Five things Mel and Mo believed about AI at work that turned out not to be true.
The full script, word for word — 299 words, about 1 minutes to read. Current as at September 2026. AI tools change quickly; if something looks different when you try it, check the product's own help pages.
Mel: This is What I Got Wrong. No tips. Just the things we believed about AI at work that turned out not to be true. Mo. Start.
Mo: I thought the trick was the prompt.
Mel: It isn't?
Mo: It's the context. A short instruction with the actual document attached beats a clever paragraph with nothing attached. I spent a year writing better and better prompts about documents I had never given it.
Mel: I thought I had to get the question or prompt right first.
Mo: You don't.
Mel: I would sit there editing the question before I sent it. Now I ask it badly, read what comes back, and question it again. It's faster, and the answer is better.
Mo: My second. I thought getting better at AI meant using it for everything.
Mel: You tried.
Mo: I did. One-line emails. Simple sums. Questions I could answer before the app opened.
Mel: What changed.
Mo: I noticed some of the shortcuts were taking longer than the work.
Mel: Your third.
Mo: I thought it was smarter than me, so I could trust whatever it gave me.
Mel: You can't.
Mo: You can't. I double-check anything important — and anything I'd be embarrassed to get wrong.
Mel: Last one.
Mo: I thought it would be objective. Because it's a machine.
Mel: And it isn't.
Mo: More often than not it agrees with me. If I put my view in the question, it comes back in better sentences, and I mistake that for confirmation. Now I debate with it. I challenge whatever it gives me.
Mel: Does that work.
Mo: It has changed more than one decision.
Mel: Because the AI disagreed.
Mo: Because I had to defend myself.
Mel: And were you wrong.
Mo: We have a title.