https://x.com/melissapan/status/1915557250141479364?s=46

· Sorted by Oldest

Jason
·
Multi-agent is a software abstraction. An abstraction either helps or gets in the way. When it gets in the way, you spend more time debugging the abstraction, than if you just built it from scratch. I think abstractions are hard to get right out of the gate (think LangChain last year). People are struggling enough getting their individual agents working, don’t think adding a multi-agent abstraction will help fix. So don’t start there to build something. But, there are a number of places I think it’s relevant. MCP integrations of 3rd party agent tools to agent mode code IDEs will probably be the most immediately relevant.

Jason
·
Multi-agent is a software abstraction. An abstraction either helps or gets in the way. When it gets in the way, you spend more time debugging the abstraction, than if you just built it from scratch. I think abstractions are hard to get right out of the gate (think LangChain last year). People are struggling enough getting their individual agents working, don’t think adding a multi-agent abstraction will help fix. So don’t start there to build something. But, there are a number of places I think it’s relevant. MCP integrations of 3rd party agent tools to agent mode code IDEs will probably be the most immediately relevant.