Information in this post reflects publicly available sources as of June 22, 2026.
A code editor called Cursor reportedly crossed two billion dollars in annual recurring revenue this year. The team behind it is somewhere around fifty people. That works out to roughly forty million dollars of revenue for every person on the payroll, a figure that simply did not exist in software before the last couple of years.
Cursor is not a fluke. Midjourney is reported to run on a team of about forty and bring in something like half a billion dollars a year, having raised no outside money at all. Perplexity reportedly served tens of millions of users with fewer than forty employees. A leaderboard now exists for the sole purpose of tracking companies that hit enormous revenue with fewer than fifty people, and it is filling up.
Something is happening to the shape of the technology company. The question worth asking is what exactly, because the headline numbers hide as much as they reveal.
The pattern people noticed
The metric that captures the shift is revenue per employee. For a healthy traditional software company, somewhere around two hundred thousand dollars per employee has long counted as solid. The best AI-native firms are reporting numbers five to twenty times that. One widely cited analysis pegged the top AI-native startups at an average of roughly three and a half million dollars per employee.
This is not the same story as "AI makes workers more productive," though that is part of it. The more striking claim is that headcount has stopped scaling with growth. A traditional company serving twice as many customers needed more support reps, more salespeople, more operations staff. Each new cohort of users pulled in a new cohort of employees. The AI-native firm breaks that link: serving the next million users is closer to a compute cost than a hiring decision.
The clearest academic treatment of this comes from a June 2026 working paper by Hyunjin Kim and Rembrand Koning, who studied Y Combinator startups and the broader universe of venture-backed firms. Comparing AI startups to non-AI startups in the same industry and founding cohort, they found AI-native firms roughly twenty-five percent smaller, with flatter hierarchies and a higher share of engineers, yet carrying comparable valuations. Smaller teams, same implied worth, much higher value created per head.
Two ways to be an "AI company"
The useful idea in that paper is a distinction that cuts through a lot of loose talk. There are two quite different ways a firm can be built around AI, and they have very different organizational consequences.
The first is the process channel: the firm's own employees use AI tools to work faster. Engineers code with Cursor or GitHub Copilot, support staff lean on a model to draft replies, analysts let a chatbot do first-pass research. This is the version most people picture, and it is real, but on its own it mostly makes an existing organization a bit more efficient. It does not fundamentally change what the firm is.
The second is the product channel: AI is embedded into what the firm sells, so the product itself performs work that used to require a human team. This is the one that reshapes the org chart. When the product generates the slide deck, runs the accounts receivable, or triages the call, the firm no longer needs to staff and coordinate the people who used to do that. Productive capability moves out of the workforce and into the thing being sold.
In the YC data, about two-thirds of the firms tagged as AI-native were embedding AI directly into their products this way. The split matters because only one of these channels actually explains the tiny teams. When the researchers tested both, internal tool use did not predict smaller firms. Putting AI into the product did.
What "small team" actually buys
The economics become clearer with a concrete contrast. A traditional services business that delivered something like tutoring, collections, or customer support scaled by adding people. Twice the customers meant roughly twice the staff, plus the managers to coordinate them. Revenue and headcount climbed together.
The AI-native version of that same business does the delivery inside the product. The Kim and Koning paper found the gap was largest exactly where you would expect: in services businesses like therapy, tutoring, and telemedicine, where the old model meant hiring people to do the work. Their illustrative pairs are stark. An AI-native exam-prep platform with seven employees sat next to a human-tutor platform with over nine hundred. An AI psychiatry platform with twenty-eight people sat next to a therapist-network firm with over four hundred.
Whether those pairs are typical or cherry-picked is a fair question, and the authors are upfront that they are extreme cases. But the direction matches what the market is rewarding. Investors increasingly treat a lean team as a signal of quality rather than a limitation, and the phrase doing the rounds is that the badge of honor has shifted from dollars raised to dollars earned.
Why the reactions split
This is where the story stops being a simple celebration, because different groups are looking at the same facts and drawing opposite conclusions.
Founders and investors mostly see opportunity. If a handful of people can build a company that previously took hundreds, the cost of starting up collapses and far more things become worth trying. Sam Altman's much-repeated prediction of a one-person billion-dollar company lives here. The optimistic version is not that there will be fewer jobs but that there will be vastly more companies.
Labor-market watchers see something more uncomfortable. The task-level research on AI has generally found that it helps less experienced workers the most, narrowing the gap with experts. But the firm-level picture runs the other way. AI-native firms in the YC data employed fewer entry-level workers and more senior ones, and skewed toward engineers from elite institutions concentrated in a few cities. A separate line of work has documented a measurable decline in early-career employment in AI-exposed occupations. If the firm of the future needs a small number of very senior people, the bottom rung of the ladder gets harder to reach.
These two readings are not actually in conflict. Each AI-native firm being smaller does not have to mean fewer jobs overall, as long as enough new firms are created to make up the difference. In the YC data, AI-native firms employed about half as many people as their peers after three years, which implies entry would need to roughly double to break even on headcount. First-deal counts for AI startups did surge over the period. Whether that entry boom is large and durable enough to absorb the displaced work is the open question, and nobody has a confident answer yet.
The skeptic's read on the numbers
There is also a quieter case that the headline figures are partly an artifact of how they are counted, and it deserves a hearing rather than a dismissal.
Most of these revenue figures are annual recurring revenue, which takes one month's revenue and multiplies by twelve. For a young company running aggressive promotions and free pilots, that can flatter the picture, since a three-month trial someone forgets to cancel still feeds the run rate. Different firms also report on different bases. Some count gross reseller spend that flows through them, others report closer to net, and the two can differ by a wide margin while wearing the same label.
Then there is the circularity in parts of the AI economy, where large model providers, their cloud backers, and their chip suppliers are investing in and buying from one another in ways that inflate reported activity above the underlying end-customer demand. None of this makes the lean-team phenomenon fake. Cursor and Midjourney are real businesses with real paying customers. But it does mean the most jaw-dropping per-employee numbers should be read as directional rather than precise, and the gap between the best case and the median case is large.
It is also worth separating the model labs from everyone else. Companies like Anthropic and OpenAI now run at enormous revenue scale, but they are not lean in the relevant sense and they sit on top of vast infrastructure spending. The lean-team story is really about the application-layer firms built on top of those models, which is where the striking revenue-per-employee figures actually come from.
Where things stand
The honest summary is that the structural shift is real but the magnitude is contested. Firms genuinely are being built with far smaller teams than the same business would have required five years ago, and the mechanism is reasonably well understood: when AI is embedded in the product, the product does work that used to live in the workforce. That is a different and more durable claim than "everyone uses ChatGPT now."
What remains genuinely unsettled is whether these tiny, flat, senior organizations stay that way as they age, or whether the old pressures of coordination and management reassert themselves once a product has millions of users and many edge cases to handle. The data so far only covers firms a few years old. It is also unclear how much of the reported efficiency survives a recession, a tightening of promotional spending, and a more skeptical eye on ARR.
Worth watching from here: whether revenue-per-employee figures hold up as these companies face their first real churn, whether the entry boom in AI startups is large enough to offset smaller individual headcounts, and whether the entry-level hiring squeeze turns out to be a transition cost or a permanent feature of the new firm.
Summary
The lean AI firm is not mainly a story about workers being more productive with better tools. It is a story about where productive capability lives. When AI moves from the employees' workflow into the product itself, the firm can deliver knowledge work without building the knowledge workforce, and that single shift explains the smaller teams, the flatter org charts, and the startling revenue-per-employee numbers better than any claim about individual productivity does. The figures are noisier than the headlines suggest and the labor consequences are genuinely uncertain, but the underlying change in what a company has to be is real, and it is the part most worth understanding.
This is a standalone post. Future posts covering AI news and releases will appear under the In Focus label (link to be added once a series index page exists).