You've probably used an AI tool at work in the last month. Maybe for a first draft, a summary of a long email thread, or a question that would have cost twenty minutes of searching otherwise. And you probably felt faster. More effective. Like something that took forty-five minutes now took ten.

Most of the people in McKinsey's 2026 State of AI survey agree with you. 1,719 respondents across 97 countries, surveyed May through June: four in five reported individual productivity gains from AI. Eighty percent. That's not a marginal finding.

The follow-up question is where it gets interesting. Did that productivity gain improve your company's EBIT — earnings before interest and taxes, the operating profit number that shows up in quarterly results?

Thirty-seven percent. Essentially unchanged from 2025.

That's a 43-point gap between feeling productive and actually moving the bottom line. And it has a specific explanation.

The contractor problem

Here's the object lesson.

Think about what happens when a construction firm buys faster power tools. The crew gets jobs done quicker — a two-day pour becomes one day, a week of finish work becomes three days. But unless faster completion lets the firm take on more projects, reduces headcount, or lets them charge more — the speed doesn't convert to profit. The tools cost something. The crew costs the same. The jobs get done faster but the margin stays flat or compresses.

AI is doing the same thing inside most companies right now. Individuals are faster. Functions are more productive. But if the company still pays for the same headcount, the same overhead, and now also the AI subscriptions on top — faster doesn't compound to richer automatically. The efficiency gains require you to do something with the freed capacity. Take on more work. Reduce costs. Charge more. Most companies haven't done that yet.

The McKinsey survey bears this out. The 6% of respondents McKinsey calls "high performers" — the ones attributing at least 5% of their EBIT to AI — are doing something structurally different. Nearly half of them skipped buying at least one software product this year because they built the equivalent internally with AI coding tools. Compare that to 31% of their peers doing the same. The high performers are converting efficiency into captured cost.

Source spread

• McKinsey — The State of AI: Global Survey 2026 [builder] — the primary source; methodology is published; 1,719 respondents, 97 countries, May 4–June 8, 2026.
• HPCwire / AIwire — Enterprise AI is a two-speed race [builder] — good framing of the high-performer vs everyone-else divergence.
• explainx.ai — McKinsey AI 2026: 80% productive, 37% EBIT [neutral] — accessible summary of the productivity/profit gap findings.
• Startup Fortune — 32% of firms now building instead of buying [skeptic] — correctly frames the build-vs-buy shift as a structural market signal, not a trend.

What's real:

• The productivity gains are real. Eighty percent is a large majority, and this isn't McKinsey asking people if AI is exciting. They asked if it improved their actual daily work output. Four in five said yes. I believe that.
• The agent adoption shift is real. Forty percent of large enterprises — the ones with more than $1 billion in revenue — are now scaling AI agents (software that handles multi-step tasks without constant human guidance) in at least one business function. That's up from 27% a year ago. Large companies are past the pilot phase.
• The build-vs-buy shift is real and significant. Thirty-two percent of organizations skipped at least one software purchase because they could build the equivalent with AI coding tools. If you work in B2B software, that number is a threat. If you work in an organization that buys a lot of software, it's an option you should probably be thinking about.

What deserves a side-eye:

• The EBIT gain number is suspiciously flat. 37% in 2026, essentially unchanged from 2025. A full year of additional AI adoption and the profit-impact number didn't move. That's the part McKinsey's own analysis calls the "productivity paradox," and it's not getting resolved quickly.
• The high-performer definition is circular. McKinsey defines high performers as companies that report 5%+ EBIT from AI. Then they show that high performers are doing things differently. But companies that have already extracted 5% EBIT improvement from AI have self-selected as the ones who figured it out — you can't straightforwardly reverse-engineer their practices and get the same result.
• "Scaling agents" is doing a lot of work. "One or more functions" and "scaling" are McKinsey survey language, not engineering definitions. A company automating email routing in one department qualifies. A company running autonomous agents across their entire operation qualifies. The same stat covers both.

What to do about it

Whether you're a worker using AI tools or a manager deciding whether to buy more of them, this data has practical implications.

• If you use AI at work: Track what you're doing with the time you save. Faster is only valuable if the capacity goes somewhere. If you're just getting the same output faster and leaving the rest of the day the same, the productivity gain is real — but invisible to anyone who has to justify an AI budget.
• If you manage a team: "We use AI and people feel more productive" is not a business case. The business case is what changed as a result — faster delivery, smaller team, more clients, higher margins. If you can't name the downstream change, you probably haven't captured the gain.
• If your company buys a lot of software: 32% of organizations in this survey skipped at least one software purchase because AI let them build the equivalent internally. Before your next renewal, ask whether the feature set you're paying for is actually buildable. The answer might surprise you.
• If you're evaluating whether AI is "working" at your organization: The question isn't whether people feel productive. The McKinsey data says 80% will say yes. The question is what the organization did with that productivity. If the answer is "nothing changed structurally," the 37% number applies to you.

Further reading

• McKinsey — The State of AI: Global Survey 2026 — primary source; methodology, full findings, high-performer breakdown
• HPCwire / AIwire — Enterprise AI is becoming a two-speed race — the high-performer vs everyone-else framing
• McKinsey — State of AI trust in 2026: Shifting to the agentic era — the governance and risk side of the same research wave
• Startup Fortune — 32% of firms now building software instead of buying — build-vs-buy angle in detail