Writing
Notes on building AI agents that are safe to run, and on testing them. Follow along with the RSS feed.
- Enforcing an AI agent's dry run inside the browser
Why a prompt that says "don't submit" is not enough, and the guard, hook and audit that make Preflight's dry runs hold.
- Testing an agent that must never submit
How Preflight's test suite proves its dry-run guard works, with a fake LLM, real Chromium and requests that never arrive.