The AI industry in 2026 is no longer a story about whether firms have a chatbot. McKinsey’s latest global survey, in the field from 4 May to 8 June 2026 with 1,719 respondents, finds organisations scaling tools across more functions while the share reporting any EBIT impact stays stuck. Stanford HAI’s 2026 AI Index adds the other half of the picture. Capability jumped. Incidents jumped with it. The useful read for an operator is the gap between those two facts, not another list of model names.
Three changes are large enough to plan around. Agents left the demo reel at large firms and stayed flat at smaller ones. Cost, including tokens, started to constrain use for about one in five organisations. And a small group of high performers kept pulling away by redesigning work, not by buying more seats.
What actually changed
Chat is the common tool. McKinsey finds 47 per cent of respondents scaling chatbots across the enterprise, and about two in ten scaling AI agents in the same way. Coding agents sit in that second band, and they are already changing budgets. Thirty-two per cent of respondents say their organisations decided against buying at least one software product or feature because they could build it internally with agentic coding tools. That is not a hobby. It is a make-versus-buy shift showing up in the survey for the first time at this scale.
The financial result did not follow. Thirty-seven per cent attribute at least some EBIT impact to AI, about the same as last year. Eighty per cent say AI improved their own productivity. The AI industry in 2026 is living in that split. Individuals got faster. The P&L mostly did not.
Gartner’s C-suite survey, published 1 September 2026 from 1,303 respondents at organisations with at least $50 million in revenue, rhymes with that. Only 22 per cent have successfully scaled AI across multiple business units or adopted an AI-first approach. Eighty-five per cent of functional leaders still plan to increase spending in 2026, after putting an average of 12 per cent of functional budgets into AI in 2025. Spend is not the same as scale. Scaling AI agents beyond pilots is the operational version of that sentence.
Capability jumped. So did the mess.
Stanford’s 2026 Index is the independent read on the technology. Industry produced over 90 per cent of notable frontier models in 2025. On SWE-bench Verified, coding performance rose from 60 per cent to near 100 per cent in a year. Organisational adoption reached 88 per cent. Those are capability facts, not a reason to copy a lab demo into a five-person office.
The same report is blunt about the other side. Documented AI incidents rose to 362, from 233 in 2024. Almost all leading developers report capability benchmarks. Reporting on responsible-AI benchmarks remains spotty. Models can take a gold medal at the International Mathematical Olympiad and still read analog clocks correctly about half the time. Stanford calls that the jagged frontier. An operator should treat it as a staffing fact. The tool is brilliant at some jobs and casually wrong at others that look easy.
NIST’s generative AI profile already named over-reliance as a configuration risk. A year of faster models did not retire that warning. It made it cheaper to ignore.
What to watch next
Watch the cost line. McKinsey finds about 20 per cent of respondents constrained by AI operating costs, including tokens. High performers report that constraint most often on coding agents. A tool that writes software all day is a token furnace. If you cannot see the bill by job, you cannot decide whether the job is worth it. Gartner found roughly 11 per cent of organisations entirely unaware of what their function spent on AI in 2025. That is the tell. You cannot manage a furnace you cannot meter.
Watch the size gap. Large organisations moved from 27 per cent to 40 per cent scaling agents. Smaller ones stayed at 22 per cent. Catching up does not mean copying their stack. It means picking one job with a hard stop, as moving agentic AI from pilot to production argued in working language. The industry story and the office story are the same plot at two scales.
Watch who is redesigning work. McKinsey’s high performers, about 6 per cent of respondents, are 3.3 times more likely to intend a fundamental business transformation within three years. Nearly three-quarters of them report redesigning workflows because of AI, up from 55 per cent last year. A quarter of everyone else says the same. Seats are not the tell. Workflows are.
Watch talent and capital, with a grain of salt on the China comparison. Stanford puts US private AI investment at $285.9 billion in 2025, and notes that private figures likely understate China’s total. The US still hosts 5,427 data centres, more than ten times any other country, and TSMC still fabricates almost every leading AI chip. Those are industry structure facts. They are not a buying guide. AI investment trends to watch in 2026 is the capital-markets companion to this operating read.
What a small firm should do with this
Do not copy the 40 per cent. Copy the discipline behind the 6 per cent. Name one workflow. Write the stop. Meter the tokens. Read every outbound sentence. Popular use cases are not the high-return ones in Gartner’s function-level data. IT leaders chase cybersecurity detection, service-desk automation, and code generation. Positive-return mentions cluster elsewhere, including asset and cost optimisation. Your office is not Gartner’s sample. The lesson is to pick the boring job that already has a test of done.
The AI industry in 2026 got faster, more expensive to run at scale, and no better at turning personal speed into EBIT. Agents are a large-firm story so far. Incidents rose with capability. The next twelve months will sort firms that meter the work from firms that buy another seat.
Key takeaways
- McKinsey: 40 per cent of large firms scale agents, 22 per cent of smaller ones, the smaller share flat. Chatbots scale at 47 per cent.
- EBIT impact sits at 37 per cent. Personal productivity sits at 80 per cent. That split is the year.
- Gartner: 22 per cent have scaled AI across business units. 85 per cent still plan to spend more.
- Stanford: incidents rose from 233 to 362. Capability reporting is common. Responsible-AI reporting is not.
- Watch token cost, the size gap, and whether anyone is redesigning the workflow. Seats are not the tell.
FAQ
What changed in the AI industry in 2026?
Agents scaled at large firms and stayed flat at smaller ones. Token costs started to constrain about one in five organisations. Personal productivity stayed high. Firm-level EBIT impact did not move.
Are AI agents now common?
Not in McKinsey’s survey. About two in ten respondents scale agents across the enterprise, against 47 per cent for chatbots. The 40 per cent figure is large firms only.
Why does EBIT impact stay flat if people feel more productive?
A faster draft is not a redesigned workflow. McKinsey’s high performers are the ones rewriting the path, not inserting a chatbot into the old one.
Should a small firm copy large-firm agent programmes?
No. Copy the stop, the meter, and one job with a test of done. The size gap in McKinsey’s 2026 survey is the warning, not a race.
What should I watch for the rest of 2026?
Token bills by job, whether incidents and evaluations keep pace with capability, and whether anyone in the firm can name a workflow they actually changed.
Sources
- McKinsey, “The state of AI in 2026: On the road to ROI”, 25 August 2026. mckinsey.com/…/the-state-of-ai
- Stanford HAI, “The 2026 AI Index Report”. hai.stanford.edu/ai-index/2026-ai-index-report
- Gartner, “Gartner Survey Finds Only 22% of Organizations Have Successfully Scaled AI Across Multiple Business Units”, 1 September 2026. gartner.com/…/successfully-scaled-ai-across-multiple-business-units
- NIST, “Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile” (NIST AI 600-1), 26 July 2024. nist.gov/publications/…generative-artificial-intelligence
This article is general information, not investment advice. See section 4 of our Terms & Conditions.
This is the kind of decision we advise on. If you are weighing it for your own business, see how we work or start a conversation.
