These 3 AI products have different answers to “what am I paying for?”
- ChatGPT → access. Go is $8, Plus is $20, and Pro starts at $100/month. You pay a fixed price regardless of how much of your allowance you use. OpenAI carries the compute risk: power users are more expensive to serve, which is why even subscriptions have usage limits. Kyle Poyar’s piece says 70–80% of AI token usage comes from just 10% of users, so that fine print is probably doing a lot of work.
- GitHub Copilot → a person. Business is $19/seat, Enterprise $39. The buyer carries the value risk: you pay for every developer whether Copilot writes great code or garbage. But Copilot now bundles AI credits into each seat (1,900 on Business), with extra usage billed separately. It’s really seats with a usage meter bolted on. Seats are getting redefined.
- Intercom Fin → an outcome. $0.99 per resolved conversation. Intercom carries more of the performance risk because unsuccessful conversations generally aren’t billable. The catch is who decides what “resolved” means. Fin can count it when the customer just… doesn’t ask for more help. That’s exactly the “relitigating every outcome” problem from the reading.
So what fits Flipper? Flipper is our AI meeting assistant for PMs: it builds the agenda and maps the meeting live. The PM is the buyer, and engineers shouldn’t need a license just to see an agenda. I’d charge per host seat, keep attendees free, and put a fair-use cap on live AI minutes, where our compute cost lives. Outcome pricing (“pay per decision”) is tempting, but we’d end up relitigating what counted as a decision every month.
