Your Next Leaders Can Prompt AI. Can They Judge It?
My Apprenticeship
Early in my career, I got my first promotion to head office. I was in trade marketing, supporting our largest sales region. Suddenly I was being asked to make decisions, build materials and analyze data for an entire region. I knew the team. I did not yet know what I was doing.
So I learned the old-fashioned way: badly, then less badly. I'd put together a draft, send it out and get it back with feedback. Then another draft. And another. After enough rounds, I'd go blind to my own changes, right up until I was standing at the front of a regional sales meeting, realizing I'd missed a number, left in a spelling error or dropped in the wrong table.
Each round taught me something no manual covered. What the region actually wanted. What they didn't. What would land with the people in the field and what would get quietly ignored. Why a number that looked fine on my spreadsheet made no sense to someone who knew the market.
It was slow and occasionally humbling. And it would kill me to ship something and then spot a mistake after it was out the door. When you're standing in front of 60-plus people and the mistake is glaring, there's nothing to do but own it and promise a corrected version. That feeling did more for my attention to detail than any training course.
Full disclosure: early on, I let a couple slide when nobody called me out.
I didn't know it at the time, but I was in an apprenticeship.
Why this matters now
Most of my leadership formation happened in an analog world: drafts, red pens, feedback in the hallway, and consequences you could feel.
Today I use AI tools constantly. The most valuable thing I bring to them was built in those early years: the ability to look at a polished answer, notice that something doesn't add up, and keep probing until I find out why. I didn't learn that from AI. I learned it from dozens of drafts that came back covered in comments.
Now a generation of leaders is coming up who may never do those drafts. AI does the first version, and often the second and third. The output arrives faster and looks better than anything I produced at their age. What doesn't arrive is the judgment that used to come with the reps.
Entry-level jobs were secret apprenticeships
Fortune put it well recently: entry-level work was always partly a training program, and AI just cut its funding. The output value moved to the machine. The training value didn't go anywhere. Someone now has to pay for it on purpose.
Recent McKinsey research sharpened the point. Two senior Microsoft engineering leaders described an "AI boost" for experienced people and an "AI drag" for juniors who don't yet have the judgment to steer or check what AI produces. The experienced get faster. The inexperienced get stuck, or worse, get confident.
Meanwhile, the bar is moving in the other direction. PwC's 2026 AI Jobs Barometer found that entry-level roles most exposed to AI are now seven times more likely to require traditionally senior skills. Organizations want judgment earlier than ever, right as they remove the work that used to build it.
The result shows up every day. A report goes out with a statistic nobody can trace. A deck reaches a client with a quote nobody said. A summary confidently cites a source that doesn't exist. Nobody checked, because it looked finished.
Why people don't check
It's tempting to call this laziness. I don't think it is. Four things are going on:
Polish looks like truth. A rough draft practically begs to be checked. A clean one doesn't, even when it's wrong.
They don't know where AI fails. Without experience, you can't tell which claims are risky. Numbers, quotes, citations, names and dates are where things most often go wrong.
Ownership gets fuzzy. When a tool helped write it, it's easy to feel the tool shares the blame.
Speed wins. When a half-day task is now expected to take an hour, checking is the first thing squeezed out.
The new BrandTruth Gap
In our BrandTruth Alignment work, the BrandTruth Gap is the distance between what an organization says and what people actually experience. Unchecked AI output is a fast new way to widen it. Every confident claim that turns out to be wrong costs trust, and customers usually find the error before you do.
Checking your work used to be a personal habit. Now it's a trust issue for the whole organization.
How to build the judgment muscle on purpose
The apprenticeship won't come back on its own. It has to be designed back in. Here's what organizations taking this seriously are doing.
Make them go first. In a recent post, BCG describes a company that requires juniors to frame the problem, test assumptions and produce their own baseline analysis before using AI to refine it. Then compare the two versions and talk through what to keep and why. That conversation is the modern red pen.
Show your red pen. Much of a senior leader's judgment lives in edits nobody sees. Narrate them: "I cut this because the client already knows it." "This number doesn't smell right." Your reasoning is the curriculum. Codifying judgement is a critical training role for leaders today.
Hunt the errors. Plant mistakes in AI output, like a fake citation, a wrong figure or an invented quote, and have your team find them. Make it a game. People remember what they caught.
Your name, your claim. Whatever tool helped, the person who publishes owns it. Scale the check to the stakes: a quick self-check for an internal note, a second set of eyes for anything going to a customer.
Compress the experience. Bank of America is building AI simulations to give interns years of on-the-job learning faster. The irony is that AI is very good at generating realistic practice cases, including ones with deliberate problems to catch.
Test without the tools, occasionally. If you only measure speed and volume, AI will make everyone look competent right up until something breaks.
The system that survives you
In my last piece, I wrote about leading through the AI reckoning, and the principle I keep coming back to: build a system that survives you. Part of that is redesigning your team for a world with AI agents in it.
The other part is harder. It's making sure there are people behind you with the judgment to run it.
For owners and senior leaders, AI adds a new layer of complexity. The work moves faster, the tools change monthly, and judgment that used to build up slowly through drafts and hallway feedback now has to be taught on purpose. But it also opens a door most owners haven't noticed yet.
Much of what made your business successful lives in your head. Why you price the way you do. Which customers you walk away from. What "good enough" looks like. The thirty-second gut call that took twenty years to learn. Until recently, getting that out of your head and into other people took a long apprenticeship, and most of it never transferred at all.
Now you can do it deliberately. Map what has made the business successful. Write down how decisions actually get made, and who makes which ones. Then use that map twice: to train your next leaders, and to show your AI tools what good looks like. Your red pen stops depending on you being in the room.
Do that, and you've built a business that survives you. It's also a business worth more to whoever takes it over, whether that's a buyer, a successor or the next generation, because the know-how they're paying for no longer walks out the door with you.
AI can produce the draft. It can't feel the sting of a mistake that went out the door, and it can't pass that feeling on to the next person. That part is still our job.
