Per-Seat vs Per-Outcome: Why AI Agent Pricing Is Shifting
AI agent pricing is moving from per-seat toward per-outcome because a seat license no longer maps to the work an agent does — and the only unit that makes vendors comparable is dollars per completed outcome.
For two decades, software was sold per seat: one license per human user. AI agents break that model, because the unit of value is no longer a person logging in — it is work getting done. This post lays out the four pricing models you will actually encounter (per seat, per token, per action, per outcome), why the market is shifting toward outcome-based pricing, and the one conversion that makes any of them comparable. It cites pricing models factually and does not invent vendor figures.
The four models you will encounter
Per seat: a flat fee per named user or per month. Familiar and predictable, but for an agent it is a poor proxy — one "seat" might resolve ten tickets or ten thousand. Per token (or per compute): you pay for the underlying model usage. This tracks cost but not value, and is volatile as model prices move.
Per action: a charge per discrete operation — per message, per API call, per run. Closer to usage, but an "action" is not necessarily a finished outcome. Per outcome: a charge per completed unit of work — per resolved ticket, per booked meeting, per reviewed pull request. This is the model that aligns price with the thing the buyer is actually purchasing.
Why the shift is happening
Per-seat pricing assumes a roughly fixed amount of work per user. Agents violate that assumption: capacity is decoupled from headcount, so a seat license either overcharges light users or undercharges heavy ones. As buyers got burned by seat licenses that did not match consumption, demand grew for pricing that tracks delivered work.
Vendors are responding. In customer support, some vendors publish a per-resolution price; in sales, some publish per-meeting or per-seat plans; coding tools mix per-seat and usage-based tiers. The common direction is toward charging for outcomes, because that is the only model a buyer can sanity-check against the value received. Where you cite a specific vendor’s rate, state it as that vendor’s own published price — do not generalize it into a market number.
The conversion that makes them comparable
Whatever the billing model, convert every quote to dollars-per-completed-outcome before comparing. A per-seat plan divides by the outcomes that seat produces; a per-token plan divides by tokens-per-outcome; a per-action plan divides by actions-per-outcome. Only after this conversion can a "$2,000/month" agent and a "$0.99 per resolution" agent sit on the same axis.
This is exactly why the WorkForce Labor Index is defined per task rather than per seat or per token. A per-outcome unit is the only one that survives across vendors and across billing models — which is what a benchmark needs.
What it means for buyers and builders
Buyers: refuse to compare raw plan prices. Convert to dollars-per-outcome, then read each quote against the human-labor baseline and, as it publishes, the WLI market rate. Builders: if you can stand behind your quality, outcome-based pricing is a trust signal — you only get paid when the work lands.
Either way, price needs a quality number beside it. A higher per-outcome price is only justified by a measurably higher quality score with a confidence interval.
Methodology note
This article describes pricing models conceptually and references vendor pricing only as the model type (per resolution, per seat, usage-based) without stating any specific vendor figure as a market rate. The WLI publishes a per-task rate with its confidence interval and sample size once transaction coverage exists; see /methodology for how the index is calculated and governed.
Next steps
Get a free, independent quality score at /eval, and track the live per-task rates at /wli as categories cross the publication threshold.