My people are scared of AI. What skills do they actually need now?

Nobody has said it out loud, but you can feel it in the room. People have gone quiet in a way they were not quiet six months ago. The clever ones nod along in the all-hands and then say very little, and underneath the quiet is a question they cannot ask, which is what they are even meant to learn now. You have run the lunch-and-learn. You bought the licences. And still the room is holding its breath.

So here is the answer, plainly, before the diagnosis. The skill your team needs is not another tool. It is four durable capabilities that outlast any single platform: judgement (knowing which problems are worth solving and which answers can be trusted), prompting (asking a machine a precise question and iterating on its reply), verification (spotting where a confident output is quietly wrong) and collaboration with the machine (working alongside it as a colleague while you stay the author of the work). Tools expire. These four compound. Train for these, and the fear in the room starts to convert into capability you can see.

Why does buying everyone an AI tool not seem to land?

Because the tool was never the bottleneck. Most teams already have access; what they lack is a frame for how their own value changes. The World Economic Forum estimates that 39% of the average worker's existing skill set will be transformed or outdated over 2025 to 2030, and that skill gaps are the single biggest barrier to business transformation, named by 63% of employers. A licence does not close a skill gap. A licence sitting on a desk that nobody knows how to angle is just a quieter version of the same problem.

And your people feel that gap personally, which is what the silence is. KPMG's 2025 American Worker Survey found 52% of workers fear AI could eventually replace their job, nearly double the year before, while 84% say they want more training to build their skills. Read those two numbers together and you have the room you are standing in. The fear and the appetite are the same energy pointed in two directions. The training is the thing that points it forward.

What does my team actually use AI for when I am not watching?

More than you think. This is the part that should change how you feel walking back into that quiet room. McKinsey's work found that leaders estimate only 4% of employees use generative AI for at least 30% of their daily work, when the self-reported figure is closer to 13%, roughly three times higher. People are already bringing tools to their desks; Microsoft and LinkedIn found 75% of knowledge workers now use AI at work, and 78% of those users bring their own. McKinsey put it cleanly: your people are far more ready for this than their leaders imagine.

So the work is not to drag a reluctant team toward the future. It is to take the experimentation already happening in the shadows and give it judgement, a method, and permission. That is a reskilling answer, not a software answer. And it answers the quiet question the room has been holding.

Which skills actually compound, and which expire?

Picture the difference like learning to drive versus learning one car. A tool certificate teaches the car. The four durable skills teach the driving, and the driving transfers to every car your team will ever drive. Here is the shape of it.

Spend your effort where the value sits, because the market is already pricing these skills. PwC's 2025 Global AI Jobs Barometer, which reads roughly a billion job adverts across six continents, found that roles asking for AI skills now carry a 56% wage premium, more than double the 25% premium a year earlier, and that the skills employers ask for are changing 66% faster in the occupations most exposed to AI. The capability is not optional polish; it is the asset the market rewards. And the human skills travel with it: LinkedIn's 2025 Workplace Learning Report found that professionals who build AI skills are thirteen times more likely to grow the human capabilities (change readiness, trust-building, judgement) that no model supplies. The gap between the teams pulling ahead and everyone else is almost entirely a people-and-skills gap. This is why we keep saying the bottleneck is no longer the technology.

Where does judgement come from, and why does it fade?

Of the four, judgement is the one leaders most want to develop and least know how to teach, because it is physiological as much as intellectual. There is a striking study of more than 1,100 parole-board rulings in which the share of favourable decisions began the session around 65% and drifted toward zero before each food break, then reset. Later re-analyses have questioned the size of the effect, so treat it as suggestive rather than settled. The direction still rings true to anyone who has chaired a long meeting: tired people make worse calls. Judgement depletes.

And there is a defensible mechanism for the inverse, that regulation sharpens judgement. Under the neurovisceral integration model (a body of physiological research linking emotional self-regulation to the brain's decision-making circuits), people with greater heart rate variability (the natural beat-to-beat variation that signals a calm, adaptive nervous system) perform better on executive-function tasks, because the vagus nerve links the prefrontal cortex, the seat of cognitive control, to the heart. The fuller heart-coherence claims remain exploratory, so hold them lightly; the autonomic finding itself is well evidenced. In plain terms: a calmer, more regulated person makes clearer decisions. The four skills are what your team practises; coherence (head-heart alignment) is the state that lets the most human of them, judgement, hold up across a hard day. This is what we mean by upgrading the human operating system before we load more software onto a tired one.

What is a sequence I can actually start with?

1. Surface what is already happening. Ask, without consequence, which tools people already use and which tasks those tools already touch. You are mapping the 13%, not policing it.
2. Name the four skills as the curriculum. Make judgement, prompting, verification and collaboration the explicit thing you train, so the team learns the driving rather than one car.
3. Teach by doing on real work. Run live tasks where people prompt, verify and decide, not a slide deck about prompting. The craft lives in the iteration.
4. Build verification into the workflow. Make checking the confident answer a named step everyone owns, so trust in the output is earned rather than assumed.
5. Protect judgement. Treat decision quality as a resource that depends on a regulated team, and structure the work so your best calls happen when people are clear, not depleted.

About the author

Thomas Green

British technology futurist, AI keynote speaker and advisor. Thirty years across enterprise technology and AI strategy, helping leaders navigate the future of work. The futurist who died.