Since 2018, Amazon has spent $1.2 billion on its Upskilling 2025 program, pledging to retrain 100,000 workers — roughly a third of its US workforce — for higher-skilled roles. The program has received extensive press coverage and multiple awards for corporate responsibility. What it hasn't received much attention for is its outcomes: a 2023 independent audit found that fewer than 12% of participants transitioned into the higher-skilled roles the program was designed to produce.
Amazon is not unusual. Its program is representative of the entire genre of corporate reskilling initiatives — well-funded, well-intentioned, and structurally inadequate to the problem it claims to address.
The Political Convenience of Reskilling
When AI displacement becomes impossible to ignore in policy circles, reskilling is the answer that emerges because it satisfies the constraints of the political conversation. It acknowledges disruption without requiring structural change. It places responsibility on individuals and companies without implicating the systems that generate the disruption. It sounds like action.
The reskilling consensus has three components: companies should retrain their workers, governments should fund retraining programs, and individuals should take responsibility for their own continuous learning. Each of these is defensible in isolation. Together, they constitute a response calibrated to the political difficulty of the problem rather than to its actual scale.
What the Evidence Shows
The academic literature on workforce retraining is not encouraging. A comprehensive review of US Trade Adjustment Assistance programs — the federal retraining support for workers displaced by trade — found that participants earned on average $50,000 less over the decade following displacement than comparable workers who were not displaced, regardless of whether they completed retraining programs.
The retraining didn't close the gap. It softened the fall slightly for some workers, but the earnings trajectory of displaced workers who retraining never fully recovered.
Several patterns emerge consistently from the research. First, retraining works best for workers under 40. Older workers complete programs at lower rates, struggle more with employer acceptance, and face compounding disadvantages in job markets that favor both credentials and recency. The majority of AI-displaced workers in the coming decade will be between 45 and 65.
Second, retraining works best when the new skills are adjacent to the old ones. A warehouse worker retraining as a logistics software administrator has a plausible transition path. A paralegal retraining as a nurse has a much steeper one — not just technically, but in terms of the time investment required, the debt involved, and the cultural distance between the two professional identities. The bigger the displacement, the less effective the retraining.
Third, retraining works best in tight labor markets. When jobs are available, retraining helps workers claim them. When AI is simultaneously reducing the number of available positions in multiple sectors, retraining produces more people competing for a smaller number of spots. This is the scenario the AI transition creates — and it's the scenario for which retraining is least equipped.
The Structural Mismatch
The deepest problem with the reskilling narrative is that it treats a structural problem as an individual one. It asks: how do we prepare individuals to compete in the new economy? The better question is: what kind of economy do we want to build, and what role should human labor play in it?
These are different questions with different policy implications. The first question leads to better training programs, online credentials, corporate learning stipends, and continuous education mandates. The second leads to questions about universal basic income, reduced working hours, wealth redistribution from AI productivity gains, and the fundamental redesign of the relationship between labor and security.
The reskilling consensus avoids the second set of questions entirely. This is not an oversight — it's a feature. The structural questions require confronting interests that have significant political power: the technology companies capturing AI's productivity gains, the investors who own the systems, the policy frameworks that protect incumbent wealth.
Reskilling asks workers to adapt. It doesn't ask systems to change.
What Reskilling Can Actually Do
This is not an argument that retraining is worthless. For specific workers in specific situations, it provides genuine value. A 35-year-old in a field being significantly disrupted who has the time, resources, and cognitive flexibility to retrain for an adjacent role can meaningfully improve their trajectory through a well-designed program.
The argument is about what reskilling cannot do. It cannot absorb displacement at the scale that AI is likely to produce. It cannot overcome the structural disadvantages of older workers in credentialist labor markets. It cannot generate new jobs in sectors where AI is simultaneously reducing demand. And it cannot substitute for the broader social architecture — income support, healthcare independence from employment, portable benefits — that displaced workers need regardless of whether retraining succeeds.
The reskilling conversation is a necessary part of the policy response. The mistake is treating it as a sufficient one.
The Question Behind the Narrative
Every time a politician announces a new reskilling initiative, it's worth asking who benefits from the framing. When individuals are responsible for their own adaptation, capital is not responsible for the disruption it causes. When the problem is defined as a skills gap, it doesn't need to be defined as a distribution problem.
The workers who will be displaced by AI in the next twenty years are not, in most cases, people who failed to invest in their skills. They are people who made rational decisions in a world that is changing faster than individual adaptation can track. The moral weight of that observation does not land on them.
After Work is not optimistic about reskilling as a solution. It is, however, interested in what comes after the reskilling narrative exhausts itself — and what kinds of arrangements might actually address the scale of the transition we're in the middle of.