I’ve got four Claude Code windows open right now and honestly it doesn’t even feel like a lot anymore - two on articles, one building a feature, one where I’m getting feedback on an RFC. Not that long ago I would’ve waited for each one to finish before starting the next, and I only changed after stumbling across an Anthropic interview where a PM mentioned “multi-clauding” as a pattern they kept seeing developers do. And honestly, once you’re running a few sessions at once, it stops feeling like using an AI assistant and more like managing a small team.
The two reasons I do it: filling the dead time while Claude works on something that doesn’t need my input, and capturing ideas that pop up mid-session that I don’t want to lose, like a quick PRD thought or a new skill idea to improve my workflow.
HOW I ACTUALLY USE MULTIPLE CLAUDE SESSIONS
- One Claude builds, another reviews. One session writes code or content, another reviews it with fresh eyes and no context from the creation process. The reviewer catches things the builder was too close to see.
- Long task in one window, quick stuff in another. If something’s going to take Claude a few minutes, I’ll open another window for smaller tasks, and even if the tasks are related, keeping them in separate sessions means you’re not polluting the context of the main work.
- Deep research in one, autonomous work in another. One session digs into documentation or explores a problem while another works through an RFC or feature build that doesn’t need much hand-holding.
- Parallel tracks that don’t depend on each other. Frontend and backend at the same time, feature and tests at the same time, different projects entirely. As long as they’re not editing the same files, they can run simultaneously and you merge at the end.
I mostly use Claude Code in multiple terminal windows, but sometimes I’ll mix in Claude Web tabs for research or writing while Claude Code handles the building. They don’t need to know about each other.
IF YOU RUN INTO MERGE CONFLICTS Developers use git worktrees to give each Claude session its own isolated copy of the codebase. Worth looking up if multiple sessions are editing the same files.
Fair warning: parallel sessions burn through tokens faster, and context-switching takes more energy than you’d expect - I’d start with two and see how it feels before adding more.