Are we, as data analysts, on borrowed time?

Can years of domain expertise and nuance be replaced by a clever prompt?
Instead of debating, I ran a real experiment: I gave an actual exam (once used at eBay) to two leading LLMs—Gemini 2.5 and Manus AI—with no guidance other than what a candidate would get.
The results were enlightening, odd (and even*) deeply revealing.

This talk walks through the test, the traps, and how LLMs fumbled or succeeded. It’s not a Luddite rant—it’s a field report on what AI can (and can’t) do when context, reasoning, and data intuition really matter.

* got it? odd? even? %2==0? Oh, nevermind