George Hanu · Coding with AI

Series

How agents work

Open the box: build the loop, then make it safe to ship.

2 guides · 14 lessons · free, no signup

Lessons + web app

Part 2 · Harden Your Agent

Six defences a real agent needs, each added to a small TypeScript agent and proved by a test you can run.

  • TypeScript
  • Switchable defences
  • Runs without an API key
  • You've built an agent, or read The Agent Loop
  1. Where untrusted text gets in
  2. Keep the agent on task
  3. What the screen and the fetch tool can give away
  4. Acting as whom
  5. Tools that must not run twice
  6. Memory you can trust
  7. The hardening test suite
Read it: Harden Your Agent Code on GitHub: Harden Your Agent

More series

Working with coding agents Get reviewable, checkable work out of Claude Code, Codex and friends. 2 guides · 13 lessons → AI in real apps Small AI features with schemas, checks and a person in charge. 1 guide · 7 lessons →