WSJ: Use AI for Mechanical Work, Keep Humans on Trust Calls

There is a useful frame in this WSJ piece on when AI does more harm than good, and it has nothing to do with which model you use. The split is between work that is repeatable and mechanical, and work that runs on values, relationships, or trust. AI belongs on the first pile. Humans belong on the second. Most teams have been quietly mixing them up.

The examples land because they are mundane. A model can scan a stack of resumes and short-list candidates, but it cannot tell you which of them will hold up when the customer call goes sideways. It can surface patterns in a strategy deck, but it cannot read how the team will react when you announce the new direction in the all-hands. The way the article puts it is that AI is best used as raw material, not as finished work. You bring the judgment about what to trust and what to throw away.

If you manage anyone, the practical version of this is a list you can write on a sticky note. Timelines, summaries, first-draft slides, number-crunching, that all goes to the model. Feedback conversations, hiring decisions, the hard call about who to promote, those stay with you. The piece also suggests a smaller habit worth borrowing: occasionally ask the model for the counterargument rather than the validation, so it widens your perspective instead of confirming whatever you already think.

The honest caveat is that the article is a framework, not a measurement. It does not tell you how often AI-assisted judgment actually backfires in practice, or whether first-line managers should draw the line in a different place from senior leaders with more reps. And the slow risk it points at, that you quietly stop noticing what you have outsourced, is exactly the kind of thing that does not show up in a productivity dashboard.

What is worth taking away is the orientation. Over the next year, the interesting move is probably being deliberate about which decisions you let any model anywhere near, rather than chasing the next smarter model.