Record lessons learned

Keep a workspace-wide record of what went wrong and how it was fixed, so the same mistake isn't repeated on another project.

Lessons Learned is a running record of problems you've solved — what broke, why it broke, and what actually fixed it. It lives at Lessons in the sidebar, under Agentic.

Lessons are kept for the whole workspace, not for one project. That's deliberate: a mistake made on one project is usually worth knowing about on the next one, and keeping lessons separately means nothing is lost if you tidy away the project that produced them. The Lessons Learned library is created for you automatically and can't be deleted.

Record a lesson

  1. Open Lessons in the sidebar.
  2. Click Record a lesson.
  3. Fill in:
    • What happened — a short summary you'd recognise later.
    • Project — optional. Tag the lesson to the project it came from so you can filter to it later. Leave it blank if the lesson applies everywhere.
    • What went wrong, and why — the cause, not just the symptom.
    • What fixed it — specific enough to act on next time.
  4. Click Record lesson.

The most useful lessons capture the cause. "The build failed" isn't much help six months later; "the migration ran before the column existed, because the deploy order wasn't pinned" is.

Find lessons again

The Lessons page lists everything recorded in the workspace, newest activity first. If you've tagged lessons to projects, use the Project filter to narrow the list to one project, or choose All projects to see everything.

Click any lesson to read it. Use Full editor → to open it as a normal file if you want to revise or expand it.

Working with AI assistants

Assistants connected to your workspace can read and add lessons themselves. That's the real payoff: an assistant starting work on one project can check what went wrong on your others first, and record anything new it works out along the way — so the knowledge accumulates instead of being rediscovered.

You don't have to do anything to enable this. Lessons are only fetched when they're asked for, so they don't slow anything down or clutter the context your assistants load.