‘Just Learn to Code’ Was Always Half a Lie: The Uncomfortable History of Reskilling

Every wave of automation arrives with the same reassuring promise attached: the machines will take some jobs, but you can always learn new skills and move up. Right now that promise wears the label reskilling for AI, and it is everywhere — in government white papers, corporate press releases, and the encouraging LinkedIn posts of people who are quietly terrified. I want to take the promise seriously, because it contains something real. But I also want to be honest about its history, because “just learn new skills” has been offered before — to miners, to factory workers, to a whole generation told to “learn to code” — and it has under-delivered often enough that we owe it a harder look.

Let me say the fair part first and mean it: retraining genuinely helps individuals. If you learn a skill that is in demand, your odds improve. That is not a trick. The problem is not that reskilling is useless — it is that it keeps getting handed to us as the whole answer to a problem it was never big enough to solve — and in doing so it quietly moves a burden that belongs to the people capturing the gains onto the shoulders of the people losing the jobs.

The promise has a long, uneven history

Reskilling is not new. It is the standard closing line of every disruption story, and it has been for two centuries. When the mines closed, the message to coal towns was: retrain, the future is elsewhere. When manufacturing hollowed out across the American Midwest and the north of England, displaced factory workers were pointed toward classrooms and “the jobs of tomorrow.” When the internet arrived, the mantra hardened into two words that became a genre of career advice: learn to code.

The uncomfortable pattern is not that these programs never worked. Some did, for some people. It is that they routinely under-delivered against the confidence with which they were sold. Studies of large-scale trade-adjustment and retraining schemes have long found modest average effects — real for a minority, disappointing for many, and rarely enough to rebuild what a closed plant took from a whole town. A laid-off fifty-year-old miner in a region with no other employers does not become a web developer because a pamphlet suggested it. The geography is wrong, the timing is wrong, and the new jobs, when they exist at all, are frequently elsewhere and pay less.

None of this is an argument against learning. It is an argument against a specific sleight of hand: taking a structural event — an industry being reorganised by people who profit from the reorganisation — and rebranding it as a personal development opportunity. This is the same move I keep seeing across the history of technology: the same move, a new machine, every time. The gains flow one way, the adjustment costs flow another, and the story we are told is that the costs are really just an invitation to self-improve.

Why “just learn to code” was half a lie

I want to be precise, because “learn to code” was not a whole lie. Plenty of people did learn, did get hired, and did change their lives. That is exactly what made it so persuasive. But it was half a lie for three reasons, and each one matters more now, not less.

It treats a structural problem as a personal one

When a factory automates or an industry offshores, the cause is a decision made at the top about how to capture more value. “Learn to code” takes that collective, engineered event and reassigns responsibility for it to each individual worker. If you retrain and thrive, the system worked. If you retrain and still can’t find work, the failure is quietly yours — you picked the wrong course, you didn’t hustle. This is a beautiful arrangement for whoever captured the gains, because the disruption produces no obligation on their side at all. The burden shifts entirely onto the person with the least power to absorb it.

Reskilling asks the person with the least power in the transaction to absorb a cost created by the person with the most.

It assumes the new jobs will absorb everyone

The optimistic version says: yes, some jobs vanish, but new and better ones appear, and displaced workers flow into them. Sometimes that broadly happens, over decades. But “eventually, in aggregate, the economy adjusts” is cold comfort to a specific person in a specific town in a specific decade. New jobs are not created in the same places, at the same time, in the same numbers, or for the same people as the ones destroyed. The average can improve while millions of individual transitions fail. Reskilling is sold on the average and lived on the individual, and that gap is where a lot of real suffering lives.

It pointed people at exactly what would be automated next

This is the part that should make anyone selling reskilling pause. For years, the advice was to funnel people into coding — and a large share of what entry-level coding actually involves is routine, pattern-heavy work: boilerplate, simple scripts, standard components, glue code. That is precisely the category current AI tools are fastest at automating. We spent a decade telling displaced workers to run toward a cliff we could already see. The “safe” skill was safe only until the next machine arrived: there is no permanently safe rung when the people deploying the technology are actively looking for the next thing to automate. I’ve written more about how technology gets captured and why the pattern repeats; reskilling is the pattern’s favourite alibi.

What reskilling genuinely can do

Here is where I refuse to be cynical, because cynicism is just another way of telling people not to try. Reskilling is real and worth doing. If you are wondering will AI take my job, learning to work alongside these tools rather than against them is a rational and often effective response. The person who understands how to direct AI, check its output, and fold it into real work is more valuable, not less. Skills compound, and I would never tell someone facing displacement to sit still.

What reskilling can do for you as an individual is genuine:

• It improves your odds. In a churning market, the person with a fresh, in-demand skill genuinely competes better than the person without one. That edge is real even if it isn’t guaranteed.
• It buys adaptability. The specific skill you learn may date quickly, but the habit of learning — of moving toward the new tool rather than flinching from it — is durable across waves.
• It restores some agency. Displacement is disempowering. Doing something concrete, even something modest, is better than waiting to be rescued.

I say all this without an asterisk. If you take one action after reading this, let it be to learn something that makes you harder to automate and easier to hire. That is not the lie — the lie is what gets bundled around it.

What reskilling cannot do

Reskilling operates entirely on the supply side of the labour market — it changes what workers can offer. It does nothing to the demand side, and nothing at all to the question of who captures the value the new technology creates. That limit is not a detail. It is the whole thing.

Reskilling cannot change who captures the gains. When AI makes a company far more productive, that surplus is real, and by default it goes to the owners of the capital and the platforms, not to the workers who retrained to stay employable. You can upskill your way into keeping a job; you generally cannot upskill your way into a share of the windfall your new productivity creates. The pie grows, and who gets the extra slices is decided by power, not by how many courses you completed.

Reskilling cannot work equally for everyone. It works best for people who are already advantaged — younger, healthier, better-educated, living where the new jobs are, with savings to live on while they retrain. It works worst for exactly the people hit hardest by disruption: older workers, single parents, people in regions with one dominant employer, people already stretched too thin to take six months off to learn a new trade. A remedy that helps the resilient most and the vulnerable least is not a general solution. It is a filter dressed as a ladder.

And reskilling cannot outrun the pace of the thing displacing it. This matters acutely now. When each new model can absorb another band of tasks every year, the half-life of a “safe” skill keeps shrinking. Telling people to perpetually retrain into whatever is briefly beyond the machine’s reach is not a stable life — it is a treadmill that speeds up. You can already see it in the vanishing entry-level job: the rungs people used to retrain onto are being pulled out first, and in which office jobs AI replaces first you can watch the same logic move up the ladder into work that felt secure a few years ago.

You can upskill your way into keeping a job. You generally cannot upskill your way into a share of the windfall your new productivity creates.

What has to sit alongside it

So the honest position is not “reskilling is a scam.” It is “reskilling is one tool, and offered alone it is half a lie — because it leaves the two hardest questions, who pays for the transition and who captures the gains, entirely untouched.” If we actually want the promise to hold, reskilling has to be surrounded by the things that make individual effort add up to a fair outcome.

• Real transition support. Not a pamphlet and a link to an online course, but funded time to retrain, income while you do it, relocation help where the jobs really are, and programs designed with the specific displaced population rather than at them. Past retraining efforts tend to work far better when they are well-funded, well-targeted, and paired with income support than when they are a box ticked on the way out the door.
• A genuine safety net. If we accept that some individual transitions will fail even when people do everything right — and history says they will — then the humane and rational response is a floor no one falls through. That is not charity. It is the cost of running an economy that reorganises itself this fast without breaking the people inside it.
• Bargaining power. Individuals negotiating alone against firms deploying automation will lose the fight over how the gains are split, every time. Collective leverage — unions, professional associations, worker representation, credible policy — is how workers have historically clawed back a share of the productivity they help create. Skills make you more valuable; power is what lets you turn that value into pay.
• Rules on sharing the gains. This is the one almost no reskilling pitch mentions, because it is the one that costs the winners something. If AI produces an enormous productivity surplus, the question of how that surplus is shared — through wages, ownership, taxation, shorter hours, or public investment — is a political choice, not a law of nature. Reskilling changes who is employable. Only the rules decide who gets rich.

Notice that three of those four have nothing to do with what any individual worker learns. That is the tell. If the entire remedy on offer is “change yourself,” and nothing is asked of the people capturing the gains, then you are looking at the oldest move there is: privatise the adjustment cost, socialise nothing, and call the whole thing empowerment.

Learn the skill. Don’t swallow the story.

I’ll hold both halves of this at once, because both are true. Go learn. Pick up the tools, get fluent with AI, make yourself harder to replace — that effort pays off more often than not. Reskilling for AI is a genuinely good individual bet in a bad structural situation.

But refuse the version of the story that stops there. When someone tells you the answer to a technology reshaping your whole industry is simply that you should retrain, ask the questions the pitch is designed to skip. Who is capturing the gains this technology creates? Who is bearing the cost of the transition? And why is the entire burden of adjusting to a change you did not choose being placed on you alone? Every time we are told the machine is just an invitation to better ourselves, someone is being spared the harder questions about where the value went. Learn the skill. Just don’t let it be used to let them off the hook.