The SECRET Claude Code Upgrade That Has Everyone Going Nuts (Full Guide) — Transcript
Full transcript
- 0:00Claude has a secret feature hidden
- 0:02behind a weight list that turns it into
- 0:04a master of cloud orchestration. You
- 0:07just talk to one agent and it controls a
- 0:09fleet of fullyfledged Claude sessions
- 0:12that run 24/7. They share context, they
- 0:16share memory and they never shut off.
- 0:18It's called Claude projects and it's the
- 0:21ultimate hack if you're someone who is
- 0:23working on complicated applications and
- 0:25you need to do so outside of your local
- 0:27environment. So today I'm going to show
- 0:29you how to get access to this feature,
- 0:31how it actually works, and then we're
- 0:32going to run through a quick demo so you
- 0:34can start using it right away. Now
- 0:36before I show you how to get access to
- 0:37Cloud Code projects, let me explain why
- 0:39you should even care about this at all.
- 0:41The big cell is that with Claude
- 0:43projects, Claude is now the project
- 0:46manager. You talk to one agent and it
- 0:48controls everything versus the normal
- 0:51setup. When we are normally inside of
- 0:53claude, when you're just sitting there
- 0:54on the desktop app or inside of the
- 0:56terminal, you are in charge of
- 0:59everything. And that means you are in
- 1:01charge of managing all of these
- 1:03different terminals and all of these
- 1:04different agents and you're also in
- 1:06charge of keeping them all on the same
- 1:07page. On top of that, unless you are
- 1:10running something either in the cloud or
- 1:12you have some sort of like Mac Mini
- 1:13setup that's always on, when you start
- 1:16shutting things down, whether that's
- 1:17your laptop or your computer, your
- 1:19machine just literally is not on, your
- 1:21work stops. Compare that to clawed
- 1:23projects, which runs in Anthropics Cloud
- 1:26and importantly also uses your usage.
- 1:30It's not a cloud situation where now
- 1:32we're paying API cost. This is pulling
- 1:34from your max plan. So we get pretty
- 1:36much all the advantages of the cloud
- 1:37with all the advantages of the
- 1:39subscription plan which hasn't always
- 1:40been the case when we talk about
- 1:42Enthropic in the cloud. So instead of
- 1:44you being in charge of managing all of
- 1:46these separate threads, guess who does
- 1:48it? Well, Claude does it. We just talk
- 1:51to one orchestrator and it then does
- 1:53said orchestration. On top of that,
- 1:56these are not sub agents. It's
- 1:58controlling full-blown Claude sessions
- 2:00that also share context and memory
- 2:02between each other and the orchestrator.
- 2:04And because it's in the cloud, it's
- 2:06always going to run no matter where we
- 2:08are, whether that's overnight. I can
- 2:10check in remotely from my phone. So, if
- 2:13you are someone who deals with
- 2:14complicated projects that you want
- 2:16access to 24/7, no matter where you are
- 2:18at, well, this Cloud project makes total
- 2:22sense. It makes it super easy to manage
- 2:24multiple agents. So, how do you get
- 2:26access? Well, right now, you need to
- 2:28submit an application on this page right
- 2:30here, and I will put a link to it in the
- 2:33description. It puts you on a wait list.
- 2:36Now, when this first came out a couple
- 2:37months ago and it was like very limited
- 2:39release, it was kind of hard to get off
- 2:40the wait list, but as of now, if you put
- 2:42your name in on this wait list, you're
- 2:43going to get access to it probably
- 2:44within 24 hours. I know it was less than
- 2:4624 hours for me. And a couple weeks ago,
- 2:48they let everybody off the wait list.
- 2:50So, there's really no excuse not to get
- 2:52this up and running on your Claude. Now,
- 2:55once you get access, it will populate
- 2:57inside of your desktop app over here on
- 2:58the left, and it says projects beta.
- 3:01Remember, we used to have projects back
- 3:03in the day, especially with the chat
- 3:04app, where those were essentially just
- 3:05like folders and you'd have
- 3:06conversations that were all about some
- 3:09specific subject. This is much different
- 3:10than that. Like I said, this is running
- 3:12in the cloud. This is all about
- 3:13multi-agent orchestration. So, before we
- 3:15jump into the next part, a quick word
- 3:17from today's sponsor, me. I just
- 3:19released an updated cloud code
- 3:21masterass, and it is the number one way
- 3:23to go from zero to AI dev, no matter
- 3:25your technical background. So, it
- 3:27doesn't matter where you are starting
- 3:28from your technical journey. It could be
- 3:30literally zero. I will bring you up to
- 3:32speed on the most powerful AI tools in
- 3:34the game today. I update this every
- 3:36single week. So, if you want to get your
- 3:38hands on this, you can find it inside of
- 3:39Chase AI Plus. I will put a link in the
- 3:42pin comment. So, if Cloud Code Projects
- 3:43is all about multi-agent orchestration
- 3:45in the cloud, when should we actually
- 3:47use this and when shouldn't we use this?
- 3:49Well, starting with the don'ts, if we're
- 3:50just talking about quick tasks that you
- 3:52can just knock out in a single chat
- 3:54session, then obviously no. We don't
- 3:56need to use cloud code projects for
- 3:57this. We also need to be wary if we're
- 3:59talking about work that needs to be done
- 4:00on our machine. If it's manipulating
- 4:03files and folders that quite literally
- 4:04live on our computer, it's not something
- 4:06that can be in the cloud. It's not
- 4:07something that can be in a GitHub repo,
- 4:09then we aren't going to want to use
- 4:10cloud code projects because this is all
- 4:12again operating inside of the cloud.
- 4:14Ideally, it's connected to some sort of
- 4:16GitHub repo. Now caveat I will say you
- 4:19do have the ability to spawn sessions
- 4:22you know that your orchestrator controls
- 4:23that do live on your local machine but
- 4:26that isn't what it's really built for.
- 4:28Also this isn't something that's on team
- 4:30or enterprise yet. So when we talk about
- 4:32you know team projects this isn't what
- 4:35it's really built for either. So what
- 4:37should we use it for? Well we want to
- 4:39use it if we have some sort of project
- 4:40or an area that we're going to continue
- 4:42to feed. If it's a complicated build
- 4:44that we can split into different parts.
- 4:46Think of all these sub sessions
- 4:47controlled by the orchestrator being in
- 4:49charge of certain sections of a task.
- 4:52And Boris Churnney, the guy who actually
- 4:54created Cloud Code, pretty much said he
- 4:56stopped managing sessions inside of his
- 4:58Cloud Code desktop app and really just
- 5:00uses projects as like this backlog, this
- 5:02dumping ground of any ideas or tasks
- 5:04that come into his head. He dumps it
- 5:05into projects because remember projects
- 5:08and all of the subsessions beneath it
- 5:10that the orchestrator controls all share
- 5:12memory, all share context. So if it's
- 5:14complicated, it's long running, and is
- 5:16it something that can work best in a
- 5:17cloud environment, we're going to want
- 5:18to use projects. Otherwise, we probably
- 5:21don't need it. So just keep that in
- 5:23mind. And lastly, before we jump into
- 5:24the demo and we actually see projects in
- 5:26action, what is really the difference
- 5:28between projects and sub aents? We keep
- 5:30talking about we have these sub sessions
- 5:32with context and memory. How is that any
- 5:34different than just one session with sub
- 5:36aents being spawned? Well, a lot of it
- 5:38has to do with persistence. When we
- 5:41spawn some sort of sub agent in a normal
- 5:43cloud code session, it's going to do
- 5:45whatever it's been tasked to do. It's
- 5:47going to create a summary of all the
- 5:48work it's done. It's going to send it
- 5:50back to that main session you're working
- 5:53in, and it's just going to continue to
- 5:54fill up the context of that one session.
- 5:56Versus in a project, that single
- 5:59session, that orchestrator up top that
- 6:01we're talking to, these quote unquote
- 6:04sub aents it's creating. And they're not
- 6:05really sub agents. They're full sessions
- 6:08because they have their own context
- 6:10window. Furthermore, they're persistent.
- 6:12They'll own their own pull requests. And
- 6:15because these workers are persistent and
- 6:17they maintain context throughout, we can
- 6:19also message them as much as we want. I
- 6:21can actually go in there and talk to
- 6:22worker one or worker two as if I'm
- 6:24talking to the primary orchestrator. So,
- 6:26there's a lot of things listed here in
- 6:28terms of the differences. The biggest
- 6:29ones you need to remember is in terms of
- 6:31who is managing it. Normally you do all
- 6:33the management in terms of projects.
- 6:35Claude is managing it all and dealing
- 6:37out the work accordingly. Secondly,
- 6:40persistence projects. Those sub workers
- 6:42are never shutting off. And so
- 6:44ultimately sessions with the sub aents
- 6:45is best when we're just doing one task
- 6:47at a time. And these projects are great
- 6:49if we have a stream of work we just
- 6:50don't want to babysit. So now let's hop
- 6:52into the demo. If you go into your cloud
- 6:53desktop on the top left, you will see
- 6:56something that says projects beta. We
- 6:58simply click on this and you will see
- 6:59this page pop up. Now, if you have
- 7:01projects that have been sitting in
- 7:02Claude for the past several years,
- 7:04potentially back from like just the chat
- 7:06days and they were just projects were
- 7:08basically just folders, it's going to
- 7:09ask you, hey, do you want to delete
- 7:10these or you want to archive all these?
- 7:12You can just archive all of them till
- 7:14you get to, you know, essentially a
- 7:15clean slate. So, we are going to click
- 7:19new projects.
- 7:21Then, we're going to give it a name. We
- 7:22will call this projects demo. You have
- 7:25the ability to give it a goal. So some
- 7:27sort of overarching thing it's trying to
- 7:30achieve. In this case, we will leave it
- 7:32blank. And I also have the ability to
- 7:33give it context. So if I already had,
- 7:36you know, certain documentation
- 7:38or files and folders that I want this
- 7:40project to already sort of bring on
- 7:42board, I can do that here as well. Then
- 7:44I'm just going to hit create project.
- 7:45And we're going to get a page that looks
- 7:47like this. Now the first thing I want
- 7:49you to do is come up here to the top
- 7:51where we have this little gear icon. And
- 7:53we're going to take a look at our
- 7:54project settings. So again, we can edit
- 7:57the goal at any point in time. We also
- 7:59have the models that we can choose from.
- 8:01And the big things we want to pay
- 8:03attention to are the coordinator model
- 8:04and the thread model. So the threads are
- 8:07essentially the workers that the
- 8:10coordinator spins up. In reality,
- 8:13threads themselves can have multiple
- 8:15workers, but just think of it as like
- 8:17coordinator and then threads/workers
- 8:20underneath it. So, the coordinator model
- 8:22is going to by default go to Opus 5.5 on
- 8:25a low effort setting. You might be like,
- 8:26"Hey, do I really want it on a low
- 8:27effort setting? What do I want to put it
- 8:28on at least like medium or high if it's
- 8:30doing all these things?" Well, in
- 8:31reality, it's not really the one doing
- 8:32the work. It's just the one coordinating
- 8:35and diving out the work. So, not really
- 8:37an issue to keep it on low. If for
- 8:39whatever reason you feel like it's not
- 8:40doing a good job of coordinating, you
- 8:42could always bump this up to medium or
- 8:43high. But if usage starts to become a
- 8:47problem because usage is something that
- 8:49projects can burn up pretty quickly
- 8:51because we could be you know similar to
- 8:53dynamic workflows you could have 5 10 15
- 8:57sessions or threads or workers all doing
- 8:59stuff at the same time 24/7. So we want
- 9:02to keep an eye on that thread model opus
- 9:055.5 on medium should be totally fine for
- 9:08you. But if you even want to switch this
- 9:09to something like sauna that works too.
- 9:11We also have the ability to have it auto
- 9:12continue when usage limits resets. So
- 9:14let's say you burn through all 100%.
- 9:16When the week rolls over and we get a
- 9:18fresh reset, well, it's just going to
- 9:19pick up from where it left off. We also
- 9:22can pause the project, restart Claude or
- 9:24archive the project from here, too. Then
- 9:27we have memory. So the big one we want
- 9:28to pay attention to is project
- 9:30instructions. This is similar to agents
- 9:31or claw.md.
- 9:33This can be very powerful because
- 9:34they're pretty much going to read this
- 9:35in every single thread. But, you know, I
- 9:40wouldn't blow this. Like, similar to how
- 9:41we approach agents.mmd these days, less
- 9:44is more. If there's something where
- 9:45you're like, I really want it to always
- 9:46do the specific thing in a specific way
- 9:48every single thread, then maybe this
- 9:50makes sense to fill it up here.
- 9:51Otherwise, if you're not sure, just
- 9:53leave it blank. It also has auto memory,
- 9:55and that works pretty much the same way
- 9:56as automemory inside of clawed code. as
- 9:58you tell it to do things throughout your
- 10:01sessions, it has its own little scratch
- 10:02pad where if it thinks it's something it
- 10:04should remember, it will just
- 10:05automatically add essentially a little
- 10:07sticky note to its memory. Next, we have
- 10:10the environment tab. So, the first thing
- 10:12we need to do is designate a project
- 10:14repository. All these workers, all these
- 10:16sessions are going to be in the cloud
- 10:17and they need a GitHub repo where they
- 10:19can communicate and send their PRs. So,
- 10:22to do that, we just need to select a
- 10:24particular GitHub repo. Mine is going to
- 10:26be AI Creator Radar because I'm creating
- 10:28this thing that goes out on the web,
- 10:30goes out on Twitter, sees trending AI
- 10:31news, and then turns that into a
- 10:33dashboard which says, "Hey, here's
- 10:34things that haven't been talked about on
- 10:35YouTube. Here's how we can turn them
- 10:37into short form. Here's how we turn to
- 10:38carousels." That type of thing. So, you
- 10:40can either create this manually inside
- 10:43of GitHub or just hop into a normal
- 10:45cloud code session and say, "Can you
- 10:47create a GitHub repository for me that
- 10:50I'm going to use inside my project?"
- 10:52Next, we have the cloud environment. So
- 10:54if you just keep this as is where it
- 10:56says select the cloud environment, it's
- 10:57just going to use the default anthropic
- 10:59cloud environment, you also have the
- 11:01ability to add a cloud environment. So I
- 11:03could give it some sort of name. I can
- 11:06define the network access everywhere
- 11:08from none to full. And then I can start
- 11:09uploading secrets and environmental
- 11:11variables. So if I am doing something
- 11:14that requires, you know, API keys and
- 11:16secrets, I'm going to need to add a
- 11:19cloud environment for it to be able to
- 11:20work there. Now, we are going to leave
- 11:23it on select a cloud environment. And
- 11:24you will see later it's going to prompt
- 11:26us to essentially change our cloud
- 11:27environment to add some keys. And I'll
- 11:29walk you through that. Next, we have
- 11:31some local settings. Let cloud use your
- 11:33device. Yes. Pre-approved folders. If
- 11:35you have a session or a worker that is
- 11:37going to need to work on your local
- 11:39computer or access access it, that's
- 11:41something we can do. By default, it's
- 11:43always going to be in the cloud, but it
- 11:44can actually, you know, go on your
- 11:46machine. And so to do that, you just
- 11:47need to select this to yes. And then we
- 11:50want to give it some sort of folder it
- 11:51can work in. In this case, I set my
- 11:53radar-local folder. Then we have a usage
- 11:55tab followed by debug access. So let's
- 11:58say you have two projects, project A,
- 11:59project B totally doing different
- 12:01things. You can let project B take a
- 12:03look at project A and vice versa to help
- 12:05debug because perhaps you're running
- 12:07into problems where maybe project A
- 12:10can't debug on its own. Maybe we have
- 12:11like some issue with the memory. Like
- 12:13something is poisoning its context.
- 12:14Well, how about we get a fresh set of
- 12:16bias to take a look at what's going on.
- 12:17We can do that by simply giving it
- 12:20access via the debug access panel.
- 12:22Probably won't use this a lot, but know
- 12:23it's there. So, let's go ahead and give
- 12:25it our prompt. So, we're saying we're
- 12:28creating the AI creator radar. We've
- 12:30given it kind of an end state we're
- 12:31looking for. Again, I want it to be able
- 12:32to scrape things off of X, and then I
- 12:34want it to be able to take a look at
- 12:35YouTube, see what's working, see what
- 12:37hasn't been touched on, see where
- 12:39there's some sort of gaps, and I wanted
- 12:40to do it on a dashboard where I can just
- 12:42see it all in one place. So, let's go
- 12:45ahead and put it to work. and see what
- 12:48happens. And as this is working, you'll
- 12:50kind of see this has begun to populate
- 12:51over here are the threads. So
- 12:53coordinator main page right here, I can
- 12:55talk to everything and like do
- 12:57everything from this one window. But if
- 13:00I want to kind of like click inside of
- 13:02here and see what's going on at
- 13:04individual thread or even worker level,
- 13:06I have that ability too. So after
- 13:08letting it run for about 25 minutes,
- 13:09here is where we are at. First of all,
- 13:12after we gave it the prompt, it split
- 13:15into five parallel threads, and each
- 13:17thread was given some piece of the pie
- 13:19that it needed to execute on. We had
- 13:21something working on the dashboard and
- 13:23hosting, one working on the Xcanner, one
- 13:25on YouTube coverage, one on content
- 13:27packs, and one on cloud code news. If we
- 13:29go over here to the right on the
- 13:31threads, we can see I have two that are
- 13:32now ready for review. I have three that
- 13:35have some sort of blocker that require
- 13:36me to come in and either make some sort
- 13:38of decision or give it some more
- 13:39information. And then we also have the
- 13:42idle thread. Now at any point in time I
- 13:45can click on any of these threads and
- 13:47see what this particular worker is doing
- 13:49at a more granular level. I can click
- 13:51here and see the blockers. And so it's
- 13:53really easy me to go really easy for me
- 13:55to go into each individual session and
- 13:57actually talk to it because as you see
- 13:59when I click into here I do have an
- 14:01ability to steer that thread. I also can
- 14:03view these threads from the orchestrator
- 14:05panel right here in the middle. So I
- 14:07just click view thread and it pops up.
- 14:10Also I can just handle everything
- 14:12through here through the orchestrator.
- 14:14So instead of going into each individual
- 14:16thread like I would with sessions, you
- 14:17know, when I'm sort of just normally
- 14:18operating inside of cloud code, I can
- 14:20just talk to the single agent and it
- 14:22will handle it all for me. And the whole
- 14:24time they're operating inside of my
- 14:26GitHub repo, which we see right here. If
- 14:28I hop into the pull requests, we can see
- 14:30what all these individual threads,
- 14:32individual workers are actually sending.
- 14:34So let's handle some of these blockers.
- 14:36The first one is related to the cloud
- 14:38environment. Remember I mentioned how
- 14:39earlier if we need API keys as part of
- 14:42our project, we need to edit the actual
- 14:44environment. Well, let's go ahead and do
- 14:45that. So go up to project settings, go
- 14:48to environments, and we're going to add
- 14:51a cloud environment. Going to call this
- 14:54demo environment. For network access,
- 14:56we'll throw it to full. And now we need
- 14:58to add a network secret. And so I put in
- 15:00all the correct information and then I
- 15:02just hit add. And now it's working in a
- 15:04cloud environment that has my particular
- 15:06secret. So then I can say, hey, I just
- 15:08updated the cloud environment with the
- 15:10YouTube key as a network secret.
- 15:14And the coordinator is going to send
- 15:16that information to that particular
- 15:18thread. And now I'm just going to repeat
- 15:20this process for the rest of the
- 15:21blockers. And that's kind of how you
- 15:22would run this. All right. So we got rid
- 15:24of the blockers and right now it's
- 15:25finishing up. One thing that's kind of
- 15:27cool is like I said, you can tell
- 15:29certain sessions that you want it to be
- 15:30local. We did that with the session that
- 15:32had to do with scanning X because X was
- 15:35actually logged into my account on my
- 15:37cookies. So it used the inapp browser
- 15:39inside of cloud code desktop to do that.
- 15:41And so it actually built a local
- 15:44scheduled routine. So you can see on
- 15:46theuler it created this routine. It's
- 15:49inside of that folder that we designated
- 15:51earlier is like the local folder it can
- 15:53play around with. And you can see kind
- 15:55of the instructions it gave it. And so
- 15:56here's a look at the dashboard it
- 15:58created. It shows me which stories are
- 16:01trending on Twitter but really have no
- 16:03coverage in the YouTube space. For each
- 16:06of these, like, hey, Claude writes about
- 16:0890% of Claude Code's code. It shows me
- 16:11potential content pack, what the video
- 16:14could look like, what a short could look
- 16:15like, what a carousel could look like,
- 16:18as well as what an X post could also
- 16:19look like. It also gives me kind of a
- 16:21breakdown of what's going on with X. So
- 16:24like the top story right now is DeepSeek
- 16:26version 4.1 Flash and GLM 5.3 Flash go
- 16:29free in open code. So it shows me like
- 16:32the posts per hour, the views. I can
- 16:34actually click on the posts themselves
- 16:36and it will bring it up. And it also
- 16:39gives me more stats about my channel as
- 16:42well as like some Clawude Code specific
- 16:45news. Lastly, at the bottom, I can see
- 16:47kind of a refresh log. And so overall,
- 16:50this is pretty solid work considering it
- 16:53was kind of a oneshot and this is just
- 16:55the first pass that Claude Code Projects
- 16:58has done on this. And the real cell for
- 17:00projects at this point isn't the
- 17:01singular thing it created. It's the fact
- 17:03that this is now something that is
- 17:05living in the cloud. I can access it at
- 17:08any time and I have all of these threads
- 17:11that still are sharing and maintaining
- 17:14context. So I can work on this whenever
- 17:16I want. I can constantly feed it new
- 17:18information. it can run even when my
- 17:20computer is shut down and I can continue
- 17:23to add more and more complexity to it
- 17:25all through a single orchestrator. So
- 17:28while I do have the ability to you know
- 17:30click on individual threads and handle
- 17:32it one at a time I can just talk to one
- 17:34agent from any place and have it control
- 17:37everything. The last thing I'll mention
- 17:39and I will also link this in the
- 17:41description is the documentation on
- 17:43cloud code projects. It is very
- 17:45extensive. So, if there's anything you
- 17:48feel like I didn't cover enough or you
- 17:50want to go deeper and I highly suggest
- 17:51you take a look at the documentation,
- 17:53but besides that, that's all I have for
- 17:56you today. Highly suggest if nothing
- 17:58else, you hop on that weight list
- 17:59because you will get access very, very
- 18:00quick. I think this is a really cool
- 18:02feature inside of Cloud Code. Kind of
- 18:04niche, really depends on the type of
- 18:06projects you're working on. And if
- 18:08you're someone who really does need that
- 18:09247 access, if that's not you, no
- 18:13problem. But if it is, it's great
- 18:14because you know in the past when we've
- 18:16had these sort of anthropic cloud
- 18:18options, they've always been draining
- 18:20from the API. Now it just comes straight
- 18:21from usage, which for many of us, I feel
- 18:24like we have a lot of usage remaining.
- 18:26Like Opus 5.5 has been pretty token
- 18:30efficient, especially compared to
- 18:31something like Fable 5.1. So having the
- 18:33option to go into the cloud, stay on
- 18:35usage, and always have complicated
- 18:37projects running is a great value ad. So
- 18:41as always, let me know what you thought.
- 18:42Make sure to check out Chase AI Plus if
- 18:44you want to get your hands on my cloud
- 18:45code masterass.
About this transcript
This page contains the full transcript of The SECRET Claude Code Upgrade That Has Everyone Going Nuts (Full Guide) by Chase AI, generated from the public captions YouTube serves with the video. The transcript has 4,073 words across 562 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.
What you can do with it
Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.
Free YouTube transcript tool
YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.