This paper is specifically disproving the efficacy of agentic frameworks like AutoGen.

Also, the built-in function-calling in GPT4 is simpler to use than AutoGen2's abstraction.