Run Claude Code, Cursor, and Codex in parallel
A workflow for running several AI coding agents at once with git worktrees and one task per agent, and why attention becomes the bottleneck.
By Piotr Jura, creator of AgentNotch. Tested on macOS with AgentNotch v1.3.1.
Claude Code writing tests, Cursor refactoring, Codex drafting a migration. On paper you tripled your output. In practice two things decide it: setting the agents up so they don't collide, and keeping them fed.
Setup is the easy half: each agent gets its own working directory and scoped task, and they run side by side (rules in the table). Starting five takes seconds; keeping five moving is the hard part. Each stalls on a review, a question, or an approval, and five idle agents are slower than two you keep fed.
Where AgentNotch fits
AgentNotch fills that gap. It watches your Claude Code, Cursor, Codex and Kimi sessions across your worktrees and surfaces the one that needs you next. You answer one signal instead of scanning panes. It runs on your Mac, never your code or keys. The worktrees run agents in parallel; the notch keeps them fed.
Side by side
| The rule | Why | |
|---|---|---|
| One working directory per agent | A git worktree or a separate clone; a Mac app like Conductor makes the worktrees for you. | Two agents in one folder overwrite each other and muddle the diffs. |
| One task per agent | A tightly scoped job on its own branch. | Keeps diffs reviewable and stops context bleeding between agents. |
Questions
- How do I run multiple coding agents in parallel?
- Give each agent its own working directory so they do not fight over the same files, using a git worktree or a separate clone per agent, and assign one task to each. Once they are running, the hard part is attention, which is where a monitor that flags the session needing you earns its place.
- Can two coding agents work in the same repo at once?
- Not safely in the same working directory, because they overwrite each other's edits and confuse each other's diffs. Use a git worktree per agent so each has an isolated checkout of the same repo on its own branch.
- What is the hardest part of running agents in parallel?
- Attention, not compute. Starting five agents is easy; knowing which of the five just finished, stalled, or needs a decision is the part that breaks down, because you cannot watch five windows at once.
- How many agents can I run in parallel?
- You can start as many as your machine and rate limits allow, but the practical ceiling is how many you can respond to. Past three or four, keeping each one moving depends on catching its idle moments quickly rather than on raw capacity.
Keep reading
See which agent needs you.
Try AgentNotchfree for 14 days, no card. Watch Claude Code, Cursor, and Codex from your Mac's menu bar.
Updated 2026-07-17