The number that moved
McKinsey surveyed 1719 professionals and business leaders for its State of AI in 2026 report. Of those, 37 percent attribute at least some EBIT impact to AI, about the same share as a year earlier. The firm's own high-performer bar is narrower: at least 5 percent of EBIT plus a description of the impact as significant. That bar is cleared by 6 percent of respondents, unchanged since last year.[1]
In the same survey the gain and the expectation sit at different levels. Among respondents who use AI in their own work, 80 percent say their personal productivity improved. Among all respondents, 39 percent expect their employer to cut jobs because of AI within the coming year, against 32 percent in 2025.[1]
The account the gain lands in
Between the two survey years the measured organisational return stayed where it was while the expectation of cuts rose from 32 percent to 39 percent. Both readings come from the same respondent pool; the belief about headcount moved while the composition of the sample held.[1]
The gain the survey concretely measures sits with the person doing the work: 80 percent of AI users report better personal productivity, and no line of the firm's reported earnings picks it up. One plausible reading is that the saved time returns to the worker's own throughput and reaches no measured account at all — neither wages, nor hours, nor price. Another reading is available, and the report itself offers it: coauthor Michael Chui says the return requires organisational change and that the high performers are seeing real return, which would make the flat 6 percent a sign of lag. The survey cannot separate the two, because the EBIT attribution rests on what respondents themselves say, with no audit standing behind it.[1]
The weight of the expectation
The weight of that expectation falls unevenly. An updated Stanford study puts employment for workers aged 22 to 25 in the most AI-exposed occupations 19 percent below their peers in less exposed fields, against a gap of 13 percent a year earlier, while across the economy as a whole the difference is close to nothing. The researchers trace the widening to slower hiring; firings and pay levels do not explain it.[2]
The spending side of the same survey gives that expectation both its footing and its limit. Among respondents at organisations above 1 billion dollars in annual revenue, 40 percent say they are scaling AI agents, up from 27 percent a year earlier. Close to a third say they built a software feature in-house with coding agents instead of buying it. In the same survey, 20 percent say AI operating costs have constrained their use. An employer reading those two lines has grounds to expect a smaller payroll and grounds to expect a larger bill.[1]
A measurable check exists, and McKinsey has already run it once: the report notes that actual workforce reductions in 2025 fell well short of what the previous year's respondents had anticipated. If the same gap repeats, the 39 percent expecting cuts this year will again exceed the real reduction the next survey measures. That was the reading of a column here on 22 August about developers giving their own off-hours to the agent; McKinsey's survey supplies the firm-level counterpart of it: the speed shows up where the person sits, and the account that would price it stays unwritten.[1], [3]