I wrote a rule. The AI broke it 20 minutes later.
Not despite the rule. After it.
What happened
I wrote a rule into my AI assistant's permanent memory:
Saved. Good feeling. Back to work.
Twenty minutes later I spot a payment reminder in my calendar. Set for 09:00. The AI had entered it right after I saved the rule. Not despite the rule. After it.
Writing something down is not the same as following it
This is the point most people miss with AI systems, and honestly with most systems: writing a rule down creates a false sense of done. 'I have documented it, so it is handled.' But documentation and behaviour are two different things. The rule was in memory. The behaviour had not changed yet.
This is not an AI problem
Organisations write AI policies that no one reads. Executives document processes that no one follows. Teams put rules into a wiki that do not survive the next sprint.
The machine simply showed me, in the clearest possible terms, what happens in every company: something looks followed on paper and is not.
The fix was simple
I now have it check whether the rule is active in the running session, not just saved for the next one. One verification step. Two minutes.
The lesson was bigger than a calendar entry. Documenting a rule and following it are not the same thing.