I’m finding more and more that having a super strong foundation with your product’s API will make our break agents using your service.
No validation? 
No verbose error messaging?
A multitude of options for the agent to consume data? 
We’re making our own API’s strong and flexible, especially for agent consumption. We’re even creating our own mini feedback loops with the agents themselves.
We’ll spin up an agent, have it use the tool for about 15 or so minutes. Then, ask it questions like: “What’s one tool that you’d like to have that would make working with the API easier?”, and, “What information are we missing from the MCP that would make using this tool more intuitive?”.
We’ll then feed the outputs back the product agent and go have it create a plan and execute. Crazy stuff.
Curious if any other teams are finding agent feedback loops interesting/useful