Guides · Why it happens
Why does my AI keep making a mistake I already corrected?
Short answer
Your correction lived in one chat, and that chat ended. The next session never heard it. Save the correction as a note, and Clyde recalls it the next time the same task comes up. Notes an agent marks wrong drop in the ranking, so bad advice stops coming back.
Make a correction stick
- When you correct the agent, tell it to remember the correction. One short note, like "tests run with pnpm test, never npm test".
- Next time, recall brings the note back before the agent repeats the mistake.
- If a recalled note was wrong, the agent says so with
feedback. Wrong notes sink, and used ones rise.
When the old advice is the problem
If the agent keeps following something that used to be true, the fix is to replace the note rather than add another one. See when an agent follows an old instruction.
Setup is on the docs page.
Questions
Will one correction fix it everywhere?
Every connected agent can recall the note, in any app, once it's saved.
What if the agent saves the wrong thing?
Tell it to replace the note so the old one stops being served, or mark it wrong so its confidence drops.
Does this work in Claude Code?
Yes, and there the hooks make the agent recall before it starts a new task.