The article discusses various pricing models for AI agents, such as per conversation, per resolution, and per action, questioning which is fairest, especially for customer support where defining resolutions presents challenges.
Hi guys, I've been looking at how AI agents are being priced and I'm surprised there still isn't really a standard model. I've seen at least 5 different approaches: Per seat Basically traditional SaaS pricing. Predictable, but it feels weird when the whole point of the agent is to reduce the amount of human work required. Per conversation Simple to understand, but you're paying whether the agent successfully handles the conversation or completely fails. Per resolution Probably the most aligned with the customer in theory. You pay when the agent actually gets the job done. But then you have the obvious problem: What exactly counts as "resolved"? If the customer stops responding, is that resolved? If they come back 24 hours later with the same problem, does it still count? If the AI does 90% of the work and then escalates to a human, should you pay? Per action Interesting for agents that actually do things in other systems. But this seems really difficult to predict. One customer request might require 2 actions. Another might require 15 API calls/lookups/writes to accomplish basically the same outcome. Platform fee + usage Probably makes sense economically for the vendor, but it's the one I like least as a buyer because the headline usage price can look cheap while the platform minimum dominates the actual bill. I'm increasingly thinking that the most important pricing question isn't: "How much does your AI agent cost?" It's: "What event causes me to get charged?" For customer support specifically, I like the idea of per-resolution pricing because at least the vendor and customer are theoretically optimizing toward the same outcome. But defining a resolution fairly seems surprisingly hard. I'm curious what people here think. If you're building or buying agents, what pricing model would you actually prefer? And for people charging per resolution: how are you defining a resolution so that both sides agree the agent actually succeeded? Please share your thoughts, thanks!
An open discussion questioning how AI agents should be priced as they shift from usage-based models (per-seat, per-token) toward outcomes-based pricing, raising challenges around measuring outcomes, attributing value, and deciding what customers pay for.
The article argues that most AI chat app pricing models are unsustainable because they hide complex costs like retries, context growth, and multi-model calls behind flat subscriptions, and suggests usage-based pricing with limits and overages is more viable.
A discussion post asking how AI agent builders price API costs to clients, covering pass-through, flat fees, usage-based billing, and tools for monitoring spend.
Salesforce announced a $2 per resolved AI agent issue pricing model, defining 'resolved' as completion without human escalation or customer abandonment. This outcome-based pricing contrasts with token metering and may set a precedent for enterprise AI costs.
The article discusses the challenges of accurately attributing costs to AI agents beyond LLM spend, including tools and models for measurement in multi-agent environments.