The AI Skill Gap Nobody Is Training For

Ask an AI tool to draft a difficult email to a client and you will have three polished versions before your coffee finishes brewing. One concise and direct. One warm and diplomatic. One that reframes the bad news as an investment in quality. All three read well, but working out which one is right for this client, this history, this relationship, is where the real work starts.

That gap, between what AI can produce and what only you can judge, is becoming the whole story of working with AI at the moment. Taste is a valuable commodity

Most AI training solves the wrong problem

Walk into most companies rolling out AI right now and the training looks the same. Prompting workshops or Copilot certifications. Maybe an afternoon on how to phrase a request so the model gives you something usable. All of that is worth doing.

None of it touches the real, observable bottleneck. Producing a competent first draft used to take hours of skilled effort. Now it takes seconds. What is hard is knowing whether the draft is any good: what to trust in it, what to question, what needs rewriting before it goes anywhere near a client. That is judgment, and a prompting course does not teach it.

AI flips the way people used to get good at their jobs

For decades, becoming an expert followed a predictable path. You start out consciously following rules, and with enough repetition, the rules disappear into instinct. A seasoned prosecutor reads a courtroom without working through the logic step by step. A good facilitator can sense the unspoken tension in a workshop before anyone says a word. Researchers call this tacit knowledge: the point where you know more than you can explain, built through years of doing the task yourself. A form of professional muscle memory.

AI reverses that direction somewhat. The model has read almost everything published and knows nothing about your client's politics, this quarter's pressures, or why a particular phrase will land badly given what happened last time. To get anything useful out of it, you have to say out loud the things experienced people usually leave unspoken: what good looks like here, which assumptions need challenging, what matters and why. The professionals getting the most out of AI are not the best prompters. They are the ones who can put their judgment into words clearly enough for someone, or something, else to act on it.

A habit worth building, not a tool worth learning

There is a useful five-step routine here, adapted from research by Harvard Business School, and it works for almost any AI-assisted task where your judgment matters.

Form a view before you open the tool. Sketch what a good answer looks like, based on what you already know, before AI shapes your thinking for you. If you cannot critique a finished version of the work, you are not oriented enough yet.

Push AI past generation. Most people only ask it to produce something. Ask it to critique its own output, compare the tradeoffs between versions, simulate how a specific stakeholder would react, and challenge its own weakest assumptions. Each of those modes drags reasoning into the open, yours and the model's. Push hard - the models are now smart enough

Map the gap. Compare your original view against what AI produced. What did it add that you missed? What did it get wrong because it does not know your context? What looks plausible but is not, the kind of error that survives a casual read and only gets caught by someone who knows the account?

Hand it over with the reasoning attached. The output goes to the client. A short note on what AI produced, what you changed, and why, goes to your manager. That note is what turns a task into a development opportunity instead of just a completed deliverable.

I'm going to add a surprise, FIFTH step here. Actively disclose AI use to the client. This transparency in a world full of AI slop will be appreciated. Your clients will likely be using AI in some way, better to have it out in the open.

The advantage for managers

If you manage people, this is where the payoff lies. Ask for that reasoning trail on every AI-assisted piece of work your team sends you: what AI produced, what they changed, and why. It gives you a concrete basis for coaching that "check this over" never did, and it forces your team to build the very skill that is becoming scarce. Don't give up on critical thinking.

The organisations that build this into how people work will develop stronger judgment faster than old-fashioned apprenticeship ever managed, because the reasoning that used to stay locked in someone's head now has to be written down.

And always remember... you could just write something WITHOUT AI support. Writing is thinking and learning. Let's not forget that.

If you are working out how to build AI judgment across a team rather than leaving it to whoever happens to be curious, that is a lot of what we do at Futureformed. Have a look at our work, or drop us a line for a chat.