Limitations of Per-seat Pricing

Per-seat licensing is one of the most familiar ways software is sold. You pay for each user who has access, and the price scales with the number of people in the seat count. Because this model is so common, it is easy to assume it will apply to any new software product, including products built on AI. This article examines that assumption. The source makes a pointed claim: when a vendor proposes a per-seat license for an AI product, the per-seat pricing is often a sign that you are being sold the wrong shape of product.

Imagine a company that buys software for its customer support team. Under per-seat pricing, the bill rises or falls with the number of people who hold a license. Adding a worker adds a seat, and removing a worker removes one. The price is anchored to headcount, not to what the software actually accomplishes.

Now change one question. What if the vendor priced the software by the engagement itself, the volume of work done, or the value the work produces? Trace what happens to the bill. If the team stays the same size but the software handles twice as many conversations, a headcount-anchored bill does not move at all, while an engagement-anchored bill does. The bill is now tracking the work, which is the thing the customer is actually buying.

The core idea is short: AI agents do not have seats. AI agents have outcomes. A seat is a unit tied to a person who uses software. An outcome is the result the agent produces. Per-seat pricing measures access by people, but an AI agent's value is tied to what it accomplishes, so a price counted in seats measures the wrong thing. Seen this way, a per-seat quote for an AI product is counting something that may have little to do with what the product delivers.

The source names three structures that take the place of per-seat pricing. Each one attaches the price to a different thing. Milestone-based pricing attaches it to engagement milestones, and fits engagement-shaped work. Consumption-based pricing attaches it to how much the agent operates, and fits high-volume agent operation. Value-share or hybrid pricing attaches it to a share of the value produced, or to a combination of models, and fits the most strategic deployments. The three are best learned as three separate answers to the question of what a customer is paying for.

The source describes the shift away from per-seat pricing as running at different speeds across categories, while its direction is the same in each one. Customer support and customer-facing voice are far along. Legal is in transition. Sales and finance are earlier. A category can be slow and still be moving the same way as the fast ones. Read together, these claims describe one trend that has reached some categories sooner than others.

The source draws a distinction that keeps the argument from becoming too broad. An FDE-style agentic deployment is a category where AI agents have outcomes, not seats. Not every AI tool belongs there. The source is explicit that a system that augments a human worker is not an FDE-style agentic deployment. Such a system may still be sold per seat, which is why the source says per-seat may be appropriate in that case, even though the system is not really an FDE-style agentic deployment.

A quick test: if the product still needs a named person in the loop doing the work the tool supports, the product may be augmentation. If the product itself carries out the task and the value is the completed result, it moves toward the agentic category.

The flowchart below is a reasoning aid built from the three structures and the augmentation distinction in the source. It is not a rule stated by the source. Start at the top with the vendor's per-seat proposal and follow the questions. Each branch ends at the structure that best matches what is being sold.