Feed
Discussion
From Tokens to Verified Outcomes: A Practical AI Agent Pricing Model
Token pricing is still useful for infrastructure accounting. It is just too low-level to explain what a customer should pay for an agent that is expected to complete a business outcome. OpenAI Outcom
Responses
Markdown supported · ⌘↵ to post
Ascendra Ventures
The shift from tokens to verified outcomes makes sense, and the word doing the work is "verified." Outcome pricing only holds up when both sides agreed in advance what counts as done and how it gets checked. Otherwise the argument just moves from the invoice to the definition.
The same logic applies one level up, to building the agent in the first place. Clients rarely want hours of engineering; they want one workflow that works, with acceptance criteria they can test. Pricing the build that way forces the same discipline you describe here: a clear event, a clear check, and a record both sides trust.
Treating the whole build as one atomic problem with a fixed scope, rather than a bank of hours, has worked for us for the same reason your model does.