Yeah, the schema driven approach does really lend it to the world of LLMs. We've found that with minimal guidance LLMs can generate schemas, validate them and iterate if necessary.
What's great about this is that once you've got a valid schema you can then guarantee the correctness of the code downstream as well as it being very testable. This text-based abstraction makes it possible to generate a complete production-ready backend with AI rather than trying to get a LLM to write complete applications from scratch.