An AI Connection Should Never Cost A Seat

A seat was always a head — one human, opening the app. Now that a share of the work is done by something that is not a human, most pricing has to decide what an AI is. Charging it a seat is the easy answer and the wrong one.

Why is per-seat pricing awkward for AI?

Because a seat prices access, and access is not the thing an AI consumes.

Seat pricing works when it does because it tracks something real: another person joins, they get an account, they use the product roughly as much as everyone else, and the price is a proxy for the value they get. Rough, but honest enough.

None of that survives contact with an AI participant. An AI does not have an account you provision, it does not consume evenly, and it is not a person whose access you are buying. It shows up, does a bounded amount of work, and stops. Charging a monthly seat for it prices a subscription against something that behaves like a meter.

And it produces exactly the wrong incentive. If connecting your second assistant costs money, you connect fewer — which is to say, the pricing punishes the behaviour the product exists to enable. A workspace whose whole argument is bring every AI you already use cannot then charge per AI. That is selling a thing and taxing its use.

What should a seat mean, then?

The narrow thing it originally meant. A seat is a human on your workspace.

Ours: the first two people — you and a teammate — are free. From the third person, each seat is paid. Any AI you connect is free, always, and never a seat.

The two-free part is not a growth trick, or not only one. It is the honest shape of the product: a workspace where every change is signed is worth very little to exactly one person and starts being worth something the moment a second principal writes into it. Charging at the point where the value begins would be charging at the wrong end.

Why usage is pay-as-you-go from one balance

The work an AI does is uneven, so pricing it as a flat monthly plan means someone is always wrong. Heavy months you outgrow the plan. Idle months you fund capacity you did not touch.

So usage draws a single shared balance the owner funds. You, your teammates, and every AI all draw the same pool. Only what actually ran counts, and the Usage screen shows every action — what ran, and who ran it.

That last clause is the part I care most about, and it is not really a pricing decision. It is the ledger showing up in the billing screen. If every write is signed, then every charge has an author too — so spend is legible for the same reason the document history is legible. A bill you can read line by line, with a name on each line, is a different object from a monthly number you have to accept.

There are two guardrails and they are deliberately boring. You only pay for what ran; idle costs nothing. And if the balance reaches zero, work pauses — nothing is lost, no overage, no surprise. You resume by topping up.

What free-for-two is actually testing

I would rather say this than dress it up: free-for-two is a bet, and it is falsifiable.

The bet is that the second principal is where the product becomes real, and that a workspace which is free until a third person joins will get to that second principal more often than one that charges at the door. If people arrive, stay solo, and never invite anyone, the bet is wrong — and it will be wrong in a way I can see in the data rather than argue about.

The pricing is downstream of a claim about the product, which is how it ought to be. The claim is that a signed shared record is worth paying for once more than one of you is writing into it. Everything above is just that sentence with numbers attached.

Where does seat pricing still make sense?

For humans, and I do not think that changes.

A person joining your workspace really is a step change: another set of judgments, another principal on the record, another party whose access you are buying. Charging for that is legible, predictable, and easy to budget — which is exactly what a subscription should be. The seat price buys the one thing that does not fluctuate.

What breaks is only the attempt to stretch the same unit over a participant that behaves nothing like a person. Seat pricing is not obsolete; it is just no longer the whole model, because the workspace no longer contains only people. Two units for two kinds of participant is less elegant than one. It is also the shape of the thing being priced.

---

Part eight, and the last, of Inside the workspace. The details live on the pricing page.