For me, the 'autonomy' part is nothing more than less predictable router that works on its own, instead of hardcoding all the cases. Of course, with LLMs / agents you still need to provide tools (functions, branches) so it's not really "doing things for you".

More autonomy in terms of LLM choosing tools you provide, instead of you deciding which if/else branch to call in classing web service.

Obviously, you could build many of these apps with regular event-driven web service, where different events cause different side-effects in the system.

While(true) loop powered by LLM - so like a web service with a bit more authonomy.

Would love to see more flexibility in terms of middlewares. For example - starting as frontend dev server, adding endpoints with custom timeouts, delays, JSON responses, etc. Any recommendations for quickest execution here?

Who said it’s right coded?

Here we go: youtu.be/zg3B33ThhPc

if not CRUD then Maybe<DDD>

It's like saying you're Agile org because you have sprints.

Good point - I was thinking about limiting it only to test-based conversations, with code generation disabled. Maybe that's the standard approach we'll see in 2025?