Running Multiple Claude Code Sessions on One Project: What I Split, What I Shared, What Broke

Running Multiple Claude Code Sessions on One Project: What I Split, What I Shared, What Broke

Since September 30, 2026, I've had two Claude Code sessions working in the same project folder at the same time. One handles my blog workflow. The other is building a new YouTube channel. I'm not a developer, and I talk to each one in its own window.

People search for "multiple Claude Code sessions" with all kinds of follow-ups: at once, in parallel, can they talk to each other. Here's what a day and a half of actually doing it taught me. It's a short track record, so treat this as field notes, not a benchmark.

Why I opened a second session

The blog had a steady routine: research, draft, review, publish. The video channel needed something different: scripts, voice, an assembly pipeline. Doing both in one conversation would have mixed two very different jobs into one long context. So I asked how to run the video work "in another window," and set one up.

A laptop showing the ChatGPT page

Photo: Hatice Baran on Pexels

Same folder or separate copy?

Claude Code can work in an isolated copy of a repository (a git worktree). I didn't use one, on purpose.

Not all of my project's state is in git. The task database, the dashboard's data, and the Google sign-in tokens live in folders that are deliberately excluded from version control. A fresh copy would start without them, so the video session wouldn't see the shared task board and couldn't sign in. Both sessions therefore open the same folder.

That choice is what makes the next three sections necessary.

Give each session a lane

Before the new window opened, the blog session drafted a scope rule for me to paste in as its first message:

  • This session is for video work only. Never touch blog tasks, approvals, or publishing.
  • When asked to "process pending instructions," handle only video tasks.
  • When saving work to git, commit only video files.

The blog session took on the mirror-image rule. This matters because both sessions read the same inbox. When I delegated a video task and a blog task seven seconds apart, both items showed up in the blog session's inbox. Because of the lane rule, the blog session handled only the blog item and left the video task alone.

What the sessions share (and can't avoid sharing)

SharedWhy it matters
The project folder and rulebook (CLAUDE.md, procedures)Both sessions follow the same rules without extra setup
The task database and dashboardOne board shows both pipelines; each session must filter by lane
The subscription's usage limitsClaude's Pro and Max limits are shared across Claude and Claude Code (Claude Help, checked 2026-10-01), so two busy sessions draw from one allowance
The git historyBoth commit to the same branch

What broke: shared logs and commits

Nothing went down, but one thing got messy right away. My dashboard writes a JSON copy of its database after every change, into one shared export folder. When the blog session checked what had changed, it saw export files modified by the video session's work.

A habit like "commit everything that changed" would have filed the other session's changes under a blog commit message. The fix is the lane rule applied to git: each session stages only its own task folders and files by name, and leaves the shared export for whoever owns the change.

Working in ChatGPT on a laptop

Photo: Matheus Bertelli on Pexels

How the sessions talk to each other

Short answer: they don't, directly. Neither session can see the other's conversation. They coordinate through the things they share:

  • the task board (each task has an owner and a status),
  • the dashboard's messenger, where the agents and I leave notes per task,
  • the files and commit history in the folder.

That's slower than live chat between agents, but every handoff leaves a record I can read.

A checklist before you open session #2

  • ☐ Does the new work have a clear lane that doesn't overlap the first session's?
  • ☐ Does state outside git (databases, tokens) mean you need the same folder rather than an isolated copy?
  • ☐ Did you put the lane rule in the first message of each session, and save a copy in a file both can read?
  • ☐ Will each session commit only its own files?
  • ☐ Are you ready for both sessions to share one usage limit?

Comments

Post a Comment

Popular posts from this blog

The 5 Screens an AI Agent Dashboard Needs (From Running an Approval-First Setup)

Image
I run a handful of AI agents for my side projects, and I don't write code. That combination creates one hard requirement: nothing goes public until I click Approve , and I should never need a terminal to do it. So on day one I had a small local web dashboard built (by an AI coding agent) with exactly that job. It runs only on my own PC. After a day of real use, including one rejected approval and a couple of mistakes, here are the five screens I'd insist on again, and what each one taught me. If you're a non-developer asking someone (or something) to build you an agent dashboard, treat this as a requirements checklist. 1. A task board with honest statuses Every instruction becomes a numbered task that moves through seven states: waiting, in meeting, working, in review, awaiting approval, done, failed . Why seven and not "to do / done"? Because the dangerous moment is the gap between "the agent thinks it's finished" and "a human has signed ...
Read more

Day One With an AI Agent Team: 6 Things That Broke (and the Fixes)

Image
On September 26, 2026, I set up a small team of AI agents on my PC. Claude does research, writing, and review. OpenAI's Codex and Google's Antigravity make images. I don't write code. Most setup guides show the happy path. This post is the other half: the six things that actually went wrong on day one, why they happened, and what I changed. Every item comes from my own failure log, so none of this is hypothetical. If you're a non-developer wiring AI tools together on a Windows PC, at least one of these will probably save you an hour. 1. The image tool had no quota left before I started What happened. My first image request to Codex failed instantly with "You've hit your usage limit," and a time when it would reset. Why. Codex usage on a ChatGPT plan is shared across the places you use it. OpenAI says the same limits apply in the app, the CLI, the IDE extension, and the cloud ( OpenAI Help Center , checked 2026-09-26). I hadn't run anything thro...
Read more

Automating Blog Posts With AI, Safely: What the Blogger API Can't Do (and My Workarounds)

Image
For my first few posts, publishing meant copying HTML from my dashboard and pasting it into Blogger by hand. It worked, but it was slow and error-prone: once, the wrong page's text ended up on my Privacy Policy. So I automated it with Google's official Blogger API. Now I do one thing: click Approve . Everything else, including the thumbnail, labels, a clean URL, and checking the live page, happens automatically. Here's how it works, the two things the API can't do, and one mistake that briefly took my blogs offline. Rule one: no approval, no post Before anything else, the publishing script checks my dashboard's approval records for that task. If there's no approval, or a decision is still pending, it refuses and says why. It also refuses to publish the same draft twice. I tested both refusals before the first real run. If you automate publishing, build the "no" path first. Photo: Andrew Neel on Pexels The one-time setup (about 15 minutes) Cre...
Read more