AI's Productivity Promise Fails Without Up-Skilling Europe
The CEPR piece on “How AI is affecting productivity and jobs in Europe” sketches the usual terrain: tasks will change, some jobs vanish, others appear, productivity probably rises. Reasonable, tidy, bloodless.
But reasonable can hide a brutal distribution fight.
Follow the money.
Who actually gets the gains?
The article talks productivity versus employment as if “Europe” were one coherent labor market. It isn’t. A hedge fund in London and a machine-parts supplier in eastern Germany do not live in the same AI story.
Where do productivity gains usually land? Where capital, skills and clients already sit. Corporations with scale soak up the best models, engineers and data. They have legal teams to negotiate cloud deals, compliance teams to handle audits, in‑house staff to tune tools against proprietary datasets.
Small firms? They get the same glossy sales decks, the same cheerful webinars, the same canned chatbots bolted onto their websites—and then wonder why their margins barely move.
Convenient, isn't it.
So one core point the CEPR framing underplays: AI is likely to widen regional and firm‑size gaps unless policy does more than offer generic “support for innovation.” The question isn’t just how much productivity grows, but where and for whom.
That gap is not just an economic curiosity. It’s political dynamite.
When gains cluster in a few metros and sectors, voters outside those islands stop caring about aggregate charts. They see headlines about “AI‑driven productivity” and then look at their own wages, their own shuttered high streets, and draw their own conclusions. This is not an academic footnote; it’s the fuel tank of populist anger.
The CEPR lens works for economists. For politicians trying to hold together coalitions across regions, it’s a map with key provinces missing.
Skills are necessary—but not sufficient
To its credit, the article leans into job transitions and training. That’s the standard answer: reskill, upskill, hope for the best.
But here's what they won't tell you: skills alone won’t turn productivity into higher living standards unless workers have bargaining power and real options.
You can train someone in Python or prompt‑engineering. If local labor markets are dominated by a few large employers, those skills still get priced on the employer’s terms. If the new jobs appear in high‑cost capitals while the worker is locked into a small town by family, mortgage and local obligations, your training voucher looks like a taunt.
Is mass retraining for downtown tech corridors really a plan for someone caring for an elderly parent in a rural area, facing high childcare costs and no spare cash for relocation? On paper, maybe. In lived reality, not so much.
So the more honest claim is this: policy has to pair reskilling with support for relocation or commuting and with institutional changes that push wages up, not just competencies.
What might those institutional changes look like in practice? Stronger collective bargaining frameworks so AI productivity doesn’t bypass pay packets. Tax incentives that actually reward firms for locating AI‑heavy operations away from the usual superstar cities. Public co‑investment in local infrastructure and applied research labs that can anchor AI work in mid‑sized regions instead of defaulting to the same few hubs.
None of that is glamorous. It’s trench warfare in policy form. But the alternative is watching AI show up in GDP tables while most people’s paychecks flatline.
The corporate choice at the center
There’s another missing piece in the CEPR framing: corporate strategy.
Large firms will decide whether AI is an augmentation tool or a replacement weapon. Those are boardroom calls, not “inevitable” consequences of the technology.
Follow the money: if swapping people for software brings higher short‑term returns and there is no countervailing regulation, union pressure or reputational cost, why would executives voluntarily choose the slower, more expensive augmentation path?
You can already see the divergence. One bank experiments with AI to help staff handle compliance and customer queries faster. Another quietly uses it to justify branch closures and headcount cuts. Same technology, different calculus.
Policy levers—the tax code, subsidy design, public procurement criteria—can tilt that calculus. Reward firms that use AI to lift existing workers’ productivity and penalize pure substitution strategies, and you get one trajectory. Pretend the choice is “technology’s fault”, and you get another.
Regulation: not just about safety
The CEPR article nods to regulation, but mostly as a brake or a guardrail. Europe “likes rules,” the cliché goes.
Here’s the twist: rules can concentrate power as easily as they can restrict it.
Design AI regulation with heavy compliance burdens and the likely winners are the firms that can afford armies of lawyers and auditors. Smaller competitors fall behind or exit. So you can have a regime that reduces some harms while quietly entrenching market concentration.
Think about it this way: the same regulation that reins in reckless uses of AI in one sector can hand another sector to a few incumbents who have the resources to manage the paperwork. Again, the question isn’t just whether you regulate, but who your rulebook structurally favors.
That’s a third missing piece: the interaction between AI rules and inequality, not just risk.
A historical echo
We’ve been here before. When automation reshaped manufacturing, the story wasn’t only “machines versus jobs.” It was which regions housed the plants that modernized, which unions had the strength to win severance, retraining, and wage growth, which governments invested in new industries instead of writing off whole towns.
AI will play out with its own twist, but the logic rhymes. Technology shifts the frontier; institutions decide who gets pushed over it.
The CEPR article charts the frontier.
The real fight in Europe will be over those institutions, and whether they’re steered by the same handful of winners—or by governments willing to treat AI not as fate, but as a contested economic choice.