"In the loop" is engineer talk for the person who signs off before a machine's work goes out. Most people hear the line as a job guarantee: keep a person somewhere in the process and the work stays yours.

Mollick means something harder. A loop only has a human in it if the human is awake.

Picture the ordinary version. The draft arrives, clean and confident. You skim the first paragraph, the numbers look like numbers, and you hit send.

You were in the loop the way a sleeping guard is in the building.

A Harvard researcher, Fabrizio Dell'Acqua, caught this with recruiters sorting job applications. The ones handed a very good AI paid less attention and made worse calls than the ones handed a weaker tool. Good software taught them to stop looking.

That is the trap. The better the machine gets, the less your check feels needed, and the more it matters on the day the machine is wrong.

Being the human does not mean reading everything twice. It means deciding in advance where you will look hardest: the figure a client will quote back to you, the clause that moves money, the name of a real person.

Pick those spots before the draft lands. A reviewer who waits until the text looks fine has already decided to check nothing.

So today: open one thing an AI wrote for you this week that you sent without a real read. Check the three claims that would cost you most if they were false.

If all three hold, you have spent ten minutes. If one breaks, you have found out who was running the loop.