Lead post-mortems to capture lessons learned and improve team process

When the agent gets stuck or takes a wrong turn, it’s frustrating — and once you fix it, it’s tempting to move on. But skipping the retrospective means repeating mistakes.

Make time to analyze what went wrong

As the tech lead, your job is to understand why the agent struggled:

• Was context missing?

• Were requirements ambiguous?

• Did the problem need to be broken down differently?

If you’re using Git (you should be!):

1. Stash your working changes — keep your progress safe.

2. Clear the agent’s context window — avoid “cheating” with the new knowledge.

3. Try adding context to any of the files we've mentioned so far:

   1. The spec for the file

   2. README.md

   3. project-architecture.md

   4. coding-guidelines.md

   5. git-usage.md

4. Run it a few times to get a feel for what context helps. Try a few different problem breakdowns!

   1. Update your specs and guidelines with what you’ve learned.

   2. If you find a better way to structure problems, add it to AGENTS.md so future tasks can follow the improved approach.

Treat failed runs as data for better prompts, stronger specs, and smarter processes so you'll learn from any mistakes and strengthen your working practices.