Eight numbers that carry the argument. The first six say something is happening at the entry level and in AI-exposed work; the seventh is the strongest evidence it is not yet economy-wide; the eighth is the capital flow driving both sides of the ledger.
Pulled straight from the Bureau of Labor Statistics on load — the official series where displacement would have to show up if it were real at scale. Two-year sparklines; year-over-year change coloured by whether the move is the bad direction for workers.
Thirty-four feeds merged, deduplicated and channel-routed — eight Google News queries plus Indeed Hiring Lab, HR Dive, HR Executive, CIO Dive, Axios, Guardian, NYT, BBC, WIRED, Ars Technica, The Register, Yahoo Finance, MarketWatch, Fortune, Quartz, Fast Company, TechCrunch, NBER and fresh arXiv economics preprints. General-interest desks are filtered to headlines carrying both an AI and an employment signal, so a tech story about whales doesn't land here. Stories that arrive with art lead the section. Refreshes on load and every ten minutes.
Ordered by observed impact, not by fear. Severe means measured job, posting or revenue loss is already on the record. High means high task exposure with early or partial evidence. Moderate means exposed on paper but slowed by licensure, liability or regulation. Insulated means physically, legally or capacity-constrained — several of these are net winners. Microsoft applicability scores are shown where they exist; read them as task overlap, not replaceability.
| Occupation | Risk | AI score | Evidence |
|---|
Five blocks is severe structural exposure; zero blocks marks a net beneficiary of the build-out. The pattern is consistent — sectors selling human hours or intermediated information are being repriced, sectors selling physical capacity are being bid up.
Live prices on the displacement trade, refreshed on load. Rel is year-to-date performance against the S&P 500 — the column that separates AI disruption from a general market move. The written thesis under each name is editorial and dated; the numbers next to it are not. Not investment advice.
| Name | Side | YTD | Rel S&P | 1yr | Trend | Thesis |
|---|
Everything above, stepped back from. Eight things the corpus actually shows — including the reconciliation of the field's most-cited disagreement, and the several places where the data cuts against the popular version of the story.
Where those eight point next, across four horizons. Probabilities are judgment; the resolve date and the falsifier on each are there so the calls can be scored rather than quietly forgotten.
2026 announcements where the employer itself named AI, automation or machine learning as a driver. Read with the caveat below — AI is a flattering reason to give for a cut that had other causes.
| Company | Month | Headcount | Functions cut |
|---|
The honest state of the record. Three findings are well established, three are genuinely contested by serious researchers, and four are methodological caveats most coverage drops.
Read as inference from the tables above, not as advice.
Every citation chip on this page resolves to one of these. Primary research first, trade press where it carried the only reported figure.
Live endpoints — keyless, fetched server-side, cached at the edge:
