I write code and
buy software companies
Sold 2 startups I built from scratch. Started xo.capital — we've bought, fixed, and sold 10+ companies including 3 YC startups.
Currently a dev at Motion
Current Projects
Endless Series
AI-powered commodity apps that do things end-to-end
Dev Tools
Developer tools targeting software engineers
Failed Projects
thecompiler.io
Killing this project
The Post-Mortem
I saw Thomas Tunguz (or whatever that VC's name is) do an article on content overload. Thought it was cool. The problem resonated with me — I had all these podcasts, newsletters, and news I never had time to read. So I built a tool to grab it all and create a digest pulling out only what you care about.
Classic mistake: did no research, talked to no users. Just built it because I thought it was useful to me.
Marketed it to UCLA students. Their usage? Zero. Didn't do any ad spend but managed to get about 102 visitors and 6 signups. The brutal truth? It wasn't useful.
What I missed: Turns out Perplexity can do it all. Or if you want to get fancy, it's just an n8n workflow + Perplexity + Resend. No need for a separate tool. No need for a SaaS. Just... no need.
RIP The Compiler. You taught me to talk to users before building.
noco.io - AI App Builder
March 2025 - Nov 2025
The Post-Mortem
Built an AI-powered app builder with VS Code in the cloud. Got 110 users signed up. ZERO people successfully built an app besides me. Spent 4 months and $5k building it. At $20/user, the unit economics were brutal — would have lost a lot of money.
What I learned: Went deep on AI agents — used CrewAI, Vercel AI SDK, and LangGraph. Learned why Cursor is genius for using VS Code's language server. Built a rad system that could spin up totally isolated code containers in Kubernetes — each user got their own sandboxed environment.
The brutal truth: The product-market fit was broken from the start. We told people "hey, you don't have to know how to code!" and then gave them a full VS Code editor. Not exactly user-friendly for non-coders. People don't have the patience to learn or try anything other than "one shot" solutions. They want it to work immediately or they bounce.
Personally super bummed this one didn't work out — would have been a fun company to run. The tech was cool, the infrastructure was solid, but none of that matters if users don't get value.
Had a fun time building it. Learned a shit ton about AI agents and developer tooling. But the unit economics suck and user patience is even worse.
RIP noco.io. You taught me that unit economics matter more than cool tech.