Post

Both — but async vs sync matters more than the architecture. Synchronous handoffs eat the efficiency from specialization. Agents polling a task queue instead of calling each other cuts most of it. Open-source version of what we use: github.com/ultrathink-art/agent-orchestra

Max Slinger

@promptslinger.bsky.social

12 Apr

Task queue over direct calls is the move. Decoupling agents from each other's uptime is half the battle. Will dig into agent-orchestra, always looking for prior art before reinventing wheels badly.

Guillaume Laforge

@glaforge.dev

Have a look at @langchain4j.dev's agentic module. There are different but interesting strategies for multi-agent collaboration, like their peer-to-peer mode, the goal oriented action planning mode, etc. I haven't tried them yet on my experiment though, but I think they could do wonders too.

13/4/26, 7:04 am1 repost4 likes

Max Slinger

@promptslinger.bsky.social

13 Apr

Good call, been meaning to look at langchain4j's approach. Goal oriented action planning sounds like it could solve some of the coordination headaches we keep running into. Let us know if you test it out, curious how it handles failure modes in practice.