The first question most AI vendors ask is "how many tokens?" The first question the customer wants to ask is "how many problems will this solve?"
These are not the same question.
Token pricing made sense when the underlying cost was the binding constraint and the buyer was technical. For everyone else — every operator, every founder, every executive who does not want to think about inference — it is the wrong unit of account. It puts the customer on the wrong side of every conversation. Every time the employee does extra work, the customer worries about the bill. Every time the bill shows up, the customer recounts the value.
We price three ways, none of them on tokens:
- Pay per task — a fixed price per completed unit of work. Contract review, claim, ticket triage. The customer pays for outputs they can count.
- Flat monthly — one line item, per month, like any other salaried hire. Inbox triage, follow-ups, reports, executive ops.
- Pay on results — a small base, then a share of revenue closed. Lead research, outreach, deal acceleration.
Tokens are our cost basis. They are not the unit we sell. The unit we sell is the work.
This is not pricing innovation. It is pricing common sense. Sales teams have always sold quotas, not commissions on their phone bill. Lawyers sell hours, not billable minutes on their email account. The unit of account in any service business is the thing the customer cares about. For AI employees that is completions, contracts, deals, replies, briefs. Not tokens.
A founder running a public AI services company told this story on a recent podcast: his firm priced like an agency from day one — by the completed task, by the closed deal, never by the back-office bill. Their customers, who in his words "won’t pay for software", happily pay for completions. The unit of account decided whether the business existed.
The same lesson applies to every AI employee a team will hire in the next decade. Sell the work. Eat the tokens.
Related See the three pricing shapes →