McKinsey has surveyed corporate AI use for close to a decade, and the 2026 edition, published August 25, reads like the ones before it: adoption up, conviction up, returns pending. Buried in the second section is the number that breaks the pattern. Among organizations with more than $1 billion in annual revenue, the share scaling AI agents in one or more business functions rose from 27 percent to 40 percent in twelve months. Among smaller organizations, it went from 22 percent to 22 percent.
The finding is not that agent adoption grew. It is that agent adoption grew in exactly one place.
What the survey found
McKinsey fielded the survey online between May 4 and June 8, 2026, collecting 1,719 responses across 97 nations, weighted by each country’s share of global GDP. Thirty-six percent of respondents sit above the billion-dollar revenue line. The authors are QuantumBlack senior partners Dan Tinkoff and Lieven Van der Veken, senior fellow Michael Chui, and associate partner Tara Balakrishnan.
The toplines move as expected. Forty-four percent say AI is scaling across their enterprise, up from 38 percent a year ago. Fifty-six percent report AI in three or more functions, up from 51. Chatbots remain the most widely scaled tool at 47 percent; agents and coding agents each sit near two in ten.
Split by company size, those same numbers pull apart. Fifty-four percent of respondents at billion-dollar-plus organizations report scaling AI enterprise-wide, against roughly one-third at smaller ones. On coding agents specifically, 31 percent of large enterprises report scaling. Every gauge shows the same shape, and the agent gauge shows it most sharply — because that is where the two groups started closest together. Five points apart last year. Eighteen now.
Why the gap widened
Three things plausibly explain it, none of them about willingness.
The first is capital. McKinsey found 28 percent of respondents now spend more than a tenth of their entire ICT budget on AI, and 60 percent expect to increase AI investment next year. Agents are not a license purchase. They are an integration project with a metered bill attached, and one in five respondents say AI operating costs, tokens included, are already constraining use. McKinsey reports that constraint as broadly consistent across company sizes — but a fixed token bill lands differently on a $50 million company than a $5 billion one.
The second is talent, downstream of the first. Scaling an agent means someone owns evaluation, rollback, and access control. Large organizations staff that as a function. Smaller ones assign it to whoever already has the most work.
The third is the coding agent itself.
The build-versus-buy shift
Nearly a third of respondents — 32 percent — told McKinsey their organization decided against buying at least one software product or feature because they could build it in-house with agentic coding tools. Among the small cohort McKinsey classifies as “AI high performers,” that figure is nearly half, against 31 percent of everyone else. The decisions cluster in technology and healthcare, then professional services and energy.
That is a software market number dressed up as an AI survey number. A meaningful slice of enterprise buyers has started treating the mid-tier SaaS feature — the reporting add-on, the workflow module, the connector — as something to generate rather than procure. The pressure lands hardest on vendors whose moat was implementation effort rather than data, distribution, or regulatory position.
It compounds the size gap too: the firms able to build instead of buy are the ones with benches big enough to maintain what they build. Smaller firms stay buyers.
What the survey cannot tell you
Every figure above is self-reported, and nobody verified that anything is actually scaling. “Scaling” is never defined; it is whatever the respondent takes it to mean. And “smaller organizations” spans a 900-person firm and a five-person one, which is a wide place to hide a trend.
More pointedly, the survey undercuts its own adoption story. Eighty percent say AI improved their individual productivity — but the share attributing any EBIT impact to AI is 37 percent, unchanged from last year, and high performers are flat at 6 percent. Last year 32 percent predicted AI-driven headcount declines; 14 percent report they materialized. To McKinsey’s credit, the authors checked that against the 552 people who took both surveys and got the same result. This year, 39 percent again predict declines.
The report puts it plainly: “Organizations’ conviction in AI is growing faster than the immediate financial returns they can attribute to it.”
Chui, pressed by The Register on that framing, defended the trajectory rather than the numbers. “Some ROI is already being achieved, and we expect more over time,” he said. “It’s a journey, not a destination.” The lag, he added, should surprise no one: “History doesn’t repeat itself, but it rhymes.”
Where the trackers disagree
Other 2026 datasets measure different things and should not be stacked on McKinsey. A widely circulated Digital Applied compilation puts 31 percent of enterprises with at least one agent in production, citing S&P Global Market Intelligence, with banking and insurance at 47 percent, healthcare at 18 and government at 14. That counts existence, not McKinsey’s scaling — and the same compilation credits its 31 percent partly to McKinsey, whose own 31 percent means something else entirely: large enterprises scaling coding agents. Gartner separately predicts more than 40 percent of agentic AI projects will be cancelled by end-2027. The three are not contradictory so much as non-comparable: one counts existence, one scale, one expected mortality.
What to watch
Three things. Whether the small-company line moves at all in 2027, or whether 22 percent becomes a durable floor. Whether the build-instead-of-buy figure surfaces in enterprise software revenue guidance, where it stops being a survey response and becomes a fact. And whether the EBIT number moves — because a flat 37 percent alongside rising adoption should worry McKinsey clients more than any gap between large and small.
“Some ROI is already being achieved, and we expect more over time. It is a journey, not a destination.”— Michael Chui, Senior Fellow, McKinsey QuantumBlack