How will LLMs affect API design?

How will APIs change if they are not primarily meant to be integrated by humans?

What APIs do may not be affected by who calls them (other services, and clients), but how they are presented may change to accommodate a new class of integrators - the LLMs.

Some thought experiments:

• As an API writer, will I write documents that are better prompts for the LLMs consuming them?
• As an API designer:
  • will I structure API endpoints differently?
  • will I add more API endpoints for LLMs to call?
• As an API consumer that uses own LLM to write clients what would I want different or extra? (An API Doc that is a friendly prompt 😉?)

Or is it all moot because .. when LLMs are smart enough to understand current way of writing APIs (documents, code, and design), why would you change anything?

Will they perform better? How do you test this?

If someone well immersed in writing prompts were to write APIs and docs, their choices might reflect their immersion. And that will be interesting to see.