When running 5+ parallel agent sessions each running their dev loop, I was getting a massive over subscription on the CPU cores causing thrashing. Creating a lease queue seems to help with the coordination and keeps different agents chugging along. Any better suggestions?
Interesting read. I feel like nobody has gotten memory right but ifl that’s what’s gonna make things like a company brain actually useful long term. I don’t think .md files are going to be durable
https://t.co/mI5Nmz9Zgf
GPT 5.6 Sol in cursor cost 180$ to test and fix one redis caching change. maybe the harness is not as token efficient as the codex harness. Haven’t tried sol in codex yet. It was pretty thorough though
Been trying to setup Hermes as my assistant to delegate all my tasks. It’s actually been quite difficult to get Hermes to delegate to Claude code at a desirable level of abstraction. It seems to fall outside the models distribution for agentic coordination