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. .

Twitter tweet

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

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.

Twitter tweet

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.

Twitter tweet

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

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.

Twitter tweet

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.

Twitter tweet

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.

Twitter tweet

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

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

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.

Twitter tweet

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.

Twitter tweet

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.

Twitter tweet

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%. 

Twitter tweet

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

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.

Twitter tweet

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.

Twitter tweet

One user made Moltbot buy him a car. The bot searched for the car, negotiated the price and even close the deal.

Twitter tweet

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:

Twitter tweet

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

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:

• Sendle, a shipping startup for small businesses, has shut down — learn more below

• The 6 GTM problems that keep repeating.

• AI agents are good enough that teams are starting to build instead of buy software — learn why this matters below

And huge thanks to today’s sponsor, Deel. Get their guide to scale a remote-first company!

A founder’s guide to scaling globally ^AD

Build and scale a remote-first company with confidence.

This guide helps founders navigate global hiring, compliance, payroll, compensation, and culture, with a practical roadmap to grow internationally without losing speed or control.

Get the Guide

This Week In Startups

🔗 Resources

The 6 GTM problems that keep repeating.

Everyone can have their own AI CRM.

It's time for agentic video editing.

Pilot helps startups run world-class finance, including bookkeeping, payroll, taxes, and fundraising, powered by dedicated US-based experts and AI-driven insights * 

📰 News

Apple is developing an AI wearable.

YouTube will soon let creators make Shorts with their own AI likeness.

ChatGPT will now predict how old you are, in an effort to protect young users.

X open sources its algorithm while facing a transparency fine and Grok controversies

💸 Fundraising

Zanskar raises $115 Million to find and develop Carbon-Free Energy Using AI.

Healthcare AI startup OpenEvidence raises $250M at $12B valuation.

French startup Mistral raises €1.7 billion, cementing AI lead in Europe

AI startup Humans& raises $480 million.

* sponsored

Fail(St)ory

The Merger Trap

Sendle spent a decade selling small businesses on shipping that didn’t make you lose your mind. 

It raised over $100m, shipped tens of millions of parcels, expanded into North America, then shut down after a 2025 merger went bad. The ugly part was how fast it all snapped once trust and credit disappeared.

What Was Sendle:

Sendle launched in Sydney in 2014 for small merchants who shipped every day and hated carrier friction. 

In Australia, the default option for most businesses was Australia Post, the national postal service. Sendle positioned itself as the easier, cheaper alternative, with a cleaner experience for e-commerce sellers.

The company didn’t build its own fleet. It ran a software layer on top of third-party courier networks. Merchants bought the Sendle interface and pricing, while other operators did the physical delivery.

That model helped it grow. Sendle claimed it shipped more than 65 million parcels across Australia, the US, and Canada.

But the tradeoff was control. Service quality, costs, and reliability depended on partners. If a carrier tightened terms, raised prices, or had a bad week operationally, Sendle still owned the customer complaint.

Sendle expanded into North America in 2019. Reporting later pointed to burn and thin margins getting worse with aggressive headcount and marketing in the US.

By 2022, the pressure was visible. The company did layoffs and raised extra runway.

In August 2025, Sendle merged with two US delivery firms, ACI Logistix and FirstMile, under a new umbrella called FAST Group. The idea was to look bigger and move more parcels in the US.

Soon after, Federation Asset Management, a key backer, said it found ACI had unpaid bills that weren’t properly flagged during due diligence. Federation also said suppliers got fed up with late payments, which hit day-to-day performance.

At that point, partners tightened terms and cash got tight fast. FAST couldn’t raise enough to keep going, so the board shut Sendle down. Bookings stopped and merchants had to scramble.

The Numbers:

• 💸 Funding raised: $100m+

• 🗓️ Time operating: 12 years

• 🏙️ Founded: 2014, Sydney

• 🌎 Expanded: US + Canada in 2019

• 📦 Claimed volume: 65m+ parcels shipped

Reasons for Failure: 

• Partner dependence turned into a credit trap: Sendle’s asset-light model relied on third-party couriers to do the real work. That meant pricing, service levels, and payment terms were largely outside its control. When suppliers worried about getting paid, they protected themselves first.

• US expansion likely amplified bad unit economics: The North America push came with higher burn and more complexity. Reporting pointed to thin margins per delivery and spending that outran contribution. If you grew by hiring and marketing before the economics held up, you needed constant funding to stay afloat. When funding tightened, the model didn’t have room to breathe.

• The merger imported hidden liabilities: FAST Group was supposed to create scale and a stronger fundraising story. Instead, a key backer said ACI Logistix had unpaid obligations that weren’t disclosed properly during diligence. That kind of surprise spooked capital and tightened supplier terms at the worst time. Once confidence broke, raising emergency cash got ugly fast.

Why It Matters: 

• Mergers changed your failure mode. A company could manage its own burn, but it couldn’t easily absorb a partner’s surprise obligations.

• Asset-light didn’t mean low risk. It shifted risk onto partners and credit terms. When those partners got nervous, the whole product wobbled.

Trend

Buy vs Build

The era of SaaS might be over. Or, not really. But something interesting is happening.

AI agents like Claude Code, Cursor, and the one I talked about last week — Claude Cowork — have made building custom tools stupidly easy. 

The question companies are asking isn’t “which SaaS should we buy?” anymore. It’s now “can we just build this ourselves?”

Why it Matters

• The build vs buy line moved. Internal tools that used to take quarters now take days. 

• SaaS pricing assumptions look fragile. If a $20k–$30k a year tool can be replaced by something internal that’s good enough, renewals get a lot harder.

What’s going on

In short: Claude Code is fricking amazing, and each week brings new examples that push the boundaries of “what this can do.”

In the last two weeks:

• A principal engineer at Google experimented with Claude Code and had it build in one day something that had taken their team a year.

Twitter tweet

• Anthropic built Claude Cowork in about 10 days using Claude Code.

• The CEO of Cursor built an entire web browser using Cursor in a single week. I’m talking 3M+ lines of generated code across thousands of files. Here you can see the agent in action.

Twitter tweet

If an agent can spit out a functioning web browser, it can absolutely build:

• a reporting dashboard

• an internal CMS

• a lite analytics pipeline

• a custom ops tool

…often faster and cheaper than buying a SaaS and onboarding it.

And the market is starting to notice

Software stocks haven’t behaved like you’d expect in an AI boom.

Traditional SaaS names are lagging, even as AI infrastructure and broader tech do fine. According to Morningstar, many software companies now trade 20% or more below historical valuation levels. Not because revenues collapsed, but because investors are nervous about AI changing demand.

Big enterprise names like Salesforce, ServiceNow, and Adobe have all looked weak on the charts. The thinking is straightforward: if teams start building smaller tools internally, renewals get harder and seat growth slows.

You also see this fear surface in real time. After the launch of Claude Cowork, both Bloomberg and MarketWatch tied the product directly to software selloffs, framing it as a sign that AI generalists could eat into specific SaaS categories

The Trend

The real shift is pretty simple.

A growing chunk of software spend was never about “software.” It was about convenience. Paying so teams didn’t have to build small, annoying tools themselves.

That tradeoff is breaking down. Agents make it cheap to build internal tools that are specific, ugly, and good enough.

This doesn’t kill SaaS. But it pressures the parts of SaaS that relied on being “easier than building.” That advantage is thinner than it was six months ago.

Help Me Improve Failory

That's all for today’s edition.

Cheers,

Nico

Hey - It’s Nico.

Welcome to another Failory edition. Today’s issue takes 5 minutes to read.

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

• Alle, a startup building an AI fashion stylist, shut down this week — learn why below

• Carta’s State of Startups 2025

• Anthropic just shipped Cowork, the Claude Code for non-tech people — learn why this matters below

And huge thanks to today’s sponsor, Playbookz. Grab their deal while it’s live!

Get 2-10 clients/mo from LinkedIn ^AD

Playbookz helps founders turn LinkedIn into a consistent source of conversations with buyers. They handle everything: weekly content, outbound DMs, and ads.

That means:

• more qualified meetings booked

• more “I’ve been following you” inbound

• more trust built before the first call

• less pressure to “go viral” to see results

Most teams see: 50k reach, 10-30 meetings, and 2-10 new customers/mo.

Book a no-obligation call here. You’ll get $1,000 off your first month!

This Week In Startups

🔗 Resources

Carta’s State of Startups 2025

LLM’s vs Marketplaces

Sales problems vs. product-market fit problems

Spark Mail stops inbox overwhelm by helping you search, reply, and act faster. * 

📰 News

Google’s Gemini to power Apple’s AI features like Siri

Google announces a new protocol to facilitate commerce using AI agents.

Slackbot is an AI agent now.

AI models are starting to crack high-level math problems.

💸 Fundraising

Manufacturing automation software startup Tulip raises $120M.

AI video startup Higgsfield hits $1.3 billion valuation with latest funding.

Robotics startup Skild AI raises $1.4B.

Quantum software startup Haiqu raises $11M to launch hardware-aware operating system.

* sponsored

Fail(St)ory

Fashion Discovery

Alle was trying to become your AI fashion stylist. 

You’d talk to it, upload photos, get outfit ideas, and shop across brands from one place. It raised money, reached scale, and still shut down after six pivots and 2.5 years of trying to make it work.

What Was Alle:

The core idea was conversational shopping for fashion: You’d tell the app what you want in plain language. A vibe, a budget, an occasion. The AI would respond with full outfit ideas pulled from more than 1,000 brands

The app leaned hard into image workflows:

• Upload a screenshot from Instagram, get similar products and cheaper alternatives.

• Upload a photo of what you’re wearing, get a fit check and styling suggestions.

• Or upload your photo and do a virtual try-on before buying something.

In December 2023, Alle raised $3M in seed funding from Elevation Capital. The company claimed more than 1.5 million Gen Z users and a fast-growing active base.

But traction never hardened into a business. Discovery was sticky, monetization wasn’t. Affiliate revenue was thin. Brands didn’t pay enough for traffic that didn’t convert cleanly.

So the pivots started. B2C tweaks first. Then early exploration of B2B angles. Six pivots in roughly two years. Each pivot came with the belief that the next version would unlock a larger opportunity.

It never did. By October 2025, the team stopped pretending otherwise. The app disappeared from app stores. Cofounder Prateek Agarwal wrote the post founders write when they’ve done the math on opportunity cost and lost.

Six pivots wasn’t grit anymore. It was a signal.

The Numbers:

• 📱 1.5M+ claimed Gen Z users

• 🧠 6 pivots in ~2.5 years

• 💸 $3M seed round raised in Dec 2023

• 🛍️ 1,000+ brands aggregated

Reasons for Failure: 

• Discovery without monetization is a trap: Alle solved a real user problem. People don’t know what to wear. But solving discovery doesn’t mean you capture value. Affiliate-style economics require high intent and clean attribution. Alle sat too early in the funnel. Users explored, saved, and bounced.

• Consumer AI demos age fast: What felt magical in 2023 felt normal by 2026. Conversational search, image matching, basic try-on. None of it stayed special for long. Big platforms shipped similar features as add-ons, not standalone products. Even general-purpose tools like the GPT Shopping Researcher now cover a lot of this by default.

• Pivoting masked the real constraint: Six pivots sounds like learning. It can also be avoidance. Each new direction assumed the problem was surface-level: pricing, segment, wrapper. The core constraint stayed the same. Fashion discovery is valuable to users, but not valuable enough to pay for or defend.

Why It Matters: 

• User love is not leverage. Millions of users mean nothing if you sit in a low-intent layer of the stack. If you don’t control checkout, you don’t control money.

• AI features decay faster than brands. What differentiates you today becomes a feature in someone else’s app tomorrow. If your moat is “better prompts,” you don’t have a moat.

• Pivot count is a signal. One or two pivots is learning. Six is the market telling you something you don’t want to hear.

Trend

Claude Cowork

Anthropic just shipped Cowork.

If you saw Claude Code and thought “cool, but I don’t like the terminal stuff” this is the version you’ve been waiting for.

It’s the same basic idea as Claude Code: give Claude real access to where the work lives, let it run a task end to end, and don’t make you babysit every step. 

Why it Matters

• UI is the unlock. The command line scares normal people. Cowork has a much more friendlier UI.

• Claude Code turned “vibe coding” into a real workflow. You steer, it grinds through the steps, you review. Shipping feels less like tool-wrestling.

• Cowork tries to bring that same feel to office work. Docs, spreadsheets, decks, and messy folders are where most companies burn time. If this clicks, “vibe working” becomes a thing

First, what is Claude Code

Claude Code is Claude in “do the work” mode.

It’s a coding agent and, according to most devs, it’s the best one right now. While other agents still faceplant on more complex tasks, Claude Code’s loop of planning, executing, and testing usually gets you to a real result.

It got so useful that Anthropic quickly realized people weren’t using it just for coding. They were using it for the stuff around coding: cleaning things up, moving pieces, turning scattered notes into something coherent, updating docs so future-you doesn’t suffer.

That’s why they built Cowork.

What Cowork is

Cowork takes the Claude Code workflow and makes it usable for non-technical people.

In Cowork, you give Claude access to a folder you choose on your computer. Once it has that, it can read, edit, and create files inside that folder.

It sounds simple, but it unlocks a bunch of tasks that normal chatbots couldn’t actually finish end-to-end.

Twitter tweet

It can also pair with Claude in Chrome, so it can handle browser work too.

And it’s built for queuing tasks. You can drop a stack of jobs on it, let it work through them, and come back to review instead of sitting there in a back-and-forth loop.

What this means

Cowork matters because Claude Code already proved the shape of this thing works. 

Cowork is that same idea aimed at everyone else. Ops, finance, recruiting, sales, anyone drowning in docs and folders and “can you turn this pile into something clean.”

It’s also weirdly early. Neither GPT nor Gemini has a comparable “give it a folder, let it do the work” product in this exact form. If Cowork lands, that’s real edge for Anthropic.

I don’t know the killer use case yet. But an agent that can work inside your file system and also move through the browser is a general-purpose lever. You’ll find reasons to use it fast.

Help Me Improve Failory

That's all for today’s edition.

Cheers,

Nico

Hey - It’s Nico.

Welcome back to Failory after our two-week Christmas break.

Here are the three key takeaways from today’s edition:

• Ynsect, an insect meat startup, has shut down — learn why below

• What's working in SaaS pricing right now

• The AI trends for 2026 — see what they are below

A huge thanks to today’s sponsor, Playbookz. Be sure to grab their deal while it's live!

LinkedIn demand gen that books meetings ^AD

Playbookz helps founders turn LinkedIn into a consistent source of conversations with buyers. They handle everything: weekly content, outbound DMs, and ads.

That means:

• more qualified meetings booked

• more “I’ve been following you” inbound

• more trust built before the first call

• less pressure to “go viral” to see results

The behind-the-scenes work includes weekly posts, targeted outreach, and DM-first ads, but the benefit is super straightforward: reliable demand gen without founder overwhelm.

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

This Week In Startups

🔗 Resources

What's working in SaaS pricing right now

Everything is Computer: the production model of the future

Authority Is the AI Bottleneck

Delve’s AI Copilot automates SOC 2, HIPAA, and ISO compliance, saving you time and accelerating audits. * 

📰 News

Disney+ is launching short-form videos this year

Amazon’s AI assistant comes to the web with Alexa.com

Nvidia launches Alpamayo, open AI models that allow autonomous vehicles to ‘think like a human’

Meta bought Manus AI

💸 Fundraising

Anthropic reportedly raising $10B at $350B valuation.

xAI Raises $20B Series E

Clinical voice AI startup Tucuvi raises $20M

AI evaluation startup LMArena raises $150M at $1.7B valuation

* sponsored

Fail(St)ory

Ynsect Meat

Ynsect was supposed to be the company that made bug protein mainstream. It raised over $600 million and still ran out of road. This week, the French startup entered judicial liquidation.

If you remember the name, that’s not an accident. Ynsect had celebrity buzz, state backing, and a climate-friendly story that fit the moment perfectly.

What killed it was a series of very expensive decisions made before the business was ready.

What Was Ynsect:

Ynsect started in France in the early 2010s, when insect protein was already a known idea but still small-scale. The company aimed straight for industrial production.

The model centered on farming mealworms in controlled facilities and turning them into protein and oil. Those ingredients were sold mainly into animal feed, with pet food as a secondary outlet.

To compete, Ynsect invested early and heavily. Vertical systems, climate control, automation, custom plants. Most of the spending happened before demand was proven.

Animal feed was the first real target. It is the largest protein market and shaped how Ynsect designed its production. Insect protein was positioned as a substitute for fishmeal and soy.

However, the company stayed in animal feed longer than the economics allowed. Sustainability mattered to investors, but feed buyers still bought on price. Insect protein added cost where there was no margin for it.

Pet food emerged as a second path. It was less price-driven and offered better margins, making it a more workable market for insect protein. Ynsect kept both markets running at the same time.

That meant one cost structure serving two different businesses. The expectation was that scale would eventually fix animal feed while pet food supported margins. It never did.

In 2021, Ynsect acquired Protifarm and moved into human food ingredients. Revenue peaked at €17.8 million that same year, far below what the infrastructure required. Losses grew as more capacity came online.

The biggest bet was Ÿnfarm in northern France. Hundreds of millions went into a facility built for high-volume animal feed. That decision fixed the company’s strategy in place.

When demand fell short, Ynsect cut jobs and shut smaller sites. The main cost base remained. Cash ran out soon after.

The Numbers:

• 💰 Capital raised: $600M+

• 📈 Peak revenue: €17.8M in 2021

• 🔻 Net loss: €79.7M in 2023

• 🏭 Big bet: Ÿnfarm giga-factory

Reasons for Failure: 

• Strategic indecision baked into the org: Ynsect never chose its customer with enough conviction. Animal feed, pet food, and human food each demand different economics and execution. Trying to keep all three alive slowed every decision. The Protifarm deal added surface area without adding near-term cash. 

• Sustainability did not change buyer behavior: The climate story resonated with investors, regulators, and the public. It did not move purchasing managers in animal feed. Price per kilo still ruled. Insect protein could not beat incumbents at scale. No amount of impact framing fixes that.

• Building the factory before proving the model: Ÿnfarm locked Ynsect into a capital-heavy path early. Hundreds of millions were spent before unit economics were stable. Once the plant was built, flexibility disappeared. 

• A correct pivot that arrived too late: Focusing on pet food was the right call. It just came after irreversible bets were placed. By the time margins were prioritized, costs were fixed. Late clarity is still clarity, but it is rarely survivable.

Why It Matters: 

• Factories freeze your customer choice. Ynsect’s plant was built for animal feed volumes, which made later shifts toward pet food harder.

• Commodity buyers surface cost problems immediately. Animal feed pricing left no room to hide unproven unit economics.

Trend

AI Trends for 2026

It’s January, which means founders and analysts are publishing 2026 predictions like it’s cardio.

Most of it is junk. AGI timelines, apocalypse timelines, and the same “AI will change everything” line with different fonts.

Still, a few predictions repeat across credible places. When the same themes show up from Microsoft, IBM, MIT, Altman and Karpathy, it’s usually because teams are already spending money in that direction.

Prediction 1: AI becomes central to research

We already have a few clean cases where AI produces outputs researchers can use without squinting. 

• AlphaFold is the obvious one: DeepMind says it has predicted over 200 million protein structures and made them broadly available through the AlphaFold database.  

• In materials, DeepMind’s GNoME work reports 2.2 million candidate crystals and roughly 380,000 to 381,000 predicted stable structures, which expands the search space for what labs might synthesize next. 

• AlphaTensor found new matrix multiplication algorithms that beat prior approaches for certain sizes.  

• AlphaDev discovered faster small sorting routines that got integrated into LLVM’s standard C++ sorting library.  

What experts are betting is that this becomes a repeatable workflow, not a set of one-off wins. Peter Lee at Microsoft Research explicitly frames 2026 as AI generating hypotheses, using tools that control experiments, and collaborating with researchers, basically “joining the process of discovery.” 

Altman puts a sharper stake in the ground: “2026 will likely see the arrival of systems that can figure out novel insights.”  

Prediction 2: Multi-agent systems move into production

I think 2026 is when multi-agent setups move from “a cool pattern” to the default for complex work, and agents start coordinating with each other through shared protocols.

This is already headed that way. 

• Microsoft’s AutoGen is explicitly built around multiple agents conversing to accomplish tasks, not one agent doing everything.  

• Anthropic has published how they built a multi-agent research system, which is basically a real-world example of “handoffs” between specialist roles.  

• Google introduced an Agent2Agent protocol (A2A) specifically for agents to communicate, exchange information securely, and coordinate actions across platforms.  

If this holds, teams stop wiring every system as a one-off. They plug agents into a common “talk to other agents” layer, then worry about governance, security, and tracing who told who to do what. 

Prediction 3: Health AI gets productized for consumers

According to a recent analysis by OpenAI, healthcare questions are one of the most common things people use ChatGPT for. They say over 230 million people globally ask health and wellness questions on ChatGPT every week. 

Before the start of the year, Microsoft researchers predicted that in 2026, “AI is poised to shrink the world’s health gap,” moving beyond diagnostics into things like symptom triage and treatment planning, and showing up in products used by millions. 

Well, it seems Microsoft was right on this one because a couple of days ago OpenAI launched ChatGPT Health. It’s a dedicated Health space inside ChatGPT where you can connect medical records and wellness apps like Apple Health and MyFitnessPal, so it can help you understand test results, prep for doctor visits, and make sense of options like insurance tradeoffs.  

I think this keeps expanding in 2026 because there’s still a lot of healthcare work that AI can realistically take on: clinic triage, documentation, follow-ups, and decision support inside existing systems. 

Prediction 4: Shopping shifts toward chat interfaces

I’ve been banging on this for a while. In November I discussed AI-Powered Shopping and said that assistants can’t be good shoppers without infrastructure like clean catalogs, reliable pricing and inventory feeds, and merchant APIs they can actually hit. 

This is already moving from theory to product. 

• Amazon built Rufus directly into the Amazon app and desktop experience, trained on its catalog, reviews, and Q&A, and positions it as a conversational way to compare products and get recommendations.  

• Google is pushing the same direction in Search: conversational shopping in AI Mode, responses that include price and reviews, and even “agentic checkout” plus an AI that calls local stores for inventory. 

MIT Technology Review’s writers are basically calling this becoming normal behavior: a chatbot that recommends, compares, finds deals, and increasingly handles the boring parts of buying. 

Help Me Improve Failory

That's all for today’s edition.

Cheers,

Nico

Hey - It’s Nico.

Welcome to another Failory edition. Huge thanks to all 35,000+ of you who have been reading Failory this year!

This will be this year’s last edition. We’ll be back next year with more content on building products from scratch and step by step workflows you can copy to validate ideas fast, ship simple products, and automate growth. Our goal is to help you turn ideas into working products without hiring and to automate growth tasks that take a lot of time and budget. Stay tuned!

Let’s get into today’s issue. If you’re short on time, these are the 3 key things you should know:

• Luminar, a public company making Lidar sensors for self-driving cars, has filed for bankruptcy — learn more below

• The best growth tactics of 2025

• Microsoft released Trellis.2, an image-to-3D model generator — learn why this matters below

This season, give yourself the gift of compliance done right ^AD

Delve’s AI Copilot makes SOC 2, HIPAA, and ISO compliance smoother than ever during the holiday rush. Copilot reads your policies, gathers evidence automatically, completes tasks, and keeps every requirement moving so your team can stay focused on shipping and growth instead of audit prep.

The proof is magical.

• Lovable achieved SOC 2 in just 20 hours.

• WisprFlow closed Mercury and Superhuman in 1 month after accelerating compliance with Delve.

• 11x unlocked $1.2 million in ARR by reaching enterprise readiness faster.

If you want to wrap up the year with confidence and a clean audit path, Delve gives you everything you need to get compliant on time.

Use code MBXMAS for $1,500 off until December 31.

Book your demo now

This Week In Startups

🔗 Resources

The best growth tactics of 2025

Who cares about Mission and Vision?

The Founder’s Playbook for Performance Culture

Stop guessing. Start learning from real customers and accelerate your path to product-market fit. * 

📰 News

Apple opens up its App Store to competition in Japan

Luma releases a new AI model that lets users generate a video from a start and end frame

Google launches Gemini 3 Flash

OpenAI released a new image generation model: GPT Image 1.5

💸 Fundraising

Vibe-coding startup Lovable raises $330M at a $6.6B valuation

Edison raises $70M seed for its AI-doing-science platform

Call center chatbot startup PolyAI raises $86M at a $750M valuation

Nuclear startup Last Energy raises $100M for its steel-encased micro reactor

* sponsored

Fail(St)ory

Eyes for Cars

Luminar was supposed to be the lidar company that made self-driving cars safer.

Last week, it filed for Chapter 11 after layoffs, lawsuits, and losing the one customer that mattered. 

What Was Luminar:

Luminar Technologies built lidar sensors for autonomous cars. 

Most autonomous systems today lean heavily on cameras. They’re cheap, familiar, and good enough for lane keeping and basic driver assist. But cameras struggle when the sun is low, when headlights flare, or when you’re asking software to judge distance at highway speed.

Radar fills some of that gap. It works in fog and rain and can track objects far ahead, but it doesn’t see shape very well. A motorcycle, a car, and a guardrail can blur together.

Lidar sits in between. It sends out laser pulses and measures how long they take to come back, building a depth map of the road in front of the car. The problem is that they are expensive, power-hungry, and hard to manufacture at scale. This was what Luminar was trying to solve.

Investors bought into that bet. Luminar went public in 2020 through a reverse merger.  It signed Volvo, pulled in Mercedes-Benz, and quickly grew into a $3 billion company on expectation more than delivery.

Volvo was the anchor. Safety is its whole identity, and lidar fit neatly into that story. What started as a small 2020 order kept expanding. First tens of thousands of sensors. Then hundreds of thousands. By 2022, Luminar was planning around 1.1 million units.

That commitment reshaped the company. Luminar staffed up, locked in manufacturing capacity, and spent heavily to meet projected demand. The business assumed Volvo volumes were real, durable, and imminent.

They weren’t. Mercedes-Benz exited in late 2024, pointing to missed requirements. That left Luminar exposed. Volvo wasn’t just a customer anymore. It was the business.

Then Volvo walked too. It terminated the supply agreement, citing contract failures and supply chain risk. It also made clear it could ship advanced safety features without lidar. 

After that everything collapsed fast. Luminar cut roughly a quarter of its staff, the founder stepped down as CEO, and last week the board chose to file for Chapter 11 bankruptcy.

The Numbers:

• 🚗 Peak valuation: ~$3B after 2020 reverse merger

• 👥 Layoffs: ~25% of staff, second cut in one year

• ⚖️ Assets: $100M–$500M

• 🧾 Liabilities: $500M–$1B

Reasons for Failure: 

• Extreme customer concentration: Volvo wasn’t just a big customer. It was the business. When Mercedes exited, Luminar had no buffer, no second anchor to absorb the shock. A single termination turned an operating problem into an existential one.

• Scaling ahead of proof: Luminar made massive upfront investments based on projected demand, not shipped reality. Equipment, facilities, and headcount were built for volumes that depended on one customer staying happy forever. 

• Leadership instability and distraction: At the worst possible moment, the company lost its center. The founder-CEO abruptly stepped down in May after an internal ethics inquiry, then stayed on the board, leaving everyone guessing who was really in charge. Months later, he launched a new venture and tried to buy the company back from the outside, while Luminar was already in distress.

Why It Matters: 

• Hardware startups don’t get mulligans on delivery. If you miss once, customers quietly design you out. 

• Safety narratives don’t override supply chain risk. Automakers will choose “good enough and reliable” over “best and unpredictable” every time.

Trend

Trellis.2

We have all been glued to image and video models lately. Sora. Nano Banana, Veo. Image and video have gotten insanely good in the last year.

But another category has been quietly catching up, mostly out of the spotlight: AI-generated 3D models.

This week, Microsoft released Trellis.2. And no, this isn’t just another model drop. It’s one of those moments where you stop and think, okay, this is starting to actually work.

That’s why it’s worth talking about.

Why it Matters

• 3D is not niche anymore. If 3D generation gets cheap and reliable, it leaks into marketing, commerce, AR, and product validation. Not just games and VFX.

• This changes who can build with 3D. You no longer need a 3D artist for every experiment. That collapses time, cost, and iteration loops for founders.

• It turns one asset into a content engine. A single 3D model can power ads, PDPs, AR previews, and social clips. That kind of reuse doesn’t exist with flat images.

What is Trellis.2

At a high level, Trellis.2 takes an image and turns it into a usable 3D object. Not a screenshot of a render. An actual 3D asset you can light, rotate, and drop into other tools.

The key difference is quality where it counts. Generated objects don’t fall apart the moment you inspect them. Materials behave like materials. Shapes don’t collapse if they’re slightly weird or asymmetrical.

Twitter tweet

This feels like a DALL·E 2 moment for 3D. Not because it’s perfect, but because it crosses a threshold. Things that were painful or impossible a year ago are suddenly easy enough to try.

The Space

But Trellis.2 isn’t the only player here. This space is suddenly full of teams shipping real 3D generation products, and it’s moving fast.

• Tencent Hunyuan 3D: A “commercial-grade” 3D generation engine that Tencent is pushing globally, with enterprise API access via Tencent Cloud.

• Meshy: One of the most practical tools right now, with APIs for text-to-3D and image-to-3D that are clearly built for integration into real apps and workflows.

• Luma AI Genie: The text-to-3D model from Luma AI.

• Tripo AI: A 3D model generator that leans hard into speed and multiple input modes (text, single image, multiple images, doodles).

The Trend

The real shift here is not “AI can do 3D now.” It’s that 3D is turning into a cheap, repeatable input instead of a handcrafted output. That unlocks very specific businesses that didn’t pencil out before.

Here are a few that feel interesting to me:

• AR without the AR team: Most companies want AR previews. Almost none want to hire AR specialists. A startup that takes a few product photos and spits out AR-ready assets for web and mobile closes that gap. Furniture, home decor, appliances, even B2B equipment. 

• The ‘Figma for physical products’ layer: Early-stage founders mock software in Figma. Physical products don’t have that equivalent. With fast 3D generation, you can sketch a product, spin it, test reactions, and kill bad ideas early. No CAD. No manufacturing. Just validation.

•  3D cleanup tooling:  Generated 3D assets often break rules: too heavy, wrong formats, missing materials, bad geometry. A startup that automatically fixes or flags issues before assets hit production saves time and money.

Help Me Improve Failory

That's all for today’s edition.

Cheers,

Nico

Hey - It’s Nico.

With 2026 just around the corner, we have some exciting plans for next year. We’ll share more hands-on content to help you become a better founder as AI keeps evolving. If there is something you want us to cover or a way you are hoping to use AI in your work, feel free to reply and let us know (we’ll read every answer).

Let’s get into today’s issue. If you’re short on time, these are the 3 key things you should know:

• Yara AI, an AI therapist, shut down because its founder considered it dangerous — learn more below

• How AI is transforming work at Anthropic

• Poetiq, a 6-person AI startup, just beat Google in one of the most challenging reasoning benchmarks — learn why it matters below

A huge thanks to today’s sponsors, Spark Mail and Delve. Be sure to grab their deals while they are live!

Stop letting your inbox run your day ^AD

Email is supposed to be where decisions get made. But often, it’s where work stalls. Threads spin out, follow-ups are forgotten, and opportunities lose steam. 



Spark’s AI Assistant helps you keep work moving. Ask a question, and it checks your emails, attachments, events, and meeting notes to find what you need — fast. 



Your assistant can:

• Summarize long threads into action items

• Find key dates, decisions, and details

• Draft replies in your voice

• Check calendars and schedule meetings

Spend less time searching your inbox and more time building your business.

Get Spark for free!

This Week In Startups

🔗 Resources

How AI is transforming work at Anthropic

The AI churn wave: A data-driven investigation into retention for AI apps

Your product ideas probably suck (that's ok)

With Delve’s AI Copilot, get compliance done right this season with the fastest way to finish SOC 2, HIPAA, and ISO before year end. Book your demo now and get 1500$ off* 

📰 News

• OpenAI fires back at Google with GPT-5.2

• Disney signs deal with OpenAI to allow Sora to generate AI videos featuring its characters

• Mistral AI surfs vibe-coding tailwinds with new coding models.

• Google will launch its first AI glasses next year.

💸 Fundraising

• AI startup Serval valued at $1 billion after Sequoia-led round to expand IT automation

• AI-for-Science startup ChemLex raises $45M.

• Oboe raises $16 million for its AI-powered course-generation platform

• Wearable startup NeoSapien raises $2M.

* sponsored

Fail(St)ory

Shrink in a Server

A couple weeks ago, Yara AI shut down. It was an AI therapist with real users, real traction, and enough runway to keep going.

But the founder killed it anyway because he decided the product was too dangerous to exist.

Its story says a lot about where AI headed, and what AI can’t handle right now.

What Was Yara AI:

Yara was a chat-based mental health companion. 

You’d open the app and start chatting about whatever was going on: stress at work, anxiety, relationship stuff. It responded with gentle prompts, CBT-style reframes, follow-up questions.

Its target wasn’t crisis care. Yara explicitly aimed at users with mild anxiety, burnout, or trouble sleeping, people already aware of their issues and just looking for coping tools.

Under the hood, it added a “clinical brain” on top of generic LLMs. That included semantic memory, safety layers, and long-term conversation tracking, so Yara could remember what you talked about last week and build from there.

It was small and bootstrapped, with a few thousand users and less than $1M raised. But the usage patterns were real. People weren’t just testing it, they were treating it like real therapy.

And that was the problem.

The product wasn’t built for people in crisis, but they showed up anyway. Some just venting. Others on the edge. You don’t get to control who opens your app.

The team tried to fix it. They built a mode-switch system: one track for everyday emotional support, another to route crisis users out to real help.

It still didn’t feel safe.

The core issue wasn’t the handoff flow. It was that LLMs are built to generate plausible-sounding text, not detect long-term emotional risk. They don’t track patterns. They can’t tell when someone’s getting worse over time. They just predict the next sentence.

By the end, the founder wasn’t reacting to something that went wrong. He was looking at what could go wrong, and decided to shut it down before it did.

The Numbers:

• 🧠 Founded: 2023

• 👥 Users: Low thousands

• 💸 Funding: <$1M (bootstrapped)

• 🛑 Shutdown: Fall 2025

• 🧰 Final move: Open-sourced safety scaffolding + mode-switch system

Reasons for Failure: 

• You don’t get to pick your users: Yara was built for low-risk, high-intent users, people looking to reflect and grow. But real usage didn’t stay in that lane. People in crisis showed up anyway. When your product talks like a therapist, people will treat it like one.

• The space got toxic: When Yara started, AI therapy was a grey area. A year later, it was a regulatory minefield. Illinois banned it. Lawsuits like Raine v. OpenAI put real names behind the worst-case scenario. OpenAI casually dropped the stat that a million people express suicidal ideation to ChatGPT every week. Suddenly, this wasn’t a cool wellness app, it was a liability with a ticking clock.

• The tech looked better than it was: LLMs are great at playing the part. They sound like they understand. That’s what made Yara compelling, but also what made it dangerous. It was just convincing enough to be trusted, but not reliable enough to earn that trust. That gap is where real harm lives.

Why It Matters: 

• Some products can’t be iterated. Most startups learn by shipping fast and fixing mistakes. Yara couldn’t afford a single one. That’s a fundamentally different game.

• The market’s already here, whether it’s safe or not. Millions are using AI chatbots for therapy today. The tools are fragile. The guardrails are weak. And there’s no regulation catching up in time.
  

Trend

The 6 person team that beat Google

Gemini 3 has been sitting at the top of every reasoning leaderboard since it launched. Everyone assumed it would stay there for a while.

But, last week, a tiny team showed up, said “hold my beer,” and quietly took the crown on ARC AGI 2, one of the most difficult reasoning benchmarks.

The twist is that they didn’t beat Google with a bigger model, but with a smarter wrapper.

Why it Matters

• Small teams can now punch way above their weight. You don’t need a GPU farm or a frontier model to move the needle; you just need a clever system that forces existing models to think harder.

• The moat is shifting to architecture. Everyone has access to the same LLMs, so the edge comes from how you structure reasoning, feedback, and search.

What Happened

Six months ago, ARC AGI 2 was a graveyard. Most models sat under 5 percent accuracy. Researchers said cracking 50 percent could take years. 

Then Poetiq showed up. A six person team, 173 days old, no custom model, no secret sauce compute deal. And they landed just over 50 percent on ARC AGI 2’s semi private set. A micro startup walked into one of the hardest reasoning tests and outscored every frontier lab.

Let’s pause here because ARC AGI 2 matters. It’s a set of tiny grid puzzles where you get a few example inputs and outputs, and you have to infer the rule that connects them.

The thing that makes it so difficult for LLMs is that every task is new: no patterns to memorize, no familiar structure to lean on, no way to bluff your way through with “vibes.” You actually have to infer the rule step by step.

And almost every model flops. They fire off one or two guesses, feel confident, and move on. There’s no systematic search, no feedback loop, no adaptation when the first attempt fails. 

Poetiq, however, was able to crack the test.

What Poetiq is

Poetiq didn’t build a new model. They built a controller that sits on top of existing ones and forces them to think in steps instead of guessing once and calling it a day.

It works with any LLM. Gemini 3, GPT 5.1, Grok, open weights. The controller just calls the model, critiques the output, and tries again. The loop is the actual intelligence, not the base model.

The process is simple: propose a move, check it against the examples, kill the bad ideas fast, refine the promising ones. A few iterations of that and the system usually lands on something consistent. No retraining needed.

Because it’s model agnostic, it upgrades instantly. Swap in a better LLM and the whole thing improves without touching a single weight. And it stays cheap because the controller keeps the search small and stops early when it’s confident.

The Trend

The real story here is that Poetiq just proved something important: small teams can push the frontier without training a frontier model. That’s new. And it changes who gets to play.

This result shows there’s real value in building meta systems and refinement layers. Not theoretical value. Not “cool research.” Actual SOTA from a 6 person team. It’s the clearest signal yet that orchestrating models is its own field.

We’re going to see a wave of small teams copying this playbook. Take off the shelf LLMs. Add structure, feedback, and controlled search. 

And this is where it gets interesting. Researchers thought 50 percent on ARC was years away. Then a tiny team did it in months. If meta systems keep improving, the gap between “expected progress” and “actual progress” is going to blow open.

If more groups start building these reasoning layers, we could see problem solving advance much faster than anyone predicted. Progress accelerates when the frontier isn’t controlled by a few companies.

Help Me Improve Failory

That's all for today’s edition.

Cheers,

Nico

Hey - It’s Nico.

Welcome to another Failory edition.

This week, we noticed we’d been paying a subscription for our testimonial wall for 2 years just to show a few tweets. To fix that, we built a simple Wall of Love Generator, which allowed us to cancel the recurring fee. Feel free to use it to easily embed your social proof.

Here are today’s top 3 highlights:

• GummySearch, a startup helping founders and marketers uncover insights from Reddit threads, has shut down - learn more below

• A guide to crafting your brand voice

• Nano Banana Pro enables founders to build tools that weren’t possible before - learn why this matters below

Today’s sponsor, Delve, helps founders automate security questionnaires with an AI browser agent that turns hours of tedious work into simple clicks.

The AI Browser Agent That Ends Security Questionnaire Pain ^AD

Ever had to fill out a gnarly security questionnaire inside a vendor’s portal and hated your life?

Delve just launched an AI browser agent that does it for you.

It finds answers from your policies, past questionnaires, and docs, then fills everything in while you click next. Hours → minutes. Switch to Delve for compliance and use this - and every other gamechanging AI features at one simple cost.

If you’re using another platform for compliance, Delve will migrate you over for free. Failory readers get $1,000 off compliance using the link below.

Set it and forget it >>> sweating questionnaires.

Book a 30-minute demo to get started

This Week In Startups

🔗 Resources

A Guide To Crafting Your Brand Voice

How to price and sell a product when creating a new category

One year with ChatGPT Pro as a first hire

How ‘fofr’ built their blog from scratch using Gemini 3

💸 Fundraising

Cybersecurity Startup 7AI Raises $130 Million in Series A Funding

AI startup Flex raises $60 million to offer finance tools for mid-sized businesses

Black Forest Labs (FLUX) Secures $300 Million Series B Funding

Fail(St)ory

When an API Pulls the Plug

Indie SaaS product GummySearch is winding down.

After four years of helping founders, marketers, and investors uncover insights from Reddit threads, GummySearch announced last month it would stop accepting new signups and renewal payments as of November 30th, 2025, and fully shut down on December 1st, 2026. Existing paying users will retain access until their billing cycles end, but the business itself is entering its final chapter.

Twitter tweet

In the announcement from GummySearch, the startup’s founder explained the reason for shutdown: GummySearch was unable to reach an agreement with Reddit to obtain a commercial license for compliant access to Reddit’s Data API.

What Was GummySearch?

Launched in 2021 by solo founder Fed, GummySearch was a lightweight audience and market research tool that pulled real-time insights from Reddit. It helped users validate ideas, monitor niche communities, study customer sentiment, and track emerging trends, all by surfacing what people were saying in subreddit conversations.

The Numbers

• Founded: 2021

• Team: 1

• Subreddits tracked: 10,000+

• Revenue: $35K MRR

• Duration: ~4 years (2021-2025)

Reasons for Failure

• No commercial API agreement: GummySearch relied on Reddit’s API for its core functionality. Despite attempts at negotiation, the company could not reach a commercial licensing agreement, meaning GummySearch would need to become a non-commercial product or shut down. As the sole founder of a for-profit platform, GummySearch’s founder chose the latter.

• The cost of being “legit”: Many products still technically use Reddit data without a commercial license, but Fed made it clear he didn’t want GummySearch to be one of them. Running a business “looking over your shoulder every day,” as he put it, wasn’t sustainable, especially when the company would be running the risk of a sudden API access cut-off affecting paying customers.

• Free tier economics: While GummySearch did offer a free tier, it relied on revenue from paid subscriptions to stay operational. Without paying users, it simply wasn’t possible for the company to continue offering a free version of the product, resulting in the full closure of all new signups and renewals.

• No viable pivot without risking quality: GummySearch was Reddit insights, and an attempt to rebuild it around other platforms or scraping alternatives could compromise quality and reliability. Fed stated he would potentially be open to selling the product to Reddit itself or to a company that has a Reddit Commercial API license to maintain the quality.

Twitter tweet

Why It Matters

APIs remain one of the most powerful yet fragile foundations of many startups. When you build on someone else's platform, you rely on their policies, pricing power, and risk factors. GummySearch wasn’t out of customers, demand, or product-market fit - it simply lost its ability to operate commercially under the constraints of its core data partner.

Indie SaaS can be resilient, but third-party platform dependency is a major risk. Even a healthy, beloved tool with a strong user base can be forced to sunset when upstream providers change course.

Transparency leads to a graceful exit. Fed’s open communication, long wind-down period, and insistence on honoring existing customers show how to handle a shutdown the right way.

Trend

AI Image Processing Goes Pro

Two weeks after its launch, Nano Banana Pro is starting to spark a quiet shift in the AI ecosystem. Not because it simply generates nicer pictures than the many other AI image tools out there, but because it enables founders to build entirely new categories of visual tools that weren’t possible before and startups to streamline their image production like never before.

For the first time, an AI model can reason about objects, layouts, labels, physics, and context with a level of precision that allows it to power real products. If earlier AI image tools were “creative assistants,” this one is closer to a visual engine - one that can take real-world inputs and turn them into structured, functional outputs.

Twitter tweet

What Nano Banana Pro Is

Nano Banana Pro runs on Google’s Gemini 3 Pro AI Model, and it’s the first widely accessible AI image processing tool that can generate, edit, adapt, and analyze images with professional reliability. It understands perspective, lighting, material physics, and spatial relationships well enough to maintain consistency across hundreds of outputs.

Some examples of what the model can potentially do:

• Read and rewrite text inside an image

• Generate accurate diagrams and step-by-step overlays

• Maintain brand consistency across thousands of variants

• Localize scenes, environments, and packaging for any region

• Infer missing details when restoring or cleaning up photos

Twitter tweet

Imagine dropping in a product shot and generating a hundred localized versions with appropriate backgrounds, signage, and cultural details, or offering a visual tech support app that provides users with step-by-step image overlays for troubleshooting and repairs.

These weren’t realistic use cases for error-prone novelty AI image generators, but they are now with Nano Banana Pro.

Why It Matters

Nano Banana Pro turns AI image generation into a reliable visual engine. It lets startups automate visuals with the same trust they place in AI models for text.

Entirely new product categories become buildable. Think real-time financial charts, global campaign localization engines, visual tech support SaaS, contextual architectural rendering tools, dynamic QR menu boards, and more.

Twitter tweet

Visual workflows can now be automated end-to-end. Small teams now have the capabilities to complete tasks that used to require large production departments.

Startups can ship features faster and validate ideas earlier. Consistent, editable, brand-safe images can now be generated on demand without sacrificing accuracy.

Visual consistency at scale becomes a default. This is critical for marketing automation, enterprise tooling, technical documentation, training content, and any product that needs structured, repeatable imagery.

Help Me Improve Failory

That's all for today’s edition.

Cheers,

Nico

Hey - It’s Nico.

Welcome to another Failory edition and Happy Thanksgiving to those celebrating!

We’ve been working on new ways to bring more value to founders like you, and this Max MRR Tool is the first step (built with Google’s AI Studio). It shows your startup’s revenue ceiling and what drives it. More tools coming soon.

Here are today’s top 3 highlights:

• goConfirm, a startup trying to make online stranger-to-stranger transactions safer, has shut down — learn more below

• The Complete Guide to AI Prototyping Tools

• OpenAI and Perplexity have launched AI shopping assistants — learn why this matter below

Today’s sponsor, Delve, helps founders automate SOC 2 compliance with AI agents that turn a months-long headache into a 15-hour sprint. Also, shoutout to Cloudways, grab their Black Friday deal for effortless site migrations and big savings.

🦃 Get SOC 2 Compliant & Stuff Your Pipeline ^AD

This Thanksgiving, we’re serving up more than turkey, we’re helping you stuff your pipeline.

Get $1,500 off compliance + a $500 Visa gift card when you become SOC 2 compliant with Delve before November 27th.

Because while your sales team is feasting, your deals shouldn’t be starving. Delve’s AI agents handle all the heavy lifting: automating your SOC 2 evidence, risk checks, and audits in record time. No spreadsheets. No manual madness. Just compliance done fast, so you can focus on your next big win.

The results speak for themselves:

• Lovable → SOC 2 in 20 hours.

• Wisprflow → Signed Mercury & Superhuman in 1mo

• 11x → $1.2M ARR unlocked.

Use code THANKSGIVING before 11/27 at 11:59 PM PST.

👉 Book your demo

This Week In Startups

🔗 Resources

Supercharge your site with high-performance cloud hosting. With Cloudways enjoy 50% off for 3 months and 50 free migrations. Activate the deal now! *

The Complete Guide to AI Prototyping Tools

Designing AI-Native Software

How to Price and Package Your Gen AI Feature

📰 News

Anthropic releases Opus 4.5 with new Chrome and Excel integrations.

Sam Altman talked about OpenAI’s forthcoming AI device.

X begins rolling out the ‘About this account’ feature to users’ profiles

Warner Music signs deal with AI music startup Suno.

💸 Fundraising

Momentic raises $15M to automate software testing

Onton raises $7.5M to reinvent the way the world discovers and decides what to buy.

AI workflow startup Model ML raises $75 million

Nuclear energy startup X-energy raises $700M in funding

* sponsored

Fail(St)ory

goConfirm

You know the feeling when you’re buying something on Marketplace or meeting a stranger from a classifieds site. The doubt. The “please don’t scam me” energy. goConfirm tried to bottle that anxiety and turn it into a business.

But they ran into a classic problem: the product everyone swears is a great idea is often the product no one actually uses when it counts.

What Was goConfirm:

goConfirm went after the oldest P2P pain point: you never know who you’re dealing with. Their answer was a portable verified identity you could bring into any transaction. Not tied to Facebook, or Kijiji, or Reddit. Just you, verified once, re-used everywhere.

The setup was simple. Scan your ID. Take a selfie. Wait a moment while the system confirmed you were real. After that you had a ConfirmID you could show to anyone, like a digital passport for internet deals.

They bundled that identity with a set of tools meant to make deals feel safer. Verified chat. A “Vault” so you weren’t sending bank details through random DMs. And a small safety net: 250 dollars of loss protection if something went wrong inside the app.

To get traction, they pushed hard into the places where scams are most common. Facebook groups for rentals. Local buy-and-sell communities. And especially event-ticket subreddits, where fake tickets were a daily complaint. They cold-messaged mods. They helped communities create “goConfirm-verified” flare

For a while, it worked at the micro level. The people who actually used goConfirm seemed happy. They told the team the product made sense. But happy users don’t guarantee real growth.

Internally, the growth chart never bent. It just crawled upward in a straight line. Slow, predictable, and far from the type of compounding curve that justifies venture funding. A few weeks ago, the team finally called it and shut the thing down.

The Numbers:

• 📍 Founded: 2022 (as Qui Identity)

• 💸 Funding: 6.5M dollar seed round

• 📱 Launch: Late 2023

Reasons for Failure: 

• It added friction, not convenience: Buying or selling on a classifieds site is already a pain. goConfirm forced people to reroute deals through an extra app. That’s friction. For many, it was too much. 

• It never got embedded where it mattered: goConfirm lived outside the marketplaces. No integration into Facebook Marketplace, rentals platforms or ticket sites. So to use it, both buyer and seller needed to care. That rarely happens. Trust works when it’s baked in. If it’s optional, it’s ignored.

• The trust problem wasn’t painful enough to justify extra steps: Most P2P transactions don’t end in scams, even if it feels like they might. That means the fear is high, but the actual loss rate is low. When the real risk is low, users won’t tolerate friction. The “better safe than sorry” pitch only works when sorry happens often.

Why It Matters: 

• “It’s such an obvious idea” is not a business model. Some ideas make perfect sense in theory but collapse when real users have to change habits to use them.

• Friction kills good ideas. People say they want safety, but they choose whatever gets the deal done fastest.

• Optional trust layers get ignored. If the marketplace doesn’t bake you in, you’re just extra work for both sides.

Trend

AI-Powered Shopping

People have been talking about AI-powered shopping for a while. Adobe’s data showed a 4,700 percent jump in generative-AI referrals to U.S. retail sites this year, which is insane. But even with numbers like that, the actual behavior always felt early. Lots of curiosity, not a real habit.

This week felt different. ChatGPT launched Shopping Research. Perplexity launched Shop with Perplexity. Both of them push people closer to a world where you ask an AI what to buy, it does the work, and you never touch a search page. It’s starting to feel like the default path, not a fringe behavior.

Why It Matters

• People want help with real decisions. Not top-10 lists. Not filters. They want something that understands constraints and narrows choices. That’s where intent is forming now.

• Memory finally matters. When the assistant uses your preferences instead of your browsing trail, the results feel relevant instead of manipulative. Intent quality goes up. Noise goes down.

• The whole funnel is compressing. Discovery, research, and checkout are drifting into one thread. If the conversation becomes the place where decisions happen, the assistant becomes the new storefront.

The Two Launches

ChatGPT Shopping Research gives users a structured way to make bigger purchase decisions. It asks for context, pulls recent data from the web, and returns a clear breakdown of options, differences, and tradeoffs. It’s built for the moments when you want to actually understand the choice rather than skim.

Twitter tweet

Perplexity’s update works in a similar way. It keeps context across questions, uses past searches as signal, and connects everything to PayPal checkout. You discover, refine, and buy without breaking the flow. It’s direct and transactional in a way that feels closer to how people actually shop.

Twitter tweet

Both launches point to the same thing: the buying process is moving into the assistant itself. Users stay in one place from “what should I buy” to “I’m done.”

The Trend

The bigger trend is simple: shopping is turning into guided decision-making. The assistant becomes the place where you figure out what you want before you ever land on a retailer’s site.

That raises the obvious question: do generalist assistants (GPT, Perplexity) win the whole thing, or is there room for vertical specialists that focus on specific niches?

There’s room for both. Generalists will dominate broad intent, but verticals can win the categories where nuance or taste matter. Home-decor, fashion, beauty, hobbies, equipment-heavy categories. These are places where you want depth, not generic reasoning.

Take Onton, for example. It focuses only on home decor. You describe a piece you’re looking for or upload a picture, and it finds products that match the style. Because it’s built around one category, the answers end up tighter than what a general model produces. Taste and constraints matter in these verticals, and specialization shows.

Founders should look at the gaps this shift is opening:

• Vertical decision engines: Pick a category with real complexity and build the expert layer general models won’t. Skincare routines, running gear, audio setups, camera systems. Depth beats breadth here.

• Taste systems: A persistent style profile that assistants can call. Fashion, furniture, decor. People want consistency, not random output.

• Budget allocators: Most people think in terms of a total spend. “Build me the best $1,500 home office setup.” This should exist everywhere, but it doesn’t.

• Local-aware shopping: Tie real-world constraints to recommendations: stock, distance, weather, timing. Assistants aren’t doing this well yet.
  

Then there’s the infrastructure layer this ecosystem will require. Models need clean catalogs, specs, pricing feeds, images, availability data, and merchant APIs. 

A few companies are nibbling at this (Manifest AI, Alhena, Amio), but the space is wide open. Opportunities include:

• Standardized product data pipelines that make catalog ingestion painless.

• Reliable inventory and pricing layers assistants can query.

• Merchant-side conversion tools that pre-answer the questions buyers always ask.

Help Me Improve Failory

That's all for today’s edition.

Cheers,

Nico

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:

• DappRadar, the Crunchbase for dApps, has shut down — learn why below

• Application Software Is Dead, Again

• Stack Overflow is remaking itself into an AI data provider.

• AI music startup Suno raises at $2.45B valuation on $200M revenue.

• Google just released Gemini 3, a new IDE and and upgraded Nano Banana model — learn more below

Today’s sponsor, Silly Money, helps founders cut their 2025 tax bill with straightforward strategies to keep more of what they earn. Also, shoutout to Cloudways, grab their Black Friday deal for effortless site migrations and big savings.

Stop overpaying on taxes in 2025 ^AD

Let’s be real: Your biggest expense every year isn’t your tools or your team.

It’s your taxes, and most founders pay wayyy more than they should.

Written by Ankur Nagpal (who sold his last company for $250M), The No BS Guide to Paying Less in Taxes breaks down the exact strategies wealthy business owners actually use:

• Smarter entities

• Deduction stacking

• Retirement tax hacks

• And moves that 17+ more tactics to cut taxes fast

With December 31st fast approaching, now’s the time to act.

Get the 28-page guide to keep more of what you earn

[Download For Free]

This Week In Startups

🔗 Resources

Host your website on lightning-fast cloud servers and scale without limits. Get 50% off for 3 months plus 50 free migrations with Cloudways. Claim your offer today! *

Application Software Is Dead, Again

The new UX era

Collaboration sucks: Be the driver

📰 News

Amazon’s Prime Video is getting AI-generated video recaps for TV shows.

Stack Overflow is remaking itself into an AI data provider.

TikTok will let you choose how much AI-generated content you want to see.

ChatGPT launches a pilot group chats feature.

💸 Fundraising

AI music startup Suno raises at $2.45B valuation on $200M revenue.

AI customer support agent startup GetVocal raises $26M.

AI cloud operator Lambda raises $1.5B+ to build data centers.

Luma AI Raises $900M to build “Supercluster” in Saudi Arabia

* sponsored

Fail(St)ory

The Crypto Tracker

Seven years of tracking on-chain activity ended with a short post on X: they were shutting it all down. DappRadar became the default tab for a lot of crypto people, and still couldn’t justify a $15k monthly burn. 

The problem was never traffic, it was how the money flowed in and out.

What Was DappRadar:

DappRadar started in 2018 as a simple idea: show what people are actually using on-chain, in one place. It turned raw blockchain noise into rankings and dashboards that didn’t require a PhD to read. Sort of like a Crunchbase for dapps.

The scale they reached was real. By 2025, they were tracking 18,111 dapps across 93 blockchains for roughly 500,000 monthly users. If you cared about on-chain games, NFTs, or dapp usage, you ended up there sooner or later. 

Monetization mostly leaned on their own token, RADAR, and their “subscription” model DappRadar Pro. 

PRO wasn’t a normal subscription though. Instead of paying monthly, PRO users staked RADAR to unlock deeper analytics and alerts.The setup felt clever on paper, but most of the economic weight stayed inside the token rather than turning into cash flow.

They also sold attention to teams looking for oxygen. Banners, sponsored writeups, newsletter spots, social pushes, the usual Web3 marketing menu.

By 2025 they had raised $7.33 million and built a loyal, steady base of users. The problem was the market around them shrinking fast. Crypto funding dropped 70 percent year over year, and their own treasury mostly lived inside a token that kept losing altitude.

The founders said the platform was “financially unsustainable”, and the numbers made the case for them. When the shutdown was announced, RADAR slid another 38 percent, erasing a chunk of the DAO’s remaining value. It turned into one of thspirals where each step speeds up the next.

The Numbers:

• 📅 Launched: February 2018

• 👥 Users: ~500,000 monthly

• 🌐 Coverage: 18,111 dapps across 93 blockchains

• 💰 Funding raised: $7.33 million total

• 🔥 Monthly burn: $15.5k

• 🪙 DAO treasury: $1.6 million total assets; only $46,162 in stablecoins

Reasons for Failure: 

• The treasury wasn’t real liquidity: Roughly 97 percent of their $1.6 million treasury sat in RADAR, a token already down more than 90 percent. Only 46k was usable stablecoins. When the shutdown news hit and RADAR dropped another 30 percent, the treasury effectively deflated in real time. 

• PRO didn’t behave like a subscription: Users unlocked PRO by staking RADAR, not by paying in stable value, which meant almost no recurring cash came in. Rewards and lock-up mechanics encouraged holding the token, not funding operations. Great for token optics, terrible for runway. 

• Their best-paying segments collapsed: DappRadar’s strongest commercial pull lived in gaming and NFTs. Both sectors shrank fast: hundreds of Web3 games dropped off the map and NFT market cap fell almost 45 percent in a month. The budgets they relied on for ads and promotions dried up right when they needed them most.

• The cost base assumed the bull market would last: Running a multi-chain data index is expensive, and by 2025 their burn was still about $15.5k a month with only three months of stablecoins left. They launched a $249 developer plan, but far too late to matter. Once most of your treasury is an illiquid token, cost-cutting becomes damage control, not strategy.

Why It Matters: 

• Your token can’t double as your treasury. If most of your runway moves with one chart, you’re not managing finances; you’re riding sentiment.
  

• A staking-based membership isn’t revenue. If cash never enters the system, you can’t pay real-world bills no matter how active the product looks.
  

• Cycles decide who survives. If your core market shrinks and your cost base doesn’t, the cliff shows up fast and there’s no graceful way down.

Trend

Google’s Busy Week

It was a busy week for Google. Not only did they drop Gemini 3, which now looks like the strongest model on the market, but they also rolled out a new agentic IDE and a serious upgrade to their image generator.

Let’s break down what shipped and why it matters.

Why it Matters

• Reasoning just leveled up again: Gemini 3’s jump in reasoning isn’t a leaderboard vanity metric. It pushes the ceiling for what you can reliably automate without babysitting. 

• The IDE war is now about who owns the whole stack: Antigravity signals that Google wants the editor, the agent, and the model under one roof. That kind of vertical control usually compounds fast.

• Image generation got a big upgrade: Nano Banana Pro behaves more like a proper creative tool: higher resolution, better control, and way fewer retries.

Gemini 3

Google released Gemini 3 on Tuesday, just seven months after Gemini 2.5. The pace alone tells you how aggressively they’re trying to close the gap with competitors. It’s their best model yet, and it landed only days after GPT 5.1, which made the comparison inevitable.

The numbers are solid: a 37.4 on Humanity’s Last Exam (previous high was 31.64) and the top spot on LMArena’s human satisfaction leaderboard. Benchmarks don’t decide everything, but they do shape defaults, and this one will become the new baseline for a lot of teams.

Early chatter is that the model feels more stable and less prone to shortcut answers, though it’s too early to tell where it outperforms in the wild.

Antigravity

Antigravity is Google’s new Gemini-powered IDE. It enters the same arena as Cursor and Windsurf, and the reaction so far is mixed. Some people say it’s exactly what they expected from Google; others say it’s not beating the incumbents yet. Fair enough for week one.

Twitter tweet

The interesting part isn’t whether Antigravity is better today. It’s how it shifts the category. Google controls the model, the agent layer, and the workspace, and that kind of vertical stack usually wins over time. 

Cursor 2.0 and Warp proved the pattern; Google is now trying to own it end to end. If they ship quickly, this becomes a real threat to everyone building agentic dev tools on top of someone else’s model.

Nano Banana Pro

A couple of days later, Google released Nano Banana Pro, its new image model built on top of Gemini 3 instead of the older Gemini 2.5 Flash. 

Twitter tweet

The upgrade shows up mostly in control and comprehension, and if you’ve wrestled with earlier image models, you can feel the difference immediately.

It now generates 2K and 4K images; you can dial in lighting, depth, camera angles, and overall scene structure without the model going off-script. It finally handles 16:9 without breaking faces or stretching artifacts across the frame.

And because it inherits Gemini 3’s reasoning, you can actually ask for edits and get revisions instead of a brand new, unrelated image.

Help Me Improve Failory

That's all for today’s edition.

Cheers,

Nico

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:

• Sonder, an Airbnb competitor worth $2B, has shut down — learn more below

• Ecosystem is the next big growth channel.

• SoftBank just sold its entire stake in Nvidia.

• Cursor raises $2.3B 5 months after its previous round.

• Spatial Intelligence seems to be the next big frontier in AI — learn more below

Today’s sponsor, Redrob, is helping teams replace generic AI outreach with real human insight through its Human Intelligence Lab. Also, shoutout to Cloudways, grab their Black Friday deal for effortless site migrations and big savings.

Your Prospects Can Spot AI From a Mile Away ^AD

That’s why your outreach feels invisible.

Everyone’s using the same AI tools, trained on the same data, writing the same “personalized” messages. Authenticity got lost in automation.

Redrob’s Human Intelligence Lab brings real researchers back into your process—uncovering paywalled data, social signals, and human context that make your outreach actually relevant and stand out.

And the best part? You can try it free, and keep using it for less than the cost of a single bad cold email.

Because real connection starts where automation ends.

Test the Difference

This Week In Startups

🔗 Resources

This Black Friday, scale smarter. Cloudways gives 50% off for 3 months + 50 free migrations — move all your sites effortlessly. Get the Deal today! *

Ecosystem is the next big growth channel.

How SaaS giants are rewriting the rules of value with AI.

The internet wants to be fragmented.

📰 News

Jack Dorsey funds diVine, a Vine reboot that includes Vine’s video archive.

Google augments AI shopping with conversational search and an AI that calls stores for you.

SoftBank just sold its entire stake in Nvidia.

Wikipedia urges AI companies to use its paid API, and stop scraping.

💸 Fundraising

Cursor raises $2.3B 5 months after its previous round.

Former Twitter CEO Parag Agrawal’s Al startup Parallel raises $100M.

Crypto startup Seismic raises $10 million to help fintechs protect customer data.

Wonderful raised $100M to put AI agents on the front lines of customer service.

* sponsored

Fail(St)ory

Hotel-Airbnb Hybrid

For years, Sonder looked like the smarter, sleeker cousin of Airbnb, a company that could finally bridge hotels and home-sharing. 

Designer apartments, digital check-ins, no hosts to text about missing towels. It was the dream version of hospitality: scalable, tech-driven, asset-backed.

And then, a couple of days ago, it all ended in a bankruptcy filing.

What Was Sonder:

Founded in 2014 in San Francisco, Sonder started with a seductive pitch: what if you got the consistency of a hotel and the personality of an Airbnb?

They leased entire apartment buildings, turned them into chic, uniform suites, and managed them like a hotel chain, but through an app. Sonder raised hundreds of millions and expanded to 37 cities worldwide, operating 9,000 apartment-style units from New York to London.

By 2021, Sonder went public via SPAC at a $2.2 billion valuation, riding the pandemic rebound and the travel tech hype. Airbnb’s stock was soaring, and Sonder seemed like the next logical bet.

And then, reality hit. By mid-2022, its stock dropped from $200 to under $50. Losses piled up. The company faced delisting threats, postponed shareholder meetings, and late financial filings. By early 2025, both the CEO and CFO had resigned, leaving a vacuum at the top.

Desperate for a turnaround, Sonder struck a deal with Marriott International in 2024, a licensing partnership that rebranded its apartments as “Sonder by Marriott.” The deal promised $126 million in potential liquidity.

But the integration became a disaster. Sonder’s tech couldn’t align with Marriott’s legacy systems. “Unexpected challenges” ballooned costs and delayed listings, slashing revenue instead of boosting it.

By 2025, Marriott declared Sonder in default and pulled its 140 properties and 7,700 rooms off its platform entirely.

The Numbers:

• 💸 Valuation peak: $2.2B (2021)

• 🏙️ Cities operated: 37

• 🏘️ Units managed: 9,000

• 📉 Stock price: $200 → <$1 (2022–2025)

• 💀 Q2 2025 net loss: $44.5M (3x prior year)

Reasons for Failure: 

• The asset-heavy model killed flexibility. Sonder’s “hotel-meets-Airbnb” pitch hid a brutal truth: owning or leasing space doesn’t scale like listing it. Every market expansion meant new leases, furniture, staff, and maintenance. When travel demand dipped or occupancy slipped, losses multiplied.

• The Marriott partnership backfired. The deal was supposed to legitimize Sonder and drive bookings. Instead, it became a tech nightmare. Integrating a nimble, cloud-native system with Marriott’s decades-old infrastructure caused endless sync failures and delayed listings. The result: months of downtime, lost bookings, and rising integration costs that drained its last reserves.

• Leadership drifted as the company spiraled. When your CEO and CFO both walk out before a liquidity crunch, investors smell smoke. Sonder’s executive churn reflected a deeper problem: a company with no clear path to profitability. It kept chasing growth metrics even as revenue per unit fell. 

• Venture-style growth economics clashed with hospitality reality. The VC world loves “blitzscaling.” Hotels don’t. Sonder tried to apply tech startup speed to a business where every unit requires physical upkeep, permits, and staff. Rapid expansion only amplified inefficiency. By the time investors realized the margins looked more like Marriott’s than Airbnb’s, the market had lost patience for asset-heavy dreams.

Why It Matters: 

• Tech doesn’t erase physical risk. Asset-heavy startups crash when they pretend they’re software.

• Partnerships can be poison. One bad integration can wipe out an entire turnaround plan.

• Cash flow still rules hospitality. Sonder learned that great branding can’t cover ugly unit economics.

Trend

Spatial Intelligence

Large language models can write symphonies in text, but can’t tell you how far the chair is from the table. They speak fluent abstraction, yet fail at physics.

The next big AI leap isn’t about bigger models or cleaner prompts, it’s about giving machines a sense of the physical world.

That’s the argument from Fei-Fei Li, one of the world’s most famous AI researchers. Her recent post, From Words to Worlds, frames “Spatial Intelligence” as the next frontier. 

The idea: today’s AI can describe the world, but not inhabit it. Even the best multimodal models can’t estimate distance or predict what happens if you push a cup off a desk. 

Fei-Fei’s fix is “world models”: generative systems that understand space, time, and physics. Where LLMs predict the next word, world models predict the next frame.

Why it Matters

• It turns AI from a text tool into an operational tool. Once models understand environments, you can automate workflows that were off limits because they involved physical context.
  

• It makes real world testing cheaper. When you can rehearse failures in simulation instead of on hardware or in the field, you cut cost and downtime immediately.
  

• New products become viable the moment consistency improves. AR, robotics, planning tools, and 3D content all get unlocked by models that keep a scene stable over time.

World Models

According to Fei Fei, a World Model is simply an AI system that can generate and maintain a consistent 3D environment over time. Instead of predicting the next word, it predicts the next moment.

The model should remember where objects are. It should understand solidity. It should know that if you open a drawer, the drawer stays open. And if you place a cup on a table, it doesn’t teleport away three frames later. Consistency is the key.

This sounds obvious until you look at what today’s models actually produce. A recent demo of an AI generated game built with Sora 2 made the rounds on X, and you could see the problem in seconds. The world kept slipping out from under itself.

Twitter tweet

The map warped unpredictably. Objects blinked in and out of existence. Nothing held together across frames. A proper world model wouldn’t behave like that because it would track physics, remember previous states, and update the environment in a stable way. It would run the simulation instead of hallucinating it.

Fei-Fei’s startup, World Labs, is the cleanest illustration. You feed their AI a photo and it generates a 3D scene you can explore. It makes up the missing geometry while keeping the whole thing stable. You walk around and nothing collapses. 

Twitter tweet

The Race to Build World Models

World Labs is not the only company chasing World Models. Here are some of the biggest players in the field:

• Google DeepMind’s Genie 3 hit in August. It generates minutes of interactive 3D at 24 fps, remembers what happened earlier, and lets agents poke the environment. Their main goal is to use generative video like a training ground for embodied agents.

• Earlier this year, Wayve released GAIA-2, a model that generates realistic driving footage to train autonomous driving systems. The goal is simple: create endless, controllable scenarios that would be impossible to collect reliably in the real world.

• And finally, Meta has also entered the World Model race with V-JEPA 2, a 1.2 billion-parameter model trained on video that enables AI agents to understand, predict and plan in the physical world.

The Trend

Spatial intelligence keeps showing up in places you wouldn’t expect. After digging through all these releases, it feels like the market is warming up for products that assume models can understand space even if they’re not perfect yet.

A few opportunities stood out:

• Workflow tools for teams adopting spatial AI. Companies will need scene-level debugging, data cleaning for 3D sensors, and simple ways to stitch world models into existing pipelines. 

• Simulation-heavy vertical apps. Warehousing, construction, and energy all need safer planning environments. A world-model driven simulator that plugs into their actual operations could have immediate ROI.

• AR and field service upgrades. Companies want on-device spatial reasoning to guide workers and verify tasks. Stable scene understanding unlocks real money here.

• Content workflows for persistent 3D. Agencies, retailers, and studios all need coherent virtual spaces, not just assets.

Most of these ideas don’t require training world models. They require meeting companies where spatial AI actually becomes useful.

Help Me Improve Failory

That's all for today’s edition.

Cheers,

Nico

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:

• K-Scale Labs, a startup building open-source humanoid robots, has shut down — learn why below

• The Benefits of Bubbles

• Shopify says AI traffic is up 7x since January, AI-driven orders are up 11x.

• Voice AI startup Giga raises $61 million to take on customer service automation

• Only 6% of big companies are seeing returns from AI adoption — learn why this might be an opportunity below

Online ads are easy to ignore. The best founders know attention comes from the real world too. Today’s sponsor, Delve, just dropped a guide to out-of-home campaigns that actually drive results.

How to run a billboard campaign 101 ^AD

You’re a fast-growing company ready to scale. You’re thinking big: billboards, airport ads, subway takeovers, all the stuff people can’t scroll past.

Delve just dropped the best guide to planning, building, and launching an out-of-home campaign that turns attention into real leads.

Learn the exact tactics they used to take over San Francisco, New York, and Texas with their viral “Done in Delve” campaign.

If you’re a scrappy startup or a Series B company ready to make noise, this playbook’s got you covered.

Watch the short film and read the guide at www.delve.co/ooh to learn everything you need to launch your first out-of-home campaign.

This Week In Startups

🔗 Resources

The Compounding Startup

The future of the web is the history of YouTube

The Benefits of Bubbles

Cloudways' Black Friday 2025 Offer: 50% off all managed hosting plans + 50 free migrations. Grab it today! * 

📰 News

Shopify says AI traffic is up 7x since January, AI-driven orders are up 11x.

Studio Ghibli and other Japanese publishers want OpenAI to stop training on their work

Canva launches its own design model

Amazon sends legal threats to Perplexity over agentic browsing

💸 Fundraising

Industrial AI startup Octonomy raises $20M in seed funding

Inception raises $50 million to build diffusion models for code and text

Voice AI startup Giga raises $61 million to take on customer service automation

Swiss Startup Mimic Robotics raises $16 million to advance physical AI for robots

* sponsored

Fail(St)ory

Robots for Everyone

For a moment, it looked like the humanoid robotics revolution might come from a house near Stanford. K-Scale Labs was the kind of moonshot startup that Silicon Valley still loves to believe in: ten people, one idea, and an open-source humanoid robot walking on two legs.

A year later, it’s over.

What Was K-Scale:

Founded in 2023 by Benjamin Bolte, K-Scale Labs wanted to build a low-cost, open-source humanoid robot, a kind of “Android for the real world.”

The company’s main product, the K-Bot, was priced around $15,000, a fraction of what most humanoid robots cost. Bolte promised accessibility, community, and speed: humanoids for hackers, not just billion-dollar labs.

The vision caught early fire. K-Scale raised $4 million at a $50 million valuation, a strong signal that investors were buying the dream. Engineers from OpenAI, Nvidia, and Amazon reportedly dropped by the house-slash-lab to see the prototypes shuffle around. Bolte seemed convinced the next $10–15 million round would be a breeze.

Then the money didn’t come.

By October 2024, K-Scale’s cash reserves were down to $400,000, barely a month of runway. The company started canceling pre-orders and refunding deposits. The team of ten shrank fast as key engineers bailed for steadier gigs. This week, Bolte sent his final email to investors: the shutdown was official.

The surprising thing was that the product wasn’t vaporware. K-Scale had already started shipping a Founder’s Edition K-Bot, demonstrating they were able to ship things fast. They had over $2 million in pre-orders, proof that someone out there actually wanted these things.

Twitter tweet

Bolte’s final act was pure hacker ethos. In his farewell note, he announced that K-Scale was releasing all of its proprietary IP (both hardware and software) under open licenses. It was a graceful exit for a company that never pretended to be proprietary to begin with. “I hope this work lays the foundation for future hackers and dreamers,” he wrote.

The Numbers:

• 🤖 Founded: 2023, Palo Alto

• 💸 Funding: $4M seed round at $50M valuation

• 🚶‍♂️ Team size: ~10 employees

• 🧾 Robot price: ~$15,000 per unit

• 📦 Pre-orders: $2M+ in deposits

• ⏳ Cash left at shutdown: ~$400K

Reasons for Failure: 

• Hardware burn with software expectations: Bolte tried to run a hardware startup like a software one: small, lean, and fast. But robots don’t iterate cheaply. Even at $15,000 a unit, the upfront costs for tooling and manufacturing were massive. Bolte admitted the unit economics “don’t make sense” without large capital to amortize those costs. That’s a death sentence in hardware, where efficiency only appears at scale.

• Misread the funding climate: K-Scale’s biggest assumption was that investor enthusiasm for humanoid robots would trickle down from giants like Figure AI and Tesla. It didn’t. The AI hype cycle isn’t the same as a robotics one, and by late 2024, even deep-tech investors were tightening their belts. Bolte’s plan to raise $10–15M right after launch was overconfident.

• Product-market mismatch disguised as traction: Those $2M in pre-orders looked good on paper—but the early buyers were mostly hobbyists and labs, not sustainable customers. The market for humanoids at $15K is still niche, more curiosity than commerce. K-Scale mistook enthusiasm for validation. 

Why It Matters: 

• Hardware dreams need brutal math. You can’t “lean startup” your way through tooling and certification, capital efficiency in robotics still means deep pockets.

• Open source isn’t a business model. It builds community, not margin, and once K-Scale gave away its IP, the only thing left was execution—and execution needs cash.

Trend

Too Much AI, Not Enough ROI

McKinsey released its annual State of AI 2025 report this week. Normally, that would be your cue to close the tab. But buried under the corporate gloss, there’s actually some signal worth hearing.

Think of it as field notes from the slowest players in the game (big companies) trying to figure out AI while the rest of us move faster.

Why It Matters: 

• Everyone’s “doing AI,” almost no one’s profiting from it. That gap means opportunity.

• The 6% Club (the few who get real returns) treat AI as product, not project. They rebuild workflows, not just wrap them in LLM wrappers.

• Their pain points are your roadmap. Where big companies choke on complexity, startups can build tools that cut through it.

The Big Picture: 

McKinsey’s survey claims 88% of companies use AI in at least one function. Sounds like mass adoption, until you realize most of it is still pilot projects. Nearly two-thirds haven’t scaled beyond experiments.

Translation: the corporate world has a lot of AI slide decks and very few working systems.

The irony is rich. After two years of “AI is eating the world,” the world’s biggest organizations are still trying to feed it spreadsheets. The report shows 39% of companies say AI has any effect on profits, and most of those admit it’s negligible. The “enterprise AI revolution” looks more like a very expensive set of demos.

The only places AI is actually driving revenue right now (marketing, sales, and product development) are the same ones that have been automating for decades. AI’s still more of a marketing intern than a CEO.

And yet, the interest in AI agents is through the roof. Sixty-two percent of respondents say they’re experimenting with them. The keyword there is experimenting. The same survey shows that only about 10% have scaled agents in any given business function. So we’re still in the “proof of concept” era.

But this is where things get interesting: in that mess of partial adoption, a few companies (6%, to be exact) are quietly making real money.

The 6% Club:

McKinsey calls them “AI high performers.” Ignore the buzzword. What they’ve figured out is painfully simple: AI pays off when it changes how the work gets done.

• They build feedback loops between humans and models. This is the “human in the loop” thing—but stripped of jargon, it just means they define where a person needs to step in. They don’t trust the model blindly, and they don’t waste humans on things the model can handle.

• They invest in real tech infrastructure. Most companies still duct-tape AI pilots onto legacy systems. High performers modernize their stack first: clean data pipelines, scalable architecture, versioned models. It’s not sexy, but it’s the foundation of every working AI business.

• They have an actual AI roadmap. That means someone decided which use cases matter, how they connect, and what success looks like. Everyone else is still running disconnected pilots with no clear outcome.

What This Means for Builders:

The most interesting part of this report isn’t how advanced AI has become, it’s how badly large companies are implementing it. They have the budget, data, and brand power, yet they can’t turn experiments into products. The report confirms a few big truths:

• AI deployment is still a systems problem, not a model problem. Everyone has access to the same LLMs, what’s missing is the infrastructure and glue that make them work in real workflows.
  

• Human feedback is the bottleneck. Everyone talks about “human in the loop,” but no one has built the perfect system for managing that at scale. That’s a product opportunity.
  

• Enterprise AI adoption is fragile. Big orgs struggle to connect models with real business value because they lack the connective tissue: data hygiene, versioning, governance.

For founders, this means there is a market for solving the boring but crucial operational pain that keeps AI from working in practice. Some obvious startup ideas that pop from this data:

• Tools for AI deployment orchestration: how to get from prototype to production without breaking compliance or pipelines.

• Lightweight “human feedback management” systems: ways to collect, review, and use human validation without slowing everything down.

• Infrastructure that tracks AI ROI per workflow: basically analytics for whether a model is actually earning its keep.

If McKinsey’s data shows anything, it’s that the enterprise world is still waiting for its “Stripe moment” for AI—something that makes complex adoption simple.

Help Me Improve Failory

That's all for today’s edition.

Cheers,

Nico

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:

• Carted, a startup that wanted to unify online shopping, has shut down — learn why below

• Non-obvious pricing advice for startups

• OpenAI completes its for-profit recapitalization

• Legaltech AI startup Legora raises $150M at $1.8B valuation

• TechCrunch Disrupt 2025 wrapped up yesterday — below are the key startup trends that stood out.

This Week In Startups

🔗 Resources

• Why pitches fail (and a template that will fix yours).

• What’s working in GTM right now.

• Non-obvious pricing advice for startups

• How founders should think about runway: A guide from Sequoia Capital

📰 News

• Two days after OpenAI’s Atlas, Microsoft relaunches a nearly identical AI browser

• OpenAI made ChatGPT better at sifting through your work or school information.

• Canva launches its own design model

• OpenAI completes its for-profit recapitalization

💸 Fundraising

• Legaltech AI startup Legora raises $150M at $1.8B valuation

• Helex raises $3.5M to advance gene therapies for kidney diseases

• AI video startup Synthesia raises $200M at $4B valuation

• AI startup Mercor now valued at $10 billion with new $350 million funding round.

Fail(St)ory

Universal e-commerce

Carted set out to unify online shopping under one “universal e-commerce API”. It then pivoted into a price-tracking app — and now, it’s shutting down.

What Was Carted:

Carted raised $13M in 2021 to solve a real, obvious pain: when you see a product online (in a video, a blog, a social feed), buying it is still very messy. You get bounced across tabs, stores, affiliate links, and sketchy Shopify pages. The founders wanted to erase all that.

They built a system that let publishers, apps, and platforms drop a universal “buy” layer right on top of the content. You could check out without ever leaving the page. Carted handled all the backend chaos: connecting to different merchants, splitting orders and tracking affiliates.

If it worked, every article, video, or post could instantly turn into a small marketplace. The idea made perfect sense, until the platforms did it themselves.

By 2023, TikTok and Instagram had rolled out native shopping. They didn’t need Carted’s API, they built their own. The platforms kept the user, the product, and the payment in-house. 

So Carted pivoted. They turned the underlying tech into a consumer app instead. Now anyone could save products from across the internet, organize them into folders, compare options, and get price drop or restock alerts. Think Pinterest meets Honey.

The app launched in 2024 and found early fans. It solved a genuine pain: the chaos of modern shopping spread across 20 tabs and half a dozen stores. But like most consumer tools that live between discovery and checkout, it was easy to like and hard to monetize.

The Numbers:

• 🗓️ Founded: April 2021
  

• 💰 Seed Round: $13M (one of Australia’s largest seeds)
  

• 👥 Team Cut: 14 → 6 during 2023 build
  

• 🧪 Pivot: API to consumer price tracker app
  

• 📲 App Launch: 2024, with engaged early adopters

Reasons for Failure: 

• Big platforms got there first: TikTok Shop rolled hard from 2023 onward, and Instagram had already put checkout inside the feed back in 2019. At the same time, publisher-side competitors showed up with on-page checkouts, like Bolt buying Tipser to let readers buy “on any digital surface.” And even where full checkout wasn’t needed, link-in-bio and affiliate pipes were good enough for creators and media: Linktree’s Store Link pushed shoppable listings right in bio.

• The pivot weakened the business model. The consumer app was clean and useful, but alerts, lists, and sharing are easy to clone and hard to monetize at scale. If you don’t control the checkout, you don’t control the take rate. Without a crisp plan for merchant SaaS, affiliate unit economics, or paid features, growth becomes marketing spend. A pivot should shorten the path to cash, not just to love.

• Timing collided with AI-led, agentic shopping. The narrative moved toward agents that generate options, compare prices, and place orders across stores. That shift made “universal API” feel like yesterday’s layer. It also hit fundraising and partnerships, because buyers and investors chase where demand is created, not where it is forwarded.

Why It Matters: 

• Don’t build on borrowed distribution. If your product depends on platforms that can replace you with a feature toggle, you don’t have a moat, you have a countdown.

• If you pivot to consumer, lock the revenue model on day one. Don’t wait for scale to set pricing. Decide your take rate, SaaS fee, or premium features early.

• Solve a pain, not a friction. Frictions get engineered away by whoever owns the platform. Pains stick around long enough to build a company on.

Trend

The Trends of TechCrunch Disrupt

TechCrunch Disrupt 2025 wrapped up yesterday, and while everyone’s talking about who won, the real story is what the 200 startups in the Battlefield revealed about where founders’ heads are right now. 

Forget the stage lights. Disrupt has always been a mirror for the broader startup psyche. This year’s reflection was clear: AI has eaten everything, climate tech is getting real, and “deep tech” isn’t dead, it’s just learned to sell.

Why It Matters: 

• AI is no longer a vertical, it’s infrastructure. The same way cloud became a baseline a decade ago, AI is now just the expected layer under everything. Startups that don’t use it stood out for being the exception.
  

• Investors are rewarding “real-world tech.” The winners weren’t building the metaverse or minting tokens. They were moving shipping containers, recycling plastic, or curing kidney disease.

1) AI Everywhere (and in Everything)

Unsurprisingly, every second booth was an AI company, even when it wasn’t. Out of 200 startups, at least 18 were formally “AI/ML,” but nearly all the rest had AI baked in somewhere. 

Finalist Elloe AI is the perfect example. They call themselves the “immune system” for enterprise AI, building a compliance layer that fact-checks and sanitizes outputs before they hit production. It’s a bet that “AI safety” will be the next compliance goldmine.

Unthread is another Top 20 startup, turning messy Slack threads into auto-resolved tickets using large language models. The pitch: stop drowning in internal noise.

Super Teacher, meanwhile, showed what happens when AI tutors actually talk back: its animated, voice-based teachers are already live in public schools for $15 a month.

AI showed up in logistics, education, and even testing software updates (CyDeploy clones entire IT environments with machine learning to safely deploy patches). 

The crypto/Web3 crowd? Nowhere. It’s clear founders have moved from tokens to transformers.

2) SaaS Still Sells

SaaS was the single biggest category at Disrupt (30 of 200 startups), but the winners were hyper-specific, built by domain insiders who know the pain firsthand.

Identifee, a finalist, is a good example. Built by ex–Wells Fargo execs, it’s an all-in-one productivity platform for bankers: CRM, analytics, and AI in one interface.

ExoMatter is another example. It is a materials-informatics SaaS that helps R&D teams search, simulate, and source alternative materials to cut cost or emissions. It’s literally a lab-insider product: founded by a PhD materials chemist and spun out of the German Aerospace Center (DLR).

3) Climate Tech Becomes Core Business

At least a dozen companies tackled sustainability directly, and one of them, MacroCycle, made the final five. Their pitch: take end-of-life plastics and textiles and recycle them back into petrochemical-grade material with zero emissions. Circular economy, finally with a business model.

Chile’s Strong by Form joined the Top 20 by building lightweight engineered wood that can replace concrete and steel. It’s cheaper, carbon-negative, and quite pretty.

COI Energy built a platform that lets businesses monetize their energy waste. Large buildings often buy excess electricity capacity they never use; COI’s system allows them to sell or share unused electricity in real time to others on the grid.

4) Deep Science Steals the Show

The Battlefield’s biggest surprise: the top two startups weren’t SaaS or social—they were science.

The winner, Glīd, built an electric vehicle that moves shipping containers from port to rail, replacing diesel trucks. Cleaner, faster, more efficient. Industrial logistics, reimagined.

Runner-up Nephrogen is working on AI-driven gene therapy for kidney disease. Founder Demetri Maxim, who actually lives with the disease, plans to join his own clinical trials. It’s biotech meets founder-market fit in the rawest way possible.

Meanwhile, Mbodi, another finalist, showed how to teach robots new tasks with plain English. Chipiron is rebuilding MRI machines to make them portable and cheap. 

All of them point to the same trend: deep tech isn’t dead, it’s finally consumerizing.

Help Me Improve Failory

That's all for today’s edition.

Cheers,

Nico

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:

• Inseco, a startup that made protein from flies, shut down — learn more below

• 32 things we’ve learned about building a startup that scales.

• Uber will offer gig work like AI data labeling to drivers while not on the road.

• Sizable Energy raised $8M to store clean energy under the sea.

• OpenAI launched Atlas, it’s first AI Browser — learn more below

Losing focus slows startups down. Today’s sponsor, TRMNL, helps you stay on track with an open source e-ink display that cuts out the noise.

No subscription, no ads, no alerts, only focus^AD

Most startups fail. There are many reasons for this, but one of them is losing focus.

So how do you keep your team on track? Meetings are OK, but they take a lot of time. Dashboards are OK, but you might measure the wrong things.

Introducing TRMNL, an open source e-ink display that helps you stay focused, without getting distracted.

No touch screen. No colors. No back light. Nothing to “play with.” Just information that’s important to your startup, on your desk or wall.

Monitor server analytics, web traffic, sales leads, git commits, and even personal data like your calendar, the weather, or the occasional The Office quote.

Get informed at a glance, then get back to work.

Stay focused. Get TRMNL.

This Week In Startups

🔗 Resources

The Ultimate Guide to the perfect SaaS pricing page.

32 things we’ve learned about building a startup that scales.

Do you have a demand problem?

Your pricing is (probably) broken.

📰 News

Amazon unveils AI smart glasses for its delivery drivers.

Netflix goes ‘all in’ on generative AI.

Uber will offer gig work like AI data labeling to drivers while not on the road.

Walmart partners with OpenAI so shoppers can buy things directly in ChatGPT.

💸 Fundraising

Tensormesh raises $4.5M to squeeze more inference out of AI server loads.

OpenEvidence, the ChatGPT for doctors, raises $200M at $6B valuation.

AI operating system startup UnifyApps raises $50M.

Sizable Energy raised $8M to store clean energy under the sea.

Fail(St)ory

Food From Flies

Inseco was supposed to be Africa’s insect protein success story. Instead, after seven years and a $5.3 million seed round, it shut down and sold off its assets.

A reminder that scaling biology isn’t just about science: it’s about execution in messy, real-world conditions.

What Was Inseco:

Inseco was born in Cape Town in 2017 from an ambitious idea: turn food waste into high-value protein. Founder Simon Hazell saw how much organic material (from wineries to food processors) was ending up in landfills. He figured black soldier fly larvae could eat that waste and become food themselves.

Inseco would collect organic byproducts, feed them to larvae, and process the grown insects into protein meal, oil, and fertilizer. The products, branded EntoMeal and EntoOil, targeted aquaculture, poultry, and pet food markets hungry for alternatives to fishmeal and soy.

At its peak, Inseco ran a 10,000-square-meter facility producing over 100 metric tons of insect meal and oil a month. Investors took note. In 2022, it raised one of South Africa’s largest-ever seed rounds: $5.3 million.

But insect farming isn’t software. You can’t “move fast and break things” when your product is alive. The larvae needed constant heat, humidity, and airflow—conditions that didn’t always cooperate.

As the company scaled, cracks started to show. Power became unreliable, capital scarcer, and patience thinner. The elegant model began to strain under real-world complexity.

By 2024, the optimism that once surrounded Inseco had faded. Operations paused, assets were sold, and the founder signed off quietly: “We got close, but ultimately ran out of time.”

The Numbers:

• 🐛 Founded: 2017

• 💰Funding: $5.3 million seed (2022)

• 🏭 Facility: 10,000 m² tech-enabled plant

• ⚙️ Production: 100+ metric tons/month

•  📉 Shutdown: 2024
  

Reasons for Failure: 

• Loadshedding crippled operations: Inseco’s production depended on stable power—something South Africa couldn’t guarantee. Loadshedding, the country’s rolling blackouts, cut electricity for hours at a time, throwing off the temperature and humidity the larvae needed to survive. “The recurring 4-hour power outages devastated operations and increased our energy costs fourfold,” Hazell said. 

• Overaggressive scaling: Inseco ramped up production before proving it could make money on each ton of insect protein. The factory was big, the costs were bigger, and the margins never caught up. Hazell later admitted they “chose an intersection that proved too aggressive”, meaning they bet that scale would fix their unit economics. It didn’t. They built a commercial-scale plant on startup-level proof.

• Weak hiring and slow pivots: When growth came fast, discipline slipped. Hazell admitted, “We made some bad hires. We scaled too quickly and pivoted too slowly.” The team spread itself thin between running a large facility and fixing its tech. Instead of doubling down on the higher-margin process they were developing, they kept chasing scale. Focus drifted, and in manufacturing, that’s fatal.

• Investor sentiment turned: The end of zero-interest-rate money hit hard across deep-tech and climate sectors. Even stronger insect startups in Europe were struggling to raise. Inseco, operating from a less mature market, couldn’t weather the drought.

Why It Matters: 

• Scale is not validation. Inseco hit industrial scale but not industrial margins. Sometimes, scaling a broken model just multiplies the burn.

• Reliability is credibility. Every blackout didn’t just stop production, it also shook investor faith. In hard-tech, operational slip-ups echo louder than financial ones.

Trend

GPT Atlas

Two days ago, OpenAI launched Atlas, an AI-powered browser built around ChatGPT. You’ve seen versions of this before: Perplexity’s Comet, The Browser Company’s Dia, Microsoft’s Copilot tab in Edge. 

Atlas just happens to be the cleanest execution yet: less friction, better context, and an agent that actually does work where you are. That’s enough to shake up the browser status quo.

Why It Matters:

• The browser is the new workspace. Atlas treats the web like one giant app. Instead of bouncing between tools, you can research, summarize, plan, and book things in one flow.

• Context finally sticks. Atlas remembers what you’ve been doing online (sites, chats, projects) and picks up right where you left off. No more hunting for that tab or file from last week.

• Agents get a real interface. Atlas gives ChatGPT a native playground. It can read a page, click links, and take action, which makes it feel more like a coworker than a chatbot.

What is Atlas:

Atlas is OpenAI’s AI-native browser, designed to make ChatGPT feel integrated instead of bolted on.

You open a page, and ChatGPT already knows where you are. You can ask it to summarize an article, compare it to what you read last week, or draft an email response based on the content in front of you. All without switching tabs or copy-pasting.

Twitter tweet

Two features matter most: memory and agent mode.

• Memory lets ChatGPT recall your browsing history and past tasks, so you can ask, “Find all the job postings I looked at last week and summarize trends.” 

• Agent mode lets it take action inside the browser: researching, booking meetings, filling forms.

The Trend:

We’ve known for a while that the browser was the next battleground of the tech world. Now it’s official. With the launch of ChatGPT Atlas by OpenAI, the fight for who controls how you access and use the web has stepped into a new phase.

The trend is simple: the browser is turning into the home for AI agents.

Instead of adding AI as a feature, companies are rebuilding the browser around it: where your assistant doesn’t just answer questions, it reads, acts, and remembers directly on the page.

The Idea:

Big companies will fight over browsers. You don’t need to. The smarter move is to build around the layer Atlas creates.

A few simple ideas worth exploring:

• AI safety shell. A tool that monitors what the browser-agent sees and does. Think analytics + alerting rather than more chat UI. Given that Atlas already faces prompt-injection risk and security warnings, this is very much needed.

• Agent-aware extensions. Atlas is Chromium-based, which means you can already build normal Chrome extensions. They can’t talk directly to the ChatGPT agent yet, but that’s coming. The moment OpenAI exposes APIs for agent communication, extensions that can feed, filter, or visualize agent behavior will explode.

Help Me Improve Failory

That's all for today’s edition.

Cheers,

Nico

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:

• Plumb, a startup trying to become the “Substack of AI workflows”, has shut down. — learn more below

• The odds of making it: A new look inside startup growth.

• Spotify partners with record labels to create ‘artist-first’ AI music products.

• Smart ring maker Oura raises $900M.

• Sora 2 videos are everywhere. — learn why this matters below

Building AI copilots or agents? Check today’s sponsor: CData Embedded Cloud powers reliable, real-time data connectivity for companies like Google Cloud and Salesforce, and it’s now available to supercharge your AI features.

Build great AI experiences with reliable data connectivity ^AD

Software leaders, as your teams work urgently to build AI copilots and agents, connectivity to your customers’ data is critical to ensure these features are not just expensive chat interfaces.

Trusted as the connectivity backbone for Google Cloud, Salesforce Data Cloud & Palantir's most demanding AI deployments, CData Embedded Cloud is now here to power your AI initiatives.

Build robust, secure, real-time connections to your customers’ data sources so they don’t have to go anywhere else to find their answers. Now, your customers can ask questions in natural language and make updates across all their data sources directly within your UI.

Transform copilots and agents from basic answer-givers into true discovery and decision partners, while CData handles the data connectivity layer.

👉 Build great AI experiences faster 

This Week In Startups

🔗 Resources

• Everyone should be using Claude Code more.

• Build a Billion-Dollar Startup with a minimal team thanks to AI.

• The odds of making it: A new look inside startup growth.

• Turn LinkedIn thought leadership into revenue (100% done-for-you). Playbookz writes your posts, 50k views guaranteed in 28 days. * 

📰 News

• Spotify partners with record labels to create ‘artist-first’ AI music products.

• Pinterest adds controls to let you limit the amount of ‘AI slop’ in your feed.

• Anthropic launches new version of scaled-down ‘Haiku’ model.

• Google released Veo 3.1.

💸 Fundraising

• Jack & Jill raises $20 million to bring conversational AI to job-hunting.

• Smart ring maker Oura raises $900M.

• Stoke, a reusable rocket startup, raises $510 million.

• AI startup Lila Sciences raises $115 million backed by Nvidia.

* sponsored

Fail(St)ory

Workflows for Sale

Last week, Plumb shut down. They nailed a clean builder and a marketplace to sell workflows, but distribution outran product-market fit. 

Here’s what they built, what broke, and what you can use.

What Was Plumb:

Plumb started as a clean, creator-friendly node based workflow editor. You wired nodes like  “Generate image,” “Parse PDF,” “LLM prompt,” “Search the web,” and created complex AI workflows fast.

While this might look like a copy of editors like Zapier, Make or n8n, Plumb’s goal was not in the tooling but rather in the distribution. Their pitch was “Substack for AI workflows”. 

When you built a workflow, the platform automatically generated a landing page for it. Creators could then tune the page, set a price for their workflow, and then share it and sell it. Other users could subscribe to the workflow and start using it, without a need to see the prompt or the logic.

Take, for example, one of the automations I found in their marketplace: A “Market Update Newsletter Writer”. You give it a PDF with market trends and it returns a newsletter ready to be sent. It’s a tiny workflow, but it was built, distributed and monetized in minutes rather than weeks. 

Another interesting thing was their product philosophy: determinism over randomness. Plumb believed that controlled workflows worked better than agents. While agents wander and hallucinate, a workflow runs a defined path, logs each step, and returns structured output you can debug. Plumb also kept humans in the loop: pause to ask for context, collect input, inject it downstream, continue. 

The Numbers:

• 📍 Founded: 2020

• 💸 Total raised: $9.8M

• 🛑 Shut down: October 2025

Reasons for Failure: 

• They invented a business model nobody asked for: Plumb built “workflow subscriptions”, a novel system where users paid recurring fees to run other people’s automations. It sounded smart, but, according to the founder, no one cared. 

• No vertical, no urgency: Plumb never picked a niche or a high-value task to solve first. They built flows that were interesting, not essential. Without a clear “this saves me money today” story, even beautiful tools collect dust.

• They locked execution inside Plumb. To charge credits and hide the workflow, every run had to start in their app. That made webhooks, MCP, and other triggers hard, because those fire from the outside. Without easy triggers you cannot plug into email, CRMs, or data pipelines. Teams need automations that fire from their systems, not a page they have to visit.

Why It Matters: 

• Distribution can’t replace product-market fit. A landing page for every flow is cute; a trigger that saves ops time is revenue.

• Pick one painful job and own it. General marketplaces impress in demos, then fade in production without a vertical wedge.

Trend

Sora 2 and the Rise of AI-Generated Media

Every few months, there’s a new “holy shit” moment in AI video. This time it’s Sora 2, OpenAI’s latest model, that makes last year’s demos look like flipbooks. 

In the last two weeks, the internet hasn’t stopped churning out clips that look so real they’ve already fooled millions.

Why It Matters: 

• AI video just crossed the realism threshold. The tech finally understands physics, continuity, and sound. The trifecta that separates “toy” from “tool.”

• Distribution and creation are merging. OpenAI launched a TikTok-style app built directly on top of the model.

• The next media gold rush is on. Meta, Google, and everyone else are racing to own the new format before Hollywood or advertisers catch up.

Sora 2

OpenAI dropped Sora 2 on September 30, calling it their “GPT-3.5 moment for video.” Not hyperbole. This thing doesn’t just render pretty frames, it simulates real-world physics. A basketball bounces off the rim. A shirt stays red across shots. Shadows behave like they’ve actually met the sun.

The biggest leap is audio. Native, synced, realistic sound baked straight into the generation process. Veo 3 had this first, but OpenAI made it default. Now, dialogue, footsteps, and background noise match the visuals perfectly.

Twitter tweet

Sora 2 also nails multi-shot continuity. Look at the following example and notice how the character stays perfectly consistent across all shots. Same clothes, same face, same expression.

Twitter tweet

That’s not just a technical milestone, it’s the point where video generation becomes commercially viable. The model can be used for ads, storyboards, product demos, even training simulations. OpenAI confirmed an API is coming soon, which means any app could spin up videos on the fly.

The Sora App

Alongside the model, OpenAI launched the Sora app, a vertical-video social feed where users generate and share clips directly from the model. Think TikTok, but every post is synthetically created.

You can write a prompt or upload an image to animate. You can remix someone else’s video. You can even add yourself to a scene through a feature called Cameos: a one-time recording that lets the model generate a full digital likeness of you, voice included. Then friends can remix you into their clips, with your consent.

What’s different here is intent. OpenAI isn’t just releasing another model, it’s building a whole platform. The social layer ensures people don’t just test Sora once and move on. They stay, remix, and share. 

The Trend: AI-Generated Media

An MIT report recently noted that AI hasn’t disrupted many industries yet except one: media. It’s not an accident. Video is fast, emotional, and monetizable. 

Advertising agencies are already using generative video to cut costs and skip shoots. The economics are irresistible: seconds instead of weeks, pennies instead of thousands.

Meta sees the same thing. Last week it rolled out Vibes, a new short-form AI video feed inside the Meta AI app. You browse AI-generated videos, remix them, post your own. It’s basically Sora with a different skin.

The Idea

The Sora 2 API will open the floodgates. When you can generate realistic video with synced audio in seconds, entire categories of startups become feasible. Here are some ideas:

• AI-native ad studios. Imagine a platform that generates and A/B tests hundreds of hyper-personalized video ads automatically, updating based on real-time performance. Every brand becomes its own mini-production house.

• Concept testers for entertainment. Make a pilot for a show or a teaser for your indie game in a weekend, post it, and gauge traction before spending a dollar on full production.

• Trust infrastructure. As synthetic media floods feeds, there’s a massive gap for tools that verify provenance and detect AI content in real time. If everyone’s making fake videos, someone has to make the authenticity meter.

Help Me Improve Failory

That's all for today’s edition.

Cheers,

Nico

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:

• Niro, a startup that offered unsecured loans in India, has shut down. Learn why below

• Handling Multiple ICPs as a Solo Founder.

• The EU will propose a unified startup regime.

• n8n bags $180M Series C.

• OpenAI launched Apps in ChatGPT (again?), opening up new opportunities to build apps that live natively inside the chat interface. Learn more below

Scale faster, build great AI experiences, keep users in your product ^AD

Software leaders, your product and engineering teams are racing to build AI copilots and agents into your products. But AI is just an expensive chat interface without access to your customers' data.

With Embedded Cloud for AI, the first managed Model Context Protocol (MCP) platform, you can provide your copilots and agents with secure, real-time access to 270+ customer data sources through one simple connection.

Your customers can ask questions in natural language and make updates across all their data sources directly within your UI.

You'll keep users in your product by surfacing real-time information they can trust and act on immediately.

Transform copilots and agents from basic answer-givers into true discovery and decision partners, while CData handles the data connectivity layer.

👉 Build great AI experiences faster

This Week In Startups

🔗 Resources

• The Ultimate SaaS Messaging Framework for early-stage Startups.

• How to turn ChatGPT into your best pipeline source.

• Handling Multiple ICPs as a Solo Founder.

• Ghostwritten LinkedIn content for founders ready to grow. Playbookz guarantees 50k qualified, targeted reach in the first 28 days of posting or you don't pay. * 

📰 News

• Anthropic to open its first India office in 2026.

• The EU will propose a unified startup regime in 2026 to replace 27 national systems and ease cross-border growth for scaleups.

• Sora hit 1M downloads faster than ChatGPT.

• Google launches its AI vibe-coding app Opal.

💸 Fundraising

• n8n bags $180M Series C, bringing its valuation to $2.5B.

• Routefusion raises $26.7M Series A to expand its payments infrastructure.

• David AI secures $50M Series B to further develop its AI audio/data tech.

• Base Power raises $1B to deploy home batteries everywhere.

* sponsored

Fail(St)ory

Fintech Without Funding

Fintech startup Niro shut down this week.

The Bengaluru, India-based embedded-lending startup ceased operations after four and a half years in business, citing difficulties in raising fresh capital and increasing regulatory challenges in the sector as the main drivers behind the decision.

What Was Niro?

Launched in 2021 by Aditya Kumar and Sankalp Mathur, Niro built a platform to embed instant personal loans into consumer internet platforms, including Snapdeal, Quikr, and Housing.com. 

The company offered unsecured loans in the range of 50,000-700,000 rupees (approx. 563-7,882 U.S. dollars), typically disbursed through partner financial institutions (e.g., L&T Finance, Muthoot Finance) without collateral.

At shutdown, Niro had raised close to $20 million from investors, facilitated $200 million in loan disbursements, and formed 30 partnerships, but execution alone wasn’t enough to keep Niro’s doors open.

The Numbers

• 📍 Founded: 2021, Bengaluru, India

• 💸 Total raised: $18.7M over 5 rounds

• 💰 Latest funding round: $4.3M, April 17, 2024

• 👥 Team: ~289 (before shutdown)

• 🤝 Partnerships: ~30

• 💵 Loan disbursements: ~$200M

• 🕐 Duration: 4.5 years (2021-2025)

Reasons for Failure

• Reduced VC funding for digital lending: Over the past eight months, VC funding in digital lending has dropped nearly 50% in India. Despite earlier raises, Niro failed to secure additional funding when it mattered most. In its founder’s own words: “Despite scouring the globe for capital … I was unable to bring this one home.”

• Regulatory tightening: The unsecured lending space in India is under growing regulatory pressure. Platforms offering instant loans without collateral are facing stricter oversight, pushing many toward secured products.

• Distribution and data challenges: Niro’s founder noted that financial institutions lack proprietary channels and differentiated data for underwriting. On the flip side, consumer Internet platforms have reach but lack financial domain expertise. Connecting the two in a highly regulated space proved harder than expected.

Why It Matters

• Embedded finance remains a compelling sector, but it’s deceptively complex. Achieving success requires striking a delicate balance between trust, regulation, risk, and distribution.

• Timing and capital can trump innovation in Fintech. Even with traction and backing, you can’t outrun a frozen funding environment or shifting industry regulations and complexities.

Trend

The Week ChatGPT Became an App Store

OpenAI dropped some major news this week in what feels like a major shift for its flagship ChatGPT platform: a new way for developers to build apps that run directly inside ChatGPT.

At DevDay 2025, OpenAI unveiled its plans for apps inside ChatGPT with the launch of the Apps SDK, giving developers full-stack access to build fully interactive applications within ChatGPT itself. Developers can use the Apps SDK to go beyond custom GPTs and create rich, functional software experiences that live natively inside the chat interface.

Twitter tweet

What Apps SDK Is

The new Apps SDK turns ChatGPT into a hosting platform for third-party tools. Developers can now build apps that render visuals, handle user input, and connect to external APIs — all from within an AI-powered chat window.

Imagine asking ChatGPT to “book me a flight to Lisbon” and instantly seeing an Expedia interface appear in the conversation — no browser tab, no redirection. Or instructing ChatGPT to “make a playlist for rainy mornings,” and the AI puts together a new Spotify playlist within your chat that you can start listening to instantly.

In the words of OpenAI CEO Sam Altman, this launch “...will enable a new generation of apps that are interactive, adaptive, and personalized, that you can chat with.”

Why It Matters

• ChatGPT just went from a conversational AI tool to an app-hosting platform. With the Apps SDK, ChatGPT becomes a host for an entire ecosystem of AI-native apps, each able to deliver full experiences through conversation.

• App distribution is being reinvented. Instead of hunting for apps in an app store, users discover them in context. You ask a question, and ChatGPT recommends the right tool. App discovery becomes intent-driven, not search-driven.

• Developers have new opportunities to monetize. Contextual placement inside the chat creates direct pathways for engagement, in-chat transactions, and subscription models. Developers who design apps that integrate seamlessly into conversations could unlock new revenue streams.

• It redefines user interfaces and experiences. Apps don’t compete for homescreen real estate anymore — they compete for conversational relevance. ChatGPT is turning into a unified workspace where interaction replaces navigation.

The Bigger Trend

This is the beginning of AI-native third-party platforms, resulting in a more cohesive digital ecosystem where conversation is the interface and intent is the input. 

We’ve seen similar trends in the past: operating systems birthed app stores and web browsers delivered extensions. Now, AI assistants are giving rise to conversational apps that live inside the chat itself, turning dialogue into a workspace and intent into action.

This hints at a future where you no longer have to “open the right app,” but rather talk with AI about what you want to do and the right app will simply appear in front of you, within a single conversational interface.

Twitter tweet

For startups, this opens up new opportunities to build apps for ChatGPT that bypass crowded app marketplaces. If you can build tools for niche workflows and intents, which slot seamlessly into the user’s conversational context, you will be well-positioned to capitalize on this new ecosystem.

Help Me Improve Failory

That's all for today’s edition.

Cheers,

Nico