Does per-seat pricing survive AI agents? What the spend data says in October 2026
Per-seat pricing is not dying: seat contracts were 65-75% of SaaS spend in April 2026, consumption 4-6%. Where AI credits creep in and what to check.
On this page
Per-seat pricing is surviving AI agents, at least so far. Ramp's analysis of its card and bill-pay data, published in April 2026, put seat-based contracts at 65-75% of SaaS spend across more than 200 vendors, flat platform subscriptions at 20-30% and consumption-based spend at 4-6%, with those shares barely moving in twelve months. What is changing is the layer on top: AI credits, usage packs and agent add-ons sold beside the seat, not instead of it.
This post looks at what the data shows, where credits are appearing in the tools small teams actually buy, and how to read a pricing page in 2026 so an agent does not quietly double your bill.
The argument that seats are dead
The case against per-seat pricing goes like this. A seat is a proxy for value: more people using the tool, more value delivered, more money paid. An agent breaks the proxy. If one person with an agent does the work three people used to do, the vendor sells one seat instead of three, and revenue falls even as the product becomes more useful. So vendors will move to pricing on usage, or on outcomes, and the seat will fade.
The logic is sound. The timing is the problem. Vendors do not reprice on a theory; they reprice when buyers accept the new model, and buyers so far have not.
What Ramp's spend data shows
Ramp sees the invoices and card charges of its customers, which makes it a useful check on the narrative. Its April 2026 piece, titled "The death of SaaS has been greatly exaggerated", reports:
| Pricing model | Share of SaaS spend (Ramp panel) | Change over 12 months |
|---|---|---|
| Seat-based contracts | 65-75% | Barely moved |
| Flat platform subscriptions | 20-30% | Barely moved |
| Consumption-based | 4-6% | Barely moved |
Two vendor examples from the same analysis stand out. Adobe has sold AI credits for two years, and 99% of its bills in Ramp's data were still seat-based. HubSpot has rolled out credits for its Breeze AI features, and its spend was described as almost entirely seat-based too. Coatue's companion chart, published the same week, put the consumption-based share of traditional SaaS spend at 3.8%, a slightly lower figure than Ramp's range but the same picture.
Ramp's own conclusion was that the "death of SaaS" narrative reflects where product leaders expect the industry to go, more than how buyers are behaving today.
Where credits are creeping in
The seat is not being replaced; it is being supplemented. The pattern across the tools we checked on 27 October 2026 is a per-seat price plus a credit pool, with more credits on the dearer plan and a top-up price for the rest.
| Tool | Seat price (yearly billing) | AI on top, as of October 2026 |
|---|---|---|
| Notion | Plus $10, Business $20 per member | Custom agents free to try, then $10 per 1,000 monthly credits; Workers (beta) also require credits |
| Linear | Basic $10, Business $16 per user | Coding sessions and Loops are marked as requiring AI credits; the pricing page gives no credit amounts or prices |
| monday.com | Basic $9, Standard $12, Pro $19 per seat, 3-seat minimum | 1,000 / 2,000 / 3,000 credits a month included; extra credits at $10 per 1,000 |
| Asana | Starter $10.99, Advanced $24.99 per user | 5 AI requests per user a month included (capped per account); packs at $0.50 a request, pay as you go $0.60 |
| ClickUp | Core $9, Business $17 per user | 1,000 / 1,500 credits per user a month pooled; add-on credits $10 per 10,000 |
| Atrix | Builder $39 for 5 seats, Studio $149 for 15 | 100 / 500 agent runs a month included; chat with the agent does not use a run; no top-up packs |
Three things to notice. First, every vendor in the table still quotes a per-seat (or per-pack) price as the headline; the credits are the second line. Second, the unit is not comparable across vendors. A Notion credit, a monday.com credit and an Asana request are different sizes, and only Asana publishes a per-request price in plain dollars. Third, the included allowance is often small enough that an active team will buy more, which is where the consumption share grows without the seat going anywhere.
Why seats are sticky for buyers, not only vendors
It is easy to assume per-seat pricing persists because vendors like it. Buyers like it too, for reasons that get stronger as agents arrive:
- Budgets are set annually. A finance team can approve twelve seats at $16. It cannot approve "whatever the agent uses". Consumption pricing moves the risk to the buyer, and most small companies have no one whose job is to watch a meter.
- Overruns are real. Secondhand reports of consumption-based AI bills running well over budget are common in 2026; we have not verified a specific figure, but the pattern matches what we see in our own cloud bills.
- Procurement is simpler. One price per person per month is a line in a spreadsheet. Credits need a definition, a monitor and a policy for what happens when they run out.
- Outcome pricing needs an agreed outcome. "Pay per resolved ticket" works for support. There is no clean unit for "planned the sprint" or "wrote the PRD", so general work tools have nowhere to attach an outcome price.
Where the seat genuinely breaks
Seats do fail in two situations, and both are worth watching.
Agents that need their own seat. Some tools now count an agent as a user. If your agent triages issues through the same login a person would use, you may be paying a human seat price for software. Check whether service accounts, bots and API tokens take a seat. In Atrix they do not: suspended members and service accounts do not consume one, and the MCP server authenticates with OAuth tokens scoped to the person who created them.
Headcount that falls while usage rises. This is the scenario the "seats are dead" argument is built on, and in a team of 30 it is plausible. In a team of five it rarely happens; the founder is still the founder, the two engineers are still two engineers. The seat count in a small company is usually the company.
How to read a pricing page in 2026
- Find the second line. Under the seat price, look for credits, requests, runs, actions or tokens. That is the variable part of the bill.
- Convert the allowance to work. Ask the vendor what one credit buys. If the answer is "it depends on the model", assume the allowance is small.
- Find the top-up price and whether it can be capped. Asana lets an admin set a spending cap on pay-as-you-go requests; not every vendor does.
- Check who or what takes a seat. People, guests, bots, agents, API tokens. Count the ones you will actually have.
- Price the year at two headcounts (now and in twelve months) and two usage levels (light and heavy). The spread between the four numbers is your real exposure.
- Prefer a price you can explain to your accountant in one sentence. For a small team, that is worth more than theoretical savings.
The flat vs per-seat pricing entry in our glossary covers the base model; this post is about what sits on top of it.
Where Atrix stands
We are not outcome-priced, and we do not claim to be. Atrix sells seat packs: Builder is $39 a month for 5 seats and 100 agent runs, Studio is $149 for 15 seats and 500 runs, and yearly billing is 20% off ($374 and $1,430). A run is one job for the AI agent, such as a Develop run that turns a brief into requirements, architecture and planned issues, or a Market run. Chatting with the agent does not use a run.
That is a seat model with a bounded usage allowance inside it, which is the same shape as most of the table above. The differences are that we do not sell top-up packs (if you run out, you wait for the month to reset or move up a plan), the pack price covers the whole workspace (work, docs, chat, calendar, CRM, hiring, people, time tracking) rather than one product, and there is no free plan: every workspace starts a 7-day trial with a card on file. Founding teams that subscribe before 31 December 2026 keep unlimited seats, which removes the seat question for them entirely. The details are on the pricing page.
The honest limit of our model is the same as anyone else's: a team of eight pays for Studio's fifteen seats. The break-even table in our earlier post shows exactly where that hurts.
FAQ
Is per-seat pricing dead for SaaS?
No. As of April 2026, Ramp's spend data showed seat-based contracts at 65-75% of SaaS spend, with consumption-based spend at 4-6% and both shares nearly flat year on year. Seats are being supplemented by AI credits, not replaced.
What is outcome-based pricing in SaaS?
Charging for a result (a resolved ticket, a booked meeting, a qualified lead) instead of for access. It works where the result is countable and attributable to the software. General work tools rarely have such a unit, which is why most of them sell credits instead.
How do AI credits work on top of a seat price?
The plan includes a monthly allowance of credits or requests. When it runs out, you buy more at a published rate, such as $10 per 1,000 monthly credits for Notion's custom agents or $0.50 per Asana AI request in prepaid packs, as of October 2026. Some vendors let admins cap the spend; check before you turn it on.
Does an AI agent count as a seat?
It depends on the vendor. Some count bot or service accounts as users; others exempt them. Ask specifically whether API tokens, service accounts and agents consume a seat. In Atrix, service accounts and suspended members do not.
Should a small team prefer consumption pricing?
Usually not. Consumption pricing moves budget risk onto the buyer, and a team of five has nobody watching the meter. A seat price with a bounded AI allowance is easier to budget. The exception is a tool you use rarely, where paying per use beats paying for an idle seat.
Is Atrix outcome-priced?
No. Atrix sells seat packs (5 or 15 seats) with a fixed number of agent runs included (100 or 500 a month). There are no top-up packs and no free plan.
Next step
Compare the base models in the flat vs per-seat glossary entry, then price your own team at two headcounts on the Atrix pricing page.
Sources
Ramp, The death of SaaS has been greatly exaggerated (23 April 2026); Coatue, Chart of the Day with Ramp (23 April 2026). Prices checked on each vendor's pricing page on 27 October 2026: Notion, Linear, monday.com, Asana, ClickUp.