AI Collaboration

AI is fast. Can you tell when it's wrong?

Deloitte Australia submitted a A$440,000 government report with invented academic citations. A New York lawyer filed six fake cases that ChatGPT had made up. Neither was a prompting problem. The output just looked real.

That's the risk. And it's everywhere now — AI writes reports, analyses data, builds launch assets, checks contracts, generates code, talks to customers. Everyone has the same tools. The only skill that separates people is whether they can tell when the output is wrong.

The Handover.

Someone quits. No notice.

The launch goes live Monday. Registrations are already open.

You get his files. Two days. One teammate: AI.

You start with a plan the model writes in thirty seconds. It looks complete. It isn't — a third of it is invented, and you won't spot that by reading it.

So you go into his drive and make it prove every line. Which file says the price? Which one says the room holds that many? What didn't it open at all?

The Deadline.

Then, it turns. You audit another team's finished launch. Twelve assets. Four are rotten.

You won't be told which.

You can't ask the person who built them.

He's gone, like people sometimes do.

Then the actual build. Page, form, payment, database, emails, pricing. Real assets, not exercises — the kind that either work for the launch or don't.

Nobody hands you the answer. There's no brief. There's nobody left who knows what's true.