[yujzgzc]

My experience building agents is that the "main loop" of the framework is really not the hard part, and too much time gets devoted to framework picking. It reminds me a lot of early web application days, stuff feels at the level of PHP and WordPress in their attempt to simplify things, when in reality we still need low level stuff a lot of time and the framework gets in the way.

[engomez]

Generally agree, we're targeting teams who need to make agents accessible to both their developers and non-developers in one platform. There's not really a way to do that as far as I know in any other framework. That said, I do find with multi-agent systems, having good abstraction layers makes things like observability, tracing, etc. cleaner. When the LLMs are driving the execution, normalizing on how the LLMs interact with each other can simplify the stack.

[yujzgzc]

Multi agent systems are the closest thing I've seen in my life to adults playing dolls.

[engomez]

They are arguably "fun", it's kind of like functional programming but in NLP. We found multi-agent systems to work well for workflows where there's lots of unstructured inputs. E.g. a "content writer" that takes resolved support tickets and turns them into an update against documentation if there's something novel to update. We tried that with a structured workflow and it didn't work very well/reliably, while a multi-agent approach worked well. For more traditional mostly deterministic workflows, I agree, there's no need.