I’ve spent months hearing the same sentence in very different boardrooms: “with AI, we’re going to need fewer people.” It’s said as calmly as a weather report. It took me longer than I’d like to see that it’s almost never an observation. It’s a decision disguised as one.
By Ángel Bonet · President and founder of ImpactCo
For months I have heard the same sentence in boardrooms that have nothing in common — different sectors, family businesses, multinational subsidiaries. The sentence is: “with AI, we’re going to need fewer people.” It’s stated naturally, like noting the weather. And I took longer than I’d like to admit to realise that it’s almost never an observation. It’s a decision disguising itself as one.
What changed my mind wasn’t intuition — it was looking at the data seriously.
The Data Doesn’t Fit the Narrative
The Stanford Institute for Economic Policy Research published a brief co-authored by Erika McEntarfer, who was Commissioner of the US Bureau of Labor Statistics until August 2025. Its finding is simple and disconcerting: since 2022, the unemployment rate of the workforce quintile most exposed to AI has risen 0.77 percentage points. The least exposed quintile has risen 0.85. Those supposedly about to be swept away are, in aggregate, doing slightly better than the rest.
It’s not an isolated figure. In the US Census survey, only 5% of firms report any effect of AI on their headcount — and that 5% splits almost evenly between those that gained jobs and those that lost them. In the Atlanta Federal Reserve survey, eight in ten executives admit their AI investments have not yet moved either headcount or productivity. And a Ramp analysis found that firms which adopted AI saw employment grow 10% over the following two years. The labour apocalypse may still arrive. In the data today, it isn’t there.
What is there is something else. Challenger, Gray & Christmas tracks announced layoffs in the US and the reason each company gives. Artificial intelligence has topped that list for five consecutive months. The tech sector alone has accumulated nearly 150,000 cuts this year — 67% more than last year.
So we have two series that don’t fit. AI barely displaces jobs in any measurable way, yet it’s the explanation most often offered when jobs are cut. And the person who said it most clearly wasn’t a technology critic: it was Sam Altman, admitting that some companies are attributing to AI reductions they would have made anyway. Marc Andreessen put it less diplomatically, calling AI the perfect excuse to clean out bloated payrolls. When the technology’s own evangelists warn you it’s being used as an alibi, it’s worth listening.
The Figure That Made Me Rethink the Whole Conversation
Up to here we might be talking about a corporate-communications problem. What turns this into a boardroom matter is a Gartner study. They surveyed 350 executives at companies billing over a billion dollars that are already deploying AI and autonomous systems. 80% had cut headcount — some by up to 20%.
Those that cut most obtained financial returns practically identical to those that cut least. In many cases, those that cut least performed better. The analyst behind the study, Helen Poitevin, sums it up: headcount reductions can create budgetary room but they don’t create return. It seems like a nuance and it isn’t: freeing cash and generating value are different operations, and I’ve seen too many committees treat them as the same. The same study notes what the companies that did get a return from their AI actually did. They invested in training, in governance, and in creating new roles that didn’t exist before. The exact opposite of cutting.
And here is what interests me most: what happens to the people who stay.
ADP surveyed 39,000 workers across 36 markets. Only 22% believe their job is safe. Fewer than one in five is fully engaged with their work. And the figure that ties the two together: someone who feels secure in their job is six times more likely to be fully engaged and 3.3 times more likely to report being highly productive.
Put it all together and you get an absurd circle. A company invests in AI. To justify the investment to the market, it announces cuts and attributes them to AI. That announcement generates insecurity across the whole workforce, starting with those who stay. That insecurity destroys engagement. And without engagement, the AI investment doesn’t pay off. You end up paying twice for the same technology: once in investment, once in destroyed commitment. The alibi ends up killing the very return the alibi was meant to justify.
The most revealing part of the ADP study is that the tool is not to blame. Among those who use AI daily, engagement rises to 30%, versus 14% among those who never use it. AI, when used to work, engages. What disengages is the story with which it’s installed.
Two Things Nobody Is Writing Down Anywhere
The first concerns young people. Unemployment among recent graduates reached 5.6% in the first quarter of this year — 1.6 points more than three years ago. Researchers are still debating whether the cause is AI, interest rates, or the post-pandemic over-hiring hangover, and honestly the debate isn’t settled. But for someone on a board, the cause matters less. What matters is that we’re eliminating the rung where you learn to judge. A junior’s tasks — reviewing, summarising, drafting, checking — are exactly what AI does well. And they’re also what forms professional judgement. If we automate the school, I don’t know where the senior of 2035 comes from. I have three children, and I don’t experience this part as abstract.
The second is subtler and worries me just as much. A paper published this year in Nature by Hao and co-authors finds that scientists who adopt AI publish more and reach further, but that science as a whole narrows its focus: fewer topics, less cross-disciplinary conversation. AI improves what each individual produces and reduces the diversity of what the group produces. Transfer that to a company: everyone performing slightly better and everyone thinking alike, because everyone consults the same oracle. For a company whose competitive advantage is its particular way of seeing the world — and every purpose-driven company is exactly that — this is not an operational nuisance.
And Now, Without a Safety Net
Many of us had marked 2 August 2026 as the date of full application of the European AI regulation. It no longer is. Regulation (EU) 2026/1744, the so-called Digital Omnibus, was published on 24 July and postponed to 2 December 2027 the obligations for the high-risk systems of Annex III. That annex includes, in full letters, employment and worker management: CV screening, performance evaluation, promotion and dismissal decisions. The transparency obligations did come into force, but the block governing algorithmic decisions about people won’t arrive until the end of 2027.
Europe has granted itself sixteen more months. The people who work in your company don’t have those sixteen months. They will form a judgement about you long before — and they’ll form it by watching what you do precisely now, while nobody is forcing you to do anything.
The Cleanest Proof of a 25-Year Argument
I’ve spent twenty-five years saying there is capital that grows roots and capital that extracts, and I think AI is the cleanest proof we’ve ever had of that difference, because the technology is identical in both cases. Extractive capital uses it to remove cost and presents that decision as an inevitable technical fact; it gets margin and, if Gartner is right, gets no return. Capital that grows roots uses it to expand capacity: more judgement per person, more time for what demands human judgement, new roles that didn’t exist before. It’s slower and more expensive at first, and it’s the only one of the two paths that the data associates with real return.
I don’t frame this as a moral argument, though it is also that. I frame it as a matter of the bottom line — the language in which things are really decided. If someone in your next board says that with AI you’ll need fewer people, the question I’d ask is very unsophisticated: what are we installing this for? If the honest answer is to reduce headcount, it’s worth knowing before signing that the data says you won’t get a return. And a second, more uncomfortable question: have we already attributed to AI a decision we’d have made anyway? The workforce usually knows the answer well before the board does.
Artificial intelligence is not destroying jobs. It is revealing, with a clarity that will cost some dearly, what each company really exists for. The ones that existed only to extract value are demonstrating it at full speed. The ones that exist to create it have in front of them the greatest opportunity for differentiation I’ve seen in twenty-five years.
At ImpactCo we support presidents, boards and funds that want to bring in artificial intelligence without emptying their organisation of meaning. If this conversation touches you closely, write to us: tribu@impactco.es.