McKinsey's State of AI in 2026, published on 25 August from 1,719 respondents across 97 countries surveyed between 4 May and 8 June, contains one number that moved and one that did not.3 explainx.ai 2026-08-25 97 countries, fielded 4 May to 8 June, published 25 August; agents concentrated in IT, customer operations and software engineering; workflow redesign as the high performer marker; 14 percent saw reductions last year versus 39 percent expecting them. Open source The share of large organizations scaling AI agents rose to 40 percent from 27 percent a year earlier.1 The Register 2026-08-25 1,719 respondents; 37 percent EBIT impact, flat; 6 percent high performers, flat; 40 percent of large firms scaling agents from 27; a third built instead of bought; 39 percent expect cuts from 32 and 2025 cuts fell short of predictions; 80 percent productivity; 20 percent cost constraint; Chui quote. Open source The share of organizations reporting any positive impact on earnings before interest and taxes stayed at 37 percent, exactly where it was in 2025, and the share qualifying as high performers, attributing 5 percent or more of EBIT to AI, stayed at 6 percent.1 The Register 2026-08-25 1,719 respondents; 37 percent EBIT impact, flat; 6 percent high performers, flat; 40 percent of large firms scaling agents from 27; a third built instead of bought; 39 percent expect cuts from 32 and 2025 cuts fell short of predictions; 80 percent productivity; 20 percent cost constraint; Chui quote. Open source 4 Value Add Pulse 2026-08-26 Confirms the core figures; frames the central tension as conviction growing faster than measurable returns. Open source Meanwhile 80 percent of individual users say AI improved their productivity.1 The Register 2026-08-25 1,719 respondents; 37 percent EBIT impact, flat; 6 percent high performers, flat; 40 percent of large firms scaling agents from 27; a third built instead of bought; 39 percent expect cuts from 32 and 2025 cuts fell short of predictions; 80 percent productivity; 20 percent cost constraint; Chui quote. Open source Our assessment, with high confidence, is that the survey documents a gap between individual productivity and enterprise profit that a year of agents has not closed; with moderate confidence, that the 32 percent of companies now building software instead of buying it is the finding with the largest consequences outside the survey; and with moderate confidence that the 6 percent who redesigned workflows rather than adding tools are the only group for whom the profit question has been answered.
The gap that did not close
The survey's central tension, as Value Add Pulse summarizes it, is that conviction in AI is growing faster than measurable financial returns.4 Value Add Pulse 2026-08-26 Confirms the core figures; frames the central tension as conviction growing faster than measurable returns. Open source Spending is up, agents are being scaled in IT, customer operations and software engineering, and four in five users feel more productive.3 explainx.ai 2026-08-25 97 countries, fielded 4 May to 8 June, published 25 August; agents concentrated in IT, customer operations and software engineering; workflow redesign as the high performer marker; 14 percent saw reductions last year versus 39 percent expecting them. Open source None of that has moved the share of companies that can point to a line in their income statement. McKinsey senior fellow Michael Chui's framing, that return on investment is a journey not a destination, is the consultancy's way of saying the same thing.1 The Register 2026-08-25 1,719 respondents; 37 percent EBIT impact, flat; 6 percent high performers, flat; 40 percent of large firms scaling agents from 27; a third built instead of bought; 39 percent expect cuts from 32 and 2025 cuts fell short of predictions; 80 percent productivity; 20 percent cost constraint; Chui quote. Open source
The high performers are the exception, and the survey is specific about what distinguishes them: fundamental workflow redesign rather than bolt on tool adoption.3 explainx.ai 2026-08-25 97 countries, fielded 4 May to 8 June, published 25 August; agents concentrated in IT, customer operations and software engineering; workflow redesign as the high performer marker; 14 percent saw reductions last year versus 39 percent expecting them. Open source That is a small group, 6 percent, and it has not grown.1 The Register 2026-08-25 1,719 respondents; 37 percent EBIT impact, flat; 6 percent high performers, flat; 40 percent of large firms scaling agents from 27; a third built instead of bought; 39 percent expect cuts from 32 and 2025 cuts fell short of predictions; 80 percent productivity; 20 percent cost constraint; Chui quote. Open source If redesigning work were easy, the number would have moved. Twenty percent of respondents cite AI operating costs as a constraint on deployment, which suggests that for a meaningful share the arithmetic is not yet positive even where the productivity is real.1 The Register 2026-08-25 1,719 respondents; 37 percent EBIT impact, flat; 6 percent high performers, flat; 40 percent of large firms scaling agents from 27; a third built instead of bought; 39 percent expect cuts from 32 and 2025 cuts fell short of predictions; 80 percent productivity; 20 percent cost constraint; Chui quote. Open source 4 Value Add Pulse 2026-08-26 Confirms the core figures; frames the central tension as conviction growing faster than measurable returns. Open source
Build instead of buy
Thirty two percent of respondents said their organization decided against buying at least one software product or feature because it could be built internally with agentic coding tools.2 Forkast via Yahoo Finance 2026-09-01 Build over buy at 32 percent, nearly half among high performers; industry breakdown; 17 percent deployed agents, 11 percent production ready; Gartner 40 percent cancellation forecast; Van der Veken quote. Open source Among high performers the figure is nearly half, against 31 percent of everyone else.2 Forkast via Yahoo Finance 2026-09-01 Build over buy at 32 percent, nearly half among high performers; industry breakdown; 17 percent deployed agents, 11 percent production ready; Gartner 40 percent cancellation forecast; Van der Veken quote. Open source The practice is most common in technology at 41 percent, healthcare payers and providers at 39, and professional services and energy at 38.2 Forkast via Yahoo Finance 2026-09-01 Build over buy at 32 percent, nearly half among high performers; industry breakdown; 17 percent deployed agents, 11 percent production ready; Gartner 40 percent cancellation forecast; Van der Veken quote. Open source McKinsey senior partner Lieven Van der Veken describes leaders asking what their organizations need to build tools themselves.2 Forkast via Yahoo Finance 2026-09-01 Build over buy at 32 percent, nearly half among high performers; industry breakdown; 17 percent deployed agents, 11 percent production ready; Gartner 40 percent cancellation forecast; Van der Veken quote. Open source
This is the finding that reaches beyond the survey population. A third of enterprises declining at least one software purchase is a demand shock for the application software industry, whose entire model assumes that building is harder than buying. Forkast's piece adds the caution that internally built systems succeed less often than vendor tools and that only 17 percent of organizations have deployed agents at all, with 11 percent at production readiness.2 Forkast via Yahoo Finance 2026-09-01 Build over buy at 32 percent, nearly half among high performers; industry breakdown; 17 percent deployed agents, 11 percent production ready; Gartner 40 percent cancellation forecast; Van der Veken quote. Open source The build decision may be cheaper to make than to live with. But the decision is being made.
Who gains and who loses
The 6 percent gain, and gain durably, because workflow redesign is hard to copy and the survey suggests most firms have not done it.3 explainx.ai 2026-08-25 97 countries, fielded 4 May to 8 June, published 25 August; agents concentrated in IT, customer operations and software engineering; workflow redesign as the high performer marker; 14 percent saw reductions last year versus 39 percent expecting them. Open source The frontier labs and coding tool vendors gain from every build decision, since building in house means buying tokens and seats from them instead of licenses from an application vendor.2 Forkast via Yahoo Finance 2026-09-01 Build over buy at 32 percent, nearly half among high performers; industry breakdown; 17 percent deployed agents, 11 percent production ready; Gartner 40 percent cancellation forecast; Van der Veken quote. Open source McKinsey gains a two year narrative arc it can sell: the journey.
Application software vendors lose, at the margin, a third of a purchase cycle, and the loss is concentrated in the industries and companies most able to pay.2 Forkast via Yahoo Finance 2026-09-01 Build over buy at 32 percent, nearly half among high performers; industry breakdown; 17 percent deployed agents, 11 percent production ready; Gartner 40 percent cancellation forecast; Van der Veken quote. Open source Workers face a widening gap between what they expect and what has happened: 39 percent of respondents now expect AI driven job cuts, up from 32 percent, while only 14 percent saw reductions in the past year and 2025 cuts fell short of 2024 predictions.1 The Register 2026-08-25 1,719 respondents; 37 percent EBIT impact, flat; 6 percent high performers, flat; 40 percent of large firms scaling agents from 27; a third built instead of bought; 39 percent expect cuts from 32 and 2025 cuts fell short of predictions; 80 percent productivity; 20 percent cost constraint; Chui quote. Open source 3 explainx.ai 2026-08-25 97 countries, fielded 4 May to 8 June, published 25 August; agents concentrated in IT, customer operations and software engineering; workflow redesign as the high performer marker; 14 percent saw reductions last year versus 39 percent expecting them. Open source And chief financial officers lose patience, because a second flat year on the EBIT question is the point at which pilot budgets start being asked for receipts.
The counter case
The assessment that the gap is a problem could be wrong if it is a lag. Productivity gains at the individual level take time to aggregate into margin: they show up first as work done faster, then as headcount not hired, then as cost. A flat 37 percent after a year of agents is consistent with a technology that is early, not one that is failing, and McKinsey's own framing leans that way.1 The Register 2026-08-25 1,719 respondents; 37 percent EBIT impact, flat; 6 percent high performers, flat; 40 percent of large firms scaling agents from 27; a third built instead of bought; 39 percent expect cuts from 32 and 2025 cuts fell short of predictions; 80 percent productivity; 20 percent cost constraint; Chui quote. Open source The build versus buy figure could also be less consequential than it looks: declining one feature purchase is a low bar, and Gartner's forecast that more than 40 percent of agentic projects will be canceled by end of 2027 suggests many of those internal builds will be abandoned and the purchases made after all.2 Forkast via Yahoo Finance 2026-09-01 Build over buy at 32 percent, nearly half among high performers; industry breakdown; 17 percent deployed agents, 11 percent production ready; Gartner 40 percent cancellation forecast; Van der Veken quote. Open source Finally, this is a self reported survey of people who chose to answer a McKinsey questionnaire about AI, and the 80 percent productivity figure in particular measures feeling, not output.3 explainx.ai 2026-08-25 97 countries, fielded 4 May to 8 June, published 25 August; agents concentrated in IT, customer operations and software engineering; workflow redesign as the high performer marker; 14 percent saw reductions last year versus 39 percent expecting them. Open source
What to watch
- The 2027 EBIT number. If next year's survey moves the share reporting profit impact above 45 percent, the lag reading holds; a third flat year would make the gap structural.1 The Register 2026-08-25 1,719 respondents; 37 percent EBIT impact, flat; 6 percent high performers, flat; 40 percent of large firms scaling agents from 27; a third built instead of bought; 39 percent expect cuts from 32 and 2025 cuts fell short of predictions; 80 percent productivity; 20 percent cost constraint; Chui quote. Open source
- Application software net revenue retention. Public SaaS vendors reporting retention below 100 percent citing customer builds, in earnings through mid 2027, would confirm the build shift is reaching their books.2 Forkast via Yahoo Finance 2026-09-01 Build over buy at 32 percent, nearly half among high performers; industry breakdown; 17 percent deployed agents, 11 percent production ready; Gartner 40 percent cancellation forecast; Van der Veken quote. Open source
- Agent cancellations. Evidence for or against Gartner's 40 percent cancellation forecast by the end of 2027 will show whether the 40 percent of large firms scaling agents are scaling something that lasts.2 Forkast via Yahoo Finance 2026-09-01 Build over buy at 32 percent, nearly half among high performers; industry breakdown; 17 percent deployed agents, 11 percent production ready; Gartner 40 percent cancellation forecast; Van der Veken quote. Open source
- High performers grow past 10 percent. If the share attributing 5 percent or more of EBIT to AI reaches double digits in 2027, workflow redesign has become learnable; another year at 6 percent would suggest it is not.3 explainx.ai 2026-08-25 97 countries, fielded 4 May to 8 June, published 25 August; agents concentrated in IT, customer operations and software engineering; workflow redesign as the high performer marker; 14 percent saw reductions last year versus 39 percent expecting them. Open source
Eighty percent of people feel it and 37 percent of companies can book it. The distance between those two numbers is the entire enterprise AI market's unfinished work.