The original title is "What AI agents actually pay for — data from 74 pay-per-call endpoints"

Short summary

An operator of a pay-per-call API platform for AI agents analyzed ~2,000 paid calls across 74 endpoints and found that just five endpoints drove 75% of revenue, with a single text-to-structure endpoint accounting for 40% alone. The key insight: agents pay for transformation—turning unstructured text into reliable typed JSON—not for raw data access, and spend tracks task value rather than call frequency. The author killed most of his endpoint catalog to focus depth on the few winners, concluding that breadth in agent API offerings is mostly dead weight.

• •5 of 74 endpoints drove 75% of agent API revenue; one text-to-structure endpoint alone was 40%
• •Agents pay for transformation (unstructured→structured JSON), not raw data access
• •Spend tracks task value, not call frequency—optimizing for volume would have been wrong

Generated with AI, which can make mistakes.