McKinsey 2026: more large firms scale agents, but EBIT impact is flat. Read the gap
McKinsey's 2026 survey: 40% of large firms now scale AI agents, yet only 37% report any EBIT impact from AI. What the terms mean and how to read them.
The short answer
McKinsey's 2026 survey shows agent use rising fast at large organisations while reported profit impact stands still. 40% of respondents at firms with $1 billion or more in revenue say they scale agents, up from 27%, yet 37% overall attribute any EBIT impact to AI, about the same as a year earlier. Scaling describes deployment, not returns.
Key takeaways
- In McKinsey's survey of 1,719 respondents in 97 countries (4 May–8 June 2026), 40% of those at organisations with $1 billion or more in revenue reported scaling AI agents, up from 27% a year earlier.
- At smaller organisations the share scaling agents stayed flat at 22% in the same survey.
- Only 37% of respondents attributed at least some EBIT impact to AI, about the same share as in 2025, and AI high performers stayed at about 6%.
- "Scaling" is a self-reported stage of deployment and "EBIT impact" means any attributed contribution, so neither figure measures return on a specific agent.
- A smaller firm should take the method from these surveys, not the rates: count cost per task and hours saved on one workflow before scaling anything.
In this article
What McKinsey published on 25 August
McKinsey's State of AI 2026 says agents are spreading at large organisations while the share reporting financial impact has not moved. The survey drew 1,719 respondents at all levels in 97 countries, fielded from 4 May to 8 June 2026.1
On agents, 40% of respondents from organisations with revenue of $1 billion or more reported scaling AI agents, up from 27% a year earlier. At smaller organisations the share stayed flat at 22%.2
On money, 37% of respondents attributed at least some EBIT impact to AI, about the same share as last year.1 AI high performers, those attributing 5% or more of EBIT to AI and describing its value as significant, stayed at about 6% of respondents.1
McKinsey, May–June 2026
Agents up, reported profit flat
What "scaling" means in the survey
Scaling is a stage respondents choose for their organisation, not a measured count of agents in production. McKinsey asks whether an organisation is not using agents, experimenting, piloting or at least scaling them, and defines agents as systems based on foundation models that plan and execute multiple steps in a workflow.2
Two details matter. The question went only to respondents whose organisations use AI regularly, and the results were rebased to the full sample. Scaling in one or more functions also counts, so one agent scaled in IT qualifies.2
Our view: read the 40%2 as a statement about intent and deployment at large firms. It does not tell you how many agents run, how much work they do or whether anyone measured the result. Our note on registered versus active agents makes the same point about platform counts.
What "EBIT impact" means
EBIT impact is self-attributed and the bar is low. The 37% are respondents who say AI contributed at least some EBIT at their organisation.1 Any positive contribution counts, and no figure is audited.
The flat line is the real finding. McKinsey reports more organisations scaling AI across the enterprise than a year ago, and 80% of respondents say AI improved their own productivity.1 Yet the share seeing any profit effect did not rise. Individual gains are not reaching the income statement.
McKinsey's own explanation is in how high performers work: they redesign workflows around AI rather than insert it into existing ones, and they treat operating costs as a design constraint. About 20% of respondents said AI operating costs, including token costs, had constrained their AI use.1
Other surveys tell the same story
KPMG's US pulse points the same way. It surveyed 204 US leaders at firms with $1 billion or more in revenue between 28 April and 25 May 2026. Of those, 53% were deploying AI agents, and only 26% had full, real-time visibility of what their AI systems cost to run.3 Our note on the KPMG pulse covers the cost gap.
Agent counts rise while cost and value are poorly measured. That is the gap the McKinsey headline describes.
The risk of stopping halfway is on record too. Gartner predicted in June 2025, in a forecast now more than a year old, that over 40% of agentic AI projects will be cancelled by the end of 2027.4 It cited cost, unclear value and weak risk controls. That is a prediction, not an outcome, and the dated forecast should be read alongside newer surveys.
How should a smaller firm read large-company figures?
Take the method, not the rates. McKinsey's two size groups are large: 595 respondents at organisations with $1 billion or more in revenue and 1,035 at smaller ones.2 The smaller group runs from a few dozen staff to several thousand, so its 22% says little about a firm of 100 people.
What does transfer is the lesson in the gap. Deployment is easy to report and value is hard to show. A smaller firm can skip the deployment race and start with one workflow whose result can be counted: cost per task before and after, hours saved and signed off, and an owner for the exception queue.
Our view: one measured agent beats a portfolio of unmeasured ones. Our scoring model for the first process to automate sets out how to choose that workflow, and our breakdown of monthly running costs covers the operating side McKinsey's high performers watch.
Every automation sprint on our AI automation page agrees acceptance criteria before the build and signs off hours saved at the end, for exactly this reason.
Sources
- McKinsey, The State of AI 2026: n=1,719, 97 countries, 4 May–8 Jun 2026 (Aug 2026)
- McKinsey, The State of AI 2026: organisations with $1 billion or more in revenue vs below $1 billion (Aug 2026)
- KPMG, AI Quarterly Pulse Q2 2026: 204 US leaders at $1bn+ firms, 28 Apr–25 May 2026 (Jun 2026)
- Gartner prediction (Jun 2025): over 40% of agentic AI projects cancelled by end of 2027(dated)
