Part 5/12:
A significant part of the conversation centers on the conceptual layering within Autogen. Chi likens language models to a layer of abstraction, akin to an interpreter or compiler in software development. On top of this, the multi-agent framework acts as an even higher-level interpreter, simplifying interactions with large models and connecting them with tools and humans seamlessly.
He describes three fundamental types of agents within the system:
Model-based agents: Mediating conversations and decision-making based on large language models.
Tool-based agents: Executing predefined functions or code without relying solely on models.
Human-backed agents: Acting as proxies for human input, enabling oversight and hybrid workflows.