I don't have to babysit AI sessions anymore.

Hi everyone,

Since Opus 4.5 came out last November, I've been babysitting AI sessions.

Several Claude Code sessions open at once. I check one, nudge another, then jump back to the first. It works, and it gets a lot done. It also drains your brain. By the end of the day I felt more tired than any normal workday made me.

A week ago I got access to the new Projects in Claude Code. It's the first tool I've used that tries to take that job off me.

You talk to one main thread. That's the coordinator.

You tell it what needs doing, and it starts a session for each task. Anthropic calls these threads. Each one is a full Claude Code session in the cloud, on its own branch. A thread can even split its task further with sub-agents.

You can jump into any thread and steer it yourself. Or you stay in the main thread and let the coordinator do the checking. Anthropic's launch post says it "keeps working after you step away from your computer."

This is the picture I've wanted for a year. Agents babysit agents, so we don't have to.

For development work it was really good. I ran it on our agent OS and on our app. Each is one GitHub repo with a clear scope. So the threads knew where to work.

I also ran it on a client project, and that went well too.

Business work is where it fell short for me. When you set up a project, you pick the folders it works in. My business work moves around, though. A sales task can need a customer folder, a lead-gen folder, and sometimes marketing too.

When a folder wasn't in the project, the work got weaker. So I still ended up checking on threads.

Some of that is on me, because I didn't add every folder I needed. Anthropic's docs also say a thread can add a repo when its task needs one.

For now, my business work stays in Claude Code in the terminal. I use it in Warp. There I jump from folder to folder and pull in the context each one has.

OpenAI has the same idea. Its always-on agents are called dots. Since last week, OpenAI says a dot "can now start work in Codex and follow up on existing threads."

Last week I wrote about Copilot. On September 25, Microsoft announced a new Copilot with Home, Code and Autopilot. Microsoft calls Autopilot "your digital teammate." It picks "a project back up days later, without waiting for a prompt."

My bet is Autopilot ends up running sessions like this too.

So the promise is huge. It hasn't fully landed yet. I do believe it's coming, though, and it will show up in every tool you use.

That matters for your team. Running all those sessions at once wore me out. Soon the coordinator carries that load. We get more done, and we save some brain cells too.

On Claude Pro or Max, open claude.ai/code. Look for Projects in the sidebar. It's a beta and still rolling out, so it may not be there yet. Team and Enterprise plans don't have it yet.

Pick one small job with a few parts, in one GitHub repo. Add the repo, then hand the job to the coordinator.

Watch how it splits the work into threads. Open one thread and steer it once.

One warning from Anthropic. Each thread is a full session, so projects "can reach usage limits faster." Start small.

What's the first job you'd hand to a coordinator?

Hit reply and tell me. I read every reply.

Thanks for reading.

Tim