Stop Importing Your AI Workforce Forecast

Somewhere this quarter, a leadership team will set its AI workforce strategy by quoting a founder.

Someone will mention that Sam Altman just said he is “delighted to be wrong” about the jobs apocalypse. Someone else will counter that Dario Amodei still expects up to half of entry-level white-collar work to disappear within five years. The room will lean toward one of them, and that lean will quietly shape next year’s headcount plan.

That is the mistake. Not which founder the room believes. The act of importing the forecast at all.

The forecast was never a planning input

A founder’s forecast is a public position, and public positions move. They move with new evidence, and they move with incentives, and from the outside you usually cannot tell which.

Consider the timing of Altman’s reversal. He spent two years warning that AI would erase most jobs, that entire categories would be “totally, totally gone.” Then, at a Commonwealth Bank conference in Sydney in late May, he walked it back, saying he no longer expects “the kind of jobs apocalypse that some of the companies in our space advocate or talk about.” The reversal landed four days after OpenAI filed confidentially for an IPO targeting more than a trillion dollars. A policy researcher quoted in the same TIME report, Peter Wildeford, put the question plainly: it is hard to know whether the industry has actually updated its forecasts or just its messaging.

Amodei, for his part, has not moved. He still holds the harder line, up to half of entry-level white-collar jobs gone within five years, unemployment reaching 10 to 20 percent.

So the two people closest to the technology now point in opposite directions, and you cannot settle the disagreement from your office. Neither can the public data, but the data is at least more useful than a tie.

What the data actually says

At the unemployment level, nothing has moved yet. The Yale Budget Lab, tracking AI’s effect on employment, found no meaningful change in unemployment for high-exposure jobs through March 2026.

But the first place an effect would surface is not the unemployment rate. It is entry-level hiring. And that has already started to move. Anthropic’s own labor researchers, Maxim Massenkoff and Peter McCrory, found that the job-finding rate for workers aged 22 to 25 in AI-exposed fields fell about 14 percent against 2022. They are careful to call the result barely significant, and on its own it would be easy to wave away.

It does not sit on its own. A Stanford team, Erik Brynjolfsson with Bharat Chandar and Ruyu Chen, working from ADP payroll records rather than survey responses, found a 13 percent relative decline in employment for the same 22-to-25 cohort in the most AI-exposed occupations, concentrated in the roles where AI automates the work rather than assists it. Two datasets, two methods, the same age band, the same direction. A finding that is barely significant alone becomes hard to dismiss when an independent one lands beside it. The aptly titled paper calls these workers the canaries, and the unemployment rate is simply the part of the mine the canary reaches first.

So the macro all-clear and the early-warning signal are both true at the same time. That is not a contradiction. It is what a leading indicator looks like before it becomes a lagging one.

The layoff headlines point the same way, for a reason worth naming. AI has been cited as justification for nearly 50,000 job cuts through April, according to the outplacement firm Challenger, Gray and Christmas. But the firm’s own Andy Challenger added the line that should make any CEO pause: regardless of whether individual jobs are being replaced by AI, the money for those roles is. The public story of “AI took the job” is often a cost decision wearing the language of a capability decision, the same substitution we traced in AI layoffs as debt, not AI. Which is exactly why a forecast built on those headlines is the wrong thing to plan against.

The rung that moves first

Here is why entry-level is the signal to watch rather than the headline to react to.

Booking Holdings CEO Glenn Fogel described the mechanism without naming it, saying “the lowest rung on the ladder has been pulled away by AI.” He was talking about customer service work being handed to automation, a narrower setting than a leadership pipeline, but the mechanism generalizes. Pull the lowest rung, and the ladder still looks intact from across the room. The damage is invisible until someone tries to climb.

Senior people are grown from junior ones. They are not bought at scale, because the supply of ready-made senior talent is thin and expensive, and because the judgment that makes someone senior is built on years of doing the junior work first. If you quietly stopped backfilling junior roles this year, you have not made a hiring decision. You have made a 2031 leadership decision, and you made it without putting it on any agenda. We mapped the front end of this dynamic in the AI hiring freeze paradox; the back end is the bench you will not have.

This is why importing a forecast is the wrong move. The forecast lives at the level where the unemployment numbers have not moved yet. The decision that actually shapes your company lives at the rung level, where something already has.

Instrument three signals you already own

The founders will keep revising. You do not need their forecast, because you have something they do not: direct access to the only workforce data that describes your company. Three numbers tell you where you actually stand, and unlike the macro, they move before the damage is locked in. At an 80 to 200 person company the samples are small, so read these as direction over a few quarters, not as precise statistics in any single one. The trend is the signal; the monthly wobble is not.

Hiring funnel composition by seniority. Track what share of your hires over the last four quarters were entry-level or junior, against the share that were senior. The absolute count matters less than the trend. A steadily falling junior share means you are buying experience you have stopped growing, and the bill arrives years later. Healthy looks like a stable or deliberate ratio. Decaying looks like a junior share quietly sliding while no one decided it should.

Internal mobility rate. Track the percentage of open roles filled by someone already inside the company. Internal mobility is the visible proof that your bench works, that people enter at one level and climb. When AI absorbs the tasks that used to be the climbing rungs, mobility stalls first, often a year or two before anyone notices the pipeline has gone dry. A falling internal-fill rate is the bench telling you it is running out of people to promote.

Skill-mix delta year over year. Track how the actual capability profile of your team is changing, not headcount but composition. Are you adding people and skills that complement AI, the judgment, the orchestration, the domain depth that makes the tools worth more? Or is the mix simply thinning as roles get cut and nothing is added in their place? A cut that reduces cost while hollowing the skill mix is the most expensive kind, because you cannot see the loss on a budget line.

None of these requires a new system. Your HRIS already holds all three. What is usually missing is the discipline to read them together, on a cadence, as a signal rather than as a year-end report.

The review that replaces the forecast

Put the three numbers on one page, and review them once a quarter, the same way you review pipeline or cash.

Ask three questions of the page. Is our junior hiring share trending where we chose, or where the budget cycle pushed it? Is internal mobility holding, or are we quietly losing the ability to promote from within? Is our skill mix getting deeper against AI, or just thinner? The point is not to predict the macro. The point is to notice your own company changing while the change is still cheap to correct.

The founders will keep revising their forecasts in public, on stages, ahead of their funding rounds. Your three numbers will keep reporting in private, every quarter, with no public narrative to manage. Only one of those is a planning input.