They told you AI would take your job. Then they changed their story.

CEOs still forecasting significant AI job cuts. Down from 46% in January 2025.

Hold the two June numbers side by side. The EY-Parthenon survey has the share of CEOs expecting significant AI headcount cuts falling from 46% in January 2025 to 20% in May 2026. Challenger has AI as the leading cited reason for US job cuts for the fourth consecutive month, 101,743 attributed cuts in the half year, roughly 23% of everything announced.

One of those series is a survey of what executives say. The other is a count of what their companies file. When the said and the filed diverge this widely, the divergence is the story, and MIT's David Autor supplied the WSJ with the two candidate explanations: the labour market is genuinely not imploding as fast as they expected, and it was simply bad business to say your great new product will destroy the economy.

Both can be true at once. The wipeout rhetoric of 2025 was itself a pitch, aimed at investors pricing margin expansion. The reassurance of 2026 is a pitch too, aimed at workers whose adoption the product now needs and at a public whose opinion of AI has turned. Neither pitch was ever a forecast in the ledger sense: dated, falsifiable, scored.

The reading discipline follows. Treat executive AI commentary the way you treat guidance from a company mid-restructure: note it, then weight the filings, the cut counts and the rehiring patterns at ten times the interview. If your board is briefed on AI risk from keynote quotes, reset the sources this quarter.

The wipeout was a pitch. The reassurance is one too.

The mechanics of the walk-back are commercial, not sentimental. An EY survey found 78% of leaders now cast AI as a growth accelerant, and growth stories need staff who lean in. Pew has 6% of workers expecting AI to bring more opportunity against 32% expecting less; you cannot deploy copilots into a workforce that believes the copilot is the layoff. The reassurance is deployment lubricant.

Underneath it, the substitution economics did not move. Oracle's filing still attributes about 21,000 reductions to AI adoption under securities liability. Microsoft confirmed 4,800 cuts the same week its peers softened. Meta cut 8,000 in May while reassigning 7,000 to AI teams: the same headcount arithmetic, renamed.

And the honest complication belongs on the record: Ramp and Revelio's study of 21,000 US companies finds heavy AI adopters holding employment about 10.2% higher than non-adopters two years in. Adoption and headcount are not simple enemies. What the aggregate hides is composition: the retained jobs are senior, technical and supervisory, while the entry rung is what gets traded for the licence fees.

That composition is the shift for this magazine's readers. The walk-back does not restore the junior job in Chicago; the regret cycle reroutes it, and Forrester's prediction names the destination and the price: offshore, significantly cheaper. Position for the rung as it is being rebuilt, not as it was.

Fifty-five per cent of employers already regret an AI layoff. The walk-back is not an apology, it is narrative management, and the numbers underneath it did not walk back: record AI-attributed cuts, three in four customer-facing deployments rolled back, and the quiet rehiring routed offshore at a discount. Trust the filings, not the interviews.

The counterargument deserves its full weight: maybe the CEOs are right now and were wrong before. The labour market has not imploded. Heavy adopters employ more people than non-adopters. The rollback wave proves the technology cannot yet hold the roles it was sold into. On this reading the 2025 doom was the error and the 2026 calm is the correction.

The reply is in the sequencing. If the calm were a correction, it would arrive with reinstatements at parity and attribution rows quietly deleted from filings. Instead the cuts continue at record AI-attributed share, the rehiring is predicted to land offshore at significantly lower salaries, and 55% of employers privately regret cuts they will not publicly reverse. That is not a corrected forecast. That is a managed narrative wrapped around an unchanged plan.

Boards in Mumbai, Dubai and Nairobi hold the practical edge here, because the reroute runs through them. The questions to put on the next agenda: which functions is our market absorbing from the regret cycle, at what discount, under whose quality regime, and does the contract price the rollback risk the first buyer ignored. Prepared is a choice you make while the story is still being rewritten.

6 rising role categories, each with a sourced hiring signal.

The week-over-week correction in the layoff panel is deliberate. Last issue carried Microsoft at 5,500 from a pre-announcement report; the confirmed action, filed 6 July, is 4,800, about 2% of the workforce, and Microsoft states the eliminated roles are not being directly replaced by AI even as its people chief says AI is changing how work gets done. Estimates get corrected here, in the open, the same way ledger entries get scored.

The rehiring data is the half of the market nobody headlines. Robert Half finds 32% of US hiring managers who cut for AI later rehired the same or similar roles; Forrester predicts half of all AI-attributed cuts return, offshore and cheaper. Every rising role this issue sits in that seam: auditing the failures, designing the handoffs, feeding the training data, running the rebuilt rung from the receiving end.

None of these roles bet against AI. All of them bet against the assumption that deployment is simple. If your role produces output an agent can produce, the move is unchanged from last issue: one layer up, before the ladder is repriced under you.