The challenge
AI had made building software cheap. My bet: if building is easy, everyone will want to build their own apps. clawy was meant to prove it. Describe the app you want in plain language, and an AI agent builds it, hosts it and keeps it running. No code, no servers, no AI accounts of your own.
What I built
A platform that gave every user their own cloud server in Germany with an AI agent on it, ready within half a minute of signing up and equipped with 45 skills: building websites and apps, research, image generation, presentations. Users described what they wanted in a web chat or on Telegram; the agent built it on their server and kept it live under their own address.
Around it sat everything a product needs: a control panel in Ruby on Rails, subscriptions and metered AI usage, single sign-on for every app, team seats, and an approval step that makes the agent wait for a human before risky actions. In June 2026 a bakers’ edition invited my sourdough audience to build their own bread calculators and baking assistants.
The result
clawy worked, and people paid for it: 30 people over its lifetime, and about 120 servers at the peak, one per user. But those 30 people came from my own circle. When I invited bakers who follow me for sourdough, they did not want to build. Not because it was hard. Building was never their wish. They wanted bread.
The real cost was never the servers. It was every month I spent on clawy instead of on the work people were already asking me for. Without demand beyond my own circle, I decided in September 2026 to shut clawy down.
My hammer is building things, and I love it too much to see the bias while I am swinging it. The question I should have asked on day one is now the first question of every AI Discovery: does anyone else want this, or do I just want it for them?
What I learned
-
Not everyone wants to bake.
Most people do not want to bake bread. They want to eat it. Most people do not want to build software either. They want the result.
-
A mirror is not a market.
The people who paid were engineers, founders and people who build for fun. They liked clawy because they were like me. That proved nothing about anyone else.
-
Sell the result, not the tool.
A car workshop told me their Google Maps rating matters more than their website. They did not want a tool. They wanted the result.
-
Ask before you build.
The week I stopped building and started asking, two hotels told me what they would pay for. One said yes the next day.
For the Head of IT
- Architecture: one cloud server per user, claimed from a pre-warmed pool and running an open-source AI agent; a central Rails control plane for provisioning, health checks, updates and billing; HTTPS for every user app through one edge server; all model calls through one metered AI gateway.
- Verify it yourself: the agent clawy ran on is open source at github.com/openclaw/openclaw, and clawy’s fork is public at github.com/Allonsygmbh/openclaw.
- Why it matters here: I have run AI agents in production for paying users, with their costs and their failure modes. That is the feasibility side of every proof of concept.