The agent pricing problem is deceptively simple to state: you have a product that delivers value, AI agents are doing work in that product on behalf of customers, and your current pricing model doesn't capture that value reliably. How do you price it?
There's no consensus answer yet, but the frameworks emerging from companies grappling with it are instructive.
Action-based pricing. Charge per meaningful action the agent takes — per record updated, per workflow completed, per decision made. This is clean and directly tied to value, but requires you to define "meaningful action" in a way that's transparent and hard to game. The risk: customers who use agents intensively get bills that surprise them, creating churn.
Outcome-based pricing. Charge for the business outcome delivered — revenue influenced, time saved, errors prevented. This aligns incentives perfectly but requires measurement infrastructure that most companies don't have. It also shifts risk to the vendor when outcomes are partially outside your control.
Capability tiers that include agent access. Add an "agent tier" above your current top tier. This tier includes programmatic access, higher rate limits, and agent-friendly API features. Customers who deploy agents buy the tier; customers who don't, don't. This is the least disruptive path.
Hybrid: seat floor + agent consumption. Maintain a minimum seat fee for governance and accountability, then charge on consumption for agent-driven actions above a threshold. This captures the baseline subscription value while monetizing agent scale.
The wrong answer: ignore it. The companies that pretend agents don't exist in their pricing model are leaving significant revenue on the table and creating a future pricing renegotiation that's much harder to navigate than a proactive redesign.
Start with transparency: know what your customers' agents are doing in your product. Then price accordingly.