The seat was a unit of furniture, not a unit of work

A desk used to be a measurement instrument. You could not weigh the work a person did, so you weighed the chair they sat in. One chair, one head, one salary, one line on the floor plan, and when software arrived, it inherited the chair. It charged per seat because the seat was the only thing anyone had ever learned to count.

That worked for as long as a seat and a worker were the same object. They aren’t anymore.

Here is the line the rest of this post defends: the seat was a unit of furniture, not a unit of work, and the moment work stops needing a place to sit, the seat stops measuring anything. I want to take that apart slowly, because the per-seat habit is so old it feels like physics. It isn’t physics. It’s a leftover from when the floor was the only ledger we had.

Where the seat came from, and why it was clever

Start with the thing it got right, because it got something right for a century.

Before software, before SaaS, before any of this, a company was a building full of desks. If you wanted to know how much capacity you had, you counted desks. If you wanted to know what it cost you, you counted salaries, one per desk. Headcount was the metric for everything: capacity, cost, status, growth. “We’re hiring” and “we’re growing” were the same sentence because a new desk and a new unit of output were, physically, the same event.

The seat was a clever proxy precisely because it was lazy in the right way. You couldn’t measure the output of a knowledge worker, there was no meter on a lawyer’s judgment or a designer’s eye. So you measured the one thing you could see: the chair was full, the salary was paid, the person was present. Presence stood in for work. It was wrong in a hundred small ways, the person at the desk who produced nothing, the one who produced everything from a café, but it was available, and an available proxy beats a perfect one you can’t compute.

So when software companies needed a price, they reached for the proxy already sitting there. One login per chair. The chair was full anyway; charge for the login that fills it. It wasn’t a pricing theory. It was a habit wearing a price tag.

The naive read on per-seat pricing is that it measures usage. It never did. It measured chairs. And it measured chairs because chairs were countable and work was not.

Why the proxy held for forty years of software

The seat survived the move from buildings to browsers, and it’s worth being honest about why, because the reason it held is the same reason it’s about to break.

It held because software, for forty years, was furniture too. A tool was a place you went and sat down, a box you opened, looked at, typed into, and closed. The spreadsheet didn’t do anything until you sat in front of it. The CRM was an empty room you furnished by hand. The dashboard was a window you had to stand at to see through. Every one of these tools shared the deep property of a chair: it was occupied, not active. It waited for a human to come fill it.

And if a tool waits to be occupied, the seat is a fair price. One human can only occupy one tool at a time. The login maps cleanly to the chair maps cleanly to the person. The proxy holds because the whole stack is built out of rooms you walk into. You’re not paying for work; you’re paying for the right to walk in and do the work yourself. The software was never the worker. It was the desk the worker sat at, rented by the month.

This is the part everyone skips when they argue about pricing. They argue about how much a seat should cost, per-user tiers, volume discounts, enterprise floors, and never ask what the seat is measuring. It’s measuring occupancy. It works as long as the only thing that can occupy the tool is a human who has to be present to make it do anything.

The bottleneck never disappears. It just waits to be renamed.

The thing that breaks it: work without a chair

Now put an agent in the room. Not a smarter tool you open, a worker that doesn’t sit.

An agent triages the overnight inbox while the office is dark and no chair is warm. It watches a contract for the renewal date that’s about to bite, with no human standing at a dashboard. It drafts the follow-up, reconciles the numbers, opens the pull request, chases the thing that fell through a handoff, and it does all of this without occupying anything. There is no chair to count. There is no login that maps to a head. The work happened, and the furniture the seat was built to measure simply wasn’t there.

This is where the proxy doesn’t just get inaccurate. It gets undefined.

Ask the naive question: how many seats is an agent? One? It can run a hundred tasks at once across a hundred corners of your business, that’s not one seat’s worth of occupancy. A hundred? It might do nothing for six hours and then deliver the one briefing that saves a customer, that’s not a hundred seats either. The honest answer is that the question has no answer, because the agent never sat down. You’re trying to count chairs in a room with no chairs. The unit isn’t wrong by some factor. It’s measuring a thing that isn’t in the building.

And the failure compounds, because the seat’s logic runs the other way too. Per-seat pricing punishes exactly the move that agents make valuable. The instinct of every per-seat buyer is minimize seats, fewer logins, fewer licenses, fewer humans touching the tool. But the whole promise of an agent is to do more of the company’s work, in more corners, more of the time. A unit that rewards using the system less is at war with a worker whose value is using it more. The proxy doesn’t just fail to measure agent work. It actively discourages it.

So the question stops being “how many seats does the agent need” and becomes the only question that survives: what did it actually do?

The unit that’s left when the chair is gone

Take away the chair and only one countable thing remains: the outcome.

Not the login, not the session, not the hour, those are all just shadows the chair used to cast. What’s left when an agent finishes is a result: the renewal caught before it lapsed, the lead created and persisted, the report composed from the company’s own knowledge, the bug filed, the deal moved forward. These are countable in a way a seat never made them countable, because for the first time the work itself leaves a trace you can point at. The agent did a thing; the thing exists; you can name it.

The naive worry here is that outcomes are too fuzzy to price, that “value” is subjective and you’ll never agree on what one is worth. But that worry is borrowed from the old world, where you couldn’t see the work, only the chair. When the work is done by a system that records what it did, the outcome stops being a vibe and becomes an entry. Suppose an agent files a hundred reconciled invoices, or surfaces a date that would have cost you a customer, or drafts the follow-up that recovers a deal, those aren’t feelings. They’re events with edges. The thing that made outcome-pricing impossible was never that outcomes are vague. It was that the worker was a human you could only watch by counting their chair.

This is the inversion the seat hid for forty years. We never priced by outcome because we couldn’t see the outcome, so we priced the chair and pretended the chair was the work. The agent doesn’t make outcome-pricing a nice idea. It makes it the only honest one, because the agent is the first worker whose output is legible enough to be the unit.

And notice what this fixes about the war from the last section. When the unit is the outcome, doing more work is doing more of the thing you pay for, the buyer and the system finally want the same thing. Use it more, get more done, the value and the bill move together. The seat made you ration the tool. The outcome makes you want to feed it.

The turn: stop paying for where your people sit

Here’s the part that isn’t about pricing pages.

For your whole career, the price of your tools has been quietly teaching you a lie: that work is a function of how many people you seat in front of a screen. Every per-seat invoice reinforced it. More work, more seats; more seats, more heads; more heads, more building. You learned to grow by adding chairs because the unit told you that chairs were the thing that did the work. They weren’t. They were just the only thing anyone had figured out how to count.

When the unit becomes the outcome, the whole shape of a company is free to change. You stop asking “how many people do we need to seat to do this” and start asking “what do we need done, and what’s the smallest, fastest thing that can do it”, and the answer stops being a hire and a desk and a license. A five-person company can run the workload of a fifty-person one not because the five are heroic, but because the work no longer requires a chair for every unit of it. The seat was always a tax you paid on the assumption that work and presence were the same. They were never the same. The desk just made them look it.

That’s the real reframe under a boring word like pricing. The unit you pay in is the unit you organize around. Pay per seat and you build a company of seats. Pay per outcome and you build a company of outcomes, and one of those is a building, and the other is just the work.

---

That’s what we’re building toward at Apollo, a system priced for what gets done, not for how many people you sit in front of it, because the worker that does the work doesn’t sit down anymore. The desk was a fine ledger for a hundred years. It’s just that the thing it was measuring finally got up and walked away from it.