| ▲ | petesergeant 3 hours ago | |
I think multi-agent (eg _different_ underlying LLMs) everything is really the future. Code produced via multi-agent workflows and reviews seems noticeably better. I've been experimenting with a multi-agent message board recently: https://github.com/pjlsergeant/dogpark | ||
| ▲ | monkeydust 2 hours ago | parent | next [-] | |
Experimented with multi-llm analysis for problem solving over summer, combined with multi-agent approaches it can tease out interesting angles to problems that I never considered. Expensive but use only for my high value problems. | ||
| ▲ | elzbardico 41 minutes ago | parent | prev [-] | |
I've built agents that call different LLMs and keep separated memories. Remember, agents are just long-running workflows with some nodes calling LLMs and that sometimes can be started as tool from other "agent". There are times when I wonder if couldn't just draw then in a BPMN designer that allowed me to write custom code for nodes. Is BPMN still a thing? | ||