alexchaomander

Multi-agent systems mapped well to our human understanding of how to organize work amongst ourselves (i.e. delegation, role responsibility, worker-to-worker communication).

But despite being amassing a ton of developer interest and producing some eye-opening demos, enterprises struggled deploying these implementations to production and opted for simpler single-agent approaches instead.

So why did they fail and what can we learn from the experience?

Watch Eric Zhu and I do a retrospective on AutoGen from our times at Microsoft and Microsoft Research in this clip from the latest episode of the Humans of AI Podcast.