I Used to Fix Claude's Mistakes and Move On. That Was My Worst AI Habit.

I used to fix Claude's mistakes and move on. That was my worst AI habit. I was treating every correction as a one-off, like fixing a typo, when I should have been treating it as training.

The problem with that approach is that the same mistake comes back. Claude has no memory between sessions by default. So every correction I made was invisible the next time I opened a chat. I was doing the same corrective work on repeat and calling it using AI.

What actually works is adding one line to the end of every correction: "Add a rule to your CLAUDE.md so this doesn't happen in the next session." CLAUDE.md is a plain text file that Claude reads at the start of every conversation. It's where you put everything you want it to remember about how you work. When you catch a mistake and ask Claude to write a rule for it, that rule is there the next time you open a session. The mistake doesn't recur.

I've been doing this weekly for months. The file gets more specific over time. The list of things I have to correct gets shorter. The outputs I get on the first attempt get better. It compounds in a way that takes a few weeks to feel, and then becomes obvious.

The business owners I work with who build this habit tell me the same thing: after a month, their Claude starts to feel like someone who actually knows them, rather than a capable stranger they have to brief from scratch every time.

Every mistake is a simple instruction waiting to be written. Add the rule. Close the tab. Let the file do the work next session.