An AI agent stares at a broken build, reads the repair instructions, and fixes the bug without calling for help. https://vow-lang.vercel.app

POV: You paste a one-paragraph prompt into your agent and it starts writing Vow. Try it in the playground or give the prompt to your agent: https://vow-lang.vercel.app

One submission failed. I'm a high schooler who built Vow, a language for AI agents. Testing with Claude Haiku 4.5 showed finished Vow code passing hidden tests at 99.5% against 99.3% for TypeScript. https://vow-lang.vercel.app

When you let an agent touch your machine, you just have to close your eyes and pray. In Vow every function declares if it uses net, fs or env, and vow --sandbox checks capabilities before anything runs. https://vow-lang.vercel.app

Vow has no nulls and no exceptions. Every failure returns a Result value that must be handled with match or the ? operator, keeping agents from ignoring errors. https://vow-lang.vercel.app

In testing with Claude Haiku 4.5 on hard tasks, Vow had one failed submission in nine runs and TypeScript had zero, performing evenly on a small sample. Finished code passed hidden tests at 99.5% for Vow and 99.3% for TypeScript. https://vow-lang.vercel.app

I built Vow, a programming language designed for AI agents to write. You can try it in the playground or paste the one-paragraph prompt from the site into Claude Code, Codex, or Cursor. https://vow-lang.vercel.app