The Skill AI Makes More Valuable, Not Less

The honest answer most are reaching for is some combination of “learn to use AI” and “double down on the things AI can't do.” Both are right, and both are usually left too vague to survive a budget conversation. It's worth being precise about which human capability actually becomes more valuable as the technology gets better — because the evidence on that is clearer than the discourse around it suggests.

What the workforce data actually says

The World Economic Forum's 2025 Future of Jobs report makes two findings that sit together awkwardly in most summaries. The first is that the skills needed to do most jobs are changing substantially over the next five years, with AI a primary driver. The second is that the single most-valued core skill — named essential by around seven in ten employers — is analytical thinking, followed closely by resilience, flexibility, and leadership.

In other words, the capabilities employers say they need most are not the ones being automated away. They are rising in value alongside the new technical skills, not being displaced by them.

That word — alongside — is the one most forecasts get wrong. The dominant employer response to AI is not replacement but augmentation: in the same WEF data, more than three-quarters of employers say they plan to reskill their people specifically to work with AI. The most visible attempt to do the opposite — a high-profile move to swap customer-service teams for AI in 2025 — was publicly reversed within the year, the company conceding it had traded quality away for cost. The direction of travel is people and machines together, with the human contribution shifting up the stack toward judgement.

The finding worth building a strategy on

The most useful piece of evidence here isn't a survey of opinion — it's a controlled experiment. In a 2025 Harvard Business School field study, more than six hundred business owners were given access to an AI adviser and tracked over several months.

The headline is counter-intuitive. On average, access to the AI made very little measurable difference. The effect was hiding in the distribution: the people who already brought strong judgement pulled further ahead, while those who didn't slipped slightly behind. AI didn't lift everyone equally. It widened the gap between people who could direct it well and people who couldn't.

A second finding sharpened the point. The people who got the most from the AI were those who interrogated its output rather than taking it at face value. Other 2025 research points the same way — the more readily people trust a confident AI answer, the less critical thinking they tend to apply to it.

Put plainly: AI is an amplifier. It multiplies whatever judgement you bring to it. Bring good judgement and it makes you faster and sharper. Bring weak judgement and it scales your mistakes with great fluency. The skill that decides which way it goes — weighing trade-offs, reading a situation, deciding under uncertainty, knowing when the polished answer is wrong — is precisely the skill AI makes more valuable, not less.

You cannot lecture judgement into people

Here is the problem this hands to L&D. Judgement is the one thing you cannot transfer through content. You can teach a framework, a model, a checklist — but the capacity to apply it well under pressure, with incomplete information and competing priorities, is built only by doing. People develop judgement by making consequential decisions, seeing what follows, and being made to account for the result.

That is the entire premise of experiential learning. And it is why, in a market now flooded with AI-generated content, the experiences that put people in the decision seat are becoming more important, not less. A slide deck about “critical thinking in the age of AI” teaches knowledge about judgement. It doesn't build the thing itself.

That takes reps.

Where the reps come from

A well-built business simulation is, at its core, a machine for generating those reps safely. Teams face decisions with real trade-offs — a specific cost, a specific consequence, a specific risk — and then live with the outcomes. They argue. They commit. They watch a choice play out, and in the debrief they have to connect what happened back to what they decided. Decide, observe, account for it: that loop is how judgement is exercised and refined.

It only works if the decisions are real. A simulation full of options that sound plausible but change nothing rehearses nothing — the value is entirely in whether the trade-offs actually bite. When they do, a simulation does something AI-delivered content structurally cannot: it forces a person to own a decision in front of their peers and defend it. That is the muscle that decides whether someone directs AI or is quietly outpaced by colleagues who can.

“If AI is so capable, why pay for a board game?”

It's a fair challenge, and it deserves a direct answer rather than a dodge. The two things are doing opposite jobs. AI is extraordinary at producing answers. A simulation is built to develop the human who decides whether the answer is any good — and what to do about it. The more capable AI becomes at the first job, the more the second one matters, because the value increasingly accrues to the people who can direct and question the machine rather than defer to it. Training that builds that capacity isn't competing with AI. It's the necessary complement to it.

The same logic, applied to how we work

We'll say the quiet part out loud, because it's the clearest illustration of the argument. The Sim Smithy uses AI heavily to build its simulations — it's how a fully bespoke design is delivered in days rather than the months a traditional consultancy takes. But that AI runs inside a framework of human judgement and 25+ quality gates, precisely because unguided AI produces plausible-looking work that doesn't hold up under scrutiny. The industry even has a name for that now: “workslop” — output that looks finished and quietly creates more work downstream.

The thing that separates a rigorous bespoke simulation from AI filler is exactly the thing that separates a strong operator from a weak one: judgement applied to the machine's output. We build the product the way the evidence says your people should work.

The durable bet

The bet underneath all of this is a simple one. As AI gets better, the scarce, valuable, defensible human contribution moves toward judgement — deciding well, under uncertainty, with a machine in the loop. You cannot download that capability or install it. You build it by practising it.

The organisations that work this out will spend less time asking whether to train their people for the AI era, and more time asking where, exactly, their people get to practise the decisions that matter — before the decisions are real.