McKinsey's 2026 AI survey, read from a 30-person firm

McKinsey published its 2026 State of AI survey on Tuesday. Most coverage will lead with how many companies now use AI, which is close to nine in ten. Two other numbers in it deserve more attention from anyone running a smaller business.

Number one: profit impact did not move

Of the 1,719 respondents, 37% reported a positive effect on EBIT from AI. The year before, McKinsey's 2025 survey put the share reporting any EBIT impact at 39%, and most of those said it was under 5% of EBIT.

So in a year when use became almost universal, the share of firms seeing it in their profit line stayed where it was. More companies use AI. About the same share make money from it.

That fits the MIT NANDA report from last summer, which found that only a small share of AI pilots ever showed up in profit and loss. Plenty of experiments, few that become the way the business actually runs.

The near-universal usage figure deserves a comment of its own. "Uses AI" covers a lot of ground, from a company-wide rollout to one person drafting emails with a free chatbot. A high usage number next to flat profit impact suggests plenty of the second kind. That effort is not wasted. It just has not yet changed how the business runs.

Number two: the size gap on agents

The 2026 survey reports that 40% of companies with more than a billion dollars in revenue are scaling AI agents somewhere in the business. Among smaller companies, the figure is 22%.

The easy reading is that big firms have bigger budgets. That is true and not very useful if you run a firm of thirty people. A more useful question is what an agent needs before it can be trusted with a real task, and whether a small firm can supply it.

What an agent needs, and who has it

An AI agent doing real work, chasing an unpaid invoice, preparing a quote, booking a technician, needs three things beyond the model.

1. Written rules for the task. Which invoices it may chase, what tone, when to stop and hand to a person.
2. Facts it can trust. Current prices, current customer details, current terms, each one checked.
3. A gate. A defined point where a person approves before anything irreversible or customer-facing happens.

Large firms have more of the first two scattered across policy documents and systems, plus people whose job is to wire them together. That partly explains the gap.

But a small firm has one advantage that rarely gets mentioned. Its rules fit in a few documents and its decision-makers fit in one room. A thirty-person firm can write down how quotes are approved, agree it, and have it in a shared company memory in a week. A large company might take a quarter to get the same page through its committees.

That is how we run Provecta. Our rules, checked facts and past corrections sit in one written memory that our AI sessions read before they do anything, and proposals and client deliverables wait in a review queue for a person. The system is small. The effort goes into keeping it current.

The caveats

This is a self-reported survey of executives. "Scaling agents" means different things to different respondents, and a company counted as scaling may have one agent in one team. Treat the gap as a direction, not a measurement of anyone's business.

And a company memory will not close the gap on its own. Agents still need connections to your systems, someone to own them, and a budget for the hours it takes to get them right. Writing the rules down is the cheap part. It is also the easiest part to skip.

Homework for the owner

Pick the one task you would most like an agent to handle. Write the rules for it on one page, list the five facts it would need, and name the person who would approve its work. Give yourself a week, and do it with the person who does the task today, not the person who manages them. If you cannot finish that page, the agent is not your next step. The page is.

Book 20 minutes at pgco.world/book.