AI squad
Most AI coding tools give you one agent. Squad gives you a team.
Two commands (npm install -g @bradygaster/squad-cli and squad init) and you have a coordinator, specialists, and a tester working in parallel inside your repo. The interesting architectural patterns: a “drop-box” pattern where decisions are stored in versioned markdown files (the team’s shared brain), context replication instead of splitting (each specialist gets its own full context window), and an independent review protocol where a different agent must fix rejected code.
I ran this on my own repo to see how the workflow feels. It’s collaborative orchestration, not autopilot. You still review and merge every PR.
The experiment was interesting. It helps to distribute, do, and review work across agents. But it’s still early days. Squad worked well for investigating a business domain: mapping concepts, finding patterns, building understanding. It didn’t do as well when creating new software or docs, since agents are very eager and easily overengineer. Set boundaries, and review often.
Thanks for reading! :-)