Parker claimed it could offer limits 10x to 20x larger than traditional cards, with some businesses getting access to up to $10 million in credit.

The more interesting part was the repayment structure. Normal business cards use monthly billing cycles. Parker thought that made no sense for ecommerce operators managing inventory cycles and ad spend.

So instead, every transaction got its own repayment window. A purchase made on March 1 could be repaid on May 1. A purchase made two days later would have its own separate due date.

If you are buying inventory before Q4 or ramping ad spend before a launch, timing matters more than rewards points. A normal statement cycle can create random cash pressure depending on when purchases land. Parker tried to make repayment timing predictable.

Parker had real traction: they claim more than $1 billion in payments volume across hundreds of brands and over 65 million in revenue.

But then everything collapsed very quickly. Parker abruptly stopped operating on May 4. The company filed for Chapter 7 bankruptcy three days later.

The Numbers:

🗓️ Founded: 2019
🚀 YC Batch: Winter 2019
💰 Funding: $200M+
📈 Claimed payments volume: $1B+
💳 Claimed card limits: up to 10x-20x higher than traditional cards
📉 Bankruptcy filing: Chapter 7 liquidation

Reasons for Failure:

The business was riskier than it looked: Parker looked like software, but it was really an ecommerce credit business built around larger limits and longer repayment windows tied to volatile cash cycles.
Ecommerce is a brutal market to finance: Parker picked ecommerce because the pain was real. The problem is that ecommerce volatility cuts both ways. Ad costs spike, inventory gets stuck, suppliers miss deadlines, consumer demand swings, marketplaces delay payouts. A lender sitting inside those cash cycles inherits that chaos very quickly.
The funding headlines probably overstated how safe the company was: Parker often talked about raising more than $200 million. But a large part of that was asset-backed lending capacity supporting the card program. That money helps finance customer spending. It is not the same thing as having hundreds of millions sitting in the bank to absorb losses or survive operational problems.
The company seems to have run out of options very suddenly: The bankruptcy filing reportedly mentioned acquisition talks, mergers, and out-of-court wind-down options before the Chapter 7 filing. Reports also suggest a potential acquisition collapsed shortly before shutdown. If that is accurate, Parker may have been depending on a deal to survive and simply ran out of time when it disappeared.

Why It Matters:

Parker proved that ecommerce brands still need financial products built around cash flow, not generic startup banking.
Huge funding totals and payments volume can hide how fragile a fintech company actually is.

Trend

AI That Interrupts You

For years, AI voice demos have all had the same problem.

You ask something. The model waits politely. Then it answers in a slightly awkward voice while pretending interruptions don’t exist. If you correct yourself halfway through, the whole thing starts wobbling immediately.

Last week this started to change. Two major labs pointed at the same problem: voice AI still handles conversation like software, not like people. The bottleneck is the interaction layer.

OpenAI launched GPT-Realtime-2, GPT-Realtime-Translate, and GPT-Realtime-Whisper. A few days later, Thinking Machines introduced what it calls “interaction models,” a research preview built around continuous audio, video, and text interaction.

Why it Matters

The friction is moving from intelligence to interaction: The models are already smart enough for plenty of workflows. The annoying part now is the conversation itself. Delays, awkward pauses, broken interruptions, losing context after tiny corrections.
Voice is becoming a workflow layer: You can see it in OpenAI’s examples: scheduling, support, meetings, translation, real estate. The AI handles the software while you keep talking. Less clicking around dashboards.

OpenAI’s Realtime Push

OpenAI launched three new realtime audio models:

GPT-Realtime-2 for live voice conversations
GPT-Realtime-Translate for live translation
GPT-Realtime-Whisper for streaming transcription

GPT-Realtime-2 can:

reason while talking,
handle interruptions more cleanly,
call tools in parallel,
and keep long conversations alive with a 128K context window.

Developers can also tune reasoning effort depending on the task and add spoken status updates like “checking that now” while the model works in the background.

The translation and transcription models are also built for live interaction instead of post-processing. Translation happens while both people keep talking naturally. Whisper streams text continuously instead of waiting for complete speech segments.

Thinking Machines’ Interactive Models

Thinking Machines is aiming at a bigger shift.

Their argument is basically that current AI interfaces still behave like software pretending to be conversational. The model waits for you to finish talking. Then it responds. Then it waits again.

Real conversations are messier than that.

People interrupt each other constantly. They hesitate. They change direction halfway through sentences. Timing carries meaning. Silence carries meaning too.

Thinking Machines built its interaction model around those messy moments.

The system processes conversations in tiny 200 millisecond chunks called “micro-turns.” It keeps listening while speaking. It reacts while watching video. It can jump in naturally instead of waiting for a perfectly clean stopping point.

Some of the demos feel surprisingly human in small ways. The model notices hesitation. It reacts to visual changes on-screen. It interrupts when someone says something incorrect.

Most current voice systems fake this behavior with external turn-detection systems sitting outside the model. Thinking Machines is trying to build interactivity directly into the model itself.

You can already see the direction this pushes things toward. The useful AI products over the next few years probably won’t feel like chatbots with better answers. They’ll feel like systems that stay in sync with you while work is happening.

Help Me Improve Failory

That's all for today’s edition.

Cheers,

Nico

Too Early To AI

Nicolás Cerdeira — Thu, 07 May 2026 17:23:55 +0000

Hey - It’s Nico.

Welcome to another Failory edition. This issue takes 5 minutes to read.

If you only have one, here are the 3 most important things:

NeuroPixel.AI, a startup building AI tools for fashion, shut down — learn why below
The stuff nobody tells you about startup marketing
An overview of the Agent-Orchestration industry — learn about the key players below

A huge thanks to today’s sponsor, SimpleClosure. When startups like NeuroPixel hit the wall, close cleanly and move forward with structured, expert-guided support.

When the market shifts, we'll help you handle closure right ^AD

NeuroPixel.AI’s shutdown is a familiar startup story: tougher distribution, stronger competition from major tech players, lost revenue, and a market that moved faster than the company could.

Across the space, founders are running into similar pressure: shrinking differentiation, harder-to-sustain product-market fit, and scale becoming more difficult to reach.

When shutdown becomes the right next step, the work is not over.

SimpleClosure helps founders bring structure to the process, from state filings and investor communications to final distributions, compliance, and asset decisions. So you can close cleanly, responsibly, and with a clearer path through.

👉 Learn more

This Week In Startups

🔗 Resources

The best AI-native GTM plays you're not running

The "AI Job Apocalypse" is a complete fantasy

The stuff nobody tells you about startup marketing

Your team isn’t using AI. Here’s why that’s your fault.

📰 News

OpenAI launched GPT 5.5 Instant, the new default in ChatGPT

Image AI models now drive app growth, beating chatbot upgrades

Amazon opens up its global logistics network to all businesses

X announces a rebuilt ad platform powered by AI

💸 Fundraising

China’s Moonshot AI raises $2B at $20B valuation

Corgi raises $160M at $1.3B valuation to expand AI-native insurance platform

CodeWords has raised a $9m seed round to make you a workflow automation genius

UK stablecoin startup OpenTrade raises $17M

Fail(St)ory

Eaten by Giants

NeuroPixel.AI shut down in April after five years building AI tools for fashion ecommerce. The company raised a lot of money, worked with brands like Myntra and Decathlon, and got into generative AI before most people even knew what “GenAI startup” meant.

Then the market shifted underneath them.

What’s interesting is that they were not wrong about the trend, they just saw it early. They built real tech for a real problem. And then larger image models got good enough, fast enough, to crush the advantage they spent years building.

What Was NeuroPixel:

NeuroPixel started in Bengaluru in 2020 with one problem in mind: Fashion catalogs are operational nightmares.

Every product needs photos. Then more photos for different marketplaces. Different body types. Different campaigns. Different countries. Different seasons. A simple shirt turns into hundreds of image assets before it ever reaches a customer.

NeuroPixel’s pitch was simple enough that buyers immediately understood it. Instead of doing endless photoshoots, brands could shoot apparel on mannequins and let AI render the clothes onto synthetic models.

This sounds very normal now. Any AI model could do it. But in 2020 things were very different: generative AI was very unreliable, image models struggled with hands, consistency, fabric textures, proportions.

NeuroPixel was one of the companies trying to solve those ugly edge cases for a specific vertical instead of building generic AI tools for everyone.

The company claimed its system could reduce catalog production costs by 30% and cut processing time by 90%. Later, it pushed further into synthetic humans, virtual try-ons, AI-generated marketing images, and model customization.

In 2021, the startup raised an $825K seed round. A year later, Flipkart Ventures put in another roughly $299K. That second round valued the company at around $5.8M.

The company reportedly worked with brands including Myntra, Fabindia, Van Heusen, and Decathlon.

According to CEO Arvind Venugopal Nair, NeuroPixel spent nearly four years building deep IP assuming the competition would mostly come from other startups.

Then frontier image models exploded.

In his shutdown post, Nair said the company got “massively outgunned overnight sometime in late 2025.” He specifically mentioned Google’s NanoBanana Pro as the moment things changed.

NeuroPixel did not miss the GenAI wave. It caught it early. Then the wave got too big.

The Numbers:

🧠 Founded: 2020
💵 Funding: around $1.2M total
⚡ Claimed it could reduce catalog production timelines by 90%
📸 Claimed image production costs could drop by up to 70%
🚪 Shut down service operations in April 2026

Reasons for Failure:

General-purpose AI models caught up fast: NeuroPixel built around the assumption that fashion-specific image generation would stay technically difficult for a while. That was true for a brief period. Then larger models improved much faster than expected.
By late 2025, companies like Google were shipping image models powerful enough to erase most visible differences between specialized tools and frontier systems. The CEO openly admitted NeuroPixel’s product “didn’t hold up” against models like NanoBanana Pro.
Distribution became more important than product quality: Nair said the company assumed competitors would be other startups. Instead, the market consolidated around larger ecosystems with APIs, enterprise relationships, massive compute access, and existing user bases. Even if NeuroPixel’s output quality stayed competitive, customers naturally gravitated toward tools already integrated into broader workflows.
One unpaid client damaged the runway at the worst possible moment: The CEO said NeuroPixel’s largest customer collapsed and failed to pay for more than six months of work. For a startup already under pressure, that can end the company by itself.

Why It Matters:

AI moats disappear fast when your edge is just model quality.
In AI, distribution is starting to matter more than product quality.
Being early helps until Big Tech finally shows up.

Trend

Who Watches The Agents?

The AI market spent the last two years building agents.

Over the last year, a second layer started forming around them: the software that manages the agents.

At first, it looked scattered. A few orchestration tools here, some observability startups there, random “AI employee” demos on Product Hunt. Now the category is mature enough to map properly.

Why it Matters

The market is shifting from “can AI do the work?” to “how do you operate fleets of agents safely?” Different product. Different moat.
The winning products increasingly look like infrastructure software, not chatbots. Dashboards, logs, policies, orchestration layers, runtime environments.
The strategic fight is moving upward. The agent itself may commoditize quickly. The management layer probably won’t.

Agent Control Planes

This is the core category.

The interface here looks less like a chatbot and more like an admin console. Policies, retries, permissions, logs, audit trails, supervision dashboards, orchestration layers.

The pitch becomes: run agents like infrastructure, not experiments.

OpenHands Agent Control Plane: Launched May 6 as an “operational layer” for managing agent fleets. Includes orchestration, security, observability, scheduling, retries, spend tracking, and audit logs.
BAND: Raised a $17M seed to build communication infrastructure for the “Internet of Agents.” Focuses on agent discovery, delegation, context exchange, and collaboration across frameworks.
Wayfound: Supervision dashboard for business users. Tracks agent behavior, evaluates performance, analyzes interactions, and suggests improvements.
TectoAI: Governance layer for agentic systems. Monitors behavior, policy alignment, misuse, unauthorized actions, and compliance risks. Basically HR plus security for AI workers.

Company Runtimes

This is the founder-friendly wrapper around the same trend.

Instead of presenting orchestration as infrastructure, these startups present it as a company. The interface looks like an org chart, workspace, or operating system for a business.

The pitch is simpler: manage the company, let the agents handle the work.

Cofounder 2: Launched this week with the pitch “Run an entire company with agents.” Coordinates agents across engineering, sales, marketing, ops, and design.
Naïve: Calls itself “autonomous company infrastructure.” Focuses on planning, memory, isolation, and coordination between agent teams.
Buda: “Recruit agents to run your company as a synchronous team.” Includes marketplaces, team coordination, and live views of agents working inside browsers and terminals.
WUPHF: Shared memory, persistent knowledge bases, agents collaborating in public channels. Basically Slack for AI employees.

Underneath the branding, these companies are still building orchestration systems. Tasks, memory, roles, coordination, visibility. The org chart is just the UI layer.

Agent Observability and Debugging

Agents fail in weird ways. They loop. Misread pages. Call the wrong tools. Spend too much money. Hallucinate state. Trigger workflows nobody expected.

Companies need to see what happened after the failure.

Laminar: Raised a $3M seed to build observability for long-running agents. Captures LLM calls, browser actions, workflow traces, tool usage, replay systems, and anomaly detection for agent behavior.
AgentOps: Agent-first debugging platform. Tracks sessions, tool calls, costs, and behavior across popular agent frameworks.
LangSmith: Observability and evals for agent teams already building with LangChain or LangGraph. Useful for tracing failures and turning messy runs into test cases.

Big Tech

The category is now important enough that Big Tech is moving into the same layer.

Microsoft Agent 365: Microsoft is positioning it as a control plane for enterprise agents with governance, observability, and security controls.
OpenAI Agents SDK: OpenAI is building orchestration primitives directly into the developer stack: handoffs, agent routing, manager-style workflows.
Anthropic Managed Agents: Anthropic is building infrastructure around harnesses, sandboxes, and long-running managed agent systems.

That creates the obvious tension.

Can startups own agent management? Or does this collapse into the platforms that already control the models, cloud infrastructure, enterprise identity, and distribution?

Because the next fight is probably not over who builds the best agent. It’s over who owns the system around the agents.

Help Me Improve Failory

That's all for today’s edition.

Cheers,

Nico

One step away from Nasdaq

Nicolás Cerdeira — Thu, 30 Apr 2026 18:50:58 +0000

Hey - It’s Nico.

Welcome to another Failory edition. This issue takes 5 minutes to read.

If you only have one, here are the 3 most important things:

EO Charging, a startup building charging infrastructure for commercial EV fleets, shut down — learn more below
How to use Claude for GTM and pricing analysis
Claude Design and Claude for Creative Work are doing for design what Claude Code did for Software Engineering — learn how below

A huge thanks to today’s sponsor, Galactic Fed. Fix inefficient growth and scale smarter with a free expert marketing audit.

Overpaying for Growth? ^AD

Get a Free Marketing Audit from the team behind 600+ growth campaigns

If revenue increases only when budget does, you’re scaling cost, not efficiency. We'll help you fix the math before scaling it further.

As a Failory reader, you've been selected for one of our limited 20-minute executive review spots.

You’ll meet with senior growth specialists who will:

Assess your strategy
Identify smarter growth levers
Deliver a custom audit + action plan you can execute immediately

10 Spots Only: Confirm Yours

This Week In Startups

🔗 Resources

How to use Claude for GTM and pricing analysis

How to Compete with Legacy Incumbents

When the Dashboard Says ‘Go’ and Your Gut Says ‘No

Finally build that app this weekend (no-code hackathon) *

📰 News

X announces a rebuilt ad platform powered by AI

Jack Dorsey-backed Vine reboot Divine launches to the public

OpenAI releases GPT-5.5, bringing company one step closer to an AI ‘super app’

Lovable launches its vibe-coding app on iOS and Android

💸 Fundraising

AI customer service startup Netomi raises $110 million

Parallel Web raises $100M to build a parallel web for AI agents

German robotics startup Sereact raises $110M

Israeli AI startup Aidoc raises $150m for medical tools

* sponsored

Fail(St)ory

IPO to Zero

EO Charging entered administration in April 2026 after a failed sale process and another liquidity crunch.

Five years earlier, it was preparing to go public at a $675M valuation.

This is what happens when you try to build a global infrastructure company before the infrastructure market is ready to behave like one.

What Was EO Charging:

EO Charging started in the UK in 2014 with a clear idea: fleets were going electric, but the infrastructure required to support them didn’t exist yet in a usable form. Logistics operators didn’t just need chargers. They needed depots that worked reliably every morning when vehicles rolled out.

EO built itself around solving that operational gap.

Instead of selling standalone chargers, the company handled the full rollout inside fleet depots. That included site planning, grid upgrades, hardware installation, charging management software, and ongoing maintenance support.

If a company wanted to electrify hundreds or thousands of vehicles, EO could make the transition actually work on the ground.

By the early 2020s the company had deployed roughly 50,000 chargers across more than 35 countries. Its customers included Amazon, DHL, Tesco, Uber, and Go-Ahead.

At the center of the system sat EO Hub, the company’s charge point management platform. It coordinated charging schedules, integrated with fleet telematics tools, and helped operators manage power usage across entire depots.

In August 2021 EO announced a planned Nasdaq listing through a merger with First Reserve Sustainable Growth Corp. The deal implied a $675M enterprise value and more than $150M in expected cash. At the time, EO presented itself as a European fleet-charging leader preparing to scale globally.

The listing never happened. The SPAC deal was terminated in March 2022 as market conditions shifted and growth infrastructure companies suddenly became harder to finance.

EO responded the way many climate infrastructure startups did during that period: It kept expanding.

In 2023 the company raised around $80M and pushed further into international markets including the US, Australia, New Zealand, and Italy. The strategy aimed to turn EO into a global fleet electrification platform before the category consolidated around a few dominant providers.

That expansion came with operational weight. Each new geography required local grid coordination, installation partners, permitting workflows, and support coverage across active depots.

By late 2025 the company started narrowing its focus again. EO exited the US market, sold its domestic charger manufacturing business, and shifted attention toward its cloud charging platform. The plan was to simplify the business and move closer to a platform-led model built around software and fleet charging management.

A few months later the company began a sale process. No transaction followed.

The Numbers:

📅 Founded: 2014
🌍 Chargers deployed: ~50,000 across 35+ countries
🏢 Major customers: Amazon, DHL, Tesco, Uber, Go-Ahead
📈 SPAC valuation attempt (2021): $675M
💵 Equity round (2023): ~$80M
👥 Employees at administration: 93
❌ Immediate layoffs: 69

Reasons for Failure:

The business model looked scalable on slides, not in depots: EO sold a clean promise. One partner handles everything. Hardware, installation, grid coordination, software, uptime support. Fleet operators loved that. Startups usually don’t. Every depot became its own infrastructure project. Custom timelines. Custom engineering constraints. Custom integrations. That slows revenue recognition and eats working capital fast.
International expansion arrived before operational leverage: PwC administrators explicitly connected losses to overseas expansion across the US, Australia, New Zealand, and Italy. Expanding a hardware-plus-installation model internationally is not like opening a sales office. It means new permitting systems. New grid relationships. New subcontractors. New reliability expectations.
The SPAC collapse removed the fuel mid-flight: In 2021 EO expected more than $150M from a Nasdaq listing. That capital would have funded deployment capacity, hiring, and inventory. Infrastructure companies scale through balance sheets as much as through revenue. When the deal disappeared in 2022, the plan didn’t disappear with it. The company kept moving as if the market would reopen. It didn’t.

Why It Matters:

Enterprise infrastructure customers scale depot by depot, not quarter by quarter. Their rollout speed quietly sets your growth ceiling.
One-stop-shop positioning turns you into the system integrator holding everyone else’s risk. Grid delays, installs, and timelines all land on your balance sheet.

Trend

Claude Goes After Designers

Claude Code already changed how a lot of people build software.

Now Anthropic seems to be aiming that same playbook at design.

In the last two weeks, it launched Claude Design and Claude for Creative Work, connecting Claude to tools like Adobe Creative Cloud, Blender, Autodesk Fusion, Ableton, SketchUp, Splice, Affinity, and Resolume.

AI design is not new. But a Claude Code-style workflow for design would be.

Why it Matters

Claude is moving into tools AI usually could only talk about: Blender, Fusion, Ableton, Adobe, and SketchUp are not simple text boxes. If Claude can help operate those tools through natural language, complex software becomes easier to use without making the software itself dumber.
The workflow is starting to connect instead of resetting every time: Most AI creative tools are impressive until you need to keep working. Claude Design plus the connectors suggests a different direction: generate the rough asset, refine it, export it, move it into Canva, Adobe, Blender, or Claude Code, and keep going.

Claude Design

Claude Design launched on April 17 through Anthropic Labs as a research preview for Claude Pro, Max, Team, and Enterprise users.

The basic idea is simple. You describe what you need, and Claude creates a visual draft. That could be a prototype, a pitch deck, a one-pager, a slide, a dashboard, a landing page, a marketing asset, or some HTML-style output.

Then you keep working on it.

You can refine it through conversation, leave inline comments, make direct edits, use custom sliders, and export the result to formats people actually use: Canva, PDF, PPTX, standalone HTML, or Claude Code.

That matters because AI design tools often break down right after the first output. The first draft looks impressive, but then the real work starts somewhere else. Rebuilding it. Editing it. Making it on-brand. Turning it into something a team can actually use.

The Claude for Creative Work Connectors

Then came Claude for Creative Work.

To me, this was the more revealing announcement, even if it got less attention than Claude Design.

Anthropic announced a new set of connectors with tools across design, 3D, music, video, and live visuals. The list includes Adobe Creative Cloud, Blender, Autodesk Fusion, Ableton, Splice, SketchUp, Affinity by Canva, and Resolume.

This is what makes the Claude Design launch feel less isolated.

Claude Design gives Claude a visual workspace. The connectors give it a path into the professional tools where the work already happens.

Adobe brings Claude into the Creative Cloud world, including tools like Photoshop, Premiere, and Express. Affinity by Canva focuses on production tasks like batch image edits, layer renaming, exports, and custom feature generation inside the app.

Blender gives Claude access to its Python API and documentation. Autodesk Fusion lets people create and modify 3D models through conversation. SketchUp can turn a prompt into the starting point for a room, furniture setup, site concept, or 3D scene.

The common thread is simple: creative work is full of tool friction.

The tools are powerful, but they are deep, messy, and fragmented. A lot of the work is not some grand act of creativity. It is changing formats, finding the right command, applying the same edit twenty times, cleaning up files, moving assets around, or turning a rough idea into something editable.

The connectors allow Claude to handle all of these. So, it seems to me that design may be getting its Claude Code moment.

As Claude Design keeps improving, and more tools get connected to it, Claude has a real shot at doing for design what it already did for software engineering: make expert work faster, more accessible, and much easier to iterate on.

Help Me Improve Failory

That's all for today’s edition.

Cheers,

Nico

The Factory Killed The Unicorn

Nicolás Cerdeira — Thu, 23 Apr 2026 21:15:26 +0000

Hey - It’s Nico.

Welcome to another Failory edition. This issue takes 5 minutes to read.

If you only have one, here are the 3 most important things:

Ascend Elements, a unicorn leading America’s battery recycling-to-materials supply chain push, has shut down — learn more below
Great companies are built in hackathons
OpenAI launched GPT Images 2.0 and its SOTA — learn more below

A huge thanks to today’s sponsor, Deel. Download their global expansion guide and scale without compliance risk.

International expansion without the compliance liability ^AD

Global growth for startups shouldn't be a gamble.

Between fragmented labor laws and shifting tax nexus regulations, the financial stakes are high.

Our Global Expansion Guide breaks down the essentials of localized contracts and entity setup to ensure your workforce scales with total audit-readiness.

Protect your bottom line from misclassification penalties and regulatory friction.

👉 Download the Guide

This Week In Startups

🔗 Resources

The slow decay of growth (and how to avoid it)

Great companies are built in hackathons

How to hire people who are better than you

Ship a production-level AI app in days *

📰 News

OpenAI releases GPT-5.5, bringing company one step closer to an AI ‘super app’

Google updates Workspace to make AI your new office intern

The App Store is booming again, and AI may be why

Zoom teams up with World to verify humans in meetings

💸 Fundraising

Loop raises $95M to build supply chain AI that predicts disruptions

Blue Energy raises $380M to build grid-scale nuclear reactors in shipyards

Cloneable Raises $4.6M To ‘Clone’ Expert Worker Knowledge With Agentic AI

AI research lab NeoCognition lands $40M seed to build agents that learn like humans

* sponsored

Fail(St)ory

America’s Battery Recycling Bet

Ascend Elements filed for Chapter 11 two weeks ago.

This was supposed to be one of the companies rebuilding America’s battery supply chain. Instead, it became a case study in what happens when a hard-tech startup scales faster than its execution, financing, and construction reality can support.

What Was Ascend Elements:

Ascend Elements started in 2015 as Battery Resourcers trying to rethink lithium-ion battery recycling.

The idea was simple: Take old batteries. Pull the valuable metals out. Turn them back into material that can go straight into new batteries.

This was not new. Plenty of companies recycle batteries. But most stop halfway.

When batteries get shredded, what comes out is a messy powder mix of lithium, nickel, cobalt, graphite, and other materials. The industry calls this “black mass.” It’s basically battery confetti.

Normally recyclers sell that powder to someone else for further processing. That’s where most of the value leaves the building.

Ascend wanted to keep going. Instead of selling battery confetti, it wanted to turn that material directly into one of the key ingredients inside new batteries. Something called precursor cathode active material, or pCAM.

Most pCAM today gets made in China. Ascend’s pitch was simple: Recycle batteries in the US and turn them directly into new battery ingredients here instead.

That idea landed perfectly during the early EV boom.

In 2022 the company renamed itself Ascend Elements. It wasn’t positioning itself as a recycler anymore. It wanted to become a domestic battery-material supplier at industrial scale.

Then came the big bet. Ascend announced plans to build a massive factory called Apex 1 in Hopkinsville, Kentucky. About one million square feet. The goal was to produce enough battery material each year to support roughly 750,000 EVs.

The timing helped. Between 2022 and 2024 the US government was pushing hard to rebuild domestic battery supply chains. Ascend fit perfectly into that plan.

Investors put in nearly $900 million. The Department of Energy awarded roughly $480 million more across expansion programs. The 2023 Series D alone brought in $542 million from major global investors.

From the outside, this still looked like momentum.

Inside the company, the Kentucky flagship was already slipping. Construction disputes surfaced. A major federal grant tied to the project was canceled after a strategy shift mid-build. At the same time the board replaced the CEO with a new one. That usually means the plan isn’t working anymore.

Finally, in April 2026 Ascend filed Chapter 11. CEO Linh Austin described the financial situation as insurmountable. The flagship factory wasn’t finished. The money was gone.

The Numbers:

💰 Nearly $900M+ in investor capital raised
🏗️ $480M in DOE grants awarded across programs
📉 $138M construction dispute tied to Apex 1 contractor claims
🔥 14 safety incidents reported at Georgia site before shutdown pause
🚗 Target output: materials for 750,000 EVs per year
📅 Founded 2015, filed Chapter 11 April 9, 2026

Reasons for Failure:

The company scaled into industrial manufacturing before locking a stable operational base: Ascend tried to do recycling, lithium recovery, pCAM manufacturing, and megaproject construction at once. That is a portfolio strategy inside a single startup. Industrial scale-up punishes that kind of parallel risk stacking fast. Even the Kentucky flagship was reportedly only about 60 percent complete when the crisis peaked, which meant capital was trapped inside a non-productive asset.
Apex 1 became the center of the problem: Everything depended on the Kentucky plant working on schedule. Once contractor disputes reached roughly $138 million and suppliers started filing liens, the project stopped looking stable to investors and partners. Industrial startups rarely recover after their main facility enters litigation territory before production starts.
The subsidy structure shifted mid-build: Ascend’s expansion depended heavily on federal support layered on top of venture capital. The cancellation of a $164 million CAM grant and later reporting around a canceled $316 million Kentucky-linked grant weakened the capital stack right when the plant needed stability. Subsidies accelerate builds but they also create fragility when assumptions change.
Market timing moved against domestic battery-materials economics: EV adoption didn’t stop, but timelines stretched. Automakers delayed plans. At the same time Chinese battery-material suppliers kept pricing pressure high. Industrial startups need a stable ramp window to survive their first large plant. Ascend never got one.

Why It Matters:

Contracts don’t matter if production isn’t running. Supply deals don’t generate cash without a working plant.
Government funding doesn’t guarantee survival. Ascend still ran out of time mid-build.

Trend

GPT Images 2.0

OpenAI released ChatGPT Images 2.0 last week.

And almost overnight, Nano Banana stopped being the reference point. This is the new model people are testing first.

Why it Matters

Text-heavy visuals just became much more usable: If a model can handle dense text, labels, hierarchy, icons, and layout without falling apart, you stop using it just for hero images and start using it for actual communication.
It is better at generating a set, not just a single image: With thinking mode on, it can create up to eight images while keeping style, characters, and objects consistent.
The workflow is shifting from prompting for images to prompting for deliverables

The Model

OpenAI released Images 2.0 inside ChatGPT last week, powered by a new model called gpt-image-2. It’s also already live in the API and Codex, which usually means they expect people to actually build with it, not just play with it.

The main thing they’re pushing is something they call image generation with reasoning. That sounds like marketing language, but there is something real underneath it.

In thinking mode, the model doesn’t just translate your sentence into an image. It expands the prompt, plans the layout more deliberately, and can even use live web context before generating. The result feels less like “prompt → picture” and more like “brief → output.”

You can see this in the examples they chose to show.

They’re not showing fantasy portraits. They’re showing menus, brochures, posters, manga pages, academic explainers, product grids, and editorial layouts. Basically the exact kinds of images older models struggled with the moment you added structure.

Text rendering is the part everyone noticed first.

Multilingual text improved a lot too. Japanese, Korean, Chinese, Hindi, Bengali. These scripts usually broke earlier models much faster than English did. This one holds together noticeably better.

Another change people keep testing is continuity across images.

With thinking mode enabled, the model can generate up to eight related images that stay visually consistent with each other. Same character, same objects, same style. That’s why manga keeps showing up everywhere around this launch. Manga is basically the hardest stress test for layout, typography, and sequential storytelling at the same time.

There are also some practical details that matter more than they sound.

The API supports thousands of output sizes, with aspect ratios up to 3:1 and sizes up to 3840 px on the longest edge. That makes it more usable for banners, portrait posters, mobile creatives, and the random awkward sizes real teams actually need.

It also supports editing with masks and high-fidelity image inputs, so this is not only about generating from scratch. It fits into “take this existing asset and adapt it” workflows too.

Overall, what actually feels new is not that it makes pretty images, but that it can reliably generate assets with structure, text, and continuity well enough to replace a bunch of lightweight design work that previously still needed a human pass.

Help Me Improve Failory

That's all for today’s edition.

Cheers,

Nico

Refusing the Trend

Nicolás Cerdeira — Thu, 16 Apr 2026 19:27:32 +0000

Hey - It’s Nico.

Welcome to another Failory edition. This issue takes 5 minutes to read.

If you only have one, here are the 3 most important things:

Spoony, a social network for neurodivergent people, shut down — learn why below
Google adds AI Skills to Chrome
Meta just launched Muse Spark, its new family of AI models — learn why it matters below

A huge thanks to today’s sponsor, Lucid Engine. Measure and boost your product’s visibility in AI search so it gets recommended when buyers are ready to decide.

Capture high-intent traffic from AI search ^AD

ChatGPT is already recommending products in your category.

Your ideal customer might be asking for a tool like yours right now.

If you’re not mentioned in those answers, you don’t even make the shortlist.

AI search is becoming the final comparison layer before a buyer clicks anything. You need to know where you stand.

Lucid Engine helps maker teams:

Track how they rank across ChatGPT, Perplexity, and Google AI Overviews
See which competitors are mentioned instead
Identify the exact prompts driving recommendations
Get a clear, prioritized roadmap to improve visibility

No hype. No guesswork. Just structured data and actionable steps.

Run your AI visibility audit →

This Week In Startups

🔗 Resources

State of AI Agents in 2026

The slow decay of growth (and how to avoid it)

Surviving AI price wars without destroying your business

How to use AI without losing your mind

📰 News

Google adds AI Skills to Chrome

Sam Altman’s house was targeted in a molotov attack

Anthropic essentially bans OpenClaw from Claude

ChatGPT finally offers $100/month Pro plan

💸 Fundraising

Glydways raises $170m to support global expansion plans

Biometric analysis startup GameRun raises $4M

AI fintech startup Round raises $6 million

Slate Auto raises $650M to fund its affordable EV truck

Fail(St)ory

A Human Social Network

Last week, Spoony shut down.

It was built for people mainstream social apps usually overlook, and it seemed to be gaining real traction. Then it ran into a brutal startup problem: the clearest paths to survival risked turning it into a different company.

What Was Spoony:

Spoony was a social app built for neurodivergent, chronically ill, and disabled people. It was meant to be a place where people who are often isolated could actually meet, talk, and feel less alone.

The idea came from the founder’s own experience. Nicholas Carlton has said chronic illness pushed him out of a job he loved and made his world much smaller. He ended up leaning on other people with similar experiences for support. That became the starting point for the company.

You can usually tell when a founder is solving a problem they actually understand. The product tends to have more signal and less fluff. Spoony felt like that. It was not trying to be everything to everyone. It picked a specific group and built around needs mainstream social apps mostly ignore.

Spoony was incorporated in December 2023, and by August 2024 it had already built a waitlist of 30,000 people. Early signs looked strong. The company said beta users were opening the app 3-4 times a day, and reported a 25% reduction in loneliness.

The public launch came in October 2024. By then, Spoony had positioned itself somewhere between a social network, a peer-support system, and a health-adjacent community product. That in-between space was the opportunity. It was also the trap waiting later.

Investors bought the early story. Spoony raised A$1 million in pre-seed funding, with backing from Antler and others. The bull case was clear: strong founder-product fit, a big underserved community, early engagement, and a product that felt more meaningful than another disposable social feed.

By the time the shutdown was announced, Spoony had reached more than 65,000 users. Carlton also claimed its retention and engagement were “best-in-class for consumer social.” The full KPI set was never made public, so that is still founder framing. Even so, the broad picture is hard to miss: people wanted the product.

The real problem was that liking a product and financing a company are two different things. Spoony seems to have built real community value without finding a business model that fit either the mission or the market. That gap was manageable while capital was loose. It became fatal once it wasn’t.

The Numbers:

📅 December 2023 - incorporated
🚀 February 6, 2024 - public waitlist launched
👥 30,000 - waitlist signups by August 2024
🔁 3-4 times a day - reported beta usage
💛 25% - company-reported reduction in loneliness
📲 October 2024 - public launch
💸 A$1 million - pre-seed funding raised
📈 65,000+ - users by April 2026
⚠️ April 8, 2026 - shutdown announced

Reasons for Failure:

The next round fell through: Spoony planned to raise again in October 2025, and that round fell apart. Once that happened, every other weakness got exposed at once. The company was still pre-revenue and still had no validated business model, which meant there was no cushion. A startup can survive a messy round when revenue is working. It usually cannot survive one when revenue is still a theory.
User love did not translate into a fundable monetization story: Spoony had growth, engagement, and what looks like genuine community value. But Carlton said the company had followed the usual consumer playbook of growing first and monetizing later, and by then investors were no longer buying that story. He also said traditional ad monetization only starts to make sense at around a million users. Spoony had 65,000+, which was good by normal standards, but not even close by social-platform standards.
The strongest revenue experiments pushed Spoony toward becoming a digital health company: Spoony did not ignore monetization. It tested things. Carlton said the team explored a therapy marketplace, a digital clinic, and an AI therapist, and that these experiments showed real signal. So the issue was not a total lack of options. The issue was that the most promising ones pulled the company away from “human connection platform” and toward a much heavier digital health business with more operational complexity, more regulation, and frankly a different soul.
It would not bend all the way into AI or healthcare just to stay alive: Spoony explored AI and other monetization routes, saw some promise, and still chose not to go all in because leadership thought it broke the point of the product. Carlton’s line was the clearest summary: “We are not an AI company. Our business is human connection, and that’s not a hot category right now.”

Why It Matters:

Mission can narrow your options. The same focus that made Spoony trustworthy also ruled out some of the clearest revenue paths.
Monetization can change the company. Sometimes the path to revenue is really a path to becoming something else.

Trend

Muse Spark

Meta is back in the race. Or at least, back in the part of the race people actually care about.

After Llama 4, Meta looked off balance. Still huge, still everywhere, still spending like crazy. But it did not feel like a company pushing the frontier. Muse Spark changes that. The main reason is not that Meta suddenly has the best model. The shift is how it wants to compete now.

Why it Matters

Meta changed its AI strategy. Llama made Meta look like the open-model champion. Muse Spark points to a different playbook: closed release, selective API access, and deep product integration.
Distribution is becoming the moat. Meta can drop AI into apps people already use all day, then layer in social context, creators, shopping signals, messaging, and eventually glasses.
Founders should pay attention to platform power. The big labs are bundling models with surfaces, workflows, and demand. If your startup lives on top of a platform-owned behavior, the pressure just went up.

Muse Spark

Muse Spark is Meta’s first major model launch since Llama 4 and the first public release from Meta Superintelligence Labs, the new group built after Meta’s earlier AI momentum faded.

Meta is framing it as the first model in a new Muse family, not the next Llama. It’s essentially a reset: new stack, new architecture, new direction.

The strategic shift is obvious. Muse Spark is not open weights. It is live inside Meta’s own products and only available through private preview API access for selected partners.

That fits the broader pitch. Meta is talking about “personal superintelligence,” which really means an assistant shaped by its own ecosystem: your messages, social graph, creators, community context, images, and eventually the world around you through glasses.

You can see that in the feature set. Meta is pushing multimodal perception, shopping help, local and social search, subagents for tasks like trip planning, health answers tuned with physician-curated data, and light visual coding for websites, dashboards, mini-games, and simulations.

The product direction is pretty clear. Meta wants AI inside discovery, commerce, messaging, and everyday consumer behavior. That is where it already owns attention.

The benchmarks help support the comeback story. Early testing says Muse Spark lands in the top tier overall, with especially strong vision performance and a huge jump over Llama 4.

But the weak spots are still there. Coding looks good, not best-in-class. Agentic performance does not lead. Harder reasoning still trails the top models.

What this Means

Muse Spark puts Meta back in the frontier conversation with a much sharper playbook.

It is not the SOTA model. Still, that misses the real shift. Meta now has a clearer idea of where this model lives and why it matters: inside products where people already spend time, make choices, and discover things.

That makes this launch bigger than a benchmark update. It points to a race shaped less by raw model quality alone, and more by who controls both the model and the surface around it.

Help Me Improve Failory

That's all for today’s edition.

Cheers,

Nico

Paid to use AI

Nicolás Cerdeira — Thu, 09 Apr 2026 20:54:17 +0000

Hey - It’s Nico.

Welcome to another Failory edition. This issue takes 5 minutes to read.

If you only have one, here are the 3 most important things:

Yupp, a startup that let users compare the responses of over 800 different AI models, has shut down — learn why below
The golden rules of agent-first product engineering
Anthropic released Claude Mythos Preview, a model so good at finding software vulnerabilities that it cant be released to the public — learn why this matters below

A huge thanks to today’s sponsor, Playbookz. Start scaling your LinkedIn personal brand on autopilot.

⚡️ Most founders are 1 system away from 100k+ reach on LinkedIn ^AD

Playbookz builds and runs your LinkedIn presence for you:

Content → your expertise turned into posts people actually engage with
Distribution → paid amplification that drives real reach
Pipeline → outbound that turns attention into buyer conversations

You stay focused on the business. They handle the execution. A system designed to grow your reach week after week.

20,000+ reach in 28 days guaranteed.

Plans start at $1.5k/month.

Get a free demo →

This Week In Startups

🔗 Resources

OpenAI’s policy proposals for the AI age

You need better storytelling

AI Adoption by the Numbers

The golden rules of agent-first product engineering

📰 News

Meta debuts the Muse Spark model in a ‘ground-up overhaul’ of its AI

Microsoft takes on AI rivals with three new foundational models

Google quietly launched an AI dictation app that works offline

Apple will release a foldable phone later this year

💸 Fundraising

OpenAI secures a 122 billion USD from Softbank

Hermeus raises $350M to build unmanned hypersonic fighters

Chinese robotics startup Spirit AI raises $145m

Trent AI raises $13M to build multi-agent security

Fail(St)ory

When judging AI became a job

Yupp raised $33M, signed up 1.3M users, partnered with major AI labs, and still shut down less than a year after launch.

They were building a layer to compare AI models side by side and sell the resulting preference data back to the labs. It made perfect sense in 2024. By early 2026, the market had already moved somewhere else.

What Was Yupp:

Yupp was trying to solve a simple problem that showed up the moment multiple strong AI models existed: Which one should you trust?

Instead of giving you one answer, Yupp showed several answers side by side. ChatGPT. Claude. Gemini. DeepSeek. Llama. Hundreds more. You asked once and compared everything at once.

The pitch was obvious the first time you used it. Don’t pick a model. Ask all of them. They called this your “council of AIs.”

Every time you picked one answer over another, you generated preference data. Quietly. Automatically. At scale. Yupp believed that kind of signal would become extremely valuable as labs raced to improve their models.

So they built a marketplace around it.

Users got access to premium models through a credit system. You spent credits to run prompts. You earned credits back by rating answers. The better your feedback, the more credits you earned.

Later in 2025 they added Cash Out. Some users could convert credits into actual money through Stripe, PayPal, Coinbase, or stablecoin rails.

Yupp wasn’t selling access to AI, they were buying judgment from users and reselling it to model builders.

The labs were the second side of the marketplace. When users compared answers, they generated structured signals about which models performed better on real prompts from real people. Yupp packaged that into evaluation data labs could use for post-training and benchmarking.

This is why they launched the VIBE Score leaderboard.

Instead of measuring models with synthetic benchmarks, they measured them using live user behavior. Which answer people picked. Which response they trusted. Which one actually helped.

If you could sit between users and models while the world figured out which AI systems worked best, you’d control a very valuable layer of the stack.

And for a while, it looked like they might.

They signed up more than 1.3M users. They were collecting millions of preference signals every month. Several AI labs were already paying for access.

But the window closed before that evolution could turn into a new product category.

The Numbers:

💰 Raised: $33M seed
📍 Founded: June 2024
🚀 Public launch: June 2025
👥 Users: 1.3M+
🤖 Models supported: 500+ at launch, later 800+
🛑 Shutdown announced: March 31, 2026
📉 Lifetime after launch: < 1 year

Reasons for Failure:

They built the right product for the wrong layer of the stack: Yupp optimized for chatbot comparison at exactly the moment AI moved toward agents. The founders said it directly: workflows shifted toward systems with tools, memory, and execution capability. Once users expect AI to complete tasks instead of produce answers, side-by-side response comparison becomes less central.
Their data marketplace thesis didn’t match where labs were spending: The company believed large-scale consumer preference data would become a core input into model training. That logic made sense in 2024. But much of the post-training market shifted toward expert-labeled datasets and specialized evaluators instead of broad consumer feedback. TechCrunch reported that labs increasingly relied on PhDs and domain specialists in reinforcement learning loops. Yupp’s signals were useful. They just weren’t essential enough to anchor a business.
Comparing models is interesting once, not forever. Users like seeing multiple answers the first time. It feels powerful. Over time, most people want one system that works. Not eight tabs of competing outputs. Yupp started adding features like “Help Me Choose” because comparison fatigue is real. The product itself hinted at the problem it was trying to solve.

Why It Matters:

Usage doesn’t equal durability. 1.3M users and millions of preferences per month still wasn’t enough. Early traction can validate curiosity without validating a market.
AI categories are collapsing faster than normal startup timelines. Yupp launched in June 2025 and shut down by March 2026. Entire product layers now appear and disappear inside a single funding cycle.

Trend

Claude Mythos

Most AI launches are easy to translate. Better chatbot. Better coding copilot. Faster workflow. This one is different.

Anthropic just told the market something a lot of people in AI and security already suspected: once a model gets really good at coding, it starts getting really good at breaking software too. Not because someone built a “cyber model” on purpose. Because the same skills that help fix code also help find ways through it.

That is the story behind Project Glasswing and Claude Mythos Preview.

Anthropic is basically saying: we built something too dangerous to release broadly, so we are handing it to a small circle of defenders first and hoping they can patch enough systems before everyone else catches up.

Why it Matters

Coding models are now cyber models. Anthropic is saying the offensive capability was not the product goal. It showed up anyway once the model got good enough at coding and reasoning.
The choke point is no longer finding bugs. AI is making discovery faster. The real problem now is triage, patching, and shipping fixes before someone weaponizes them.
The advantage goes to whoever gets access first. Glasswing is basically a defender-first rollout. That only works if this capability spreads soon, which is exactly what Anthropic seems to believe.

The Launch

On April 7, Anthropic announced Project Glasswing. The core of it is Claude Mythos Preview, an unreleased frontier model that Anthropic says is its strongest yet for coding and agentic work. It is not being released publicly. Anthropic says the model can produce dangerous cyber outputs and its safeguards are not good enough yet for broad access.

Anthropic claims Mythos can identify and exploit zero-days in every major operating system and major browser. It says the model found subtle bugs that had survived for 10, 20, even 30 years. In one browser case, it reportedly chained four vulnerabilities into a working exploit. In Linux, it found multiple chains that ended with full root access.

The benchmark numbers are even more blunt. On a Firefox JavaScript-engine exploit benchmark, Anthropic says Opus 4.6 managed working exploits only twice across several hundred attempts. Mythos got to working exploits 181 times, plus many more cases where it reached partial control.

In broader testing across roughly 1,000 open-source repos and around 7,000 entry points, Anthropic says older models mostly found low-severity crashes. Mythos produced hundreds of more serious crashes and 10 full control-flow hijacks on fully patched targets.

That created an obvious problem. If the model is this good, why release it widely at all? Anthropic’s answer is Project Glasswing. Keep Mythos gated, give it to a selected set of defenders, and use the window before broader proliferation to find and patch vulnerabilities in critical systems.

The launch partners are a who’s who of big tech, cloud, cyber, and critical infrastructure: AWS, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA, Palo Alto Networks, and Anthropic. The company also says it extended access to 40-plus additional organizations that build or maintain critical software infrastructure.

The Reception

The reaction has been a mix of fascination, relief, and unease.

On one side, big security and cloud players are leaning in:

AWS says it has already tested Mythos on critical codebases.
Microsoft says it saw substantial gains on real-world detection engineering tasks and is reworking internal remediation processes for AI speed.
Palo Alto and CrowdStrike are effectively saying the same quiet thing: attackers are going to get this too, soon enough.

On the other side, the controversy is obvious. Anthropic is asking the market to accept a strange premise: the responsible way to handle dangerous cyber capability is to build it first, keep it gated, and trust a private company to decide who gets the head start. That is not regulation. That is self-appointed governance with good branding.

There is also a concentration problem. The first beneficiaries are giant incumbents in cloud, finance, and cybersecurity. Anthropic can point to open-source funding and extra access for maintainers, which helps. Still, the center of gravity is clear.

And the timing is awkward. This launch landed after leaks around the model’s earlier codename and amid broader scrutiny of Anthropic’s own operational security and governance. “Trust us to tightly contain the dangerous system” always sounds weaker when the company has already shown a few cracks.

Help Me Improve Failory

That's all for today’s edition.

Cheers,

Nico

AI Ate Discovery

Nicolás Cerdeira — Thu, 02 Apr 2026 20:12:22 +0000

Hey - It’s Nico.

Welcome to another Failory edition. This issue takes 5 minutes to read.

If you only have one, here are the 3 most important things:

abillion, a startup building a discovery platform for vegan products, just shut down — learn more below.
OpenClaw: The complete Guide
There’s a new web being built. A web for agents — learn what the opportunities are below

A huge thanks to today’s sponsor, Playbookz. Start scaling your LinkedIn on autopilot with content, amplification, and built-in lead generation.

✈️ Scale your LinkedIn personal brand on autopilot ^AD

LinkedIn is the highest-ROI channel for B2B founders. Playbookz builds it for you:

Content — they turn your expertise into targeted reach and followers.
Amplification — they put their own ad budget behind your best posts
Warm leads — they automate outbound to start actual conversations

You don't touch any of it. 20,000+ reach in 28 days guaranteed.

Plans start at $1.5k/mo.

→ Get a free demo here

This Week In Startups

🔗 Resources

OpenClaw: The complete Guide

The internet ruined customer service. AI could save it.

2026 State of GTM

📰 News

Google is launching Search Live globally

Stanford study outlines dangers of asking AI chatbots for personal advice

Mistral releases a new open source model for speech generation

Microsoft takes on AI rivals with three new foundational models

💸 Fundraising

Mistral AI raises $830M in debt to set up a data center near Paris.

Robot Startup Galaxea AI Raises $291 Million

AI chip startup Rebellions raises $400 million at $2.3B valuation

Autonomous ship startup Saronic raises $1.75 billion

Fail(St)ory

The Vegan Yelp

abillion shut down last week, after eight years spent trying to make sustainable consumption feel easy, social, and scalable.

It started as a vegan food discovery app, then grew into something much more ambitious: reviews, social posts, a marketplace, brand tools, donations, even community ownership.

In the end, none of that saved it. The company said it hit its best metrics ever and still couldn’t make it through the fundraising market.

What Was abillion:

abillion launched in 2018 as a tool for people trying to find decent plant-based food without doing detective work every time they left the house. The first use case was simple enough: open the app, find vegan dishes, read reviews, avoid wasting money on sad cauliflower.

That wedge came from founder Vikas Garg’s own frustration. He had been moving toward veganism for years and kept running into the same problem many niche consumers know well: demand exists, but discovery is broken.

The product kept widening. What began as restaurant and dish reviews expanded into cruelty-free beauty, fashion, packaged goods, and other sustainable products. Over time, abillion looked less like Yelp for vegans and more like a mashup of Tripadvisor, Instagram, and a mission-driven marketplace.

Its engine was user-generated content. People posted photos, ratings, recommendations, and reviews, and those reviews became the inventory that made the app useful. To incentivize this, users earned credits for posting reviews, and those credits could be routed to charities and sanctuaries.

Then came the harder part: turning that engagement into a business. abillion added a SaaS layer for brands, letting them claim listings, add purchase links, and pay for extra visibility, marketing tools, and analytics.

In 2022, it pushed further and launched a peer-to-peer marketplace, starting in Singapore. Users could buy and sell sustainable goods inside the app, including secondhand items and homemade products, while abillion took a cut on each transaction.

They also tried something most startups never touch: community equity. Users could put credits toward ownership in the company, not just donations. The idea was clear enough: if the community creates the value, maybe it should own some of it too.

It was a smart idea but, by that point, abillion was trying to make reviews, commerce, brand software, donations, and shared ownership all work inside the same product.

And just as it got more complex, the whole discovery layer started getting weaker. When people can ask AI what to eat, what to buy, and which brand to trust, standalone curation apps start to look a lot less defensible.

The Numbers:

💸 Funding raised: roughly $17M
📅 Launched: 2018
📍 Base: Singapore
📲 Downloads: 1M+
🏪 Businesses listed: 1.3M
❤️ Donations generated: $2.8M+
🧾 SaaS customers: 4,700+
🤝 Community investment round: $500K+ from 290 members

Reasons for Failure:

AI started eating the discovery layer: abillion was built around curation. That used to be a solid wedge. Now a lot of that discovery is getting pulled into AI, where people can ask one question and get a clean answer without opening a separate app. abillion did not say AI killed the company, but it was operating in exactly the part of the stack that AI is starting to flatten.
The product kept expanding faster than the core model matured: The company moved from vegan food discovery to product reviews, then social features, then SaaS, then marketplace, then community equity. Each addition made the story bigger and the execution harder. There is a difference between layering revenue streams and piling on unresolved businesses. abillion may have ended up managing five half-proven models instead of one sharp one.
Mission created attention, but monetization lived in a narrower lane: The mission was great, the actual monetization points were more ordinary: SaaS subscriptions, transaction fees, and commerce tools for brands. People love mission-driven products until you ask who pays enough, often enough, to support venture economics. abillion built a strong moral loop for users, but the business loop looked much less convincing.

Why It Matters:

AI is absorbing discovery. If your product lives in the “help users find the right thing” layer, AI is coming for your wedge faster than most founders want to admit.
More product layers do not fix weak monetization. Adding commerce, SaaS, community features, and new loops can make the story bigger, but it does not solve the core question of who pays, why, and how much.

Trend

The Pumbling of the Agentic Web

There’s a new web being built. A web for agents.

This isn’t exactly new. It’s been taking shape for at least a year. But lately it’s picking up speed. A new type of startup is showing up: companies building not for humans, but for agents.

So let’s look at that layer.

Who’s building the plumbing behind the agentic web, and what exactly are they building.

Why it Matters

The opportunity moved down-stack. The leverage is shifting from building agents to building what agents depend on.
The category is still fragmented, which is good news for builders. Nobody owns the full stack. One startup handles browser sessions, another structured extraction, another inboxes, another payment credentials. That usually means the market is still early enough for sharp products to wedge in.
Software is starting to get a second interface. The first interface was UI for humans. The next one is structured access for agents. Platforms that expose clean authentication, scoped permissions, and reliable actions will be easier to integrate into agent workflows.

The Startups

The easiest way to see the pattern is to look at what agents currently can’t do reliably yet.

1) Browsing the web

Agents struggle with finding pages, loading them, and interacting with real interfaces.

Some startups are breaking browsing into primitives to help agents:

Browserbase launched APIs that let agents search the web, retrieve page contents, and run browser sessions programmatically.
Firecrawl lets agents click buttons, fill forms, and navigate interactive websites instead of just scraping static HTML.

2) Accessing reliable web data

Agents don’t just need information. They need structured inputs.

Nimble raised $47M to build infrastructure that searches the web in real time, validates results, and converts them into queryable tables for agents

3) Operating inside communication channels

Most workflows still happen inside inboxes and messaging threads.

Startups are giving agents endpoints there:

AgentMail provides inbox infrastructure for agents
Linq embeds assistants directly inside SMS, iMessage, and WhatsApp
A1Base provides agents multi-channel communication APIs across email, Slack, SMS, and messaging platforms

4) Paying and proving identity

Agents can’t transact without credentials and authorization.

Recent examples:

Stripe introduced payment tokens that let agents initiate purchases without accessing card details
Browserbase + Fingerprint introduced cryptographic identification for authorized agents.
World AgentKit links agents to verified human approval

The Opportunity

Cloudflare CEO Matthew Prince said AI bot traffic could exceed human traffic online by 2027.

There is clearly a market here, and we are still very early. The useful question for builders is simple: what can humans do easily on the web that agents still struggle with?

We covered the obvious ones already: browsing, authentication, messaging, payments, structured data.

But the opportunities are truly endless.

A good example is Sapiom.

It’s building financial infrastructure that lets agents buy their own tools and services. In practice, that means giving agents controlled access to budgets so they can decide which APIs, compute, or software they need to complete a task.

Something humans already do easily on the web, agents still can’t do reliably yet.

That’s where the opportunity is.

Every time you notice one of those gaps, you’re probably looking at a future startup category.

Help Me Improve Failory

That's all for today’s edition.

Cheers,

Nico

Acquihired And Killed

Nicolás Cerdeira — Thu, 26 Mar 2026 17:13:09 +0000

Hey - It’s Nico.

Welcome to another Failory edition. This issue takes 5 minutes to read.

If you only have one, here are the 3 most important things:

Clockwise, a startup building a smart calendar, got acquired by Salesforce and immediately shut down — learn why below
How the top AI-native startups launch and grow.
Google relaunched Google Stitch and started a new era of vibe design — learn more below

A huge thanks to today’s sponsor, Stripe. Get the free playbook on scaling infrastructure like AI’s fastest-growing companies.

Inside the growth engine of AI’s breakout companies ^AD

Companies like ElevenLabs, Runway, and Leonardo AI maintain rapid growth while others stall on complexity.

As businesses grow, payments break, billing gets messy, and engineers stop building to fight fires. Stripe’s playbook reveals how these leaders built infrastructure that scales.

👉 Get the Playbook

This Week In Startups

🔗 Resources

How the top AI-native startups launch and grow.

What we wish we knew about building AI agents.

Technical leaders make these 4 common storytelling mistakes

Why AI will consolidate markets and create the biggest businesses of all time

📰 News

Sora is Shutting Down.

Claude can now remotely control your computer.

Online bot traffic will exceed human traffic by 2027.

Snapchat’s new ‘AI Clips’ Lens format turns photos into five-second videos

💸 Fundraising

AI productivity startup Highlight AI raises $40M

Thai AI startup Amity announces US$100 million

Agents for finance startup Zalos raises $3.6M

Qualified Health locks in $125M to scale enterprise AI at health systems

Fail(St)ory

A Smarter Calendar

Clockwise spent nearly a decade trying to fix the modern workday. It raised $76M, worked with companies like Uber and Notion, and processed hundreds of millions of meetings.

Then a couple days ago, they announced the whole team was joining Salesforce and the product would shut down days later.

What Was Clockwise:

Clockwise was an AI scheduling startup that tried to fix one of the most annoying parts of modern work: a calendar that looks technically “free” but is actually completely unusable.

You know the type of day:

meeting at 10:00
another at 11:30
one more at 2:00
another random one at 4:00

So yeah, on paper you have “free time.” In reality, your whole day is shredded into useless little chunks. You are never fully in meetings, but you are also never fully able to focus on anything meaningful.

That was the problem Clockwise was built to solve.

Its core idea was simple: instead of just showing your schedule, your calendar should actively improve it.

So Clockwise connected to your calendar and automatically rearranged certain meetings to make your day less fragmented and more workable. The goal was to create bigger blocks of uninterrupted time, what the company called Focus Time, by clustering meetings together instead of letting them wreck your entire day one by one.

The timing helped. During the 2020 remote-work surge, meeting fragmentation became a visible problem. Clockwise launched Clockwise for Teams and reported signups up 87 percent early that year.

The product kept evolving toward AI-native workflows. In 2023 it launched a conversational assistant for scheduling. In 2024 it introduced Prism, positioned as a more intelligent calendar experience. In 2025 it launched an MCP server so AI agents could reason about human time directly.

But last week Clockwise abruptly announced the team would join Salesforce and the product would shut down on March 27. Just like that, a product used by 40,000 organizations disappeared in eight days.

The Numbers:

💰 $76M total funding raised
🏢 40,000+ organizations using the product
📅 23M meetings automatically rescheduled
🧠 8M+ hours of Focus Time created
👥 ~50 to 70 employees at shutdown

Reasons for Failure:

The category got absorbed by bigger platforms: Clockwise tried to create a new layer of infrastructure around time coordination. That works until the platform owners decide scheduling intelligence belongs inside their own stack. Calendars live inside Google and Microsoft environments. Once AI assistants become native features there, standalone scheduling optimization becomes harder to defend. The CEO’s comment that “AI changes what markets can sustain stand-alone products” is basically a public admission of this shift.
The product kept expanding because the wedge was too narrow: Clockwise started as a background optimizer. Then it added team analytics, enterprise controls, booking links, conversational AI, Prism, and agent infrastructure. That sequence looks like a company widening its footprint to become a platform instead of a utility.
Category creation produced adoption but not independence:
Clockwise was not selling a simple feature. It was trying to change how teams think about time. That explains why it introduced concepts like flexible meetings and Focus Time as system-level primitives. But creating a category often benefits the ecosystem more than the creator. Once users expect smarter scheduling everywhere, the advantage moves toward whoever controls the default calendar surface.

Why It Matters:

Building a category does not guarantee you own it. Clockwise helped define Focus Time and flexible meetings as first-class concepts. The winners may still be the platforms that ship those ideas natively.
Infrastructure features are fragile outside platform control. If your product sits on top of someone else’s calendar, inbox, or workflow surface, your leverage depends on their roadmap.

Trend

Design With Your Voice

Google wants you to talk to your design canvas. Not metaphorically. Literally talk to it, ask it for critique, tell it to try three new directions, and watch it turn that back into UI.

Why it Matters

The first version gets cheaper. Judgment gets more expensive. Getting from idea to screen is becoming easier. The hard part moves to taste, product sense, and knowing when the output is good enough versus generic AI slop.
Vibe Design is getting real. This is no longer just prompt-to-mockup gimmicks. The design tool itself is starting to behave more like a collaborator you brief, react to, and steer in real time.
The gap between idea and test keeps shrinking. When one tool can generate screens, suggest flows, make prototypes, and carry context into the next step, you get to validation faster.

Google Stitch

Google Stitch is the cleanest example of this shift right now.

Last week, Google relaunched Stitch and framed it as an AI-native software design canvas.

Stitch is not just “make me a screen.” It has an AI-native infinite canvas, a design agent, an agent manager, instant prototyping, voice features, and export tools that connect with the rest of the workflow.

The product is clearly built around intent, not manual layout. Stitch can take text, images, or code as context, so you do not need to start from a blank page. You can start with scraps. A screenshot you like. A rough idea. Existing code. A vague direction.

Google also makes the workflow pretty explicit: you start by explaining the business goal, the user feeling, or examples that match the direction you want. You are not really drawing the first version anymore. You are describing it.

The voice part is what makes this feel fresh. Stitch lets you speak directly to the canvas. You can ask for critiques. Ask for variations. Ask it to give you three menu options. Ask to see a screen in different color palettes. Google even says it can interview you and turn that conversation into a landing page.

This is real vibe design. You are talking through what you want and shaping the output as it responds.

A few other interesting things:

Stitch’s design agent is meant to reason across the whole project as it evolves, not just generate one isolated output and forget everything.
The agent manager lets you explore multiple ideas in parallel, which is useful because good product work is messy and branching anyway.
Stitch can also turn static designs into clickable prototypes instantly and generate next screens based on user interactions.

Other Signals

Stitch is the centerpiece, but it is not showing up alone.

Yesterday, Figma announced that agents can now design on the Figma canvas in beta. Through its MCP server, AI agents can write directly into files while using the team’s existing components, variables, and tokens.
On March 12, Figma had expanded its Code to Figma workflow, tightening the loop between rendered UI, editable design frames, and code.
Last week, Gamma launched Gamma Imagine, a prompt-based system for creating brand-specific visual assets like charts, social graphics, and collateral.

The Trend

Vibe designing is getting real.

This means the rough draft gets cheaper, faster, and accessible to more people. The valuable work moves up the stack toward system thinking, taste, quality control, and product judgment.

Help Me Improve Failory

That's all for today’s edition.

Cheers,

Nico

No More Withdrawals

Nicolás Cerdeira — Thu, 19 Mar 2026 19:29:01 +0000

Hey - It’s Nico.

Welcome to another Failory edition. This issue takes 5 minutes to read.

If you only have one, here are the 3 most important things:

BlockFills, an institutional crypto firm, filed for bankruptcy — learn more below
An in-depth guide to coding agents for data analysis.
We are entering the era of proactive AI — learn why this matters below

A huge thanks to today’s sponsor, Deel. Get their free startup guide to global hiring.

Hiring globally doesn’t have to be complicated ^AD

Hiring globally can unlock growth, but local laws, payroll, and compliance often slow startups down.

Our free guide breaks down what an Employer of Record (EOR) is, how startups use EORs to hire internationally without opening entities, and when it makes sense to use one so you can scale with confidence.

👉 Get the Guide

This Week In Startups

🔗 Resources

How PostHog built a marketing engine that actually drives growth.

What’s working to improve free-to-paid conversion.

An in-depth guide to coding agents for data analysis.

How to design a seed pitch deck investors actually read.

📰 News

Gemini gets into Google Maps with Inmersive Navigation.

Mistral bets on ‘build-your-own AI’ as it takes on OpenAI in the enterprise.

Elon Musk says xAI must be ‘rebuilt’.

The dictionary sues OpenAI.

💸 Fundraising

A startup transforming carbon emissions into fashion textiles raised $7.5M.

IT automation startup Standard Template Labs raises $49M.

French startup U-Space raises €24 million for smallsat manufacting.

Verifiable AI startup Axiom raises $200M to prove AI-generated code is safe to use.

Fail(St)ory

Crypto Keeps Failing

Another week, another crypto company blowing itself up.

This time, it was BlockFills, an institutional trading firm that went from big volume and global expansion to frozen withdrawals and Chapter 11 in a matter of weeks.

What Was BlockFills:

BlockFills started in Chicago in 2018 and went after a very specific slice of crypto: institutions that wanted access to the market without building everything themselves. Hedge funds, brokers, asset managers, prop shops, miners.

For a while, it looked like it was working. The company said it had been profitable since 2019. It kept rolling out more products, pushed deeper into software, and expanded across geographies.

They went from 900-plus institutional clients in 2022 to more than 2,000 by 2025. Trading volume went from $10.6B in 2023 to $57.2B in 2024, then $61.1B in 2025. On the surface, that looks like a machine that found product-market fit.

But the issue was where the business was going. By 2023, BlockFills was not just facilitating trades anymore. It got deeper into lending, credit lines, and mining-related exposure. By 2024, it was borrowing more capital so it could extend more credit to clients.

That changes the business completely. Now you are not just helping clients trade. You are taking on credit risk, funding risk, and liquidity risk.

That seems to be what caught up with them. Later reporting tied the company to loan losses and roughly $30M in mining losses. There were also reports of accounting inaccuracies. That is a very different picture from the one the company had been selling.

And still, right to the end, the public story stayed upbeat. On December 30, 2025, BlockFills published a year-in-review post celebrating record volume, global expansion, and an “exciting and prosperous 2026.”

Then, six weeks later, withdrawals were frozen.

The Numbers:

💰 Funding: $44M raised
📈 Volume (2025): $61.1B traded
🌍 Customers: 2,000+ institutional clients across 95+ countries
⚖️ Balance sheet: $50M–$100M assets vs. $100M–$500M liabilities
❗ Shortfall: ~$77M hole

Reasons for Failure:

It stopped being an infrastructure company: BlockFills sold itself as an agency model. Execute trades, take a spread, stay neutral. Then it layered on lending, credit lines, and mining exposure. That turns a low-risk business into a fragile one fast. When markets dropped, those positions didn’t unwind cleanly. They blew holes in the balance sheet.
Liquidity mismatch killed it: Short-term liabilities to customers, long-term or illiquid assets on the other side. When clients rushed to withdraw, the money wasn’t there. The “temporary” freeze is always the tell. Once withdrawals stop, the ending is already written.
Weak controls and accounting issues: This wasn’t just bad market timing. Reports point to accounting inaccuracies and internal awareness of financial problems months before the collapse.

Why It Matters:

If you need to add lending or balance sheet risk to hit growth targets, your core business is weaker than you think.
The moment you rely on customer funds to support the business, even indirectly, you’re running on borrowed time.
Volume is the easiest metric to inflate and the least useful when things start breaking.

Trend

Proactive AI

Since OpenClaw, I’ve been noticing more founders pitching something very specific: AI that does not wait. AI that stays on, keeps context, watches what is happening, and jumps in on its own. Not just a copilot sitting there until you type the magic prompt. More like a system that hangs around in the background and keeps moving.

Why it Matters

The pitch is changing. For a while, AI products sold speed. Write faster, code faster, research faster. Now more of them are selling initiative.
It changes what users expect. Once people get used to software that remembers context and takes action on its own, a normal chatbot starts to feel weirdly passive.
It creates a whole new trust problem. A proactive AI can be useful in a way chat never was. It can also mess things up way faster.

OpenClaw RECAP

I talked about OpenClaw a couple of weeks ago but, in case you missed it, their pitch was very simple: an AI that actually does things. It can control the browser, run shell commands, manage files, connect to a bunch of tools, and basically operate like a 24/7 agent on your machine.

It acts on its own, and can make decisions without human interaction. It is proactive instead of reactive. Of course, this led to a bunch of funny (and even dangerous) moments like this one:

The Signals

At the time of OpenClaw’s viral success, I thought it was mostly just a toy. Fun for sure, but too erratic, too dangerous, too messy to become something substantial.

But since then, I’ve been seeing more and more startups pitch this same kind of proactive agent.

Polsia is probably the clearest example. Its pitch is basically: AI that runs your company while you sleep. Planning, coding, marketing, ops. The whole thing.

You can watch it work live here. I’ve probably spent more time than I should watching it these last few days. It is weirdly compelling. Seeing it create tasks, solve them, hit problems, try to fix them, even tweet on its own, without ever really stopping.

Now, to be honest, I do not think it can fully automate any complex business. At least not yet. But that is not really the point. The point is where the pitch has moved. A couple of months ago, this kind of always-on, huge-context agent barely existed in public.

Another example is Enia Code, which takes the same idea and drops it into coding. Its whole pitch is that most coding tools wait for you to ask. Enia does not. It watches your code in real time and flags bugs, performance issues, weird architecture decisions, and refactoring opportunities before you prompt it. Again, proactive instead of reactive.

Then you get the startups building around the breakout product itself. That is maybe the strongest signal of all.

KiloClaw is basically hosted OpenClaw. Same idea, just managed for you. No weird setup.
CoChat does something similar for teams. It takes the OpenClaw-style agent model and adds approvals, logs, shared workflows, scheduled tasks, and the kind of controls you need once more than one person is involved.
Happycapy goes one step further and starts talking about an “agent-native computer.”

The Trend

So here’s my read.

Since OpenClaw, I think more founders have realized that a very strong AI pitch is: I keep working nonstop.

OpenClaw made that idea feel exciting enough that people wanted more of it. Then the follow-ons started showing up.

This is still early, and a lot of it is definitely hype. Some of these startups are selling a vision way ahead of the product. And the trust issues here are very real. If your product is proactive, it can screw things up proactively too.

Still, I can’t unsee the pattern now. I think we are entering the era of proactive AI.

If you are building in AI, I would keep this in mind. People are getting used to the idea that software should have initiative. More users now expect the AI to keep context, spot what matters, and do more than just sit there looking smart.

The bar is quietly moving from “answer me” to “handle it.”

Help Me Improve Failory

That's all for today’s edition.

Cheers,

Nico

Another Hyperloop Crashed

Nicolás Cerdeira — Thu, 12 Mar 2026 17:41:41 +0000

Hey - It’s Nico.

Welcome to another Failory edition. This issue takes 5 minutes to read.

If you only have one, here are the 3 most important things:

Hardt Hyperloop, a dutch startup trying to build hyperloop technology, has shut down — learn more below
The world needs an AI lab — for Data
This week, the biggest AIs got into your office tools — learn why this matters below

A huge thanks to today’s sponsor, Playbookz. Turn LinkedIn into a consistent client-acquisition channel without creating content or managing DMs.

Turn LinkedIn into a steady source of buyer conversations ^AD

Playbookz helps founders use LinkedIn to generate qualified meetings, without writing posts, managing DMs, or running ads themselves.

They handle: founder-led content, targeted outreach, paid amplification, and a setter who turns conversations into booked meetings.

That means:

more qualified meetings
more inbound from people already familiar with you
more trust before the first call
less pressure to post constantly or chase reach

The result is simple: steady demand gen without adding more to your plate.

Book a no-obligation call to see if it fits. Failory readers get $1,000 off their first month.

This Week In Startups

🔗 Resources

The world needs an AI lab — for Data

Six breakthrough business models reshaping global growth

AI Safety Has 12 Months Left

Catch up on today's AI stack (OpenClaw, Claude Code, agents) in four hours. First 10 get 50% off with code: FAILORY *

📰 News

ChatGPT can now create interactive visuals to help you understand math and science concepts.

Meta acquired Moltbook, the AI agent social network.

Anthropic launches code review tool to check flood of AI-generated code.

X is testing a new ad format that connects posts with products.

💸 Fundraising

Cryptio raises $45 million to help big firms keep track of digital assets.

French AI startup AMI raises $1B to develop 'universal intelligent systems'

Chinese brain interface startup Gestala raises $21M.

Kai raises $125M to build an agent-driven AI security platform.

* sponsored

Fail(St)ory

Sci-Fi Trains

A hyperloop startup went bankrupt last week, which feels a bit like hearing that someone’s flying car company had a tough quarter.

Still, Hardt was not just selling sci-fi with a seed round. It spent years building test infrastructure and trying to prove that this very dumb-sounding idea might, somehow, work.

What Was Hardt Hyperloop:

Hardt was a Dutch startup building hyperloop tech. The basic idea, in case you’ve successfully avoided this category for a decade, is simple: put a vehicle inside a low-pressure tube, cut drag, use magnetic levitation and electric propulsion, and move people or cargo very fast.

Sort of a train, except much more expensive, much harder to build, and wrapped in sci-fi branding.

The bigger promise was classic frontier-tech stuff. Less flying. Less road traffic. Faster trips between cities. Cleaner transport. Europe, but compressed.

Hardt spent years trying to turn that idea into something more solid than renderings and conference slides. It wanted to show hyperloop could work as an actual network, not just one straight tube for a flashy demo.

So it focused hard on lane switching. Hardt argued that if vehicles could not move between routes, the whole thing stayed a toy. A transport system has to behave like one.

And yes, the company did build things. Its main asset was the European Hyperloop Center in Veendam, a 420-meter test facility. There it tested levitation, propulsion, vacuum conditions, and later lane switching.

In 2024, Hardt finished building the test center. Later that year it announced its first successful test and then a full lane-switch test. In 2025, it said it had run more than 750 test missions and hit 85 km/h.

So this was not fake. That matters. But 420 meters is still 420 meters. At that scale, you are proving that pieces of the concept work. You are nowhere near proving that the business works.

That is what made Hardt interesting. It had enough real engineering to avoid the usual vaporware jokes. It still looked painfully far from anything commercial. The company was making technical progress while the actual finish line barely moved.

Then, on March 4 2026 the money finally ran out and Hardt Hyperloop was declared bankrupt.

The Numbers:

🚄 Founded: 2016 in the Netherlands
💰 Capital raised: $36M
🇪🇺 Major public backing: €15M from the European Commission in 2021
🧪 Main test site: 420-meter European Hyperloop Center
⚙️ Testing claim:750+ missions in one year by 2025

Reasons for Failure:

The capital needs were brutal: This is the main cause. Hardt run out of money. That is what happens when every step takes years, costs millions, and still leaves you far from revenue. Deeptech is hard. Infrastructure deeptech is a different level of pain.
The company stayed stuck in prototype land: A test track can prove the tech is moving. It does not prove the business is moving. Hardt could show progress without getting meaningfully closer to commercial deployment.
Hyperloop still makes ugly economic sense: The concept sounds great when you strip away reality. Then you add infrastructure, regulation, safety, land, politics, and adoption. Suddenly you are not building a startup. You are trying to brute-force a new transport layer into existence.

Why It Matters:

Milestones can lie. A startup can look alive because the product keeps moving. What matters is whether each milestone makes funding, sales, or deployment easier.
You need a wedge, not just a vision. Big future markets sound great. You still need something smaller that customers will pay for now.

Trend

The AI Office Takeover

We get a new AI model every week now. Sometimes every six minutes. This is great news for the benchmark addicts. Less relevant for the millions of people doing actual work inside spreadsheets, docs, inboxes, and project tools.

This week, the thing that mattered was much simpler: AI got pushed deeper into work software. It was already there, obviously. But now it is getting its hands on the actual job.

Dark times for spreadsheet masochists. Good news for everyone else.

Why it Matters

The best position in AI now is owning the workflow. The product that already has the files, messages, meetings, tasks, and permissions has a real advantage. Once AI is wired into that context, a standalone assistant starts to look pretty small. Looking at you, OpenAI
Spreadsheets, calendars, and ticket queues matter more than another writing demo.

Google Workspace

This week Google updated Workspace, pushing Gemini deeper into the ecosystem.

Docs got a stronger drafting engine. Gemini can build a document from a prompt using your files, emails, and the web, then keep rewriting inside the doc. It can tighten sections, match your writing style, and even copy the format of a reference doc so the output looks like something your team would actually use.

Sheets is the big one. Gemini can now create, organize, and edit full spreadsheets from a prompt, pulling context from your files and inbox. Google also launched “Fill with Gemini”, which can populate tables with summaries, categories, generated text, and live info from Search.

That matters because a ridiculous amount of office work still lives in spreadsheets. Budgets, planning, reporting, forecasts, dashboards, all the usual corporate suffering. If Google makes building, filling, and cleaning sheets easier, that is a real piece of office work getting automated.

Slides got upgraded too. Gemini can now create slides inside a deck, match the presentation’s theme, edit through prompts, and turn rough inputs into editable diagrams and layouts.

Microsoft Copilot

A few weeks ago I discussed Claude Cowork, one of the clearest examples of AI moving from chat into actual delegated work.

Now Microsoft has brought that idea straight into the Office ecosystem with Copilot Cowork. You give it a task, it turns that into a plan, pulls context from your emails, meetings, messages, files, and data, and keeps working in the background while you approve changes along the way.

Microsoft’s examples are very Office-core: clean up your calendar, reschedule meetings, block focus time, prep a customer meeting, build the deck, write the follow-up, pull company research, even package the outputs into docs, email, and Excel files.

And yes, Anthropic is directly in the mix. Microsoft says it integrated the technology behind Claude Cowork into Microsoft 365 Copilot. So the product here is basically Claude Cowork’s multi-step delegation model, dropped into Outlook, Teams, Excel, and the rest of Microsoft 365, with Microsoft wrapping the whole thing in enterprise security and governance.

Notion and Others

Notion and Atlassian are also part of this trend.:

Notion’s Custom Agents focus on recurring operational work: internal Q&A, task triage, standups, status reports, inbox cleanup.
Atlassian’s agents in Jira bring agents directly into tracked workflows so teams can assign work to them, mention them in comments, and run them inside the system where work is already coordinated.

The Trend

AI is getting embedded into the software people already use to work.

Which means normal people are about to start using agents without thinking of them as agents. Your aunt will use Gemini inside Google Docs. Your mother will use Claude Cowork inside Office and never once wonder who Anthropic is. This stuff is no longer just for developers messing around.

That has big consequences for the AI war. Once Gemini is built into your docs, spreadsheets, slides, and Drive, and Microsoft brings Claude-style delegation into Outlook, Teams, and Excel, the standalone chatbots like GPT start to feel a bit naked. Owning the enterprise software may end up mattering more than owning the best model.

Help Me Improve Failory

That's all for today’s edition.

Cheers,

Nico

One Patient, One Drug

Nicolás Cerdeira — Thu, 05 Mar 2026 15:36:12 +0000

Hey - It’s Nico.

Welcome to another Failory edition. This issue takes 5 minutes to read.

If you only have one, here are the 3 most important things:

EveryONE Medicines, a startup trying to cure rare diseases with personalized drugs, has shut down — learn more below
The AI native growth team
Guide Labs released an interpretable model — learn what interpretability is and why it matters below

A huge thanks to today’s sponsor, Oceans. Hire U.S.-caliber marketing talent at up to 80% less cost with their help.

The Secret to Consistent Growth: Hire Global with Oceans Talent ^AD

What do Soapbox, Magic Spoon, and Cheers Health founders have in common? They hired experienced, vetted global talent to own their marketing function.

They hired with Oceans Talent.

Oceans Talent screens 1,500+ applicants each month and matches you with experienced, professional marketers who own their channel end to end: influencer management, paid search, paid social, lifecycle email, SEO, content, brand management or creative ops.

You get:

Skills plus personality vetting
An integration plan to support the first 100 days with defined expectations, success metrics, and recommendations
Ongoing support and upskilling so execution stays on track

86% of first matches are the right fit. That's why 400+ founders trust Oceans Talent marketing specialists.

Meet & Hire Oceans Talent Marketing Specialists 👉

This Week In Startups

🔗 Resources

The AI native growth team

The Software Shakeout: What is durable and what is not in the Age of AI?

Strategic choices: When both options are good

AI image and video as a core business skill, live *

📰 News

Google settles with Epic Games, drops its Play Store commissions to 20%.

Apple had a big week of product launches.

Claude Code rolls out a voice mode capability.

ChatGPT uninstalls surged by 295% after DoD deal.

💸 Fundraising

Startup making AI chips more power-efficient raises $500 Million.

UK self-driving startup Oxa raises $103 Million.

Data management startup Validio raises $30M.

Neura Robotics raised €1 Billion.

* sponsored

Fail(St)ory

Truly Personalized Care

This week, EveryONE Medicines shut down.

They were trying to build a company around custom drugs for people with ultra-rare genetic diseases. And by custom, I mean genuinely custom: in some cases, one drug for one patient.

The rough part is that they shut down just as regulators were finally starting to move in their direction.

What Was EveryONE:

Most biotech startups are built around one drug, or one group of drugs, for a defined set of patients. EveryONE was trying to do something different.

Their idea was to treat ultra-rare cases almost like a manufacturing problem. A child shows up with a mutation that is either unique or close to it. You figure out whether that mutation can be targeted. If it can, you design a drug specifically for that kid. Then you test it, make it, get it into the clinic, and somehow do all of that fast enough to matter.

That’s what made the company unusual. It wasn’t really selling a single medicine. It was trying to build the machinery for making lots of tiny custom medicines, each one built for a different patient.

The company focused on ultra-rare pediatric neurological diseases, which is about as easy to build around as it sounds. Tiny patient populations. Very high urgency. Hard manufacturing. No settled reimbursement model. Regulators still figuring it out in real time.

Still, EveryONE did make significant progress. In the UK they were involved in the Rare Therapies Launch Pad effort, and by late 2025 they had gotten MHRA approval for a master protocol that could cover multiple diseases and medicines under one framework. Then in January 2026, they reached the point of treating a patient in the UK with a custom drug they developed.

The Numbers:

🚀 Founded: 2020
👷 Team size: fewer than 10 full-time employees as of mid-2024
✅ Key milestone: MHRA-backed master protocol approved in October 2025
💥 Shutdown: March 3, 2026

Reasons for Failure:

They were trying to scale something that still behaves like a one-off case. That’s the core tension. EveryONE wanted to turn custom therapies into a repeatable category, but each case still came with a lot of bespoke work. Different mutation, different patient, different clock ticking in the background. You can standardize parts of the system, sure, but this still isn’t SaaS with prettier margins.

Reimbursement was probably the real landmine: Making a custom therapy is hard. Getting somebody to pay for it, repeatedly, at scale, is arguably worse. EveryONE clearly knew this, which is why reimbursement showed up as such a central part of its pitch. But the whole field still seems stuck in a world where payment is negotiated case by case, often with a lot of institutional improvisation.

The rules were improving, but not fast enough: In late February, the FDA introduced draft guidance for a new “plausible mechanism” pathway for individualized therapies, basically an attempt to create a more workable approval route for treatments aimed at very specific genetic conditions. That mattered because the old model treated each custom therapy too much like a standard drug program, which is painfully slow for these cases. But draft guidance is still draft guidance, and for a startup burning cash, “the rules may improve soon” is not the same as having a business now.

Why It Matters:

If your unit of delivery is a custom job, you’re running a services company until you prove otherwise.
Category-building doesn’t mean value capture. EveryONE helped push the world toward individualized therapies… and then died right before that world could pay them back.

Trend

Interpretable AI

Since GPT first rolled out, AI has been facing the same problem: the models keep getting more useful but we still do not really know what is going on inside them.

That trade was easy to accept when AI was writing emails and summarizing PDFs. It gets a lot less cute once these models start writing code, handling support, reviewing contracts, or making decisions inside actual products.

Last week, Guide Labs released an 8B model that is a big step forward in making AI more interpretable. So today, I wanted to use this as an excuse to talk about AI interpretability: what it is, why it matters, and where AI seems to be heading.

Why it Matters

This changes what “trustworthy AI” could mean in practice. If a model is going to sit inside a real workflow, you eventually need more than “it usually works.” You need to know what pushed it toward an answer.
Interpretability is turning into a control layer. Anthropic’s persona vectors work shows that behaviors like hallucination and sycophancy may be tied to internal patterns you can actually monitor and influence.
This is starting to look like a category, not a side quest. Goodfire just raised $150 million at a $1.25 billion valuation to build around interpretability and model control. Once capital starts piling in, it usually means the market sees a real bottleneck.

The Problem

The core issue here is that today’s language models often arrive at answers through internal processes that are still mostly opaque. You get the result, but not the mechanism.

For a while, chain-of-thought made it feel like models were becoming easier to understand. Ask them how they got the answer and they give you something neat and convincing. The catch is that this explanation is not always a faithful readout of what actually happened internally. Anthropic has shown that models can produce clean-sounding reasoning while the real computation underneath is more tangled.

So the field has been stuck with an awkward reality. The models are already useful enough to deploy. They are not understandable enough to fully trust. That gap is where all this recent work sits.

The Progress

The freshest signal is Guide Labs. On February 23, it released Steerling-8B, which it describes as the first inherently interpretable 8B language model. The company says the model was trained on 1.35 trillion tokens and built around explicit concept modules, so outputs can be traced back to prompt pieces, concepts, and training origins. .

Anthropic has been pushing this from another direction, and in a more sustained way. In March 2025, it published work on tracing the thoughts of a large language model, showing internal reasoning paths and concept-level traces inside Claude. In May, it open-sourced circuit tracing tools. In August, it published persona vectors, tying internal activations to behaviors like hallucination and sycophancy and showing they could be steered.

OpenAI is also exploring something called “sparse circuits.” This is not a new GPT model or product feature, but a research effort to understand how language models work internally.

Today’s models are extremely dense: information flows through thousands of interconnected pathways at once, which makes it hard to trace which parts actually produced an answer.

Sparse circuits explore whether models could be organized more cleanly, with only a small set of pathways activating for a given task. If that works, researchers could more easily map which internal circuits correspond to specific skills or concepts.

The Trend

This seems pretty clear now: a lot of serious players are working on interpretability.

That usually means the problem is real. And it usually means the next wave of models will not just compete on benchmarks, speed, or price. They will also compete on how well you can inspect them, steer them, and trust them inside real workflows.

That matters even more if you build in regulated or high-stakes spaces. Finance, healthcare, law, government. In those markets, black-box AI has always had a ceiling. Interpretable AI breaks this ceiling.

Help Me Improve Failory

That's all for today’s edition.

Cheers,

Nico

4 Years in Limbo

Nicolás Cerdeira — Thu, 26 Feb 2026 19:25:35 +0000

Hey - It’s Nico.

Welcome to another Failory edition. This issue takes 5 minutes to read.

If you only have one, here are the 3 most important things:

Kintsugi, a startup building an AI depression detector, has shut down — learn more below
The engineeringification of everything.
Perplexity Computer is the new OpenClaw - learn more below

A huge thanks to today’s sponsor, Oceans. Hire U.S.-caliber finance talent at up to 80% less cost with their help.

Hire U.S. caliber Finance talent for up to 80% less ^AD

Skip partial finance support. You need someone who closes the books, tracks cash weekly, and makes sure there are no surprises.

Your baseline requirements:

On-time month-end close
Weekly cash and runway tracking
A forecast you can run the business on

Oceans Talent places finance pros who think ahead, reconcile accounts, manage cashflow, and stay ahead of reporting deadlines.

400+ founders have hired Accountant, Financial Controllers, FP&A Analysts, and more with Oceans Talent.

Find your finance pro 👉

This Week In Startups

🔗 Resources

Becoming an AI-native operator.

The engineeringification of everything.

How to describe your product (quickly).

How AI agents actually work (and how to build one), live *

📰 News

Anthropic adds plug-ins for finance, engineering and design.

Guide Labs debuts a new kind of interpretable LLM.

Google’s new Gemini Pro model has record benchmark scores.

Meta’s metaverse leaves virtual reality.

💸 Fundraising

Nvidia challenger AI chip startup MatX raised $500M.

Wayve raises $1.2B to scale embodied AI for autonomous driving.

Music startup Mogul raises $5M.

AI marketing startup Gushwork raises $9M.

* sponsored

Fail(St)ory

Listening between the lines

Kintsugi shut down after nearly four years chasing FDA clearance for an AI tool that listened to your voice and flagged depression and anxiety.

It raised $28M, published real studies, ran pilots, and then ran out of oxygen anyway.

What Was Kintsugi:

Kintsugi started in 2019 with a pretty crisp thesis: your voice carried measurable signals of mental state. Not the words, the acoustics: pitch, cadence, pauses, energy, that subtle drag you heard when someone wasn’t okay.

The idea: build an AI that “listened between the lines.” You gave a short voice sample, and the system tried to detect whether you were depressed or anxious. It didn’t focus on what you said. It focused on how you said it.

Of course, Kintsugi knew that an AI tool that flagged depression in a clinical setting was going to be treated like a medical device. That meant the FDA. And the FDA meant time, paperwork, studies, and a whole new set of costs that don’t show up in normal startup decks.

So the team did the thing investors always say they want. They generated evidence. They published peer-reviewed results showing their model could screen for depression from short voice samples with sensitivity and specificity in the low 70s.

Funding follows. They raise an $8M seed in 2021, then a $20M Series A in 2022, The pitch is easy to underwrite: huge prevalence, underserved detection, AI cost curve, clear ROI if you can route people to care earlier.

Then the timeline broke. FDA clearance became the critical path, and the critical path became the company. Founder Grace Chang said it took nearly four years waiting for clearance, with more than $16M spent on the process and another $1M to $4M needed to finish.

Meanwhile the cost structure didn’t care about any regulatory Gantt chart. Keeping AI engineers on $400K to $500K compensation packages turned into a burn-rate hostage situation.

They tried the obvious escape hatches. They explored fundraising, and they considered acquisition offers, and they still ended up choosing neither.

So they shut down and published the work. Kintsugi released its models and research as open-source, basically donating years of effort to the ecosystem.

The Numbers:

🚀 Founded: 2019
💰 Total raised: $28M (Seed $8M in 2021, Series A $20M in 2022)
🧾 FDA spend: $16M+ already, plus $1M to $4M more to finish
📚 Evidence: large-scale peer-reviewed study showing ~71% sensitivity and ~74% specificity for depression screening from short voice samples

Reasons for Failure:

Venture math didn’t fit regulatory latency: Kintsugi was a venture-backed company operating on a venture clock, but it picked a product path with a regulator’s clock. That mismatch was the real failure mode. When your go-to-market is gated by FDA clearance, “runway” becomes a policy decision you don’t control.
They tried to Pivot but it was too late: Kintsugi eventually launched a second product called Kintsugi Signal, a deepfake voice detector meant to tell if a voice was human or AI-generated. It was a clear attempt to find revenue outside the FDA bottleneck. Signal may have worked, but it lived in a different world: new buyers (security and fraud teams), new procurement paths, and zero free distribution from their healthcare relationships.
They anchored too hard on “clinical-grade” and left money on the table. Kintsugi chose the hardest version of the product early: a screening tool that looked and smelled like a medical device. That decision forced them into FDA timelines before they had a durable cash engine. They could have packaged earlier versions as stricter “wellness” tooling, employer mental health screening, coaching triage, or even analytics for existing behavioral health providers, then used that cash to fund the clinical path.

Why It Matters:

FDA-gated products need decade-capital, not Series A optimism.
Don’t wait for clearance to figure out cash flow, build a real revenue line early.

Trend

Perplexity Computer

Yesterday, Perplexity launched their own OpenClaw competitor.

It’s called Perplexity Computer and it’s designed to unify 19 different AI models into one coordinated system. According to Perplexity, the orchestrator can run for hours or even months.

Why it Matters

Orchestration becomes the moat. Perplexity is not claiming the best model. They are routing tasks across 19 models and stitching outputs into finished workflows.
Bundling across providers creates lock-in. If one system can fluidly use Claude-class coding, GPT-class reasoning, and Gemini-style search inside a single workflow, you stop caring which vendor powers which step.
Sandboxed agents are the enterprise path. Long-running automation scares companies when it touches local machines. A contained cloud environment is easier to approve, audit, and standardize.

What is it

Perplexity calls it a “general-purpose digital worker.”

Say you ask it to “build a competitive analysis site for AI coding tools.”

First, a research agent scans the web and pulls current data. Then a synthesis agent structures that into categories, pricing tiers, feature matrices. A writing agent drafts the copy in your specified tone. A coding agent generates a working web app. Another component tests and refines. The system then packages and delivers the final asset.

You did not have to micromanage those steps. The system delegated internally. That is the multi-agent layer.

Instead of one giant model pretending to reason, search, code, and remember all at once, the system splits the task into subtasks. Each subtask gets routed to a specialized component.

But the most important part is that Perplexity Computer can orchestrate across 19 models from different providers. So the deep research step might be assigned to Gemini. The coding step might go to Claude. And the summarization might be handled by GPT.

You are not picking models one by one. The system chooses the best AI for the task.

In theory, that means you get best-of-breed performance for each slice of work without manually stitching APIs together.

The other important piece is persistence.

This is not a session that dies when you close a tab. Computer can run for hours or months. You can tell it to monitor competitor pricing weekly, update a dashboard, and notify you when something changes. It operates on a schedule.

The Trend

To me, Perplexity Computer sits in the same philosophical lane as OpenClaw.

Both position themselves as digital workers that “get things done”. You hand over an objective, they decompose it, operate across tools, and return something finished. Multi-step reasoning. Workflow execution. Persistence over days, not minutes.

The main split is where they do the work:

OpenClaw is self-hosted and close to your environment. That proximity gives power and flexibility, but it also raises the stakes on security and deployment.
Perplexity keeps everything inside a managed cloud sandbox. This is much safer, more standardized and easier to pitch to an enterprise buyer.

In a sense, it is a more packaged, less chaotic version of OpenClaw. And that is probably what many users wanted after OpenClaw went viral a month ago. The ambition, without the operational anxiety.

I think we are starting to see a trend here. Call it openclawfication (bad name).

Over the next months, I expect more products to move in this direction, orchestration layers that sit on top of multiple models and tools. Systems built around delegation, execution and persistence.

Help Me Improve Failory

That's all for today’s edition.

Cheers,

Nico

Right When It Worked

Nicolás Cerdeira — Thu, 19 Feb 2026 19:27:48 +0000

Hey - It’s Nico.

Welcome to another Failory edition. This issue takes 5 minutes to read.

If you only have one, here are the 5 most important things:

Botkeeper, an AI bookkeeping startup, shut down a week ago — learn why below.
The AI pricing and monetization playbook.
China’s top AI labs just dropped a bunch of AI models — learn which ones matter the most below.

A huge thanks to today’s sponsor, Deel. Win $1M for your startup with a 2-minute pitch!

Win $1M for your startup – Pitch in 2 minutes ^AD

The Pitch by Deel is here.

We’re giving away $1M in investment to founders with the most promising ideas.

No warm intros and no fees. Just a pure competition for the best entrepreneurs.

Whether you're just starting or scaling, this is your chance to get the capital you need. Pitch in 2 minutes and join the competition that’s changing how startups get funded.

Apply Now

This Week In Startups

🔗 Resources

The AI pricing and monetization playbook.

The B2B revenue opportunity hiding in plain sight.

In Defense of Vertical Software.

What AI image and video generation unlocks for lean teams *

📰 News

Cohere launches a family of open multilingual models.

OpenClaw founder Peter Steinberger joins OpenAI.

GPT‑5.2 derives a new result in theoretical physics.

Gemini can now create music with Lyria 3.

💸 Fundraising

New AI startup Ineffable Intelligence reportedly raising $1B funding round.

World Labs raises $1 Billion to scale spatial AI.

Biotech startup raises $175 million to advance Alzheimer’s antibody therapy.

AI chip startup Taalas raised $169 million to take on Nvidia

* sponsored

Fail(St)ory

AI Bookkeping

Botkeeper is dead.

After 11 years and almost $90 million raised, the AI bookkeeping startup shut down a week ago.

Botkeeper was one of the better-known “AI for accounting” bets of the last decade. And according to its CEO, the tech worked. The market just moved faster.

Let’s unpack that.

What Was Botkeeper:

Botkeeper was founded in 2015 by Enrico Palmerino. The pitch was simple: bookkeeping is repetitive and rule-based, so machines should handle most of it.

From day one, they branded themselves as “the future of bookkeeping.” The mission was to modernize a conservative, compliance-heavy industry that still runs on spreadsheets and manual reconciliation.

The core product automated transaction categorization, reconciliations, and the monthly close. The goal was to remove the grunt work and let accountants focus on higher-margin advisory.

But this was never pure SaaS.

Botkeeper ran a hybrid model: machine learning plus a layer of skilled accountants who handled exceptions, reviewed outputs, and kept accuracy high. Automation did the heavy lifting. Humans made sure it didn’t break.

By late 2021, the company had raised $90 million and claimed 200+ accounting firms and 5,000+ small businesses on the platform.

Then came the product evolution.

In 2023–2024, they launched Botkeeper Infinite, a tech-only version of the platform. No bundled outsourcing. Just the software.

And by the end of 2025, Palmerino claimed they had finally cracked it. The system could clean up years of messy books in minutes, code more than 80 percent of transactions with 98 percent accuracy, reconcile accounts autonomously. The original vision from 2015, machines doing the work, was finally real.

Then, almost immediately, they shut it down.

Not during the messy early years. Not when the model was half-baked. Right when the product was supposedly at its peak.

So if the tech finally worked, what actually broke?

The Numbers:

💰 Total funding: ~ $89–90 million
📅 Founded: 2015
🏢 Customers claimed (2021): 200+ accounting firms, 5,000+ SMBs
💵 Infinite pricing: starting at $69 per client

Reasons for Failure:

Industry consolidation hit their biggest clients: This is the CEO’s explanation. In his shutdown letter, Palmerino says a wave of unexpected consolidation in late 2025 impacted their largest accounting firm clients and changed their financial outlook in a matter of weeks. When those firms merged or restructured, revenue shrank fast, and the company couldn’t course-correct in time.
The product was strong. The fit wasn’t strong enough: Palmerino admits it directly: “we did not reach a level of product-market fit strong enough to withstand rapid industry shifts or changing market conditions before our time ran out.” The tech improved, but it wasn’t embedded deeply enough to survive client mergers and budget resets. If customers can drop you during consolidation, you are useful, not indispensable.
The business was mid-transition: They had spent years selling a hybrid model of software plus human support, then shifted with Infinite toward a tech-only platform. That kind of model transition takes time to reshape margins, positioning, and customer behavior. The product may have been at its strongest, but the business model was still evolving when the revenue shock hit.

Why It Matters:

Hitting the original product vision is not the same as building a durable company.
Customer structure can be a bigger risk than technical risk.

Trend

China’s AI Sprint

It’s Lunar New Year.

While a lot of people are offline, China’s AI labs decided to ship half the future.

Within days, the biggest players dropped new flagship models. LLMs, Video generators and Agent upgrades.

Lets unpack them.

Seedance 2.0

ByteDance (TikTok) dropped Seedance 2.0, a next-gen audio plus video generation model with text, image, audio and video inputs.

What stood out was how much they leaned into control. Seedance is built around references and editing workflows. You can steer it. Iterate. Adjust style, motion, cuts. It feels less like a slot machine and more like a creative tool.

That’s why it went viral. When creators feel they can direct the output instead of gambling on prompts, they share results.

The positioning is clearly production-oriented. Motion quality. Scene interaction. Coherence. Controllability. These are the pain points that made earlier text-to-video models unusable in real workflows.

Qwen 3.5

Alibaba Group just released Qwen 3.5, the latest version of its open-weight large language model. If you don’t follow Qwen closely, here’s what matters.

First, it’s open under Apache-2.0. That means you can download it, fine-tune it, host it yourself, and ship products without waiting for API access. That alone drives fast adoption.

Second, Alibaba pairs that with a hosted “Plus” version that offers long context and built-in tools. So hobbyists and startups can experiment for free, and serious teams can pay for a stable, managed version.

They also push the “agent” angle hard. Qwen 3.5 is framed as capable of operating across mobile and desktop apps, handling bigger workloads, and executing multi-step tasks.

GLM-5

Zhipu AI introduced GLM-5 as a new flagship model.

It’s released under an MIT license and positioned for coding, long-horizon tasks, and agent workflows.

GLM-5 is showing up at the top of open-weight leaderboards and posting strong results on coding benchmarks like SWE-bench Verified. That puts it in serious territory for developers evaluating open models.

In practical terms, this means you now have an open model that competes closely with top closed systems on engineering-heavy tasks.

A lot of people are already comparing it to Claude Sonnet 4.5-level performance on coding and agent benchmarks, which is really good news for the Chinese startup.

Help Me Improve Failory

That's all for today’s edition.

Cheers,

Nico

The British SpaceX

Nicolás Cerdeira — Thu, 12 Feb 2026 19:08:01 +0000

Hey - It’s Nico.

Welcome to another Failory edition. This issue takes 5 minutes to read.

If you only have one, here are the 3 most important things:

Orbex, a British rocket startup, filed for insolvency this week — learn why below
The hidden danger of shipping fast.
OpenAI rolled out Ads in the free and Go tiers — learn why this matters below

A huge thanks to today’s sponsor, Oceans. Reclaim 10+ hours a week with a vetted Executive Assistant.

An Oceans EA+ can give you your week back ^AD

If your calendar runs you, your business pays the price. If you live in inbox triage, strategy takes a step back.

That’s where Oceans Talent comes in.

We match you with experienced, trained, rigorously vetted Executive Assistants who operate with judgment, not reminders.

Here’s what you get:

Inbox triage with clear next steps
Calendar scheduling and meeting briefs
Follow ups, vendor chasing, and coordination
SOPs for recurring workflows
Daily and weekly exec updates you can scan fast
Core competencies in Marketing, Finance, Ops, and more

400+ founders delegate their to-do list to their Oceans EA+, so they can stay focused on strategy.

👉 Reclaim 10+ hours a week with an Oceans EA+

This Week In Startups

🔗 Resources

The Lighthouse Playbook: How founders make friends and influence people.

The hidden danger of shipping fast

The skills that still matter as AI progresses, and how to develop them.

AI doesn’t reduce work — it intensifies it.

📰 News

Reddit looks to AI search as its next big opportunity.

Crypto.com CEO buys AI.com For $70 Million.

Threads’ new ‘Dear Algo’ AI feature lets you personalize your feed.

Spotify says its best developers haven’t written a line of code since December.

💸 Fundraising

Humanoid robot startup Apptronik has now raised $935M at a $5B+ valuation.

Human brain-based AI startup raises £6m.

AI fragrance startup Osmo raises $70M.

Runway raises $315 Million to advance AI video and world models.

Fail(St)ory

Close to Orbit

Orbex was the UK rocket startup that was supposed to prove Britain could launch its own satellites from home.

It built its own rocket and got close to first test launches, but the runway disappeared right before the moment that mattered.

What Was Orbex:

Orbex started in 2016 with a pretty clean bet: small satellites needed more launch options, and Europe didn’t have enough. It wanted to be the “book a launch like a service” company for small payloads.

Its flagship vehicle was Prime, a two-stage microlauncher around 19 meters tall.

It wasn’t trying to be SpaceX. SpaceX optimized for big rockets, big payloads, and shared rides. Prime aimed at the opposite end: smaller satellites that wanted a dedicated slot and a schedule they could actually plan around.

Prime also came with two clear angles.

One: sell a greener rocket, with bio-propane and low-carbon branding baked into the pitch.
Two: sell a British launch capability, with UK-built hardware and launches from UK soil.

Orbex raised real money and got meaningful public support, including a £20M government investment. But it was still a capital-intensive grind, and by late 2025 it was signaling it needed a lot more cash to reach flight and scale.

So it went shopping for a rescue.

The most serious option was The Exploration Company (TEC), a Franco-German startup building Nyx, a reusable capsule for cargo to orbit and back. Orbex had the rocket. TEC had the thing that rides on top.

They signed a letter of intent in January. Right around the same time Orbex shut its rocket engine factory in Denmark, cut 90 people, and that Danish unit filed for bankruptcy.

When the TEC deal fell apart, Orbex said it had no funding or M&A options left. It filed to appoint administrators soon after. Prime was still pre-first-flight, with test launches planned, but it never got the runway to prove itself.

The Numbers:

🚀 Founded: 2016
💰 Total raised: ~$163M
🏛️ UK public support: £20M government investment (convertible-style) + ~£26M government loans
🔥 Funding need signaled: ~£120M additional over coming years

Reasons for Failure:

They fell into the “scale-up funding gap”: Orbex’s CEO basically described the trap in plain terms: rockets take a long time, cost a lot, and there’s a nasty “scale-up” gap where you’re past early R&D but not yet flying. That’s exactly where Orbex ended up. You’ve built real hardware, headcount is high, suppliers want cash, and revenue is still theoretical until you launch. Orbex was signaling it needed around £120M more to finish development plans, and that kind of number only comes from investors who believe you’re close enough to de-risk.
“Sovereign capability” attracted money, then constrained options: Being a political project helped Orbex win public backing and attention. It also made the rescue path narrower, because any rescue deal had to satisfy “taxpayer value,” not just founder logic. Reporting said the UK government declined to back the TEC-Orbex deal, and that support was important to making it financeable.
The acquisition was a Hail Mary with too many moving parts: TEC was building a reusable cargo capsule, and buying Orbex would’ve meant controlling its own path to orbit. But it needed new money, fast execution, and confidence Prime could fly on schedule. Then Orbex shut the Denmark engine factory, cut 90 people, and that unit filed for bankruptcy right as the LOI landed. From there, it stopped being a strategic buy and became a distressed one, and those deals rarely close in time.

Why It Matters:

Vertical integration is not a lifeboat if both halves need cash.
Government money is not only runway, it is also scrutiny.
The “almost there” stage is where companies quietly die.

Trend

AI Ads are Here

We knew they were coming for a while. Anthropic warned us at the Super Bowl. Sam Altman got pretty upset about it. But now they’re finally here.

As of 3 days ago, ChatGPT is officially showing ads in the Free and Go tiers.

Ads have arrived in AI.

Why it Matters

“Intent” just moved upstream. Search catches you when you already know what to ask. Chat catches you while you’re still figuring out what you want.
The product is now the business model. When monetization depends on what happens inside a conversation, the boundary between “helpful” and “profitable” gets politically fragile.

The OpenAI Ads

After weeks of speculation, we finally see the format.

Ads are clearly labeled as sponsored. They sit at the bottom of answers, visually separated from the organic response. Not blended into the text. Not disguised as advice.

OpenAI is making a few explicit guarantees.

Ads do not influence answers. The model’s responses are optimized for what’s most helpful to you, not what pays the most.
Advertisers don’t get your chats. They receive aggregate performance data like views and clicks, not conversation history, memories, or personal details.
Targeting still exists. OpenAI decides which ad to show by matching advertiser submissions with your conversation topic, your past chats, and your past interactions with ads. The platform uses the context. The advertiser does not see it.

On paper, it seems pretty good. Clear labels. Separation from answers. Privacy guardrails.

But structure matters less than incentives. And incentives are where things get interesting.

The Debate

The pro case is straightforward:

Free access needs funding. If you want millions of people using powerful models without a paywall, someone has to underwrite it. Ads are the most obvious lever, and arguably the most inclusive one.
Also, intent in chat is absurdly high. A user asking “What’s the best X for my situation?” is basically a pre-qualified lead. They’re not scrolling. They’re not browsing. They’re asking for a decision.
OpenAI’s final pitch is that these ads can be less obnoxious than feed ads. Contextual, labeled, and only when relevant. If they stay at the bottom, stay rare, and stay honest, many users will tolerate them.

Still, there are many things to be concerned about:

Ads introduce conflicting incentives. Even if responses stay untouched, the company now monetizes what happens around the answer. Over time, that pressure can bleed into product decisions.
There’s also manipulation risk. Chat is intimate. People share doubts, financial stress, health worries. If revenue ever depends on nudging behavior at those moments, trust erodes quickly.
And then there’s the enshittification fear. Today the ads are clearly labeled and sit below the answer. Tomorrow, the pressure is to push them higher. Then maybe inside the response. Then maybe not so clearly labeled.

The Trend

I think AI ads are here to stay.

The economics demand it. These models are expensive to run. Usage keeps growing. Subscriptions alone won’t carry the whole thing. If you control high-intent traffic at global scale, you monetize it.

I also think they will quickly become one of the most important marketing channels.

Because people are already using GPT to decide what to buy. I am. Most of my recent purchases started with a prompt. Tools. Software. Even random consumer stuff. The research phase is collapsing into a conversation.

If that same interface starts attaching clean, relevant buy links to those moments, conversion rates will be absurd.

Help Me Improve Failory

That's all for today’s edition.

Cheers,

Nico

Fuel at 50% Off

Nicolás Cerdeira — Thu, 05 Feb 2026 20:09:07 +0000

Hey - It’s Nico.

Welcome to another Failory edition. This issue takes 5 minutes to read.

If you only have one, here are the 3 most important things:

Koko, a clean-cooking startup in Kenya, has shut down — learn why below
How to build AI product sense.
Google just released Project Genie, an AI model that creates interactive worlds — learn why this matters below

A huge thanks to today’s sponsor, Oceans. Build a high-output global team at up to 80% less cost with their help.

Build a high-output team without U.S. headcount costs ^AD

Did you know one U.S. hire can cost the same as a small global team?

That’s where Oceans Talent comes in.

We help startups hire vetted global talent across ops, finance, marketing, EAs, and more at up to 80% less than U.S. hiring costs.

Here’s what you get:

Deep discovery on your goals and role requirements
A match sourced from 1500+ monthly applicants
Vetting beyond resumes across 12 hard and soft skills
Integration plan to support the first 100 days with defined expectations, success metrics, and recommendations
Dedicated account managers help you scale your Oceans Talent team as you hire more.

400+ companies use Oceans Talent to do more while spending less.

👉 Get Started with Oceans Talent

This Week In Startups

🔗 Resources

How to build AI product sense.

A new look at free-to-paid conversion.

The automation curve in agentic commerce.

Pilot helps startups run world-class finance, including bookkeeping, payroll, taxes, and fundraising, powered by dedicated US-based experts and AI-driven insights *

📰 News

OpenAI launches a way for enterprises to build and manage AI agents.

Anthropic releases Opus 4.6 with new ‘agent teams’

OpenAI launches new macOS app for agentic coding.

Amazon to begin testing AI tools for film and TV production next month.

💸 Fundraising

Waymo raises $16 billion investment round.

Lawhive, a startup using AI to reimagine the general practice law firm, raises $60 million.

ElevenLabs raises $500m Series D.

Business identity startup Duna raises €30m.koko

* sponsored

Fail(St)ory

Subsidizing Fuel With Credits

Koko ran a clean-cooking network in Kenya. It sold bioethanol for everyday cooking to households that normally relied on charcoal or kerosene.

In early 2026, it shut down almost overnight.

What made Koko interesting wasn’t the product. It was the business model. The company sold fuel and cookware far below cost, took on serious debt, and expected to make its money from carbon credits sold to buyers overseas.

When that plan broke, everything else followed.

What Was Koko:

If you live in Europe or the US, a “clean-cooking startup” sounds niche. In Kenya, it isn’t. A large share of households still cook with charcoal or kerosene, especially in low-income areas. These fuels are smoky, unsafe indoors, and tied to deforestation.

Koko tried to replace charcoal and kerosene with bioethanol, a cleaner fuel that burns without filling the room with smoke.

The company built a nationwide network of automated fuel-dispensing machines. Think ATMs, but for cooking fuel. You brought your container, tapped in, refilled, and went home. By the time Koko shut down, there were more than 3,000 of these machines spread across Kenya.

What made Koko unusual was the business model. It sold bioethanol at roughly half the market price, around KES 100 per liter when the going rate was closer to KES 200. It also sold a proprietary cookstove for about $12, even though the reported cost to make and deliver that stove was closer to $115.

Koko was losing money on every customer, on purpose. The company never expected to make money from Kenyan households. The assumption was that these families could never pay the “real” price for clean cooking, and pretending otherwise would kill adoption.

Instead, Koko planned to get paid through carbon credits. When a household switched from charcoal or kerosene to ethanol, it avoided emissions. Those avoided emissions could be measured, verified, and sold to buyers overseas.

The easiest way to picture it is this: imagine you’re an airline in Europe. You can’t easily cut your emissions, so you buy carbon credits to offset them. Koko wanted to sell you those credits, at a relatively high price, and then use that money to subsidize cooking fuel for families in Kenya.

In practice, the carbon buyer paid for the clean cooking. The Kenyan customer just saw cheap fuel and a stove that worked.

That setup made Koko more like a climate-finance machine than a normal consumer startup. It also meant the whole model depended on being allowed to sell those credits internationally. When that assumption started to wobble, the rest of the business had very little margin for error.

The Numbers:

🗓️ Founded: 2013
🏠 Customers: ~1.5M households served
🏧 Network: 3,000+ fuel ATMs
🧑‍🏭 Team: 700+ employees laid off
💸 Funding: $100M+ debt + equity
🧾 Debt: $60M+ by early 2026

Reasons for Failure:

The entire business depended on one government letter: Koko needed a Letter of Authorization from the Kenyan government to sell its credits into the compliance pathways it was built around. Without it, the company couldn’t claim the “high-price” carbon revenue that funded the subsidies. When the refusal came in January 2026, the model stopped working immediately.
The unit economics were intentionally negative, and they stayed that way: Selling ethanol at half price and stoves far below cost can work if there’s a reliable subsidy source that scales with usage. Koko’s subsidy source was carbon, but the company didn’t control access to that market. So every new household was both a win and a bigger cash drain. At 1.5 million households, you’re carrying a national-scale subsidy program on a startup balance sheet.
Debt turned policy delay into a fatal cliff: Koko financed the gap with debt tied to future carbon revenue. Koko had neither. Once the government refused the LoA, $60M+ in debt became unserviceable, fast. There’s no graceful unwind when your cost base is physical infrastructure and your revenue is blocked by policy.

Why It Matters:

Koko shows what happens when your revenue depends on policy instead of customers.
Subsidies that scale faster than certainty can kill you before they ever pay off.

Trend

Text-to-World

You’re probably numb to “AI can generate X” headlines by now.

So here’s the one that actually made me sit up a bit: Google’s Project Genie doesn’t just generate a scene. It generates a place. A place you can walk around in, mess with, and turn into a tiny game with arrow keys.

I talked about world models a couple months ago. Back then it was mostly theory and demos on Twitter. Now Google ships a thing you can actually try. And it’s a little unsettling.

Why it Matters

It’s a real stepping stone toward AGI. An agent needs to predict what happens next when it acts, and world models are how you train that without wrecking real stuff.
It’s a simulation shortcut for industries that already pay for simulation. Driving, robotics, training, QA, safety drills, anywhere “let’s test it in a fake world first” is already a budget line.

What it is

Project Genie is Google’s new experiment for building interactive virtual worlds from prompts.

You type “enchanted forest” or “ancient city” or whatever your brain comes up with. You can also feed it an image. Then you drop in a character and start moving around. Arrow keys. Camera controls. The usual.

If this was just video generation, it would be cute and forgettable. The wild part is that it stays coherent while you play.

As you walk forward, the model generates what’s ahead in real time. When you turn around, the stuff behind you doesn’t randomly morph into something else. If there’s a wall, you don’t phase through it. If you hit an object, it reacts like an object, not like a glitch. It’s doing enough physics to feel like a world instead of a slideshow of vibes.

That’s the mind-bending shift: the model isn’t only drawing frames. It’s tracking state.

It has to “remember” what it already showed you, predict what should appear next, and keep the whole thing consistent across time.

Google calls it a “world model.” While LLMs predict the next word, world models predict the next moment. They simulate how an environment evolves and how your actions change it.

The Trend

So why does this matter?

The obvious answer is video games. Since Project Genie came out a lot of gaming companies have seen a sharp decline in their stock price: Take-Two gets whacked more than 9% in a day. Roblox drops around 12%. Unity eats the worst of it, down about 30%.

But gaming is just the loudest first-order effect. The bigger story is that a stable, interactive world model is a new kind of infrastructure. Once you have it, it leaks into everything.

A couple months ago I wrote about Fei-Fei Li’s “Spatial Intelligence” thesis. The point was simple: today’s AI is great at describing the world, terrible at inhabiting it. LLMs can talk about a kitchen perfectly and still fail at basic stuff like distance, occlusion, and what happens when you push a cup off a counter.

Her argument was that the next leap isn’t more words. It’s giving models a sense of space, time, and cause and effect.

That’s why world models matter. They’re the bridge between “knows language” and “can operate.”

If you want AGI, you don’t just need a model that can answer questions. You need a system that can do the boring parts of being competent in the real world.

World models are a way to train that competence without burning through real-world trials. You can’t teach an agent to be useful in warehouses, hospitals, factories, or roads by letting it crash into things until it learns.

You need a safe place to practice.

That’s what world models are: a training sandbox where actions have consequences and the world stays consistent. Project Genie is a playable demo of that idea.

Help Me Improve Failory

That's all for today’s edition.

Cheers,

Nico

Crypto + AI?

Nicolás Cerdeira — Thu, 29 Jan 2026 19:51:24 +0000

Hey - It’s Nico.

Welcome to another Failory edition. This issue takes 5 minutes to read.

If you only have one, here are the 3 most important things:

Entropy, the “zappier for crypto”, has shut down — learn why below
What's working in AI for GTM
Moltbot, the AI agent that “actually does things”, is going viral — learn all about it below

A huge thanks to today’s sponsor, Playbookz. Grab their $1k deal while it lasts!

Turn LinkedIn into a steady stream of buyer conversations ^AD

Playbookz helps founders, CEOs, and exec teams turn LinkedIn into a predictable growth channel.

They run the entire system for you:

weekly content that builds authority
outbound DMs that start real conversations
ads that amplify what’s already working

The outcome is simple: more qualified meetings booked, more inbound from people who already know who you are, and stronger credibility before the first call.

Most teams see: 150k reach, 300+ leads, 45 calls & 9 new clients in the first 90 days

Book a no-obligation call here. You’ll get $1,000 off your first month.

This Week In Startups

🔗 Resources

What's working in AI for GTM

10x job posts for 10x engineers

How to understand customers

Pilot helps startups run world-class finance, including bookkeeping, payroll, taxes, and fundraising, powered by dedicated US-based experts and AI-driven insights *

📰 News

OpenAI’s Sora app is struggling after its stellar launch.

Chrome takes on AI browsers with agentic features for autonomous tasks.

OpenAI launches Prism, a new AI workspace for scientists

China’s Moonshot releases a new open source model Kimi K2.5 and a coding agent.

💸 Fundraising

The counter-drone startup Frankenburg raised $50M at a $400M valuation.

Crypto payments startup Mesh raises $75M at $1B valuation.

CVector raises $5M for its industrial ‘nervous system’.

Jelou raises $10M to turn WhatsApp chats into transactional AI hubs

* sponsored

Fail(St)ory

Putting Agents in Crypto

Entropy shut down this week. It was an a16z-backed crypto custody startup that slowly turned into “Zapier for crypto” with AI and fancy security sprinkled on top. You can probably guess how that ended.

Four years in, the founder pulled the plug, laid out the pivots, and said the quiet part out loud: they couldn’t find a venture-scale business model.

What Was Entropy:

Entropy started in July 2021, right when “crypto infra” was a cheat code for fundraising.

At the beginning, Entropy tried to replace the “custody company” model. Normally, if a business doesn’t want to lose millions to one misplaced key, it pays a custodian to hold the assets and approve withdrawals. That’s convenient. It also creates a giant target and a single place where things can go wrong.

Entropy’s approach was basically: don’t let any one person or one system have full control. The “key” was split into pieces, and moving funds required multiple approvals working together. So even if one piece got compromised, it wasn’t enough to drain everything.

That was the wedge. But the wedge had a ceiling.

By 2021-2025, custody got crowded and entrenched fast. Big players owned the institutional market. Wallets on the retail side grew like weeds. Entropy could build something safer and still struggle to convince people to switch, because switching custody isn’t like switching a note-taking app. It’s scary, slow, and full of committees.

So they pivoted.

Instead of only securing assets, they tried to make the wallet do work. They leaned into automation: software that could run actions for you in crypto, under rules you set. The company described it as basically Zapier or n8n for crypto, with signing built in so workflows could actually move funds without you clicking through prompts all day.

And because it was 2025, the automation pitch came with the full AI garnish. “Agents.” “Autonomous.” “Context-aware.” The idea was that you wouldn’t just click buttons in a wallet anymore. You’d delegate intent. An agent would watch markets, hop across chains, route transactions, rebalance, maybe even trade, all inside guardrails you defined.

In their version of the future, your wallet stopped being a vault and started acting like a junior analyst that never slept. It would “optimize” your holdings, look for opportunities, and execute based on your strategies and risk limits.

The Numbers:

💰 Total raised: ~$27M
🧾 Biggest round: $25M seed (2022) led by a16z crypto
📆 Founded: July 2021
📉 Market context: crypto venture deal count fell ~60% YoY in 2025 (about 1,200 deals vs 2,900+ in 2024)

Reasons for Failure:

The model never matched the cap table: The founder said they ran an “initial feedback request” and learned the business model wasn’t venture scale. That’s the whole story right there. You could ship something real and still not have a path to a massive outcome.
They kept pivoting into bigger, harder problems: Entropy started with custody via threshold signing. Then it became an agent network. Then it became “Zapier for crypto” plus AI plus TEEs plus automated signing. Each step sounded more powerful, but it also widened the surface area and pushed them further into “platform” land.
The market had already picked winners in custody: Institutional custody had been dominated by players like Fireblocks and Copper. Retail wallets like MetaMask and Phantom had expanded aggressively. Entropy’s differentiation was deep security architecture, but custody buyers rarely switched for elegance.
Crypto infra got hit when funding dried up: Entropy shut down in a year when crypto venture activity dropped sharply, with deal counts down about 60% year over year in 2025. That kind of environment punished long-horizon bets.

Why It Matters:

Buzzwords don’t add up to demand. Crypto + agents + AI sounded cool. It still needed a buyer with a budget.
Custody is a trust market, not a features market. “Safer” didn’t beat defaults, audits, and existing relationships.
Test venture-scale early, not after years of building. Product feedback is cheap. Business model feedback decides the company.

Trend

Moltbot

There’s a new open-source AI agent going viral these last few days. It’s called Moltbot, and the pitch is basically “AI that actually does things.”

It became one of the fastest growing projects in Github history, crossing 85.000 stars in just two weeks, and it feels like giving a hyperactive kid the keys to your laptop: useful, fast, and a little terrifying.

Why it Matters

Power and risk ship together now. Moltbot is fun because it can actually do things: run commands, touch files, click around the web, move money if you let it. That’s also why it’s scary, because one bad config or one sneaky prompt injection turns “personal assistant” into “remote-control chaos.”
Distribution is hiding in plain sight. Moltbot doesn’t win by having a shiny new UI. It wins by living inside WhatsApp and Telegram and Slack, where you already spend your time. That’s the real growth hack.

What it is

Moltbot is an open-source, self-hosted “personal assistant” you run on your own machine or server.

You talk to it through chat apps you already live in (like Wahtsapp or Telegram), so it feels like texting a capable assistant who has access to your computer.

It’s agentic in the literal sense. It can run shell commands, read and write files, automate the browser, trigger webhooks, and run scheduled jobs. That’s why people call it persistent and 24/7. It’s not a tab, it’s a process.

It’s also not “one model.” It can route work across different model providers and different agents or workspaces depending on your setup. The core vibe is a local gateway that handles sessions, channels, tools, and events, plus a growing “skills” layer that expands capabilities fast.

Why it went viral

First, because it’s a new magic trick: texting your computer.

That sounds small, but it hits hard. People don’t want to “open an AI app.” They want to send a message like “handle this” and get a result back. Moltbot makes that feel normal in about 30 seconds.

Second, because some of the stories are ridiculous in a way the internet loves.

One user made Moltbot buy him a car. The bot searched for the car, negotiated the price and even close the deal.

And the one that should be printed on a warning label, where the bot buys a $3K “build your personal brand” course while the user sleeps:

Those stories do more marketing than any landing page. Because they’re not about intelligence. They’re about agency. The bot isn’t impressive because it talks well. It’s impressive because it can move money, click buttons, and make decisions you forgot you delegated.

Of course, this is also extremely dangerous.

Because the moment your agent can run commands and read files, you’ve created a new attack surface. Not just “someone hacks your server.” More subtle stuff too. Any untrusted thing the bot reads can mess with it: emails, web pages, attachments, DMs, random pasted text.

This is why the whole thing is so watchable. It’s cool. It’s scary. It’s both at the same time.

The Trend

I think Moltbot is telling us a few things about where AI agents are heading:

Chat is the control plane. The killer move is hiding an agent inside WhatsApp/Telegram/Slack so it feels like texting, not “using software.” That’s why it spreads. It feels like texting a human assistant.
People care about side effects, not IQ. The posts that travel aren’t “look how smart it is.” They’re “it did something real” and sometimes “it did something real I didn’t want.” That’s the new bar.
The market is going to split into two camps. One camp builds “spicy” agents that move fast and break things. The other builds the boring layer that makes them safe: permissions, sandboxing, secrets hygiene, skills you can actually trust.

Moltbot is basically a preview of the next year: more power, less friction, and a lot more ways to shoot yourself in the foot.

Help Me Improve Failory

That's all for today’s edition.

Cheers,

Nico

Failory

Fintech on Fumes

What if 60% of your support tickets never reached you? AD

This Week In Startups

🔗 Resources

📰 News

💸 Fundraising

Fail(St)ory

Money Out, Not Money In

Trend

Claude Fable 5

Why it Matters

The Coding Hype Has Substance

The Guardrails Are Everywhere

Mythos Already Changed the Security Workload

Help Me Improve Failory

The Pizza Robot Graveyard

Demystify AI for Your Business in Just 4 Steps AD

This Week In Startups

🔗 Resources

📰 News

💸 Fundraising

Fail(St)ory

Robot Aquarium

Trend

Microsoft Build 2026

Why it Matters

Microsoft IQ makes company context the product

Scout is Microsoft’s always-on work agent

GitHub Copilot is getting a new app

Microsoft is reducing its OpenAI dependency

Help Me Improve Failory

Turned Into A Feature

When It’s Time to Close, Do It Properly AD

This Week In Startups

🔗 Resources

📰 News

💸 Fundraising

Fail(St)ory

Personalized Podcasts

Trend

Burning Tokens

Why it Matters

Tokenmaxxing

Why it Broke

Help Me Improve Failory

23M Users. Only 75 Paid.

Closing Well in Uncertain Times AD

This Week In Startups

🔗 Resources

📰 News

💸 Fundraising

Fail(St)ory

The Cost of Free

Trend

Google I/O 2026

Why it Matters

Gemini 3.5 Flash

Spark

Daily Brief

Omni

Search and Commerce

Help Me Improve Failory

Sudden Bankruptcy

Sending Emails but Getting Zero Replies? AD

This Week In Startups

🔗 Resources

📰 News

💸 Fundraising

Fail(St)ory

Parker

Trend

AI That Interrupts You

Why it Matters

OpenAI’s Realtime Push

Thinking Machines’ Interactive Models

Help Me Improve Failory

Too Early To AI

When the market shifts, we'll help you handle closure right AD

This Week In Startups

What if 60% of your support tickets never reached you? ^AD

Demystify AI for Your Business in Just 4 Steps ^AD

When It’s Time to Close, Do It Properly ^AD

Closing Well in Uncertain Times ^AD

Sending Emails but Getting Zero Replies? ^AD

When the market shifts, we'll help you handle closure right ^AD

Overpaying for Growth? ^AD

International expansion without the compliance liability ^AD

Capture high-intent traffic from AI search ^AD

⚡️ Most founders are 1 system away from 100k+ reach on LinkedIn ^AD

✈️ Scale your LinkedIn personal brand on autopilot ^AD