People are already finding jobs to give AI. The conversation I want to have with teams is about what happens when the result comes back looking finished.
An answer can read well and still be wrong about the product. A tidy report can have the wrong attribution window. A set of ad variants can all look plausible and carry the same claim nobody checked.
The work either side of the prompt
The input starts with knowing what the job is. If I ask for an explanation of yesterday’s acquisition numbers, someone needs to say which numbers we trust, which comparison is meaningful and what the team needs to decide. That is part of the brief, even when it’s easier to type “what happened yesterday?”
AI can do a lot of the work in between. It can assemble the first pass, point out odd patterns, or produce versions worth looking at. Afterwards, a person needs to check the parts that matter. Is the suggested explanation supported by the data? Does the line in the ad match what the product can do? Is the result ready to put in front of someone else?
The test I keep coming back to is whether the person sharing the work can explain why they trust it and what they would still want to check.
What leadership can make easier
“Human in the loop” can mean a lot of things. A vague request to check everything can leave people doing a quick read and hoping someone else has looked at the important bits.
I’d rather agree on what needs checking, who checks it and when the stakes call for a second person. An internal outline and a customer-facing claim deserve different levels of scrutiny. So do a rough creative idea and a number that might affect how much money the business spends tomorrow.
It also helps to show the team work in progress. Share the brief, what AI produced and what you changed before sending it. Seeing someone catch an error or discard a neat-looking answer makes the judgement visible. A polished demo usually skips that bit.
Where my thinking is today
I want teams to feel comfortable giving AI the heavy lifting. I also want the person who sends, publishes or acts on the result to know which parts they’re putting their name to. Those two things get easier together when the expectations are clear.