Has anyone had multiple agents work in the same code base with different tasks?
I know you could have cursor rules applied to a certain composer agent, but what about a more “native” workflow where they interact with each other almost as humans do with PR requests?
do you mean multi-agents interact with each other by multi-servers?
Not necessarily. I see that concept as a “swarm” of agent to agent communication. In this case the agents are working independently until they are ready to “cheek code in”. They likely have their own branch. A human in the loop is just a pull request reviewer. When a PR is accepted, all other agents review their PR to see if there are merge conflicts. If so, evaluate how to address given original instructions. .
Actually found an article of someone thinking through the same thing but even deeper: Multi Boxing LLMs
Multi-agent Management System in Cursor - Showcase - Cursor - Community Forum
I know what you mean right now, i am still digging into this working pattern, probably this article might help