Although agents are alluded to in the lesson, it appears that in this lesson there is no entity called an agent. They are nodes of a graph. Or I’m wrong? It seems that the state is more associated with: a) the graph, not an agent, b) the inventory of the points of possible states of the graph scenario, and c) the state of the graph at an instant of time according to dynamics between nodes and their interrelations. Or maybe a graph can be understood as some kind of behaviors flow schema of an agent?. I am learning LangGraph, I could be wrong, I would like to hear/read the community’s opinion.
I’m not sure if I understand your question. From my understanding, the ensamble of nodes, that includes an LLM, creates the kind of recursive autonomous work flow that would be called “Agent”. This would be an abstraction, used to communicate the kind of capabilities that the software would have, so an end user would have some understanding regarding what to expect from it (e.g.: “Oh it’s an agent?, that probably means it can do complex work on its own!”). However, in practice this agent is the sum of a bunch of components that make possible that functionality.
This agent can be decomposed into different parts (nodes), that make possible it’s effectiveness in solving complex tasks, that it would otherwise struggle with by using solely the regular chatGPT interface. I understood that the key difference of this LangGraph implementation, is the possibility to integrate the use of LLMs in combination with other tools, to iterate for as long as needed to determine and solve all the steps required to complete the initial user request.
In summary, this “Agent” extends the capabilities of LLMs, to solve complex tasks more effectively by autonomously designing and implementing a work flow that extends for as long as required, and that is capable of including external tools in the process.
Thanks @chipsi21 for your time, and for your detailed answer and perspective. Then there can be a point of view from Agents perspective as a useful abstraction or call it Agents when, LangGraph and LLMs are involved.