Earn the Tool by Catching Its Mistakes

Three agency leaders got on a panel at ADWEEK House this week and said something I have been trying to say for years, just with better timing. Adweek covered the session: Gogi Gupta, Jeremy Heilpern, and Emily Garvey talked about growth, and the conversation turned to AI and new hires. Gupta called the risk cognitive surrender. I know that condition. I watched people surrender to a tool long before anyone could type a prompt, back when a shortcut was all it took. The diagnosis is right. Gupta Media has put a gate on access for new hires, and I understand why. The real question is what should open it.

What did the agency leaders actually say about AI and new hires?

Gogi Gupta of Gupta Media said new hires get limited access to AI tools for their first six months, to curb what he termed "cognitive surrender." Adweek reports that Jeremy Heilpern of Ammunition has discussed barring new hires from AI for their first few years. He also said Ammunition has an AI innovation unit that evaluates its tools. Emily Garvey of Zulu Alpha Kilo said AI cannot replace good judgment.

Read side by side, all three are talking about the same thing: judgment, and where a new hire gets it. Gupta's answer is a limited window. The far end of Heilpern's thinking is keeping the tool out for years. Garvey wants younger talent where they can see judgment in practice. I agree with the concern. I part ways on what opens the door.

Is cognitive surrender a real risk, or an old problem with a new name?

Cognitive surrender is real, and I do not think AI invented it. I watched editors trust an auto-sync without checking the slate or watching the lips. I watched designers ship whatever a template defaulted to. Hand someone a tool that lets them skip the thinking and the thinking gets skipped.

I came up producing, directing, editing, and designing brand systems before I spent about 17 years building and leading a creative team, and I saw versions of this on every one of those jobs. A colorist accepting the auto match without walking the shot by eye. A junior editor trusting the scene-detect cuts without watching the footage through. A sound designer leaving the preset EQ on instead of letting their ear make the call. None of that needed artificial intelligence. It only needed a shortcut and someone willing to stop thinking once they had one. The pattern is old. The stakes are the same: whether a person builds the judgment that lets them know when to trust their own eye and when a tool has gone wrong.

Does limiting a new hire's AI access for a set period build the judgment to catch its mistakes?

Time does not build judgment. Judgment is built with the tool in front of you, on a decision you own and must defend out loud. A window that keeps the tool at a distance delays the one skill the job will demand: catching it when it is wrong. You learn by working next to the output while someone with judgment shows where it goes wrong.

Heilpern said it best: "You need to be able to know when AI is lying to you." You cannot learn to spot a lie without the thing telling it sitting right there. A time-based gate does not touch that skill. Six months can pass and the door opens, and if that person never had to catch the tool in a mistake along the way, the calendar moved and the instinct did not. Keep the tool out for a few years and you have only delayed the day they get fooled.

How should a new hire earn wider access to AI tools?

Keep the gate, but let proof open it, not the calendar. Give the new hire a real decision to own and defend out loud. Put the tool's output next to their own first take. Have someone with judgment narrate exactly where the AI went wrong. Widen access as they show, on real work, that they can catch the errors themselves.

In practice, the shape looks like this:

• Early on, the new hire's instinct goes next to the tool's answer, and someone experienced walks them through the gap.
• On each decision they defend out loud, they build the pattern recognition that makes catching errors an instinct, not a checklist.
• Access widens as they prove, on real work, that they can spot where the tool went wrong without being told.

Gupta and Heilpern are right that something gets lost when a new hire defers to the machine before they have earned the judgment to catch its mistakes. The fix is handing them a real decision, the tool included, and making them prove they can catch its errors.