What's the worst "done" an AI agent ever gave you?

Hi, I'm Oleks. I build kwant, and it started with an AI agent telling me "all tests pass" when the tests had never run. Another one sent me a link to back up a fact, and the page said something else.

So kwant has one rule: a task turns Verified only when kwant runs the check itself, not when the bot says it's done.

We launch here tomorrow, Nov 3. Before that, I'd love to hear yours: what did an agent say it finished, and how did you find out it hadn't? I'll reply to every story.