For two years, "AI skills" was the safest line on a CV. Sprinkle it near the top, mention a chatbot or two, and recruiters took notice. Entire careers pivoted on the assumption that this demand would keep climbing forever.

Then, in January, it peaked. And started falling. And almost nobody noticed, because the drop didn't make headlines — it showed up quietly in the job postings themselves.

One of the largest HR platforms on earth, tracking skills across hundreds of employers' job requisitions, found that demand for basic AI skills — the general "I can use the tools" variety — rose steadily through 2025, crested in January of this year, and has fallen by a quarter since. Not stabilised. Fallen. While the think-pieces kept repeating last year's advice, the market had already moved on.

It's worth pausing on how invisible this kind of shift is. Nobody emails you when your skill starts depreciating. There's no notification, no warning light on the dashboard of your career. The announcement doesn't say "we no longer value the thing you learned" — the postings just quietly change what they ask for, and the people holding the old version find out at interview time, or worse, at layoff time. Skill devaluation is the only kind of inflation where the people most affected are the last to know.

The Skill Didn't Die. It Split.

Here's what actually happened, and why it matters whether you understand it: "AI skills" was never one thing. It was two very different capabilities wearing the same label, and this year the market finally priced them separately.

The first kind is familiarity. Knowing the tools exist, using them for the obvious tasks, listing them honestly on a profile. This is the version that peaked in January — and the reason it peaked isn't that it became useless. It's that it became common. When everyone in the applicant pool can produce a decent AI-assisted first draft, producing one stops being a skill and starts being a setting. Same trajectory as "proficient in Microsoft Office" twenty years ago: real, expected, and worth approximately nothing as a differentiator.

The second kind is construction. Not using AI — building with it. Connecting it to actual workflows, automating the repetitive machinery of a job, making it do the specific work of a specific role. And here's the part that should make you sit up: while demand for the first kind fell by a quarter, demand for the second kind — hands-on, build-with-AI capability — rose by half over the same period. Same dataset, same employers, two lines crossing in opposite directions.

The salary data confirms it from the other side. In the UK, job postings mentioning prompt engineering have multiplied — but the median pay attached to those postings has dropped by around a fifth. More postings, lower prices. That's not a growth market. That's a commodity market forming in real time, while the premium quietly moves elsewhere.

And this is happening everywhere at once — a global analysis across 27 countries found demand for AI skills rising in every single market studied. Not a Silicon Valley trend or a tech-sector bubble: a worldwide repricing. The countries differ in pace, but the direction is unanimous, and the direction is away from generic familiarity toward applied capability.

The Gap Nobody Is Coming to Close

There's a second finding buried in the same research, and it's the one that should decide what you do next. Asked whether they know what skills they need to stay relevant, nearly eight in ten workers said yes — they already know. Asked whether their employer is actually helping them build those skills, only two in three said yes.

Read that gap properly. Most people already know the ground is shifting. Most employers are not helping. The distance between "what I know I need" and "what my company will teach me" is now a permanent feature of working life — and it means the reskilling question is no longer whether your employer will develop you, but whether you'll develop yourself before the market finishes repricing your current skillset.

Meanwhile the jobs themselves are being rewritten around the people who can build. Four in ten business leaders now expect AI to extract more value from the employees they already have — not to replace them. Translation: the expectation has moved inside the job description. Your role isn't going away; it's quietly absorbing capabilities nobody told you were now part of it.

The cost of misreading this moment is asymmetrical. If you over-prepare — learn the applied version when the market only wanted familiarity — you've wasted some weekends. If you under-prepare — keep polishing the familiar version while the premium moves — you've potentially mortgaged years of career trajectory on a bet that's already losing. The people caught out aren't stupid; they just kept following advice that was correct eighteen months ago, from a market that no longer exists.

Why Most People Are Still Learning the Wrong Version

Watch how people respond to "learn AI skills" and you'll see the same mistake at scale. They learn the demo version.

They collect tool names. They learn five clever prompt tricks. They put a certificate next to the skill on the profile. All of which would have worked in 2024, and all of which lands in the commoditised half of the market today — the half whose demand is falling. They're studying for the version of the test the market stopped giving in January.

The builders, meanwhile, are doing something less visible and more valuable: they're learning how AI actually gets wired into work. Not "how do I chat with it" but "how does this remove the twenty hours a month my team wastes on X". Not prompt trivia — automation. The skill employers are now paying a premium for isn't knowing about AI. It's knowing what to do with it, in context, on real work.

That distinction is the whole career question right now. When a skill splits into a commoditised half and a premium half, you don't get to stay in the middle. You either climb into the version that's appreciating, or you get averaged into the one that's deflating.

What the Builders Actually Look Like

The most useful way to see the difference is to watch both versions in the same job. Two people, same title, same tools.

The first uses AI the way the demo showed them — a better search engine, a faster first draft, an occasional summary. The work gets marginally quicker. Nothing about the job changes. If the tool vanished tomorrow, they'd grumble and go back to doing it the old way. This is familiarity, and it is precisely what the market stopped pricing.

The second person looks at the job itself — the recurring report nobody wants to compile, the information that lives in three places and has to be reconciled by hand, the process that eats Friday afternoons — and quietly rebuilds it. The tool isn't a novelty to them; it's a component in a system they've constructed. If the tool vanished tomorrow, their replacement wouldn't be "do it by hand" — it would be "build the thing again with something else," because what they actually learned wasn't the tool. It was the building.

Employers can't always articulate that difference in an interview question — but they can smell it. One candidate describes using AI. The other describes what changed because of how they used it. The second conversation is the one that ends with an offer at the top of the band.

Which Side of the Repricing You're On

And the window matters. The build half of the market is still early enough that the people in it stand out — hiring data shows the demand curve is climbing, not saturated. That won't last. Every premium skill follows the same arc: rare, then valuable, then expected. The people who moved on cloud skills in 2015, data skills in 2018, prompt skills in 2023 all captured the expensive part of the curve. The ones who arrived when it was "essential" captured the commodity part. Where you land on that arc is almost entirely a question of when you start.

The encouraging part: the build half isn't engineering. Nobody's asking you to train models or write neural networks. The skills in demand are applied — automating workflows, connecting AI to real processes, turning a general tool into a specific advantage. It's a learnable, structured capability — closer to a craft than a computer science degree.

Our AI Automation course is built for exactly that side of the line: a daily-email course that takes you from using AI to building with it — real automation, applied to real work, in a format designed to finish. No hype, no tool-of-the-week — the durable version of the skill the market is actually paying for now.

The market repriced "AI skills" while nobody was looking. The only question left is which price you're holding.