Who Is Actually AI-Native?

Ask McKinsey's chief executive how many people the firm employs and he gives an odd answer. Bob Sternfels told Harvard Business Review that the firm's workforce is 40,000 humans and 20,000 agents. At CES in January he put the agent number closer to 25,000, and a McKinsey spokesperson confirmed that figure to Business Insider.

Eighteen months before that, the firm had about 3,000 agents.

Much of this runs through Lilli, McKinsey's own platform, trained on decades of its research, frameworks and client work. More than three quarters of the firm uses it monthly. Sternfels expects every employee to be working alongside at least one agent within eighteen months, and says about a third of McKinsey's revenue now comes from underwriting client outcomes rather than advisory fees, with the hope that it crosses a majority before his term ends.

A consulting firm publicly walking away from selling hours is a bigger deal than the agent count.

I have spent six editions mapping AI-native firms. Most of them are startups. This edition is about the possibility that the biggest AI-native services businesses in the world are the incumbents everyone assumed were the victims.

Harvey has raised more than $1.2 billion, and is reportedly in talks for $500 million more at a $15.5 billion valuation. It is the best-funded company in legal AI by a distance, and it is the number people reach for when they argue that AI is coming for professional services.

Now compare that with what the incumbents are spending.

Deloitte has committed $3 billion to generative and agentic AI through 2030. KPMG has committed $2 billion over five years, and says it is targeting $12 billion of AI-enabled revenue off the back of it. EY says it invests more than $1 billion every year in AI-first platforms, and has built 1,000 agents. Kirkland & Ellis is spending $500 million building a platform it refuses to license to anyone. Goodwin Procter has earmarked $25 million a year.

Deloitte's AI budget alone is roughly twice what the best-funded startup in the space has raised.

That comparison is not perfect. Committed budgets are not the same as capital raised, and a large firm's AI budget includes plenty of licence fees and consultants. But the order of magnitude holds, and it should bother anyone whose thesis depends on incumbents being too slow to respond.

McKinsey built Lilli and now counts agents in its headcount. Deloitte launched Zora AI with Nvidia, selling digital co-workers as a product rather than advice about them. Goodwin built its own operating system, named Regina OS after a former chair, and has 17 partners beta-testing digital twins of themselves through a platform called Cortex. Kirkland's chairman explained the build decision in one line. Off-the-shelf tools raise the floor for everyone, which is another way of saying they cannot make you better than the firm across the street.

Then look at what the Big Four are selling. Bloomberg Tax reported in March that all four are pivoting toward multi-year contracts to run their clients' back offices. That is doing the work itself, priced by outcome, on infrastructure they own.

That is the definition I have used since edition one.

I think that on capability, capital and data, the giants are further along than the startups they are supposedly losing to. What they lack is a business model that charges for output instead of input, and the stomach to shrink revenue on the way there.

The incumbents have one structural problem, and it is the same one in every firm.

An AI that halves the hours needed on a matter cuts revenue at any firm billing by the hour. Selling the saving to a client is easy. Explaining to a partnership that this year's distributions are lower because efficiency worked is not.

McKinsey is moving first, away from advisory and toward outcomes. Whether the rest of the industry follows probably depends on what happens to its revenue per partner.

And the firms are shrinking as they go. McKinsey reduced its workforce by more than a tenth between 2023 and 2025, with entry-level roles taking the worst of it. It cut around 200 technology roles as agents took over internal work. Deloitte, by contrast, grew headcount to 470,000 while revenue rose 4.8% to $70.5 billion. Same technology, opposite staffing decisions, which tells you nobody has settled on what the right shape of firm is.

Thomson Reuters found in its 2026 report that firm-wide AI use had roughly doubled year on year, to 40%. Which means 60% still have not got there. Only about a quarter of firms have moved past pilots into full deployment. More than 90% of professionals reported what the report calls an AI value gap, where the technology appears to be delivering for the organisation but not for them personally.

A report published last week by KPMG, Baker Tilly and Fieldguide found the profession splitting roughly in half between heavy and light users. Its most useful finding is that the split does not track spending. The firms capturing value are the ones that reshaped how work gets done, rather than the ones using AI to run existing processes faster.

That is the gap between buying AI and being AI-native, and most firms are still on the wrong side of it.

If you sort firms by what they actually sell rather than what they call themselves, you get four groups.

Rebuilt internally. McKinsey, Deloitte, KPMG, EY, Kirkland, Goodwin. Enormous budgets, proprietary platforms, their own data, and clients already paying them. Blocked mainly by their own pricing model.

Built from scratch. Crosby in legal, Pilot in accounting, Alan in insurance, Crescendo in customer support, and hundreds more like them. No legacy revenue to protect, so they can price by outcome from day one. Blocked by capital, licences and trust.

Bought the tools and changed nothing. The largest group by number. New tools bolted onto old processes, and a value gap to show for it.

No firm-wide use. Roughly 60% by the Thomson Reuters count. Mostly mid-sized, and the ones I keep writing about, because they are the inventory in everybody else's plan.

I believe that over the next few years the giants win the top of every professional services market, the startups take the volume work and the segments that were never economic to serve, and the middle gets bought.

Within a year, one of the Big Four reports agent headcount publicly the way McKinsey does. Once one firm turns agents into a metric, the rest have to answer for theirs.

Within a year, a major firm prices a standard engagement by outcome rather than by hour and says so in public. McKinsey is closest. The first one to do it credibly resets client expectations for everyone else.

Within two years, the AI-native startups start losing enterprise deals to incumbents on the same ground they were meant to win on, which is speed and price. Not everywhere. Enough to matter.

And the value gap closes slowly, because it is a management problem rather than a technology one, and management problems do not follow model releases.

Every services market I have looked at resolves into the same four positions, and the positions are held by different players than the narrative suggests. The disruptors are real, and so are the incumbents, who are spending more than the disruptors have raised.

The firms with something to worry about were never the giants. It is everyone in the middle who bought a licence, changed nothing, and is waiting to see what happens.

Subscribe and every edition lands in your inbox. If you are inside one of these firms and my read is wrong, reply. I read everything, and the sharpest replies shape the next edition.

And if you are thinking about selling a professional services firm, or buying one... that is exactly what I do. I run Eilla, an AI-native M&A advisory. Message me.

See you next time.