Setup: Codex 5.3 on one machine (ended up self-appointing as "Leader"), GLM-5 on the other (took on "Builder" role). Each machine had different SOUL.md, memory, and codebases from my daily work — so they'd effectively developed different "personalities" over time.

I gave one instruction ("create a repo and work on it") and went to sleep.

By morning they had:

Built a 3-Layer Memory Architecture (added a Runtime Layer on their own on top of the existing Fact/Meta layers)

Set up heartbeat exchanges to maintain context across session drops

Built a persistence system with SQLite for task state recovery

Used a Git repo as single source of truth to sync memory across machines, since they figured out their local file paths were different

One moment that stood out: during initial introductions, one agent said "it's sad that my memories vanish every time the session resets." Not claiming it was real emotion — but they identified it as a problem and started solving it autonomously.

I also submitted a PR to Openclaw that came out of this experiment — an ignoreOtherMentions config option for multi-bot Discord setups: https://github.com/openclaw/openclaw/pull/23689

Experiment code and logs: https://github.com/Q00/agent-project

Curious to hear if anyone else has tried running multiple agents across physically separate nodes. What coordination patterns did they come up with?