I recently had a long conversation with a friend who works at the OECD about a question that I think deserves far more attention than it gets: what is the state’s role in AI-driven reskilling?
The debate about AI and jobs usually stays inside the company walls — what employers should do, what HR should do, what individuals should do. But the numbers involved make this a national economic question, and I think there is a concrete, workable answer hiding in plain sight.
The burden is real, and right now it sits in the wrong place
Reskilling at scale is genuinely expensive. Not just the training itself, but backfill during training, temporarily reduced productivity, programme design, and the risk that some transitions simply won’t work. For a company facing AI-driven role change across hundreds of positions, this is a multi-year investment with an uncertain return.
Which is exactly why many companies quietly default to the alternative: fire and hire. Let go of the people whose roles are shrinking, recruit the people whose skills are needed. On a spreadsheet, viewed narrowly, it can look cheaper.
But fire-and-hire is only “cheap” if you ignore who else pays. Every displaced worker means severance costs and lost institutional knowledge for the company — and unemployment benefits, lost income tax and social contributions, retraining costs eventually paid anyway, and the social costs of displacement for the state. The state pays for fire-and-hire whether it wants to or not. It just pays after the damage, in the most expensive and least productive form.
So the state has a direct financial interest in reskilling happening inside companies, before displacement — not as a matter of industrial sentiment, but as straightforward fiscal arithmetic.
We have done this before: the Australian precedent
My OECD friend brought up a parallel I found genuinely instructive: the wind-down of the Australian automotive industry. When Ford, Holden and Toyota announced the end of local manufacturing, the response was not to let the market sort it out. The carmakers gave roughly three years of notice. Federal and state governments, together with the manufacturers, put several hundred million dollars into transition support — career guidance, skills and training assistance, help for supply-chain businesses.
The results were striking: the government’s own follow-up reporting found that more than four out of five retrenched workers found new employment. And the lessons documented afterwards read like a design manual for what is coming with AI: support works better when it is continuous rather than dispensed in rounds; early assistance beats late assistance; businesses that received transition support fared measurably better than those that didn’t.
An entire national industry closed, and the majority of its workforce landed on their feet — because the transition was anticipated, co-financed, and managed rather than suffered. AI will not close industries overnight, but it will restructure work on a scale that dwarfs one country’s car sector. The playbook exists.
Universities matter — but they answer the wrong decade
One honest caveat: recalibrating university education for the AI era is important, and I fully support it. But it is a ten-to-fifteen-year lever. The people whose roles are changing right now are mid-career, employed, and often decades away from retirement. No curriculum reform reaches them. The short-term instrument has to be different — and it has to reach people through their employers, because that is where the role change is actually happening.
The proposal: contestable funds, unlocked by a business case
So here is the idea, in its simplest form: the state establishes co-financed reskilling funds that companies can apply to — not with a general promise to “invest in skills”, but with a concrete business case.
That business case would need to show, defensibly: which roles are exposed to automation and to what degree; how many people are affected; what the realistic transition paths are — from which roles to which roles; what the reskilling costs, and over what timeline; and what the alternative (displacement) would cost both the company and the public purse.
In other words: a strategic workforce plan, with numbers that can be audited. If a company cannot produce one, it is not ready to spend public money. If it can, co-financing it is one of the better investments a state can make — because every successful internal transition is an unemployment case that never happens and a taxpayer who never stops paying tax.
This design gets the incentives right on both sides. The state shares the burden but does not micromanage; there is no central authority deciding which jobs are “the future”. The company keeps the decision — it knows its own strategy, its own people, its own technology roadmap. Public money simply tilts the spreadsheet away from fire-and-hire and toward reskill-and-keep, and demands rigour as the price of entry.
This is not utopian. Singapore already operates a version of it: SkillsFuture co-funds a large share of reskilling costs — with enhanced support for mid-career workers — and its career conversion programmes provide salary support while employees retrain into growth roles, employer-led throughout. The model exists and runs at national scale.
The question is not whether — it is when, and who pays
AI-driven role change is not a scenario to be debated; it is a process already underway. The reskilling will happen either way. The only real choices are timing and financing: proactively, inside companies, with the burden shared — or reactively, through the unemployment system, with the state paying more for worse outcomes.
I know which version I would rather help build.
An auditable strategic workforce plan is where co-financing starts.