Talking with Adrianno Esnarriaga — and how good it is to talk with someone who’s actually implementing agents — about what I find craziest in the multi-agent discussion… People are trying to replicate processes inside multi-agent architectures, and not the other way around.
I say this because we have decades of experience building — deterministic — process systems, and we’re not taking advantage of it. It’s much simpler to have a more deterministic architecture defined inside a process tool than to try to recreate it in the #genAI world.
Imagine… if the first agent is right 70% of the time and hands off to another agent that’s right 90%, and then to another that’s right 98%, this multi-agent system will be right about 61% of the time! (if I still remember anything about probability lol)
How about keeping whatever doesn’t need anything that sophisticated as “recipe-like” as possible? Otherwise, build one single “big agent” and use that — its success rate may well be higher. :)
Originally written in Portuguese and published on LinkedIn on January 2, 2025. I translated it into English with the help of AI (Claude): it let me bring my LinkedIn posts to people who don’t read Portuguese without rewriting each one from scratch. I reviewed the translation to keep my original meaning and tone.