The Usage Trap: Why AI Pricing Isn't Ready to Grow Up

There is a tell hiding inside every AI pricing page, and it has nothing to do with the number after the dollar sign. It is the unit being counted. Seats. Logins. Users. In an industry that talks endlessly about autonomous agents, the invoice still arrives addressed to a human being who logged in and did something.

This is not an oversight on the part of pricing teams. It is an accurate description of how the work actually gets done.

Consider Salesforce's Agentforce, or Intercom's Fin. Both are marketed as agentic products, built to act rather than merely assist. Yet the pricing underneath them still tracks something closer to activity than achievement: a person queries the system, reviews what comes back, approves or corrects it, and does so again tomorrow. That repetition is the product's real usage pattern, and seat-based pricing is simply the honest reflection of it. The vendors are not behind the technology. They are pricing the labor that still exists.

The alternative, outcome-based pricing, has become one of those ideas that every pitch deck treats as self-evidently inevitable: pay for the resolved ticket, not the seat; pay for the result, not the login. It is a clean story, and clean stories tend to get told well before the conditions that would make them true.

The conditions are not yet in place

Those conditions are specific, and they are not yet in place. Outcome-based pricing does not simply require better AI. It requires less human involvement in the process the AI performs, which is a much higher bar. Usage frequency has to fall, not merely hold steady, before an outcome becomes the only thing left to bill against. That fall depends on a kind of trust that has not been earned: trust that an agent can run unsupervised, make judgment calls, and be wrong occasionally without triggering a full human review every time. Most enterprise buyers are nowhere near extending that trust, and for reasonable reasons. The cost of an unsupervised error is not symmetrical with the cost of an unsupervised success.

There is also a quieter, more structural obstacle: revenue predictability. SaaS finance teams did not fight for years to build forecastable ARR just to hand that discipline back to variance. An outcome-based model, however elegant in theory, introduces exactly the kind of monthly unpredictability that both vendors and buyers have spent a decade trying to eliminate. Nobody wants to relearn how to budget for software.

None of this means outcome-based pricing is wrong. It means it is early, and mistaking early for wrong, or wrong for early, is the actual failure mode to avoid.

Hybrid is not a compromise

What fills the gap in the meantime is not a compromise. It is the correct architecture for the moment: a hybrid model, combining a predictable base with usage or seat components for the tasks still requiring a human hand, and a smaller outcome-based layer for the narrow slice of work where the agent has already proven itself. This structure is not a temporary bridge to something better. It is the accurate pricing of a system in transition, one that lets the outcome-based portion expand exactly as fast as trust actually grows, without forcing a renegotiation event every time a vendor's confidence gets ahead of its customers'.

Watch usage frequency, not announcements

The signal worth watching, then, is not the next pricing-model announcement. It is usage frequency itself. When it begins to fall, not flatten, but genuinely fall, that will mean agents have started absorbing tasks that used to require a person at the keyboard. That is the moment outcome-based pricing stops being aspirational and starts being accurate. Anyone pricing purely on outcomes before that shift is not ahead of the market. They are simply pricing against the evidence.

Seat-based pricing is not outdated. It is, for now, correct. Outcome-based pricing is not the future arriving early. It is a reward that most categories have not yet earned. And hybrid pricing is not the safe, unambitious middle path it is often mistaken for. It is the only model that prices the technology honestly, for where trust it's actually earned.