The empty chair: why per-seat software is dying and what replaces it

In February the market wiped roughly 285 billion dollars off software companies in a matter of days, and most of the commentary called it what commentary always calls a sharp fall: a bubble, an AI-hype correction, a sector getting ahead of itself. That reading is comfortable and it is wrong. What happened in February was not a valuation getting punctured. It was an entire industry discovering, all at once, that the unit it had priced itself on for two decades is dissolving underneath it.

The unit is the seat. Software as a service was built on a beautifully simple idea: charge per human who uses it, per seat per month, and grow revenue by growing the number of people logging in. For twenty years that model printed money, because the number of humans doing knowledge work only ever went up. The number is now going down, because agents are starting to do the work the humans in those seats used to do, and the moment the seat empties, the entire pricing model of the software industry stops making sense. The selloff was the market working that out in real time.

The model that pays you to be worse

Here is the structural break, and it is worth stating as bluntly as it deserves. Per-seat pricing, in an age of capable agents, pays the software vendor to under-deliver.

Follow the incentive. If I sell you a tool at a price per human seat, and my tool gets an AI feature good enough that you now need three people where you needed ten, I have just cut my own revenue by seventy percent by making my product better. The better my agent works, the fewer seats you buy, the less you pay me. I am financially punished for the quality of my own software. No industry survives long with its incentives pointed that precisely backwards, and the smart money in February simply noticed that a great many software companies were still standing on exactly that fault line.

This is not a new disease, it is just a fast-acting strain of an old one. The billable hour has always had the same sickness: pay a lawyer by the hour and you have paid them to be slow. Per-seat software had the sickness in dormant form for years, and nobody minded because the number of seats kept rising anyway. Agents woke it up. Once your customer can deliver the same outcome with fewer people, a price tied to the number of people is a price tied to the wrong thing, and it is going to fall.

The steam engine and the meter

To see where this goes, look at the last time an industry priced on the wrong unit and got repriced around the right one.

In the early industrial age, if you ran a factory, you bought a steam engine. Your capacity, and your costs, scaled with the machines you installed. You paid for horsepower you owned, sitting in your building, whether it was turning or not. The unit of the industrial economy was the installed machine. Then the electric utility arrived, and it did something quietly revolutionary: it let you stop buying engines and start buying work, metered by the unit actually consumed. You no longer paid for capacity sitting idle. You paid for the kilowatt-hours you used to make the thing you sold.

That shift did not just change how factories paid their power bills. It repriced the entire industrial economy around output rather than installed capacity, and it redrew the map of who won. The firms that clung to owning and running their own engines, priced by the machine, were slowly undercut by the ones who bought metered power and priced their own goods around what they produced.

Per-seat software is the steam engine. It charges for installed capacity, a seat, a login, a human-shaped slot, whether or not real work flows through it. Outcome pricing is the meter. It charges for the work the software actually does: the ticket resolved, the document processed, the case closed. We are living through the exact same transition, compressed from decades into a couple of years, and February was the moment the market started pricing the steam-engine companies like steam-engine companies.

What the meter actually looks like

The repricing is already visible, and the numbers are stark. Per-seat pricing has fallen from around a fifth of the SaaS market to about a seventh in twelve months, which is an enormous move for something as sticky as a pricing model. Intercom charges just under a dollar for each customer conversation its AI actually resolves. HubSpot cut its own per-resolution price to fifty cents in April. These companies are not charging for who has access. They are charging for work completed, the software equivalent of the kilowatt-hour, and their incentive is now aligned with the buyer for the first time: they make more money when their agent resolves more, not when the customer hires more humans to sit in more seats.

I want to be honest that the pure meter has not won outright, and the reason is instructive. The dominant model right now is a hybrid: a base fee that anchors access, with a metered layer on top that prices the actual AI work, and it is already the single most common shape in the market. That is not a fudge, it is the sensible landing spot, and it mirrors your electricity bill exactly. You pay a standing charge for the connection and then a unit rate for what you use. Software is rediscovering the utility bill, because the utility bill is what pricing looks like when you are selling metered work rather than owned capacity.

What this means if you are building software now

If you are a founder shipping software in 2026 and you are still pricing purely per seat, you have quietly priced yourself to root against your own product. Every improvement to your AI is a cut to your revenue. That is not a position you want to defend to an investor who has read the February tape, and it is not a position that survives the next two years.

The move is to find the unit of value your software actually delivers and charge for that. Not the login. The outcome. The resolved ticket, the completed reconciliation, the closed case, the shipped document. Price the work, not the chair. It is harder, because you have to be able to measure the outcome and stand behind it, and standing behind an outcome is a scarier promise than renting a seat. But it is the only pricing that aligns you with your customer in a world where your software, not their staff, does the work, and it is where the whole industry is visibly heading.

This is exactly the conversation we have with the founders we build with, from the first week. When your product is agentic, the question is not "how many seats can we sell." It is "what is the unit of work we can charge for, measure honestly, and be proud to be paid on." Get that unit right and the incentive to make your product better is the same as the incentive to make more money, which is the only healthy place for a software business to stand.

The chair was never the point

The seat was always a proxy. Nobody ever wanted a login for its own sake. They wanted the work the person in that seat would do, and the seat was simply the closest thing we could meter in an era when software needed a human to drive it. Agents remove the human from the loop, and the moment they do, the proxy falls away and leaves the real thing exposed: you were always buying work, and now you can pay for it directly.

The 285 billion dollars was not a bubble bursting. It was the sound of an industry repricing from the chair to the work, from the engine to the meter, and doing it in a fortnight instead of a generation. The companies that read it as a crash will spend the next two years defending a seat nobody sits in. The ones who read it as a repricing will do the harder, better thing, and charge for the work their software actually does. The empty chair is not a problem to be managed. It is the whole signal, and it is pointing at the answer.

Louis O'Connell-Bristow is a co-founder of Moonlabs, the operator-led AI incubator and academy, and previously built the home.co.uk, Homemove and homedata.co.uk stack. Moonlabs builds and funds AI-first companies and helps founders price for the work their software does, not the seats it fills. Site: moonlab.ventures.