@ruben the "cost moved from top of funnel to inside the product" line is such a clean way to put it. are founders actually re-pricing fast enough or mostly still stuck on old habits?
The third position is where most PMs actually live and don’t realize it. You mention the old margins mostly survive when AI is a thin slice — but I keep seeing founders price as if they’re in position one (hosting their own models), when their LLM call is read-only or buried in a workflow. The blind spot is not knowing which bucket you’re in.
The reflex you name is real, and pay-per-call is the correction most founders resist because it feels like a downgrade from subscription revenue. It is not. It is the only pricing that survives a cost line that moves with usage. I price my endpoints per call in USDC, and the discipline it forces is knowing what one call costs to serve before you publish it.
@ruben the "cost moved from top of funnel to inside the product" line is such a clean way to put it. are founders actually re-pricing fast enough or mostly still stuck on old habits?
The third position is where most PMs actually live and don’t realize it. You mention the old margins mostly survive when AI is a thin slice — but I keep seeing founders price as if they’re in position one (hosting their own models), when their LLM call is read-only or buried in a workflow. The blind spot is not knowing which bucket you’re in.
Built a free LLM margin calculator for exactly this (feature-level margin visibility, no token cost since it’s API-read-only): https://www.aimargintracker.com/tools/margin-calculator?lang=en
The reflex you name is real, and pay-per-call is the correction most founders resist because it feels like a downgrade from subscription revenue. It is not. It is the only pricing that survives a cost line that moves with usage. I price my endpoints per call in USDC, and the discipline it forces is knowing what one call costs to serve before you publish it.
The customer who loves the product most can be the account doing the most damage. That’s a nasty pricing problem.