Hey, it’s Nico.

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

Here are the 3 most important things of today’s issue:

• Val Town, a platform for making small tools on the browser, shut down — find out why below

• How PostHog cuts AI costs with decision models

• OpenAI launched Businesses in ChatGPT, integrating more of the sales funnel into their platform. — learn why this matters below

Today’s newsletter is sponsored by Cadabra, the platform helping founders monetize their startup’s data for AI training.

Your dead startup's data could be worth up to $2M+ ^AD

You shut it down. The Slack threads, docs, code and support tickets are still there, and AI labs pay for that kind of data.

Cadabra values your company's internal records and connects you with micro1's AI training data programme. micro1's published range is $100k to $2M+ per approved data package. Closed and winding-down companies have their own defunct-company track.

Still running? Companies with 30+ employees qualify for the main programme.

• Everything is de-identified before it's used

• You approve the scope in writing

• You keep ownership

Find out what your data could be worth →

Free, confidential valuation. 10 questions, about 60 seconds, no email required. Estimates are indicative. Payouts aren't guaranteed and depend on micro1's review.

This Week In Startups

🔗 Resources

Five B2B growth playbooks from Clay, Granola, Vercel and more

How PostHog cuts AI costs with decision models

How to get your startup recommended by AI through third-party sites

Which part of your GTM is dragging? HubSpot for Startups’ free 10-minute audit finds it and builds your 90-day plan.*

📰 News

Anthropic releases Claude Fable 5.2 and Mythos 5.2

Thinking Machines launches Inkling 2 Small on Tinker for inference and fine-tuning

Jensen Huang resigns as CEO of NVIDIA

OpenAI’s quarterly revenue reportedly crosses $10B for the first time

💸 Fundraising

Mirendil, a startup that builds AI agents that write code and run research experiments, raised $1B in Series A funding

Mercor, a startup that uses human expertise to train and evaluate AI models, raised $750M in Series D funding

Fractile, a startup that develops chips for running AI models, raised $720M in Series C funding

Databento, a startup that provides real-time and historical financial market data through an API, raised $100M in Series C funding

*sponsored

Fail(St)ory

The Team Upgrade

Val Town let developers write small applications in a browser and immediately share a working URL. On November 10, it announced its wind-down, with hosted services scheduled to end January 10.

The founder said the company couldn't compete in cloud computing. His earlier updates describe months spent helping individual developers find a reason to bring the product into their teams.

What Was Val Town:

A small internal tool can need a surprising amount of setup: somewhere to run code, store data, schedule jobs, and check what broke. That work can make a useful automation feel like another project to maintain.

Val Town put those pieces together. Developers could write JavaScript or TypeScript, save it, and get a running application with a web address. Storage, email handling, logs, and version history came with the platform.

Its AI assistant, Townie, could write code, inspect errors, and change scheduled jobs. Developers could also connect the coding agents they already used to Val Town's hosting.

Kilo Code said it built a support agent in 89 minutes with Val Town's founder. The application received a ticket, sent its contents to a model, and posted a suggested response as an internal note.

Individuals could start free. New Pro subscriptions cost $25 a month, while Business started at $200 for collaboration, access controls, and higher allowances. That's a substantial step for a two-person startup considering its first shared tool.

Val Town targeted small technical startups and narrowed its sales focus to workflows such as qualifying incoming leads.

Behind the interface, the company operated hosting, storage, email, and AI tooling. Keeping those services running came with ongoing costs. Val Town needed small teams to find uses valuable enough to justify a recurring $200 subscription.

The Numbers:

• 💰 Funding: $7M across two rounds

• 👥 Team: 6 in July, according to the company

• 💳 Monthly plans: $25 Pro / $200 Business

• 🎯 April team sales: 4 against a target of 14

• 📈 July ARR growth: 6%, company-reported

Reasons for Failure: 

• A steep jump to Business: Moving from Pro to Business meant going from $25 to $200 a month, a jump the founder called too steep. The company was struggling to sell that $200 plan: zero new subscriptions in March against a target of 11, followed by four in April against a target of 14.

• Winning customers could create ongoing work: The founder warned about drifting into consulting. Building software for customers left his team responsible for maintenance when nobody at the customer took ownership. He began requiring customers to participate in development, an attempt to keep each sale from creating another application to look after.

• Its customers were shutting down, too: Val Town targeted early startups, whose subscriptions depended on those businesses surviving. In June, three of the five Business customers it lost had shut down. Overall ARR fell 1% that month.

Why It Matters: 

• Track how often a personal project becomes a paid team workflow, and whether that upgrade happens without a founder explaining the use case on a call.

• When onboarding includes building software, identify who will maintain it afterward. Include that continuing support work when judging what the subscription earns.

Trend

After the Recommendation

On Tuesday, OpenAI launched Businesses in ChatGPT, giving eligible U.S. ecommerce and online service businesses a way to publish an agent that answers customers’ questions inside ChatGPT.

When a business appears in an answer, you can open its profile and start a separate, clearly labeled conversation with its agent. You could discover a product, compare options, and ask the seller about its return policy without leaving the platform.

OpenAI is making room for the sales conversation that follows a recommendation. That could move some of the work of convincing a customer into the same place that introduced them to the business.

Why it Matters

• OpenAI could shape the shortlist and host the seller’s pitch, bringing more of the buying decision inside ChatGPT.

• Ads can lead directly into a business’s agent, letting OpenAI monetize the introduction and host the sales conversation.

ChatGPT Hosts the Conversation

Through Business Hub, verified businesses can update their information, supply product feeds, and see aggregated mentions, website clicks, and discovery topics. Eligible businesses can also publish a Business Agent using information and guidance they provide, then sponsor it as an ad destination.

OpenAI had already started testing Sponsored Agents in September with selected U.S. advertisers. This week adds business profiles and an organic route from ChatGPT’s answers into a branded conversation. A customer can choose to engage with the business after it appears in an independent answer.

The profile gives the business a presence it can manage inside ChatGPT, with its description, logo, website, and delivery and return policies. Retailers can upload a product feed containing prices and availability. Shopify merchants can connect their store, with eligible product information already kept up to date through Shopify Catalog.

The reporting makes that presence easier to measure. Businesses can review the past 28 days of activity: how often ChatGPT references their website, how many clicks it sends, and the common topics in answers where they appear. They get a first-party view of how often they’re surfacing and which subjects bring them into the conversation.

The Trend

Google’s Business Agent on Search already lets shoppers ask brands product questions without leaving Google. Its Business Agent for Leads beta in India puts an agent inside an ad to answer questions and qualify prospective customers.

In Adobe’s October survey of 1,011 U.S. consumers who had used public AI assistants for shopping, 40% said answers to specific brand questions encouraged them to visit a website.

If customers choose to engage with businesses’ agents, more of that explanation and persuasion could happen even before they reach the company’s site. A platform could introduce the business, host the sales conversation, and send over someone who has already decided what to buy.

That could make businesses more dependent on OpenAI for a part of the customer relationship they currently manage on their own sites. OpenAI would set the rules for how their representatives appear, how customers reach them, and what the business gets to learn from those interactions.

Help Me Improve Failory

That's all for today’s edition.

Cheers,

Nico