Experiments

Things I'm trying, written up as I go. Some will work.

• Agentic built, but who confirms it works? 27 October 2026

  Coding agents write the code and the tests, and the basics nobody prompts for go unchecked. An eval that runs the app and holds it to the industry's own rules, with reproducible proof for every failure.

• Two green checks and a bug 11 October 2026

  An AI code review and an agentic tester both went green over a data leak. How do you trust anything when it's AI all the way down?

• Watching an agent work inside a company 1 October 2026

  Running qm, YC's open source CompanyOS harness, against a synthetic company from Restaged, with an LLM proxy showing every model call.