Being "in the loop" sounds like a safeguard. Mollick means it as a job.

The phrase gets borrowed by compliance decks and vendor slides, where it means a person clicks approve before anything ships. That version is passive — a signature at the end of a process you no longer understand. Mollick's is the opposite. The human in the loop is the one who could still do the work themselves, who reads closely enough to catch the confident error, who owns the output when it leaves the building.

The failure here is quiet. You never decide to stop paying attention. You just approve the draft you skimmed, merge the code you could not have written, forward the summary you did not check. Each one felt reasonable — the model is usually right. Usually is the trap. AI is most dangerous exactly where it is most fluent: a wrong number stated as calmly as a right one, a source that does not say what the paragraph claims it says.

This is why the tool rewards people who already know things. Expertise is what lets you notice the seam. Hand over the judgment along with the typing and you don't just risk one bad output — you stop building the very skill that would have caught it. The loop is not where you supervise the machine. It is where you stay sharp enough to.

So today, take one thing the AI made for you — a summary, a draft, a block of code — and verify a single claim in it by hand. Trace one source. Redo one number. Not because it is probably wrong. Because being in the loop is a muscle, and today is a rep.