Big Idea

The argument in one line.

Claude Code Projects turns Claude into the project manager: you talk to one orchestrator that runs persistent, memory-sharing cloud sessions on your subscription, which beats babysitting terminals for long-running, splittable work.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…

You run a complicated build that splits into parts and you're tired of managing a pile of terminals and agents by hand.
You want Claude work to keep running overnight or while your laptop is shut, without paying per-token API costs.
You already live in a GitHub repo workflow and want workers that open their own pull requests.

SKIP IF…

Your work is quick, single-session tasks that finish in one chat.
Your files live only on your machine and never touch a GitHub repo.
You're on a Team or Enterprise plan, which the video says isn't supported yet.

TL;DR

The full version, fast.

Claude Code Projects is a beta feature, gated by a quick waitlist, where one coordinator agent runs full Claude sessions in Anthropic's cloud. Those sessions share context and memory, persist, own their pull requests, and draw on your subscription instead of API billing. It fits long-running, splittable work tied to a GitHub repo, not quick tasks or local-only files. The demo shows the coordinator splitting a content-research dashboard into five parallel threads, surfacing blockers, and picking up a secret you add to the cloud environment.

Free for members

Chat with this breakdown — free.

Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.

Create a free account →

Chapters

Where the time goes.

00:00 – 00:36

01 · Cold open

Pitch: one agent controls a fleet of 24/7 sessions that share context and memory.

00:36 – 03:42

02 · What Projects does

Claude becomes the project manager, runs in Anthropic's cloud on your usage, and access comes through a waitlist that now clears fast.

03:42 – 05:29

03 · When to use it

Use it for backlogs and splittable builds; skip quick tasks, local-file work and team plans.

05:29 – 06:51

04 · Projects versus sub-agents

Persistence, separate context windows, owned pull requests, and who does the managing.

06:51 – 12:44

05 · Setup and project settings

Create a project, pick coordinator and thread models, set instructions, memory, repo, cloud environment and local folders.

12:44 – 17:37

06 · Demo: five threads and blockers

The prompt splits into five parallel threads; two ready for review, three blocked; adding a secret to the cloud environment clears a blocker.

17:37 – 18:49

07 · Result and wrap-up

The dashboard, a local scheduled routine, why persistence is the real value, and the usage tradeoff.

Atomic Insights

Lines worth screenshotting.

With Claude Code Projects, you stop managing sessions yourself: one coordinator manages them and keeps every worker on the same page.
Projects workers are not sub-agents. They are full sessions with their own context windows that persist, own pull requests, and can be messaged directly.
Cloud sessions in Projects draw from your subscription usage, not API credits, which removes the usual cost penalty of running Claude in the cloud.
A sub-agent returns a summary and fills your main session's context. A project worker keeps its own context and never shuts off.
The coordinator defaults to Opus 5.5 on low effort because it only divides up work. Spend your effort budget on the thread model instead.
Projects can burn usage fast, since five to fifteen workers can run around the clock. Watch the usage tab before you scale up.
Keep project instructions short. Like any agents file, less is more, and every thread reads it.
Because the sessions run in the cloud, a project needs a GitHub repo where workers can push work and send pull requests.
Work that needs your local files does not belong in a cloud project, though you can pre-approve a local folder for specific sessions.
Auto-continue lets a project resume on its own after your weekly usage limit resets.
Rule of thumb: separate pieces of work get a project, one thing gets a session.

Takeaway

Let one agent run many cloud sessions

WHAT TO LEARN

Claude Code Projects pays off when work is long-running, splittable and repo-based, because one coordinator manages persistent sessions that never shut off.

02What Projects does

Projects runs full Claude sessions in the cloud on your subscription usage, so you get always-on work without the per-token API bill.
Applying through the waitlist currently gets access within about a day, and the feature shows up as Projects (beta) in the desktop app.

03When to use it

Pick a project over a session when work splits into parts or keeps getting fed; pick a plain session for one quick task.
Skip it for work tied to local-only files, tangled code with shared files, or Team and Enterprise plans.

04Projects versus sub-agents

Workers differ from sub-agents because they keep their own context, persist, own pull requests, and accept direct messages.

05Setup and project settings

Keep the coordinator on low effort and put medium effort on the threads, since the coordinator only divides up the work.
Watch usage closely, since many workers running around the clock can drain a plan fast; auto-continue resumes after a reset.
Connect a GitHub repo, and add a cloud environment with secrets when threads need API keys.

06Demo: five threads and blockers

Blockers surface per thread, and you can clear them through the coordinator instead of opening every session.

07Result and wrap-up

The lasting value is a living cloud project whose threads keep sharing context, not the first-pass output.

Glossary

Terms worth knowing.

Claude Code Projects: A beta feature in the Claude desktop app where one coordinator agent runs multiple persistent cloud Claude sessions on a shared goal, memory and GitHub repo.
Coordinator: The top-level agent you talk to in a project. It splits work into threads and routes your messages to the right worker instead of doing the work itself.
Thread: A persistent worker session under the coordinator with its own context window, its own pull requests, and a direct chat you can steer.
Cloud environment: The settings a project's cloud sessions run in, including network access level and secrets such as API keys.
Debug access: A project setting that lets one project inspect another so a fresh set of eyes can find problems like a poisoned context.

Resources

Things they pointed at.

02:30linkClaude Projects waitlist ↗

18:05linkClaude Projects docs ↗

03:20productChase AI Plus Claude Code masterclass ↗

Quotables

Lines you could clip.

00:05

“You just talk to one agent and it controls a fleet of fully-fledged Claude sessions that run 24-7.”

One-sentence promise of the whole feature→ IG reel cold open↗ Tweet quote

01:20

“It's not a cloud situation where now we're paying API costs. This is pulling from your max plan.”

Cost angle most people care about→ TikTok hook↗ Tweet quote

18:20

“The real sell isn't the singular thing it created. It's the fact that this is now living in the cloud and the threads keep sharing and maintaining context.”

States the actual value over the demo→ newsletter pull-quote↗ Tweet quote

The Script

Word for word.

Read-along

Don't just watch it. Burn it in.

See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.

Claude has a secret feature hidden behind a waitlist that turns it into a master of cloud orchestration. You just talk to one agent and it controls a fleet of fully -fledged Claude sessions that run 24 -7. They share context, they share memory, and they never shut off.

It's called Claude Projects, and it's the ultimate hack if you're someone who is working on complicated applications and you need to do so outside of your local environment. So today I'm going to show you how to get access to this feature, how it actually works, and then we're going to run through a quick demo so you can start using it right away.

Now before I show you how to get access to Cloud Code Projects, let me explain why you should even care about this at all. The big sell is that with Cloud Projects, Cloud is now the project manager. You talk to one agent and it controls everything.

Versus... the normal setup. When we are normally inside of Claude, when you're just sitting there on the desktop app or inside of the terminal, you are in charge of everything.

That means you are in charge of managing all of these different terminals and all of these different agents, and you're also in charge of keeping them all on the same page. On top of that, unless you are running something either in the cloud or you have some sort of like Mac mini setup that's always on, when you start shutting things down, whether that's your laptop or your computer, your machine just literally is not on, your work stops.

Compare that to Cloud Project, which runs in Anthropix Cloud and importantly, also uses your usage. It's not a cloud situation where now we're paying API costs. This is pulling from your max plan.

So we get pretty much all the advantages of the cloud with all the advantages of the subscription plan, which hasn't always been in the case when we talk about Anthropix in the cloud. So instead of you being in charge of managing all of these separate threads, guess who does it? Well, Claude does it.

We just talk to one orchestrator and it then does said orchestration. On top of that, these are not sub -agents. It's controlling full -blown Claude sessions that also share context and memory between each other and the orchestrator.

And because it's in the cloud, it's always going to run no matter where we are, whether that's overnight, I can check in remotely from my phone. So... If you are someone who deals with complicated projects that you want access to 24 -7, no matter where you are at, well, this cloud project makes total sense.

It makes it super easy to manage multiple agents. So how do you get access? Well, right now you need to submit an application on this page right here, and I'll put a link to it in the description.

It puts you on a wait list. Now, when this first came out a couple months ago, when it was like very limited release, it was kind of hard to get off the waitlist. But as of now, if you put your name on this waitlist, you're going to get access to it probably within 24 hours.

I know it was less than 24 hours for me. And a couple weeks ago, they let everybody off the waitlist. So there's really no excuse not to get this up and running on your cloud.

And once you get access, it will populate inside of your desktop app over here on the left. And it says projects beta. Remember, we used to have projects.

back in the day especially with the chat app where those were essentially just like folders and we have conversations that were all about some specific subject this is much different than that like i said this is running in the cloud this is all about multi -agent orchestration so before we jump into the next part a quick word from today's sponsor me i just released an updated cloud code masterclass and it is the number one way to go from zero to ai dev no matter your technical background so it doesn't matter where you are starting from your technical journey it could be literally zero I will bring you up to speed on the most powerful AI tools in the game today.

I update this every single week, so if you want to get your hands on this, you can find it inside of Chase AI+. I will put a link in the pinned comment. So if Cloud Code Projects is all about multi -agent orchestration in the cloud, when should we actually use this and when shouldn't we use this?

Well... Starting with the don'ts, if we're just talking about quick tasks that you can just knock out in a single chat session, then obviously, no, we don't need to use Cloud Code Projects for this. We also need to be wary if we're talking about work that needs to be done on our machine.

If it's manipulating files and folders that quite literally live on our computer, it's not something that can be in the cloud, it's not something that can be in the GitHub repo, then we aren't going to want to use Cloud Code Projects because this is all, again, operating inside of the cloud. Ideally, it's connected to some sort of GitHub repo.

Now... caveat i will say you do have the ability to spawn sessions you know that your orchestrator controls that do live on your local machine but that isn't what it's really built for also this isn't something that's on team or enterprise yet so when we talk about you know team projects this isn't what it's really built for either So what should we use it for?

Well, we want to use it if we have some sort of project or an area that we're going to continue to feed. If it's a complicated build that we can split into different parts, think of all these sub -sessions controlled by the orchestrator being in charge of certain sections of a task. And Boris Cherny, the guy who actually created Cloud Code, pretty much said he stopped managing sessions inside of his Cloud Code desktop app and really just uses...

projects as like this backlog this dumping ground if any ideas or tasks that come into his head he dumps it into projects because remember projects and all of the sub sessions beneath it that the orchestrator controls all share memory all share context so if it's complicated it's long running and is it something that can work best in a cloud environment we're going to want to use projects otherwise we probably don't need it so just keep that in mind and lastly before we jump into the demo and we actually see projects in action what is really the difference between projects and sub agents we keep talking about we have these sub sessions with context and memory how is that any different than just one session with sub agents being spawned a lot of it has to do with persistence when we spawn some sort of sub agent in a normal cloud code session it's going to do whatever it's been tasked to do it's going to create a summary of all the work it's done it's going to send it back to that main session you're working in and it's just going to continue to fill up the context of that one session versus in a project

that single session, that orchestrator up top that we're talking to, these quote -unquote sub -agents it's creating, and they're not really sub -agents, they're full sessions because they have their own context window. Furthermore, they're persistent. They'll own their own pull requests.

And because these workers are persistent and they maintain context throughout, we can also message them as much as we want. I can actually go in there and talk to worker one or worker two as if I'm talking to the primary orchestrator. So there's a lot of things listed here in terms of the differences.

The biggest ones you need to remember is in terms of who is managing it. Normally, you do all the management. In terms of projects, Claude is managing it all and dealing out the work accordingly.

Secondly, persistence, projects, those sub workers are never shutting off. And so ultimately, sessions with the sub agents is best when we're just doing one task at a time. And these projects are great if we have a stream of work, we just don't want to babysit.

so now let's hop into the demo if you go into your cloud desktop on the top left you will see something that says projects beta we simply click on this and you will see this page pop up now if you have projects that have been sitting in cloud for the past several years potentially back from like just the chat days and they were just projects were basically just folders it's going to ask you hey do you want to delete these or you want to archive all these you can just archive all of them until you get to you know essentially a clean slate so we are going to click new projects Then we're going to give it a name.

We will call this project demo. You have the ability to give it a goal. So some sort of overarching thing it's trying to achieve.

In this case, we will leave it blank. And I also have the ability to give it context. So if I already had, you know, certain documentation or files and folders that I want this project to already sort of bring on board, I can do that here as well.

Then I'm just going to hit create project. And we're going to get a page that looks like this. Now, the first thing I want you to do.

is come up here to the top where we have this little gear icon and we're going to take a look at our project settings so again we can edit the goal at any point in time we also have the models that we can choose from and the big things we want to pay attention to are the coordinator model and the thread model so these threads are essentially the workers that the coordinator spins up in reality threads themselves can have multiple workers but just think of it as like coordinator and then threads slash workers underneath it.

So the coordinator model is going to, by default, go to Opus 5 .5 on a low effort setting. You might be like, hey, do I really want it on a low effort setting? Why don't I want to put it on at least medium or high if it's doing all these things?

Well, in reality, it's not really the one doing the work. It's just the one coordinating and divvying out the work. So not really an issue to keep it on low.

If for whatever reason you feel like it's not doing a good job of coordinating, you can always bump this up to medium or high. But if usage... Starts to become a problem because usage is something that projects can burn up pretty quickly because we could be, you know, similar to dynamic workflows.

You could have 5, 10, 15 sessions or threads or workers all doing stuff at the same time 24 -7. So we want to keep an eye on that. Thread model, Opus 5 .5 on medium should be totally fine for you.

But if you even want to switch this to something like sauna, that works too. We also have the ability to have it auto -continue when usage limits reset. So let's say you burn through all 100%.

When the week rolls over and we get a fresh reset, well, it's just going to pick up from where it left off. We also can pause the project, restart, clod, or archive the project from here too. Then we have memory.

So the big one we want to pay attention to is project instructions. This is similar to agents or clod .md. This can be very powerful because they're pretty much going to read this in every single thread.

But, you know... I wouldn't blow this. Like similar to how we approach agents .md these days, less is more.

If there's something where you're like, I really want it to always do the specific thing in a specific way every single thread, then maybe this makes sense to fill it up here. Otherwise, if you're not sure, just leave it blank. It also has auto -memory, and that works pretty much the same way as auto -memory inside of Cloud Code.

So as you tell it to do things throughout your sessions, it has its own little scratch pad where if it thinks it's something it should remember, it will just automatically add essentially a little sticky note to its... memory next we have the environment tab so the first thing we need to do is designate a project repository all these workers all these sessions are going to be in the cloud and they need a github repo where they can communicate and send their prs so to do that we just need to select a particular github repo mine is going to be ai creator radar because i'm creating this thing that goes out on the web goes out on twitter sees trending ai news and then turns that into a dashboard which says hey here's things that haven't been talked about on youtube here's how we could turn them into short form here's how we turn into carousels that type of thing so you can either create this manually inside of github or just hop into a normal cloud code session and say can you create a github repository for me that i'm going to use inside my project

Next, we have the cloud environment. So if you just keep this as is where it says select the cloud environment, it's just going to use the default anthropic cloud environment. You also have the ability to add a cloud environment.

So I could give it some sort of name. I can define the network access everywhere from none to full. And then I can start uploading secrets and environmental variables.

So if I am doing something that requires, you know, API keys and secrets, I'm going to need to add a cloud environment for it to be able to work there. Now, we are going to leave it on select the cloud environment, and you will see later it's going to prompt us to essentially change our cloud environment to add some keys, and I'll walk you through that.

Next, we have some local settings. Let cloud use your device. Yes.

Pre -approved folders. If you have a session or a worker that is going to need to work on your local computer or access it, that's something we can do. By default, it's always going to be in the cloud, but it can actually go on your machine, and so to do that, you just need to select this, do yes.

And then we want to give it some sort of folder it can work in. In this case, I set my radar -local folder. Then we have a usage tab followed by debug access.

Let's say you have two projects, project A, project B, totally doing different things. You can let project B take a look at project A and vice versa to help debug because perhaps you're running into problems where maybe project A can't debug on its own. Maybe we have like some issue with the memory, like something is poisoning its context.

Well, how about we get a fresh set of eyes to take a look at what's going on? We can do that. by simply giving it access via the debug access panel probably won't use this a lot but no it's there so let's go ahead and give it our prompt so we're saying we're creating the ai creator radar we've given it kind of an end state we're looking for again i want it to be able to scrape things off of x and then i wanted to be able to take a look at youtube see what's working see what hasn't been touched on see where there's some sort of gaps and i wanted to do it on a dashboard where i can just see it all in one place so let's go ahead and put it to work and see what happens and as this is working you'll kind of see this has begun to populate over here are the threads so coordinator main page right here i can talk to everything and like do everything from this one window but if i want to kind of like click inside of here and see what's going on an individual thread or even worker level i have that ability too so after letting it run for about 25 minutes here is where we are at first of all after we gave it the prompt it split

into five parallel threads and each thread was given some piece of the pie that it needed to execute on we had something working on the dashboard and hosting one working on the x scanner one on youtube coverage one on content packs and one on cloud code news if we go over here to the right on the threads we can see i have two that are now ready for review i have three that have some sort of blocker that require me to come in and either make some sort of decision or give it some more information and then we also have the idle thread now at any point in time i can click on any of these threads and see what this particular worker is doing any more granular level i can click here and see the blockers and so it's really easy me to go really easy for me to go into each individual session and actually talk to it because as you see when i click into here i do have an ability to steer that thread i also can view these threads from the orchestrator panel right here in the middle so i just click view thread and it pops up also i can just handle everything

through here through the orchestrator so instead of going into each individual thread like i would with sessions you know when i'm sort of just normally operating inside of cloud code i can just talk to the single agent and it will handle it all for me and the whole time they're operating inside of my github repo which we see right here if i hop into the pull requests we can see what all these individual threads individual workers are actually sending so let's handle some of these blockers the first one is related to the cloud environment remember i mentioned how earlier if we need api keys as part of our project we need to edit the actual environment well let's go ahead and do that so go up to project settings go to environment and we're going to add a cloud environment i'm going to call this demo environment for network access we'll throw it to full and now we need to add a network secret and so i put in all the correct information and then i just hit add and now it's working in a cloud environment that has my particular secret so then i can say hey i just updated the cloud environment with the youtube key as a network secret and the coordinator is going to send that information to that particular thread and now i'm just going to repeat this process for the rest of the blockers and that's kind of how you would run this

right so we got rid of the blockers and right now it's finishing up one thing that's kind of cool is like i said you can tell certain sessions that you want it to be local we did that with the session that had to do with scanning x because x was actually logged into my account on my cookies so it used the in -app browser inside of cloud code desktop to do that and so it actually built a local scheduled routine so you can see here on the scheduler it created this routine it's inside of that folder that we designated earlier is like the local folder it can play around with and you can see kind of the instructions it gave it and so here's a look at the dashboard it created it shows me which stories are trending on twitter but really have no coverage in the youtube space for each of these like hey claude writes about 90 of claude code's code it shows me potential content pack what the video could look like what a short could look like what a carousel could look like as well as what an x post could also look like it also gives me kind of a breakdown of what's going on with x so like the top story right now is deep seek version 4 .1 flash and glm 5 .3 flash go free and open code so it shows me like the post per hour the views i can actually click on the posts themselves and it will bring it up and it also gives me more stats about my channel as well as like some

cloud code specific news lastly at the bottom i can see kind of a refresh log and so overall this is pretty solid work considering it was kind of a one shot and this is just the first pass that cloud code projects has done on this and the real sell for projects at this point isn't the singular thing it created it's the fact that this is now something that is living in the cloud i can access it at any time and i have all of these threads that still are sharing and maintaining context.

So I can work on this whenever I want. I can constantly feed it new information. It can run even when my computer is shut down and I can continue to add more and more complexity to it all through a single orchestrator.

So while I do have the ability to, you know, click on individual threads and handle it one at a time, I can just talk to one agent from any place and have it control everything. the last thing i'll mention and i will also link this in the description is the documentation on cloud code projects it is very extensive so if there's anything you feel like i didn't cover enough or you want to go deeper and i highly suggest you take a look at the documentation but besides that that's all i have for you today highly suggest if nothing else you hop on that waitlist because you will get access very very quick i think this is a really cool feature inside of cloud code kind of niche really depends on the type of projects you're working on and if you're someone who really does need that 24 7 access if that's not you no problem but if it is it's great because you know in the past when we've had these sort of anthropic cloud options they've always been draining from the api and now it just comes straight from usage which for many of us i feel like we have a lot of usage remaining like opus 5 .5 has been pretty

token efficient, especially compared to something like Fable 5 .1. So having the option to go into the cloud, stay on usage, and always have complicated projects running is a great value add. So as always, let me know what you thought.

Make sure to check out Chase AI Plus if you want to get your hands on my Cloud Code Masterclass. And besides that, I'll see you around.

The Hook

The bait, then the rug-pull.

A secret feature behind a waitlist, and the waitlist now clears in about a day. Chase AI shows what Claude Code Projects does, who should use it, and builds a five-thread dashboard with it.

Frameworks