If you've tried to make a JSON API with Phoenix, it can be pretty cumbersome to generate an OpenAPI spec for the project. It's a very manual and tedious process, even when working with OpenApiSpex (the go-to Elixir library for generating an OpenAPI specs). And if your code implementation changes, then you often have to update your spec to match it.

With Ash, the same data used to model your application is also used to derive the data needed to build the OpenAPI spec. So there's a real value proposition there IMO. It eliminates much of the problems of keeping your spec in line with your code, since they are both modeled in a single location.

Disclaimer: I only learned this by working through the book, so I haven't actually gotten to experience anything off the happy path.