Revenue Per Employee Hit $2.7 Million at AI-Native Teams. They Are Still Hiring.

Lovable reported $400 million of annual recurring revenue in February 2026 with 146 employees. That is roughly $2.7 million of revenue per employee. The median private software company manages $141,125. The gap is real, and it is mostly explained by three structural choices and one accounting shift, not by a team that somehow works harder.

The short answer on revenue per employee at AI-native teams

AI-native software teams are running $1 million to $3 million of revenue per employee. Traditional private SaaS runs $141,125 at the median. The mechanism is self-serve distribution, a narrow product surface, and compute cost that lands in gross margin instead of payroll. The team is smaller. The cost base is not smaller in proportion.

The framing in circulation is "ten people, ten million". That works out to $1 million per head. It is a fair description of the top AI cohort's average and a poor description of the market. Bessemer's fastest ten companies average $1.13 million per employee. Everybody else is closer to $141,000.

So the interesting question is not whether the model exists. It plainly does. The question is which parts of it transfer to a company that is not growing at four times a year, and which parts are a side effect of the growth itself.

What the verified record actually shows

Very few companies publish both revenue and headcount on the same date. Most of the widely quoted ratios are two estimates from different sources divided by each other. That distinction decides how much weight any single figure can carry.

Microsoft sits at roughly $1.8 million of revenue per employee and Meta at $2.2 million, both from audited filings, per Forbes analysis of Redpoint Ventures data. Two of the largest employers in software already clear the bar that AI-native teams are praised for clearing.

Lovable is the only one of these with dated, repeated disclosure

Lovable published $100 million of ARR in July 2025, $200 million in November, $300 million in January 2026 and $400 million in February. Headcount was 45 at the first milestone and 146 at the last. Those are company claims rather than audited results, and they are at least internally consistent and repeatedly restated.

One detail is worth naming because it shows how loose these ratios get. TechCrunch publishes $2.77 million per employee. Divide $400 million by 146 and you get $2.74 million. The difference is trivial. It also means the published ratio and the published inputs do not quite reconcile, which is normal and is exactly why you should treat one decimal place as noise.

Midjourney is the number everyone quotes and nobody can check

The most repeated statistic in this whole category is Midjourney at roughly $200 million of revenue with 11 employees. Trace it and the sourcing dissolves. Headcount estimates across public sources run 11, 40, 60, 107 and about 160. Revenue estimates run $200 million, $300 million and $500 million.

Pick the friendliest pair and you get $18 million per employee. Pick the least friendly and you get $1.3 million. The company has taken no outside funding, so there is no investor reporting to constrain any of it. I would not put that figure in a board deck, and it is in most of the ones I have seen.

The median software company is nearly 20 times lower

SaaS Capital fielded its fifteenth annual survey in March 2026 across more than 1,000 private SaaS companies. The median came in at $141,125 of revenue per employee, up from $129,724 a year earlier. That is an 8.8% improvement in one year, which is healthy and is not a step change.

The band detail matters more than the headline. Companies between $1 million and $3 million of ARR run a median of $109,644 per employee. So the "ten people, ten million" model implies $1 million per head, which is 9 times what a company at that revenue scale typically achieves.

One finding cuts against the venture narrative directly. Bootstrapped companies post higher revenue per employee than equity-backed ones at every ARR level. At $5 million to $10 million of ARR that is $177,240 against $152,295. Capital efficiency is not a new discovery. It is what companies do when nobody hands them money to hire with.

Why the two numbers are not comparable

Revenue per employee is a ratio between a fast-moving numerator and a slow-moving denominator. Revenue can quadruple inside a year. Hiring runs on notice periods, visas and onboarding, so headcount lags by two or three quarters.

A company growing at four times a year therefore posts an inflated ratio for as long as the growth lasts. This is not a structural property of the company. It is a measurement artifact of the growth rate. SaaS Capital makes the reverse point in its own data: equity-backed companies grow faster despite lower revenue per employee, so a low ratio is not an indictment of raising money.

The correct comparison is against companies at the same growth rate, and almost nobody publishes it that way. When you see $2.7 million per employee, read it as evidence of a growth rate first and an operating model second.

The headcount did not disappear. It moved into cost of goods sold.

Here is the part the tiny-teams coverage keeps leaving out. Bessemer's supernova cohort averages $1.13 million of ARR per employee at roughly 25% gross margin, and the report notes those margins are often negative. Its slower cohort runs $164,000 per employee at 60% gross margin.

Traditional SaaS sits at 75% to 80%. So the fastest AI companies are converting a quarter of each revenue dollar into gross profit where a normal software company converts three quarters. What used to be a salaried engineer is now an inference bill, and it sits in a different line of the P&L. The composition of that line is covered in the breakdown of what actually sits inside an AI company's cost of goods sold.

Gross profit per employee is the honest version

Multiply each cohort's revenue per employee by its gross margin and the picture changes shape. The supernovas produce about $283,000 of gross profit per employee. The slower cohort produces about $98,000. The advantage falls from 6.9 times on revenue to 2.9 times on gross profit.

That is still a large advantage and it is a different claim. A 2.9 times edge in gross profit per head is a good business. A 6.9 times edge in revenue per head is a headline. Founders who plan hiring against the headline will underfund the company by roughly the difference.

The margin gap is closing, slowly

ICONIQ surveyed around 300 executives building AI products in April and December 2025. Its snapshot projects average AI product gross margins reaching roughly 52% in 2026, with the strongest results at companies that chose a clear point of differentiation. That is real improvement from a 25% base and it is still well below classic software.

Until that gap closes, revenue per employee flatters AI-native companies relative to SaaS by construction. The same effect running through pricing and packaging is examined in the margin trap inside AI unit economics, and the underlying cost driver in what inference costs do to gross margin.

The structure, function by function

Strip out the growth-rate effect and the accounting shift, and something genuine remains. AI-native companies allocate people differently, and the difference is concentrated in go-to-market rather than engineering.

These shares come from ICONIQ's State of Software 2025 benchmark, built on 127 software companies with Q2 2025 data, as reported in secondary analysis rather than read from the report itself. The support comparison draws on a separate scan of 22,988 go-to-market job postings in Q1 2026, which measures hiring intent and not confirmed hires. Treat the direction as reliable and the decimal places as not.

The shape is the finding. Sales share falls by eight points and post-sales share rises by roughly the same. These companies are not removing go-to-market headcount so much as moving it behind the point of purchase, because the purchase happens without a person involved.

Lovable adds around 1,500 paying customers a day with no traditional sales team, according to Forbes analysis of the company's figures. That is the actual structural claim. A product that sells itself does not need a pipeline organisation, and the org chart follows the pricing model rather than the other way round.

Support is the function that actually shrank

Support falls from 6.6% of go-to-market headcount to 2.2%, a reduction of about two thirds. This is the one place where AI substitutes for people cleanly, because deflecting a support ticket is a well-bounded task with a checkable answer.

Notice what that implies about everything else. If support is the function AI genuinely compresses, and support is 6.6% of go-to-market headcount at a traditional company, then automating it entirely removes a single-digit percentage of the org. The rest of the revenue-per-employee gap has to come from somewhere else, and mostly it comes from not building the sales organisation in the first place.

Generalists hired for slope, not for a title

Anton Osika has described Lovable's hiring filter as four traits: slope, breadth, curiosity and a bias to build. He told Business Insider that he cares more about how fast someone learns and adapts than where they are today, and that he prefers people who can do a bit of everything across design, code and product thinking over someone world-class at one thing.

Read as an operating decision rather than a hiring philosophy, this is a bet on lower coordination cost. Specialists create handoffs. Handoffs create meetings, tickets and waiting. A ten-person team of generalists can hold the whole product in a room, which is worth more than the specialist depth it gives up, up to some size.

Where that size sits is the unanswered question. Nobody has published a study of when the generalist model breaks, and every company running it is still small enough that it has not had to find out.

What happens to the model when it scales

The most useful fact about Lovable is not its revenue per employee. It is the headcount trajectory underneath it.

Run the arithmetic. At $100 million with 45 people the ratio was $2.22 million. At $400 million with 146 people it was $2.74 million. Revenue grew four times and the efficiency ratio improved by 23%. The team tripled.

The company was also advertising 70 open roles at the time of that reporting. Filling them takes headcount to 216 and, at flat revenue, drops revenue per employee to $1.85 million. The AI-native operating model as practised by its best-known example involves hiring aggressively.

Gartner's March 2026 predictions release, as reported in trade coverage rather than read directly, forecasts a wave of unicorns running $2 million of ARR per employee by 2030. Lovable cleared that in 2026. It cleared it while adding roughly two people a week.

Where this argument is weakest

Three problems, and the first one is severe enough that it should change how you read every number above.

The denominator is not honest

Revenue per employee counts full-time staff. It does not count contractors, agencies, offshore delivery partners or the outsourced functions that used to be internal. A company can improve the ratio by 30% without changing anything except the contract type on its invoices.

None of the companies here publish contractor spend. So a ratio that looks like an efficiency breakthrough may partly be a reclassification, and there is no public way to tell the two apart. That is a serious gap in a metric being used to justify hiring decisions across the industry.

The same problem runs the other way through compute. A model provider's engineers built the capability that replaces your engineers. They are still employed, just on someone else's payroll and inside your COGS line. The industry-level headcount saving is much smaller than the company-level ratio suggests.

The strongest counter-argument to everything I have written is that this has always been true and it did not stop revenue per employee from being useful. Microsoft at $1.8 million per employee has enormous contractor exposure and the figure still tells you something real about the business. Fair. It tells you less than one decimal place of precision implies.

The best available data also describes the best available companies. Bessemer studied 10 supernovas. ICONIQ draws heavily on its own portfolio, which skews to top performers by construction. These are benchmarks of the winners, presented as benchmarks of a category. The median AI-native company does not appear in any of them.

Then there is the cost that does not show up in the P&L at all. UC Berkeley researchers spent 8 months inside a 200-person company running more than 40 interviews, and found fatigue and burnout rising among the heaviest AI users as organisational expectations for speed climbed with the tooling, according to TechCrunch's reporting on the study. One engineer put it plainly: you do not work less, you work the same or more. A ratio that improves because 10 people absorb the workload of 30 is not the same achievement as a ratio that improves because the work got smaller. The gap between reported usage and reported trust in these tools is examined in the developer trust data.

What a 10-person company can actually copy

Sorted by how transferable each choice is, rather than by how often it gets written about.

The uncomfortable read of that table is that the top two rows are distribution decisions and the bottom row is luck. Very little of the advantage is about how the team works internally. It is about what the company chose to sell and to whom.

That conclusion is consistent with what happens one tier down. The economics of very small software businesses are set by the same variables, as covered in the unit economics of a one-person software business, and by the growth rate itself, which is the subject of the analysis of the fastest revenue ramp on record.

Frequently asked questions

Where to start this week

Start by recalculating your own ratio properly. Take trailing twelve month revenue, divide by full-time headcount, then divide again by the number including contractors and agencies on retainer. If those two figures differ by more than 20%, the first one is not measuring what you think it measures.

Then run the version that matters. Multiply revenue per employee by your gross margin to get gross profit per employee, and compare that against the $283,000 the fastest AI cohort produces. That is the number that pays salaries. Revenue per employee only tells you how much money passed through.

If the gap is large, look at your pricing page before you look at your org chart. Almost every company in this piece got its ratio from selling without a sales conversation, not from a clever team design.

Related reading