Per-seat pricing looks innocent on the invoice. It is not. It is the single most reliable way an AI services contract turns into a multi-year tax on a thing the client does not own.
The shape of the trap
The pitch sounds reasonable. A modest monthly number per user — call it $60. The pilot is ten seats, $600 a month, nobody argues. Then adoption "succeeds." Six months in, the seat count is 50. Twelve months in, the contract auto-renews for two more years at a 7% escalator. The client now has a thing they cannot leave (Class C — observable in the signed order form).
The mechanics are not exotic. Four small clauses that almost always travel together.
The seat count only ratchets up. The order form has a true-up: exceed the count in any month and you pay the overage and the new floor for the rest of the term. It does not ratchet down.
The data lives inside the seats. Prompts, fine-tunes, chat histories, eval traces, and integration glue accumulate in the platform. The export, if it exists, is a JSON dump with no schema and no way to replay the workflows.
The integrations are theirs, not yours. The Slack hook, the CRM connector, the SSO bridge — every one is vendor infrastructure that disappears the day the seats lapse.
Termination is asymmetric. The vendor can terminate for "abuse" with 30 days notice. The client can terminate at the renewal anniversary, in writing, 60 days before, after paying outstanding overages.
Each clause is rational for the vendor. Stacked, they form a contract whose dominant strategy for the client is to keep paying.
The arithmetic on a 50-seat, 3-year deal
The numbers are worth doing in public. Assume $60 per seat per month, 50 seats, 36-month term, 7% annual escalator, and a true-up that lands the client at an average of 58 seats over the term.
Year one at 50 seats: $60 × 50 × 12 = $36,000. Year two at 58 seats, 7% escalator: $64.20 × 58 × 12 = $44,683. Year three, 7% escalator again: $68.69 × 58 × 12 = $47,809. Three-year total: roughly $128,500.
For $128,500 the client owns nothing. The artifact at termination is a JSON export and a list of seats that have stopped working. To rebuild on a different platform costs another implementation project, which the original vendor will quote (Class C — disclosed pricing pattern, not a worst case).
The same $128,500, spent on an engagement that produces a runbook the client owns, looks different at month 37: the client still has the thing, and the next $128,500 is not owed to anyone.
Why the market keeps choosing the trap
Two reasons, both honest.
Seat licensing is cheap in month one. The pilot is $600. The owned-artifact engagement is more. A buyer optimizing for the next quarter's budget picks the cheap option, and the cost of the trap lands three quarters later, on a different budget owner.
The hiring market is downstream of the same incentive. Jay Kumar Chimata, who runs a 22,000-user AI talent platform, did an interview describing what employers in 2026 are actually willing to pay for versus what candidates are studying: Meet Jay Kumar Chimata: JobFirst.ai and the Real AI Job Market (Class E — primary-source interview with the operator of a hiring marketplace).
Our reading, in our own words: hiring managers pay a premium for people who can ship and operate an artifact end-to-end, and they discount people whose skill set is bounded by one vendor's UI. The seat-licensing engagement is the contract version of that same discounted skill set — a great outcome for the platform, a bad outcome for the team's next job interview.
The outcomes-based alternative
The counter-model is not complicated. Scope the engagement to an artifact the client owns: code, prompts, eval set, deploy scripts, and a written runbook a person outside the engagement can re-run. Bill for the engagement, not for ongoing access to a UI. List dependencies with substitution notes. If the engagement fails, the falsifier goes in the shared ledger by Friday.
That model is described in long form on /anti-extraction-engagement-model and in the contract language on /partnership-vs-vendor-contract.
Four questions worth asking any vendor — ours or anyone else's — before signing a seat-licensed contract:
- If I stop paying, what do I still own at the end of the term?
- Can someone outside this engagement re-run the workflows from a written runbook?
- What is the true-up clause, and does it ratchet both ways?
- Show me the export. Open it in something that is not your product.
A contract whose answers to those four are evasive is the trap. The test is not whether the vendor is polite; the test is whether the answers are concrete.
If you want to walk a current or proposed seat-licensing contract through those questions with us, the intake call is on /workshop — the same call whether we end up scoping an engagement or telling you the current one is fine.