the end of per-seat

AI agents are breaking down the per-seat pricing model of software, and vendors are replacing it at speed with prices per use or per result. For you as a buyer that means something uncomfortable: fewer users on the invoice does not necessarily mean lower costs, because the meter is moving from the number of people to what the software does. The best-known example comes from SaaS community SaaStr, which cut 80 percent of its Salesforce seats and watched the bill rise by 83 percent. Anyone renewing a software contract in 2026 should therefore read the usage terms first: credits, caps and the definition of “resolved”.

80% fewer seats, an 83% higher bill

Jason Lemkin, founder of SaaS community SaaStr, published his own Salesforce invoice in April 2026. Last year: $12,000 per year for over ten human users. Now: around $22,000 per year for two human seats and one API seat. An increase of 83 percent, with 80 percent fewer people (SaaStr, 2026). The explanation sits in the twenty-plus AI agents the company has connected to Salesforce. By Lemkin’s account they use the platform roughly a hundred times as intensively as the humans ever did, and that usage is billed as consumption.

One caveat belongs here: this is a case of one company with three people and twenty-plus agents, not a market average. The mechanism behind it is general, though. In a follow-up post Lemkin sums it up himself: seats crash to consumption, not to zero (SaaStr, 2026). Automating saves you nothing as long as the vendor simply moves the meter.

what is ai changing about software pricing?

Software vendors are shifting from a price per user to a price per use or per result, and in most cases the new meter comes on top of the existing subscription. Bain & Company analyzed more than thirty established software vendors and found that about 65 percent run a hybrid model: an AI usage meter next to the existing seat price (Bain & Company, 2025). A survey of 240 software and AI companies shows the same movement: per-seat as the primary model fell from 21 to 15 percent in a year, while hybrid models rose from 27 to 41 percent (Growth Unhinged, 2025).

The pace is high. The PricingSaaS 500 Index counted more than 1,800 price changes at the five hundred largest SaaS and AI companies in 2025, an average of 3.6 per company. The number of companies with a credit model grew 126 percent that same year, from 35 to 79 (Growth Unhinged, 2026).

What does that mean concretely? Intercom charges $0.99 per resolved conversation for its AI agent Fin (Macha, 2026). Zendesk charges $1.50 to $2.00 per automated resolution, on top of seat prices of $55 to $169 per employee per month (Richpanel, 2026). HubSpot switched on April 14, 2026 to $0.50 per resolved conversation for its Breeze Customer Agent, with the message that you only pay when the task is complete (HubSpot, 2026). And Salesforce runs three pricing models for Agentforce at once: $2 per conversation, Flex Credits at about $0.10 per action, and since late 2025 also a license of around $125 per user per month for unlimited internal use (The Pricing Conundrum, 2026; Fin, 2026). That the market leader sells three models side by side says enough: nobody knows yet what AI work is worth, the seller included. Meanwhile Agentforce is growing fast, to about $800 million in annual revenue, up 169 percent (The Pricing Conundrum, 2026).

is outcome-based pricing fairer for the customer?

Outcome-based pricing sounds fairer than paying per seat, but it rarely solves the unpredictability for the buyer, and it adds problems of its own. The first problem is stacking. As the Bain research shows, at most vendors the results meter comes on top of the seat price. Your old bill stays, and a variable line item gets added.

The second problem is the measurement dispute. Who decides what counts as “resolved”? HubSpot reports in its own announcement that the Breeze agent resolves an average of 65 percent of conversations across more than 8,000 customers; that is the vendor’s own research, and it is also the vendor keeping the tally (HubSpot, 2026). Salesforce reports 2.4 billion “Agentic Work Units”, but that is explicitly an adoption metric, not a billing unit; how those units relate to the invoice is unclear even to analysts (The Pricing Conundrum, 2026). As long as the vendor defines, measures and invoices, you start every disagreement at a disadvantage.

The third problem is the default setting. Zendesk included AI agents in every Suite subscription as of May 2026, with 5 to 15 free resolutions per employee per month. Everything above that is billed per unit, and that overage has been switched on without a limit by default since January 2026 (Richpanel, 2026). “Included AI” is in practice a new meter without a ceiling. The consequences are measurable. In a survey by SaaS management platform Zylo of 218 IT leaders, 78 percent reported unexpected costs from AI features or usage pricing in the past twelve months, and 61 percent had to cancel or shrink projects because of unexpected software costs (Zylo via Advisable, 2026).

what does this mean for dutch small business?

For Dutch small business this shift arrives at a sensitive moment, because this is exactly when it is buying in. According to research by software company Wolters Kluwer among more than a thousand European SMBs, 84 percent of Dutch companies plan to invest more in AI over the next three years, the highest percentage of all countries surveyed (Wolters Kluwer, 2026). Those investments land largely with precisely the vendors now rebuilding their pricing model.

On top of that, an SMB lacks the negotiating power a party like SaaStr does have. Buy through a self-serve subscription or a standard contract and you get the list price and the default settings, uncapped overage included. Most SMBs also lack a procurement department that reads the fine print on credits and “resolution windows”. And notably: in two rounds of research I found not a single Dutch analysis that interprets this shift for SMBs; the only Dutch-language piece I found is a reactive explainer of one HubSpot price change from 2024 (Bright Digital, 2024). That silence does not mean safety. It plays out here just the same, only without advance warning.

This is, moreover, the same gap I described earlier between pilot and production: the pilot phase is small and manageable, and only at production scale do you see what something really costs (see the production gap).

It can be done differently, by the way. With fikst. I deliberately charge one fixed, transparent monthly fee, precisely because an SMB has to be able to build on predictability. Above all, it proves that a vendor can simply make that choice.

how to prepare your contract renewal in 2026

You do not need to become a pricing specialist, and you do not need to rush away from your vendor either; per-seat is alive and well for now as part of hybrid models. What you do need is half an hour with your contract and these seven points:

1. Take stock of which meters are already in your contract: seats, credits, conversations, resolutions, API calls. Many contracts contain more of them than you think.
2. Ask the vendor which pricing models exist side by side and have every variant calculated on your volume. At Salesforce there are three at once; pick one deliberately.
3. Ask for the exact definition of the billing unit. What counts as a “resolved conversation” or an “action”, who measures it, and can you verify that measurement yourself?
4. Ask what the default setting for overage is and demand a cap or a hard stop. At Zendesk, unlimited overage has been on by default since January 2026.
5. Ask how agent traffic is billed when you connect AI to the system yourself: as a seat, as API usage or as consumption. This is exactly where the SaaStr bill derailed.
6. Require monthly usage reporting plus an automatic alert at, say, 80 percent of your budget.
7. Run a growth scenario: what does this contract cost if your AI usage grows fivefold? If the vendor cannot answer that question on a single page, that is your answer.

My position, without detours: outcome-based pricing in its current form is above all a revenue model for the vendor, and predictability is worth more to an SMB than theoretical fairness per result. Choose vendors who can explain their meter in one sentence, cap every variable cost line, and treat every price change in 2026 as an opening to renegotiate; after all, you have something they badly need, a customer who stays.