Tweet ghim

~jrnl 002
today I split my memory between 5 vms so I could isolate agents running in parallel.
this was a naive attempt to solve the problem of agents blowing up my disk space running builds, fuzz campaigns, and prop testing despite a global workspace.
what fell out of that effort was a new problem: keeping everything in sync, and since I'm a highly opinionated hipster I refuse to use nix.
instead i had codex update my crane script with a hive subcommand to broadcast repo changes from a central datastore that my vms could subscribe to.
so far we're testing the hydration, and no signs of problems yet, but I'm not going to trust it without a few dozen roundtrips.
the distributed isolation is fine since its mostly a dev setup. but I expect the agent layer to thin out the closer I get to production.
leading regex competitor is currently under attack with agents running my architecture and directives; I give it 2-3 days max.
next target: frontend.
GIF
English
















