My Complete Parallel Coding Agent Setup

Every config, command, and hook I use to run five coding agents at once without losing track. Copy it in an afternoon.

My most-read article this year explains why you should run coding agents in parallel and the ideas that make it manageable. This post is the other half: the complete setup, exactly as it runs on my machine.

The original article was a high-level overview of my setup. This is a more in detail view into exactly how I do things, so you can copy it over to your own workflow. This is precisely the tooling, ideas, and techniques that I apply on a daily basis as a Data Science Lead shipping AI systems.

What's inside

1. The worktree layout that isolates each agent, with the exact commands

2. My terminal layout rule (tabs, panes, and when to use the agents view)

3. The sound-hook config so agents interrupt you, not the other way around

4. The two CLAUDE.md rules that make returning to a thread take seconds

5. The verifiable-task prompt template so agents check their own work before reporting

6. How I pick which tasks to parallelize (and which never to)

7. My cleanup routine for stale worktrees and dead sessions

If you set this up once, running five agents feels like running one. Let's go through it.