There is a particular silence in a room of policymakers when someone shows them the displacement numbers. Forty percent of customer service. Thirty percent of paralegal work. The first cohort of junior software engineers who will not be hired this year because the work they would have done now happens between a prompt and a response. The slides are clean. The forecasts are sourced. Nobody argues with the numbers. The silence is what they do next, which is nothing, because the numbers describe a problem that does not yet have a constituency. The people who will lose these roles do not yet know they are losing them. The people who could speak for them do not yet exist as a group.
This is the failure mode that matters. Not the displacement itself. The displacement is real, but it is also the part that gets attention, because compression is visible. Compression has a name and a face and a press cycle. What does not have a face is the thing forming on the other side of it. A new layer of work is appearing above the automated layer. Coordination, judgment, taste, accountability. The orchestrator role. It is appearing now, in real time, inside the companies that have already started using these tools at scale. It is also being staffed entirely by people who were already on the inside of those companies. Everyone else is reading the displacement story.
The mechanism here is not new. Every general-purpose technology in economic history has done the same thing. The spreadsheet did not kill bookkeeping. It killed manual tabulation and created financial analysis, corporate development, modelling consultancies. The web did not kill retail. It killed classified ads and created the entire architecture of digital commerce, growth, product management, devrel. Each abstraction layer compressed the work beneath it and opened a larger surface above it. Each transition produced winners who could ride the new layer and losers who could not. The losers were not stupid. They were not lazy. They were standing on the wrong side of a wall that the policy class had not yet noticed was there.
The wall this time has a name. It is the gap between the moment a role disappears and the moment the role that replaces it becomes legible enough for an ordinary person to train for it. The mathematics of this gap are unforgiving. If displacement runs at, say, five years and category creation lags by three, the area between those two curves is human suffering measured in tax base, in dignity, in the slow-motion radicalisation of people who can see they have been left behind but cannot see what they were left behind from. The endpoint optimists are right about where we end up. The lag pessimists are right about everything that happens in between. Both arguments are correct. The policy class keeps trying to choose between them.
I learned this in a smaller way running businesses through earlier transitions. When I was supporting Avado, I watched two cohorts of mid-career professionals try to retrain into digital work at the same time. The ones who started while they still had their jobs got through. The ones who started after they had lost them, mostly did not. The difference was not aptitude. The difference was time, money, and identity. By the time the second group arrived, they had spent six months in the version of themselves that had been made obsolete, and the work of becoming someone new was now compounded by the work of unbecoming someone old. Reskilling is not a curriculum problem. It is a sequencing problem. You have to start before the displacement, not after it.
The lesson is one sentence and the rest of this is the cost of not learning it. Reskilling has to begin before the role is lost, not after, because the gap is not a gap in skill. It is a gap in identity, and identity does not retrain on a deadline.
What this means for someone in a policy seat right now is specific. It means the orchestrator role has to be named, formalised, credentialed, and made legible before the displacement curve peaks, which on current trajectories is inside this decade. It means the people who will need this training are not the people currently asking for it, because the people currently asking for it are already on the ridge, they can see and touch the new AI world of work now. It means the budget line for reskilling has to be funded out of the surplus the new technology is generating, captured at the point of capital accumulation, and routed to the workers whose labour is being absorbed into the model weights. None of this is novel as policy. All of it is overdue.
The hard part is that the constituency for this work does not yet exist, and by the time it does, the work will already be late. This is the structural problem with every transition: the people who need the help cannot ask for it in language the system recognises until the help is no longer the help they need. Policymakers who wait for the demand signal will arrive on time for the wrong intervention. The ones who move before the signal will be accused of solving a problem that does not yet exist, right up until the moment it does.
Every economic transition in history has produced two groups of people. The ones who reached the new ground and the ones who did not. The line between them is almost never talent. It is almost always whether someone built them a bridge while the old ground was still under their feet.
I asked AI to solve the timing issue, here is what it recommended:
https://claude.ai/public/artifacts/c607c539-1332-46b7-97cc-3647479fa178