I've been testing AI tools long enough that not much catches me off guard anymore. You develop a feel for what these things can and can't do, and most of the time the output lands somewhere in the range of what you expected.

Then last week I was researching scooter battery specs — nothing complicated, just cross-referencing some numbers I'd saved — and I fed the information into a custom chatbot I'd built. It didn't just pull the answer. It flagged an inconsistency between two of my saved documents that I'd completely missed and had been working off the wrong number for two weeks.

No prompt asking it to check my work. It just did it.

I'm not saying AI is magic. I've written enough about what these tools can't do to know better than that. But every once in a while something happens that shifts your sense of where the ceiling actually is. That was one of those moments.