It assumes the new skill will sit inside the same workflow. AI restructures the workflow. The skill survives. The role doesn't. Reskilling alone won't save it.
Every AI policy paper, every corporate L&D budget, every World Economic Forum panel converges on the same answer: reskill. Learn the new tool. Stay relevant.
The advice is plausible and incomplete. It defends against task substitution. It doesn't defend against the workflow restructuring that's the actual threat.
Reskilling assumes three things. One: the new skill will be scarce long enough to earn from. Two: the workflow that uses the skill will still exist when you've learned it. Three: the role that bundles the workflow will still have economic logic. AI breaks all three.
When AI restructures workflows, the new skill is also routed through the model. You're sprinting toward a moving target on a track that's being torn up. The typist who reskilled into "faster typing with the new word processor" wasn't wrong about the skill - they were wrong about the system that used to need it.
The honest move is to first identify which constraint in your system AI hasn't dissolved - and probably can't. Reskill in service of that constraint. The skills follow the constraint, not the other way around.
When reskilling can't save the role because the role's logic has collapsed.
In the late 1970s, as word processors began appearing in offices, the standard advice to typists was to "upskill" - learn shorthand, specialise in transcription, become a faster typist with new tools. All sensible-sounding moves.
None of them saved the role. Because the threat wasn't that typists were too slow. The threat was that the constraint that bundled the role - expensive edits, manual revisions, formatting limits - had collapsed. Once those went, the system no longer needed a dedicated typing role at any skill level.
The pattern is repeating now. Reskilling junior associates to "use AI better" won't save junior associate roles when the system no longer needs a three-year cadre at all. The honest answer is rebundling, not reskilling.
Reskilling is what you do inside a rebundling strategy. Done in isolation, it's an L&D budget line that doesn't change anyone's economic fate.
One of the big four launched a multi-million-dollar programme to train its junior associates on prompt engineering and AI tools. Sensible. Productivity per associate is up.
But the junior associate role itself is shrinking - there are fewer of them, doing more, billing less. Reskilling made each one more capable. It didn't save the role's economic position in the system. The bundle is dissolving regardless.
"Radiologists will be replaced by AI" was the 2016 consensus. Image analysis is exactly what AI does. Nearly a decade later, demand for radiologists has grown.
Why? They rebundled their role around constraints AI can't resolve - multi-domain judgement, accountability for edge cases, contribution to tumour-board deliberation. They didn't reskill into "AI-enhanced image readers." They redefined what a radiologist's job is.
Look at your firm's L&D and AI-training budget. Then ask:
Reskilling is useful in service of a rebundling strategy. In isolation, it defends the wrong threat and steadily wastes the budget.
Reskill toward the constraint. Not toward the tool.
An advisory engagement that redesigns the operating model around constraints AI doesn't dissolve - and aligns the workforce, skills, and roles around those constraints rather than around the latest tool.