Agentic Coding Agent

Pick a coding problem. An AI writes a Python solution, the code runs in a locked box with no internet, and it's checked against tests the AI never saw. If it fails, the AI is shown what went wrong and tries again, within hard limits so it can't loop forever.

  1. AI plans
  2. AI writes code
  3. Sandbox runs it
  4. Plain code checks every test
  5. Retry on failure, max 3

1. Choose a problem

5 runs per visitor per day

Want to see it fix its own mistake? Open a recorded run on a twisted problem where the first two attempts fail.

2. Watch the run

Choose a problem and press Run the agent.
Each step appears here as it happens.