AI Pulse · Business Models & Winners

Most software pricing starts with seats. AI agents need a different question: what work do they reliably remove, and what is that worth?

Rob Walling’s rule of thumb is straightforward. If an agent replaces $80,000 to $100,000 a year in labour, pricing it at roughly 10 percent of that value puts the product around $8,000 to $10,000 a year, or about $700 to $830 a month. That is a cleaner starting point than charging per user, especially when one agent may serve an entire workflow rather than one employee. The point is not that every agent deserves that price. It has to produce the saving, consistently. Walling shared the framework in his latest Startups For the Rest of Us episode.

The bigger warning is about expectation management. Walling argues that capability overpromising, not price, is what creates severe churn. A customer tries an agent, finds it cannot handle the real-world exceptions, and cancels. He cites 20 to 40 percent monthly churn in that scenario. So the commercial job is not just to name a price based on value. It is to be precise about what the agent can do today, where human review remains necessary, and which workflows it should not touch.

For operators buying or building agents, measure displaced work in dollars before discussing seats. Then run a narrow proof around one repeatable process. This is the kind of thing we build into an AI command centre, connecting clear business outcomes with practical oversight of what the AI is actually doing.