If the doco we're talking about is specifically an API reference, then the documentation should be written first. Generate code stubs (can be as little as an interface, or include some basic actual code such as validating required properties are included, if you can get that code working purely with a generated template). Then write your actual functional implementation implementing those stubs.

That way you can regenerate when you change the doco without overriding your implementation, but you are still forced to think about the user (as in the programmer implementing your API) experience first and foremost, rather than the often more haphazard result you can get if you write code first.

For example, if writing a web API, write documentation in something like OpenAPI and generate stubs using Swagger.