Coaching for Judgment: The Skill AI Makes Scarce

A junior marketer can now produce ten campaign concepts before lunch. An analyst can generate a full market sizing in minutes. The output that used to prove someone was good at their job is suddenly cheap, and managers are finding that volume tells them very little about who is actually getting better.

What still separates people is judgment. Picking the one concept worth testing. Spotting the assumption that quietly breaks the market sizing. Knowing that a polished answer is wrong. That skill was always important. What changed is that it is now most of the job, and it does not develop by accident when the machine does the first draft.

This piece covers why judgment gets scarcer as AI spreads, how it is actually learned, and the coaching practices a manager can use every week to build it in a team.

Why cheap output makes judgment more valuable

The clearest way to see this comes from economics. Agrawal, Gans and Goldfarb frame recent AI progress as a drop in the cost of prediction, and they define judgment as the work of determining payoffs: deciding what each possible outcome is worth to you. Their conclusion is that prediction and judgment are complements as long as judgment is not too difficult (NBER Working Paper 24243, 2018). When one gets cheaper, demand for the other goes up.

Generative AI extends the same logic from prediction to production. For many everyday work products, such as text drafts, code snippets, slide outlines and summaries, the cost of a first version has collapsed. The expensive part moves to the edges of the work: framing the right question at the start, and deciding at the end whether the output is good enough, true enough and worth shipping.

Employers are already saying so. In the World Economic Forum Future of Jobs Report 2025, analytical thinking remains the most sought-after core skill, with seven in ten companies calling it essential, and creative thinking is expected to keep rising in importance through 2030.

What judgment means in practice

Judgment sounds abstract, which is why it rarely gets coached. Broken into parts, it becomes something you can observe and give feedback on:

• Framing. choosing the problem worth solving and the question worth asking, before any tool is opened.
• Standards. having a clear picture of what good looks like for this audience, this brand, this decision. Taste is the name we give to well-calibrated standards, and you know it is calibrated when it keeps predicting how the audience or the result will actually respond.
• Verification. checking claims, numbers and sources instead of trusting fluent prose.
• Trade-offs. weighing speed against quality and risk against upside, and saying out loud which one you picked.
• Knowing when to stop. recognising the point where another iteration adds polish but no value.

Each of these is visible in how someone talks about their work. That is the opening for a coach.

The trap: AI can erode the skill you need most

The risk is not that people stop working. It is that they stop thinking at the moments that matter. A study of 319 knowledge workers, who shared 936 real examples of using generative AI at work, found that higher confidence in AI was associated with less self-reported critical thinking, while higher self-confidence was associated with more (Lee et al., Microsoft Research, CHI 2025). The finding is a correlation, not proof of cause, but the same study found that critical thinking shifts toward verifying information, integrating responses and stewarding the task. The work does not disappear. It changes shape, and people need help noticing that.

The cost of misplaced trust is measurable. In a preregistered experiment with 758 BCG consultants, those using GPT-4 completed 12.2% more tasks, 25.1% faster and at higher quality on tasks inside the AI’s capability frontier. On a task deliberately chosen to sit outside that frontier, consultants using AI were 19% less likely to reach the correct solution than those working without it (Dell’Acqua et al., Harvard Business School, 2023).

The researchers call this a jagged frontier: AI helps on some tasks and hurts on others of similar apparent difficulty. Nobody hands you the map. Knowing which side of the line a task falls on is itself a judgment call, and it is one of the most useful things a manager can teach.

How judgment is actually learned

Daniel Kahneman and Gary Klein spent years on opposite sides of the debate about expert intuition, then wrote a joint paper on where they agreed. Their conclusion: whether an intuitive judgment can be trusted depends on how predictable the environment is and whether the person has had the opportunity to learn its regularities. They add that subjective experience is not a reliable indicator of judgment accuracy (Kahneman & Klein, American Psychologist, 2009). In the paper they spell out that learning needs feedback that is both rapid and unequivocal.

That gives managers a practical brief. Feeling sure is not evidence of good judgment, for your people or for you. Judgment grows when someone makes a call, sees the result quickly, and can tell clearly whether they were right. Most workplaces are poor at this. Feedback arrives months later, or never, or is too vague to learn from. AI makes it worse, because the person who accepted a flawed output often never finds out it was flawed.

So the coach’s real job is to engineer feedback loops: shorter, clearer and tied to specific decisions.

Six coaching practices that build judgment

1. Ask for the reasoning, not just the result. in every review, ask "why this version and not the others?" and "what did you check?" Someone who can explain what they cut and why is building judgment. Someone who cannot is forwarding output.
2. Make predictions before outcomes. before a campaign launches or a proposal goes out, have the owner write down what they expect to happen and how confident they are. Compare when results arrive. A simple decision log turns vague experience into the rapid, clear feedback that Kahneman and Klein describe.
3. Compare versions side by side. put two drafts, two AI outputs or two strategies next to each other and ask which is better and exactly why. Comparison trains standards faster than critique of a single piece, because the differences become concrete.
4. Run premortems on the big calls. Gary Klein’s premortem, described in Harvard Business Review in 2007, asks the team to assume the project has already failed and explain why. It makes it safe to voice doubts, and it rehearses exactly the scepticism people drop when an AI answer looks confident.
5. Map the frontier as a team. keep a short shared frontier map: the tasks where AI output has proven reliable and the tasks where it needs a human check. Update it when someone gets caught out. This turns individual mistakes into team judgment.
6. Delegate decisions, not just tasks. hand over a real decision with clear bounds ("choose the subject line, stay within brand tone, budget is fixed") and let the person own the outcome. Judgment does not grow in people who are only ever asked to execute.

What to stop doing

Some common management habits quietly block judgment from developing:

• Silently rewriting their work. if you fix it without explaining why, they learn nothing except that you will fix it.
• Praising volume. celebrating how much someone produced with AI rewards the part that is now cheap.
• Approving without asking. a quick "looks good" on something you did not interrogate teaches the team that review is a formality.
• Answering every question. when someone asks what you would do, ask what they would do first, then compare.

How to tell it is working

Judgment is hard to measure directly, but the signs show up within a few months:

• Fewer rework cycles. drafts arrive closer to right the first time.
• People catch errors before you do. including errors in AI output they generated themselves.
• Better-calibrated predictions. the decision log shows confidence levels that match actual hit rates more closely over time.
• Sharper disagreement. people push back on your view with reasons, not just preferences.

A 30-day starting plan

You do not need a programme. Start small and make it routine:

• Week 1. add "why this and not the alternatives?" to every one-to-one and review.
• Week 2. start a shared decision log for any call above an agreed threshold, with a predicted outcome and a confidence level.
• Week 3. run one side-by-side comparison session and one premortem on an upcoming launch.
• Week 4. draft the team’s first frontier map, and hand one real decision to someone who has not owned one before.

If you are still getting the team comfortable with the tools themselves, start with our guide to making a team AI-ready. Coaching for judgment is the next layer: once people can use AI, the question becomes whether they can tell when it is right.

Output is becoming a commodity. The people and teams who pull ahead will be the ones who know what to keep, what to question and what to throw away. That does not happen by default. It happens because a manager decided to coach for it, one question at a time.

If you want help building judgment into how your managers lead and review work, that is the kind of work we do at Gaveau Strategy.