PyData Boston/Cambridge Talk @ Moderna: What makes an agent?

Take 2 of what defines an LLM agent

Hello fellow datanistas!

Here is take 2 of “what makes an agent”, in which I recently delivered a talk at PyData Boston/Cambridge. It was the first time we hosted it at Moderna as well! I wanted to share the talk material here.

Right from the get-go, I let the audience know that they were going to be left with more questions than answers. and is based heavily on my previous post. In the talk, I explored the anatomy of LLM bots, using LlamaBot as a pedagogical tool, and discussed the concept of structured generation. We also examined the role of agents in decision-making and external interactions, using examples like a restaurant bill calculator and a travel planning assistant. Throughout the talk, I noted down discussion points from the audience that helped highlight the complexities and varying definitions of agency.

I think the tl;dr from these two takes on Agents is that even though the major players have invested in agent-oriented workflows, a well-defined philosophical definition of an “agent” isn’t going to be clear anytime soon. As such, anytime you hear the term “agentic”, it’s likely marketing material.

If you’re curious, please check out the full blog post here! If you find it insightful, please consider forwarding it to others who might benefit from these discussions.

Cheers,
Eric