working on a .net 9 backend and considering spinning up a small python service just for the ai/llm stuff. keep the c# side clean, let python handle prompt logic and model calls since the tooling there is just better for that usecase.
curious if anyone's done this in practice. how did you handle the communication, grpc or just http? was it worth the added complexity or did you end up just calling the api directly from c#?
just actually sitting and deciding things myself. like how do i want this structured, what pattern makes sense here, how should the dependencies flow. nobody talks about that part. the thinking before the typing. ai just skips it for you and you dont even notice its gone until you sit down alone with a blank file.
did it today for a side project. set up DI myself, thought through the architecture, wrote constructors without prompting anything. took longer obviously but it felt like actually building something instead of just reviewing code someone else wrote.
feels like a skill thats quietly disappearing and most people wont notice until its completely gone.
anyone else still do this on purpose?