A Markdown File Is an Employee: $15M ARR on 2–3 People

Clip via a16z. Analysis by FourWeekMBA.

Garry Tan’s line — $15 million in ARR, four months, two or three people, because “a markdown file is an employee” — is doing two jobs at once, and both are bigger than the soundbite. First: headcount is no longer the variable you scale. Second: the new unit of labor is a prompt with a file system, not a person with a seat.

The number only works under one condition, and it deserves to be stated plainly rather than admired. $15 million of ARR on three people requires the markdown to be doing closed-loop work — running the inbox, writing and shipping the code, moving the CRM — not producing decks a human then acts on. The file has to be labor, not output. If it is generating artifacts that people still have to process, the leverage evaporates and you are back to a normal small team with good tooling.

This is where the framing gets its edge. For a decade, the SaaS growth story was “software that helps your employees.” Tan is describing “software that is the employee” — and the difference is not marketing. It changes the cost structure, the org chart, and the defensibility all at once. If a markdown file can hold a role, the moat is no longer the headcount you can afford; it is the quality of the closed loops you can encode.

Enterprise is selling the identical story with a procurement wrapper and a per-seat line item. That is the tension worth naming: the incumbents are pricing agentic labor as a subscription, per seat, wrapped in compliance; Tan is pricing it as a file you can fork. Those are two very different futures for who captures the value — the vendor metering access, or the operator who owns the file.

But the risk is the exact one DHH names in his clip, and it is not a side note. If nobody reads what the markdown produces, you have not hired an employee — you have automated error and relabeled it a headcount cut. A closed loop with no reviewer is not efficiency; it is unaudited action at scale, and it fails in the direction of confident wrongness.

So the honest read of Tan’s line is that it is simultaneously the most exciting and the most dangerous claim in the batch. $15M on three people is real leverage if the loops are genuinely closed and someone competent is still in the loop to catch the failures. It is a landmine if “a markdown file is an employee” becomes a license to stop looking.

For daily structural analysis of the AI economy, subscribe to The Business Engineer.