I think just having devs attempt to feed an agent openapi docs as context to create api calls would do enough. Simply adding tags and useful descriptions about

I think just having devs attempt to feed an agent openapi docs as context to create api calls would do enough. Simply adding tags and useful descriptions about endpoints makes a world of difference in the accuracy of agent's output. It means getting 95% accuracy with the cheapest models vs. 75% accuracy with the most expensive models.