Linear Hit $100M ARR With 160 People. Agents Now Do Half the Work.

On August 26, Linear announced a $99 million employee tender offer at a $2.5 billion valuation. The valuation is the headline everyone ran with. It is not the interesting number.

Four paragraphs down the same announcement, under a heading about agent-driven product development, Linear published this: agents are now installed across 95% of paid Linear workspaces, and the share of work those agents create has grown from 3% a year ago to 50% today. Since the start of 2026, the number of issues with a pull request attached has grown sevenfold.

Half the work in a product used by OpenAI, Cursor, Cognition and Salesforce is now created by agents. It took twelve months. The headcount did not move.

The four numbers, and the one that is derived

Linear published these itself: it passed $100 million in annual recurring revenue earlier this year, has more than 40,000 paying customers, runs 177% net revenue retention, and is cashflow positive with more cash in the bank than everything it has raised to date.

Gergely Orosz added the number Linear left out. 160 employees.

Divide one by the other and you get roughly $625,000 of revenue per employee. Hold onto that figure, because it is about to break a definition.

Note the honest caveat first: Linear did not publish headcount, and its own announcement says it has more than 30 open roles. So $625K is a snapshot of a company that is deliberately hiring, not a steady state. It is still five to ten times what most scaling software companies run.

The company that refused the money

The structure of the deal matters more than the valuation. Linear did not raise a primary round. It ran a $99 million tender so employees and former employees could sell vested equity, with Accel and 01A joined by Salesforce Ventures and S32. The previous tender was at $1.25 billion alongside the Series C, so this marks a doubling in about a year with no new capital taken in.

Linear's own explanation is the most quotable paragraph in the announcement: "Fundraising can become a habit. A company raises a round, then starts preparing for the next." Then it declines to.

Terrence Rohan read the same event from the other side of the table, and his point is the sharper one.

Capital is not scarce right now. It is badly allocated. Which means the scarce thing is not money, it is a company shape that does not need any.

The definition of AI-native excludes Linear

On August 12, Nikhil Krishnan asked his followers a straightforward question: what actually makes a company AI-native, and the more specific the better. He was writing a piece cataloguing how many different ways he had heard the term used. That is the state of the phrase. It needed crowdsourcing.

The sharpest answer in the thread came from Evan Armstrong, and it is the only version anyone has put numbers on.

Now run Linear against it. Revenue per employee lands around $625,000, under the $1 million bar. The company is seven years old, so a twelve-month test does not apply. And it is conventionally structured, with engineering, product, design, sales and support functions, so the third test is a no.

Three for three, and the conclusion is not the one the arithmetic suggests.

Linear is not a laggard. It runs 177% net revenue retention, 40,000 paying customers, is cashflow positive with more money banked than it has ever raised, and does $625,000 per head, which is roughly five to ten times what a typical scaling software company manages. Half the industry is actively trying to copy its shape. If your definition of AI-native excludes that company, the definition is what is broken.

And it is worth being precise about what the definition is: one reply, in one thread, written in a few minutes by someone doing the category a favour by arguing in numbers at all. It was never meant to be a standard. The interesting thing is that even the best available attempt fails to catch the obvious case, which tells you the term is measuring the wrong quantity.

The test worth stealing

The mismatch is the useful part, because it tells you the $1 million bar describes an aspiration rather than the current frontier. If Linear sits under it, your mid-market company is not behind, it is in good company.

So use the number that does catch Linear, and make it the fourth test: what share of the work in your system was created by an agent?

It is a better test than revenue per employee for three reasons. It moves inside a quarter instead of over years. It is not confounded by pricing power, industry margin structure or an installed base that predates AI. And unlike a ratio you can only report, it points directly at the intervention, because the way you raise it is by giving agents somewhere to file work.

Karri Saarinen's own reaction to the number is the tell.

Read that last clause again, because it is the architecture and not the sentiment. The same system where people plan and build. The agents did not get their own tool, their own dashboard or their own review queue. They got write access to the place the work already lived. That is why the share went from 3% to 50% instead of stalling at a pilot.

What this means if you are not Linear

Measure agent share of work this quarter. Pick your system of record, whichever one holds the unit of work your company actually ships, and count what fraction of new items were created by something other than a person. Most companies will find a number under 2%. That is the baseline, and it is the only AI metric on this list your board will understand without a preamble.

Give agents write access to the system you already have. The reason Linear's number moved is that agents file into the same place humans do. A separate agent console produces a demo. The record you already run produces a share of work.

Do not read 160 people as an instruction to stop hiring. Linear has 30-plus open roles and says outright that scope expanded and quality at scale means growing the team. The lesson is not a smaller company. It is that output and headcount stopped moving together, which is a different and better claim.

Stop using revenue per employee as a qualifier. It is a destination, not an entry requirement. If you use it to decide whether you are allowed to start, you will never start, because the ratio only moves after the work is done.

The valuation will be forgotten by Christmas. The 3% to 50% will not, because it is the first public number that puts a date on how fast the agent share of real work moves once you let it into the system of record.

Twelve months. Same team.