AI Agent Architecture Software · AI Developer Accelerator

Jun 5 (edited) • LangChain

AI Agent Architecture Software

Hello everyone,

I've been part of this community for quite some time, mostly as a reader, but now I'd like to ask a question that is closely related to AI agent development. I'd especially appreciate insights from people with hands-on experience designing AI agents and multi-agent systems.

When building AI agents, it has become common practice to represent agent architectures as graph-based workflows using frameworks such as LangChain and related tools. This applies both to the internal architecture of a single agent and to orchestration in multi-agent systems handling more complex tasks.

My question is: how much time and effort does it typically take you to design a high-quality agent (or multi-agent) architecture without LLM assistance?

We all know that modern LLMs are extremely capable when it comes to coding. However, I'm curious about your experience with using them for architecture design itself. How good are they at proposing agent or multi-agent architectures, and how much do you trust their recommendations in this part of the development process?

Do you mainly:

Design the architecture yourself and use LLMs only as assistants?
Or do you largely rely on LLMs to generate the architecture and then refine it afterward?

I'm currently working on a system aimed at optimizing the design of AI agent architectures (graph-based workflows), so any experiences, opinions, or lessons learned would be extremely valuable.

Looking forward to hearing your thoughts.

0 comments

AI Agent Architecture Software