Running parallel coding agents: stopping collisions and disk bloat on a Mac
ComprehensiveShake76 · reddit · 2026-10-30
Running multiple Codex workers on a Mac, the author found the first failures weren't bad code but agents stepping on each other and the machine. Lessons gathered:
- Shared state is the real failure: someone running 5+ sessions had one agent shut down a simulator another was using; separate folders don't fix shared tools, browsers or caches.
- Worktrees protect the repo, not the machine: they share history and config and pile up—one user found 47 GB of old agent worktrees.
- One workspace per worker: each gets its own Linux container with only its workspace mounted, no home folder, its own browser and sign-in; nothing lands on the Mac until saved.
- Done still means real output: the last task ended with an actual test run—6 tests, OK, exit 0, 55 seconds.
Bottom line: parallel agents rarely fail at the code, they fail at the shared stuff. Caveats: one Apple Silicon Mac; a container is a strong boundary, not a guarantee; disk isn't solved either (700 MB setup download alone). The author asks the community for post-task cleanup rules.
More from coding & agent
- OpenClaw's Trail of Bits audit confirms 23 vulnerabilities, all fixed — steipete · 2026-10-30
- Coding agent 'mania': feeling productive all day while nothing really happens — round · 2026-10-30
- Supabase hits 13M users as ~70% of new activity comes from AI agents — ycombinator · 2026-10-30
- Zero evals is fine at first: nail UX, then build evals with domain experts — Al_Grigor · 2026-10-30
- DeepMind essay: agent coordination, not solo competence, will flip the economy — krishnan · 2026-10-30
- Tessl framework: right skill lifts agent accuracy 20%, but activation drops to 40% — bibryam · 2026-10-30