One hundred and twelve thousand, seven hundred and thirteen. That is how many US job cuts employers have pinned on artificial intelligence so far this year — more than double the total they blamed on AI for all of 2025. Amazon, Microsoft, Oracle, HSBC: the announcements arrive weekly now, each one explaining that AI has made some portion of the workforce unnecessary.

It is a compelling story. It is also, according to the people who actually measure the labour market, mostly not true.

The same data that produces the scary headline produces the contradiction. AI was cited in roughly a quarter of announced cuts this year — but total announced cuts are down forty-one percent compared to the same period last year. If AI were eating jobs en masse, the total would be rising. It is falling. Something else is going on, and it is worth understanding what — because how you read these announcements changes what you should do about them.

The Stated Reason Is a Message, Not a Measurement

Here is the first thing to know about that 112,713 figure: it counts the reasons companies give, not the reasons things happened. When a company announces a restructuring, it picks a story for the announcement — and "AI transformation" is the best story available in 2026. It sounds strategic rather than desperate. It tells shareholders the company is lean and forward-looking. It converts a cost-cutting exercise, which markets often punish, into an innovation narrative, which markets often reward.

Economists have a name for the gap between announced reasons and actual causes, and the gap is wide. A company that over-hired in 2021 and needs to cut eight percent of staff can say "restructuring" — or it can say "AI-driven efficiency." Both describe the same org chart. Only one of them makes the stock go up.

The labour data tells the less cinematic version. Roles are being rewritten faster than they are being removed. Demand for people who can simply use AI peaked months ago; demand for people who can build with it keeps climbing. The work is changing shape underneath the job titles, which is harder to announce and much harder to blame.

To Be Fair: Some of the Displacement Is Real

The sceptical read can be taken too far, and it is worth saying so plainly. Some announced AI cuts are exactly what they claim to be. When a company deploys a customer-service agent that genuinely handles sixty percent of tickets, the forty agents let go are not a narrative — they are a deployment that worked. Certain categories of work are visibly shrinking: first-line support, basic content production, routine document processing, tier-one analysis. The data doesn't show a mass extinction. It does show targeted displacement, concentrated in roles built entirely out of repeatable, low-context tasks.

The distinction that matters is not "AI cuts real or fake." It is which jobs are exposed and which are merely scared. A role composed of tasks that can be specified end-to-end — predictable inputs, predictable outputs, no judgement calls in between — is genuinely at risk regardless of what any announcement says. A role composed of coordination, judgement, trust, and messy context is not — regardless of what the announcement says. The scary number flattens that difference. Your own task list restores it.

The Tell: No Named Work

There is a reliable way to separate a real AI transition from AI-washed cost-cutting, and it takes about ten seconds. A genuine AI deployment can name the work it changes. Not "efficiency" or "transformation" — actual work. These tickets get auto-triaged. This report writes its first draft. That reconciliation runs without a person. If an announcement cannot point to the specific tasks that now happen differently, there is no deployment behind it. There is a budget cut wearing a costume.

This is the pattern economists keep flagging: workplace AI initiatives fail when they start from "what can we automate?" instead of "what problem are we solving?" The first question produces a tool in search of a purpose and, eventually, a layoff announcement with AI in the headline. The second produces something smaller and less newsworthy — a piece of work that genuinely runs differently.

The uncomfortable corollary is that companies announcing AI layoffs are often describing their own failure to deploy AI well. They bought the tools, never redesigned the work, and cut headcount anyway. The technology took the blame for a planning gap.

Run the Audit on Your Own Week

The "name the work" test works in both directions, and the more useful direction is inward. Take your last working week and list the ten tasks that consumed the most hours. Then apply the deployment question to each: could this be specified precisely enough that a machine could execute it — clear inputs, clear output, no judgement calls in between?

Most people find their week splits into three rough piles. The first pile — status updates, routine scheduling, first-draft production, data wrangling — is automatable now, today, with tools that already exist. The second pile is partially automatable: work where a machine produces the draft and a human applies the context. The third pile is the work only you can do: the judgement calls, the relationships, the decisions where being wrong is expensive.

That three-pile map is your actual risk assessment — far more accurate than any headline. Someone whose week is ninety percent pile one is exposed no matter what the labour data says. Someone whose week is mostly pile three has a headline problem, not a job problem. And everyone has a pile two, which is where the next five years of careers get built: the human who specifies the work, reviews the output, and owns the result. If your pile two is thin, our AI Automation course teaches exactly that operating layer — and if the audit left you staring at pile one, our Jobsearch Strategies for the Digital Age course is the hedge worth holding while you rebuild.

Reading the Next Announcement Like an Analyst

The next AI layoff headline will arrive within weeks. When it does, run it through three questions before it rents space in your head. First: is specific work named? "AI-driven restructuring" means nothing; "the support team is moving to agent-first handling" means something — for that company and, more importantly, for every company with the same work. Second: does the number match the claim? A company citing AI while cutting two percent is trimming; one cutting thirty percent is either automating an entire function or collapsing. Third: what happened to total headcount over two years? The company that hired ten thousand people in 2021 and cut eight thousand in 2026 did not undergo an AI transformation. It underwent a correction, and AI took the meeting.

This matters because the announcement cycle is designed to make you feel helpless — a vast force, arriving everywhere, indifferent to your choices. The analytical read produces the opposite: specific companies, specific functions, specific work. Fear generalises. Analysis locates. Only one of those tells you where to stand.

What This Means If You Work for a Living

None of this means the risk is imaginary. Cuts are real, and the roles most exposed are the ones built entirely out of work AI can already do unaided. But the exposed roles and the feared roles are not the same list. The layoff announcements are a poor map of actual risk — they tell you which companies needed a good story, not which jobs are vanishing.

The better map is the audit you just ran. The people who come out ahead are not the ones who fear AI or the ones who ignore it. They are the ones who can name the work it should change — and position themselves on the right side of that line before someone else draws it. Every week you operate pile-two work competently, you accumulate evidence that the line should be drawn around you, not through you.

That is a learnable skill, not a talent. It is also, quietly, the skill that separates the person designing the automation from the person being automated. Both might use the same tools. Only one decided what the tools are for.

And if you manage people: your team is running this same audit on themselves right now, whether you know it or not. The managers who keep their best people are the ones who make the pile-two work visible — who can look a report in the eye and say "here is the work only you do, and here is what we're automating so you can do more of it." In a market where every headline says jobs are disposable, the manager who names the irreplaceable work is doing retention.

So this week's action is concrete, and it costs one hour. Run the three-pile audit on your own work. Write down your pile one and pile three. Then decide — before anyone decides for you — which side of the automation line you're standing on. The 112,713 figure will keep growing, and the headlines will keep making it sound like weather. It isn't weather. It's a set of choices being made by people who can name the work. Become one of them.

If you want to build that skill properly — how to audit your own work, identify what should be automated, and wire up the automations without writing code — our 7-day email course AI Automation walks through exactly that: workflow identification, no-code pipelines, multi-step AI chains, and the judgement calls in between. And if the demand-side picture interests you, read our piece on AI job requirements tripling in non-tech roles — the other half of this story. And if you're already automating but your instructions still need cleanup, our Prompt Engineering for Professionals course sharpens the layer underneath.