Hiring for AI fluency is easy. Growing someone you can trust with an outcome is the hard part, and the way we used to do it just got automated.

Key Takeaways

• Fluency with tools is cheap to acquire and easy to fake. Judgment is neither.

• Judgment was learned through repetitive low-stakes work, which is exactly the work that no longer exists.

• Small teams can rebuild the ladder on purpose, and they have an advantage over large firms in doing it.

The Fluency Test

UBS made news last week for a hiring rule. Graduates and interns joining its banking division in 2027 will have to show, in the interview, that they can work with AI. Not that they have opened a chatbot, but that they have handed a real task to a model and can explain how the result beat what they did before. Santander has a similar rule for some trainees.

You may run a ten-person company and think this has nothing to do with you. But the same logic is already in your job posts. "Comfortable with AI tools" is on the list. You want people who arrive able to use the machine.

Fair enough. But notice what the test measures. It measures whether someone can operate a tool. It says nothing about whether they can tell when the tool is wrong. And the second skill is the one you actually pay for.

Never-Skilling

For as long as there have been professions, people learned judgment by doing the boring work first. The junior analyst built the model badly, got corrected, built it better. The new account manager wrote the weak email, watched it fail, wrote a stronger one. The apprentice swept the floor and, somewhere in the sweeping, learned how the shop worked.

None of that work was valuable on its own. It was valuable because it was cheap to get wrong. You could fail a hundred times at low cost and arrive, eventually, at the instinct that lets a senior person glance at a plan and say "that won't work" before anyone has spent a dollar.

That repetitive work is precisely what the machines now do. Coverage of the UBS decision called this "never-skilling," and the phrase deserves to stick. It is not deskilling, where a skill you had erodes. It is never getting the chance to build the skill at all, because the rungs were removed before you arrived.

Large firms can paper over this for a while with senior headcount and training budgets. You cannot. In a small business the gap shows up fast, as a bright new hire who can generate anything and evaluate nothing.

Where Judgment Actually Comes From

Judgment is not knowledge. It is a record of having been wrong in a specific domain often enough that your errors became predictions. The senior marketer who knows a subject line will fail is not reciting a rule; she is remembering forty that failed.

That record is built from three ingredients. Repetition: you do the thing many times. Feedback: you find out, quickly and clearly, whether it worked. Low stakes: the failure has to be cheap or nobody lets you fail.

The old ladder supplied all three by accident. Tedious work was repetitive, the boss checked it, and nobody cared much if the intern's first draft was bad. Now the draft is instant and passable. The repetition is gone, the feedback loop is gone, and the stakes have moved up, because the first thing a new hire touches is a real thing.

So the question for anyone leading a team is no longer how to hire fluency. It is how to supply repetition, feedback, and safe failure on purpose, now that the work no longer supplies them for free.

The Practice Ledger

Here is a diagnostic you can run in a week. Take one role on your team, ideally the most junior. Write down the five judgment calls a great person in that role makes that an average person gets wrong. Which customers to chase. When to say no. What "done" looks like.

For each call, ask two questions. How did the last person who was great at this learn it? And does that path still exist? Most of the paths ran through work that is now automated. That is your list of missing rungs.

Then, for each missing rung, design a replacement with the three ingredients. A weekly review where the new hire predicts which of three campaigns will perform best, then sees the result. A rule that the agent's output goes to the junior first, who marks what they would change and why, before a senior person looks. A monthly hour where the most experienced person on the team walks through one decision they got wrong.

None of this costs money. It costs attention from senior people, which is the point. Seth Godin has argued for years that the work that matters is the work that requires a person to care. Teaching someone to see is that kind of work, and it cannot be delegated to the tool that created the problem.

From Insight to Action

1. Rewrite one job post this month so that it asks for a judgment story, not a tool list. "Tell us about a time you overruled a recommendation and were right" beats "proficient with AI."

2. Give every new hire a prediction habit in their first week. Before any result comes in, they write down what they expect and why. The gap between the guess and the outcome is the curriculum.

3. Route automated output through the least experienced person first, with the explicit job of finding what is wrong with it.

4. Protect one hour a week of senior time for teaching. Put it on the calendar before the client work fills it, because the client work always fills it.

5. Keep a shared log of decisions that went wrong and what they taught. Make it safe to add to and boring to read, which is how you know it is honest.

6. Measure the team on outcomes it can defend, not on volume it can generate. Volume is now free.

A company that hires for fluency defaults to a team that can operate tools. A company that designs for practice defaults to a team that can be trusted with outcomes.