// transcript — 270 segments
0:00 Introduction to Claude Cowork Dispatch
0:01 Enthropic just shipped a feature called claw dispatch which lets you essentially
0:05 run and manage an AI agent through your phone. In this video, I'm going to show
0:08 you guys how you can use it, why it's significantly better, and probably more
0:11 preferable from a security perspective than a lot of the pre-existing solutions
0:15 like OpenClaw at all, and then how to set it up on your end so you guys can
0:18 run everything that you need from the convenience of your device. So, as some
0:21 of you guys will know, I'm a big fan of using claude code to do a lot of
0:25 economic knowledge work for me. This video is not trying to say that co-work
0:28 is better than cloud code, but it is a different and easier interface that a
0:31 lot of less technical people can use to get up and running. So in cloud code, I
0:34 have a bunch of skills which are essentially tasks that I fire off one at
0:39 a time. I say something like, "Hey, scrape me 100 HVAC leads in Texas." And
0:43 this will actually go through and then literally contact the internet, run a
0:47 bunch of scrapers, and then return them back to me totally autonomously. The
0:50 issue is obviously using my phone with a setup like this is annoying. It's often
0:53 pretty fragile and then it takes a fair amount of setup. So, Claude Co-work
0:57 solves all this for us completely out of the box. Okay, so totally hands-off
1:01 demo. I have co-work set up with cloud code dispatch. It's right over here. And
1:04 uh I just want to scrape 200 agency owners in California with a 1 to 10ish
1:06 Using Claude Dispatch on the Go
1:08 size. You can see the second that I entered that in that was immediately
1:12 reflected over here on my screen. And the reason why is because this is now
1:15 running locally. Now, I don't know, I'm just on the go, right? I'm waiting for
1:18 some train or something like that. And I say, "Hey, I could probably get a couple
1:22 of phone calls in." So I launch it from here. Immediately after we see actually
1:26 see a window open under dispatch over here um as a new tab and sort of our
1:31 claude co-work ongoing conversations. So what it's doing now is it's actually
1:33 running the skill. Okay, it's doing a bunch of the specific SOPs inside of the
1:37 skill like generating search keywords, running a test scrape and so on and so
1:40 forth. And then I'm going to see the results back here on my phone the second
1:43 that it finishes. Okay, and then just a few minutes later I have all of them on
1:47 my phone here. I don't know if you guys could see this but um it's like data set
1:50 leadsfinder. I have all the phone numbers. So, if I just wanted to, you
1:53 know, click on one of these and maybe give them a ring, start doing some cold
1:57 calling, I could. Um, I just initiated and then completed that loop entirely
1:59 from my phone, which is pretty sweet. But obviously now, that's not the only
2:02 thing that I can do. Um, I have this pretty cool inbox cleaner flow, which I
2:05 like running. And, um, what I'll do now is I'm just going to initiate that inbox
2:08 cleaner again from my phone. So, I'll say run inbox cleaner. And then my hands
2:13 are up here just so you guys could see. Um, it's going to start a conversation.
2:17 Um, go down to that tab there. And then just like it did before, it'll it'll run
2:21 my own skills on my computer using um you know, Claude Co-work. And just
2:23 because I want to do multiple things simultaneously, I'm also going to have
2:26 it run another skill. Uh what I'm going to do is hypothetically, let's say I
2:30 have found a YouTube thumbnail that I really like and I have a video that's
2:34 coming up soon. So, you know, let me just make this a little bit bigger.
2:37 Okay, there's this awesome one here from um I don't know, uh Don't Create an LLC.
2:41 It's by Jasmine Deluchi. She's this person that uh I don't know, creates
2:44 content all about accounting and stuff like that. So, what I'm going to do is
2:47 I'm just going to share this. I'll copy the link and then I'm also going to feed
2:53 this directly into Claude um dispatch and I'm going to say, "Hey, recreate
3:00 this thumbnail but with my face." Then I'm going to paste in the link. And now
3:05 we should also pass the next message in Q. And what this is going to do is it'll
3:09 launch an additional dispatch conversation that uh will allow us to
3:13 run multiple of these things. Okay. And then if you see here, we're running all
3:16 three accounts in parallel, fetching, classifying, then marking red for each.
3:21 Over here, we're actually reproducing a thumbnail using my thumbnail generation
3:25 skill with which has some face swapping uh functionality built in. And I'm going
3:28 to get it all on my phone in just a few seconds. Now, in my case, I didn't uh
3:31 realize, but one of these skills actually required a little piece of
3:34 access from my computer that I hadn't given it ahead of time. And so, when
3:37 that occurred, I just had a little window that popped up at the bottom of
3:40 my Cloud Dispatch, and it said, "Hey, you know, we're trying to reach XYZ. uh
3:44 do you want to allow this request? So I'm the one that just instantiated the
3:46 request. Obviously I'm allowing it. This sort of like security and then allow
3:50 listing style permissions is what makes this 10x more safe uh you know secure
3:55 and then scalable for business and then enterprise applications than something
3:58 like openclaw. Okay. And then in this case it's now created a bunch of files.
4:02 I don't think this is the best or most uh realistic look that I've ever had
4:07 before but still not pretty bad. And uh you know we initiated this from my phone
4:10 and it's not just available over there. It's also available over here. So,
4:13 dispatch is reporting back right now, letting it know that, you know, it uh
4:16 just finished the thumbnail thing. And now it's asking me to allow it
4:20 permissions to go and grab those thumbnails, which I will do in a second.
4:23 And now it's actually showing them to me all on my phone. So, you know, I have
4:26 them right over here, too, right? Can actually take a look at them. And then,
4:29 okay, I guess I can't scroll through them, but I don't know. That's another
4:32 one right over there. That's another one right over there. And so, I just
4:35 generated all these from my phone. Um, basically through like an app, right?
4:37 This app. But I didn't actually have to like whip up an app for it. I could just
4:40 route this to my computer, which is pretty sweet. And then the inbox cleaner
4:43 also finished. And in my case, basically what this does is it just automatically
4:47 marks as read anything that uh I would consider to be some form of spam, some
4:50 people soliciting me, that sort of thing. And then if there is any like
4:54 important email or whatever, it goes through and then it pre-drafts a reply
4:57 based off of a bunch of old replies that I've sent. And then it just allows me to
5:00 Inbox Cleaning Automation
5:01 QA and sort of human in the loop. So I then go through my inbox, but I
5:05 basically save like 90% of my time. So that's pretty neat, right? Like it's
5:08 done my personal Gmail, it's done a couple of other things. Um, simple
5:11 systems like this essentially allow me to save two or three hours a day. And
5:14 before I was constrained with doing all this stuff on my computer. I had to lug
5:17 around my big fat laptop with me. But now, you know, I could set up my laptop
5:20 or I could set up another computer or some sort of Mac Mini in the corner and
5:24 then I could just like sort of port in run whatever skills, run whatever
5:27 workflows I need even while I'm over here. And then, you know, naturally I I
5:30 receive and then can send additional data follow-ups and everything like that
5:33 Enhancing Workflow Efficiency
5:33 from here while it's still super secure. So, in my mind, this has solved
5:36 everything. I'm sure we're just a few stones throws away from having this sort
5:40 of stuff in Telegram. Let me show you guys how this all works under the hood.
5:43 So, if it isn't clear, I think that most of these autonomous agent tools are
5:47 pretty broken. And they're broken for a couple of reasons. The first is these
5:52 Open Claw and DIY agent stacks, which I think everybody in their mom is going to
5:54 try and tell you is like the future and stuff like that. Um, they are terrible
5:58 from a security perspective. We've had tens of thousands of API keys leaked. uh
6:03 very elementary sort of security errors here have hit news headlines over and
6:06 over and over again. It got so bad they had to change your name multiple times
6:09 and stuff like that. And I mean like their core functionality is pretty cool,
6:13 right? Which is that I get to text my model basically from my phone and then
6:16 have it do economically valuable tasks for me. And the reason why beginners and
6:19 you know people that aren't very tech super into this sort of tech love it is
6:22 because it's like hey I you know I talk to my friend in Telegram and then
6:25 immediately underneath I have my Claudebot in Telegram and that's
6:27 awesome, right? Right? It's just like good from a user experience perspective.
6:31 But all of that isn't really worth risking, you know, your API keys, your
6:34 credentials, your credit card numbers, your identity. Right? In addition to
6:37 obviously massive security issues, there's also like a context explosion
6:40 issue where you are paying for the most part using your um API pricing instead
6:45 of like a monthly recurring subscription, which Anthropic subsidizes
6:49 heavily and tends to be somewhere between like four to 8x cheaper. What I
6:53 Security Concerns with Current Solutions
6:53 mean by this is you are doing direct API calls to their backend and every single
6:57 time you make an API call um it bills you usage wise like that's you know
7:02 every message 50 you know every time a skill runs that's $2 every single moment
7:06 that you're having a conversation with models on you know openclaw whatever
7:09 unless they're running locally or running using something like Neotron
7:12 what you're really doing is you're just constantly like burning your money in
7:16 the background and then you know ultimately this pay per token plus the
7:20 security problems in my view makes this like a non-starter, you know, if I'm
7:23 using this in a serious business application, I'm basically never going
7:26 to recommend something like open cloud or whatever. And so when this came out,
7:28 you know, I made a bunch of videos talking about how horrific it was and
7:31 all of the API keys that got leaked. And I was like, you know, I bet it'll just
7:34 take a couple of months before anthropic launches a native feature that does all
7:37 this stuff. And that's what cloud dispatch is. So with cloud dispatch, you
7:41 basically connect automatically to whatever your desktop is, your computer.
7:44 Um, so your skills, you know, all of the pre-existing MCP connectors and all of
7:48 your files. But it's also sandbox and local. And I mean like you know if
7:51 there's a company that I trust to understand how to rein in you know the
7:57 downsides of like AI agent flexibility it's the companies that have developed
8:00 the best AI agents out there like anthropobic. They've spent you know
8:03 hundreds of millions if not billions of dollars doing this exact thing. So I
8:06 trust them to you know sandbox and make things pretty local. Um it's also uh
8:11 like a flat monthly charge because it's technically used as part of your
8:13 subscription. Now they're talking about how they're going to make this available
8:16 to pro plans pretty quick. I think right now as of the time of this recording
8:18 it's on max plans but they're they're going to want to try and get everybody
8:21 on it. And so what that means is instead of paying you know I don't know like it
8:24 really depends on your usage and I don't want to pretend like I know exactly how
8:27 much you're us using. But when I tried open claw I spent like a h 100red bucks
8:30 or so in just my first few days and then I was like okay I'm going to turn this
8:32 off. Instead of spending that you know you you'll probably just spend 20 to
8:33 Advantages of Claude Dispatch
8:36 whatever the the the subscription amount is and it's going to be heavily
8:39 subsidized. And then, you know, it says same power here, but I want you guys to know
8:44 realistically speaking, you're probably going to get significantly more usage
8:48 out of co-work because out of cloud dispatch because they're just going to
8:51 like build in more and more and better functionality. This is a company that
8:54 has access to basically all of the information on planet Earth at this
8:57 point. Um, I trust that they're going to do a much better job with this sort of
9:00 infrastructure and architecture than some random third party. In terms of how
9:04 Dispatch specifically works, um, it's sort of like a multi-layer system. you
9:07 know, down at the bottom its brain is claude co-work and uh you know, I
9:11 usually make videos not on co-work because that's the non- tech version of
9:16 this intelligence and and agent tech model. I make more videos on cloud code,
9:19 but cloud co-work and cloud code are basically you know the exact same thing.
9:22 It's just like support for a wrapper that wraps around skills and then you
9:26 know pre-established SOPs. So basically what happens is um Claude Co-work which
9:30 is your base sort of agent here has access to your skills library and a
9:33 bunch of MCP servers to do things for you. Dispatch on your phone basically
9:38 just um it communicates through layers with you know your MCP servers your
9:41 skills back to Cloud Co-work. Cloud Co does something and then it kind of
9:44 communicates back up the ladder to dispatch. And that's what I was talking
9:46 about with um you know the fact that this is like an agentic thing. You're
9:49 you're launching an agent. Those agents can launch sub agents and and so on and
9:52 so forth. All right. So, I'm going to show you guys how you guys could set up
9:55 a Claude dispatch in just 5 minutes. It's really easy. We're just going to
9:59 update Claude desktop or just reinstall. We're then going to update the Claude
10:02 mobile app. If we see dispatch in the sidebar, we're good to go. We can
10:05 actually enable it and then turn all of this stuff on right now. If not, you
10:08 know, we're just going to restart our computers until we see, you know,
10:11 whatever the the update logic is. Um, this is literally it's probably not even
10:14 going to take five full minutes. Let me show you guys how all this stuff works.
10:17 In terms of how to set this up, this is a pretty new feature. It was launched
10:21 sometime in the last 24 hours or so. And so many of you will not have this
10:26 dispatch tab added to your Claude Co-work. That's okay. If you don't
10:29 already have this, what you need to do is you need to update the application.
10:34 So just for all intents and purposes, in case you don't have Claude desktop set
10:37 up, it's as easy as heading to claude.com/down, then going from desktop over here,
10:44 selecting your operating system, and then choosing either Mac OS, Windows, or
10:48 Windows ARM 64. You'll also have to pair with the desktop app on your mobile. So,
10:53 choose iOS or Android here. Once you're done on desktop, you'll have an
10:57 interface that looks like this. Now, I should note that there's actually three
11:00 different sections to cloud desktop. There's chat, co-work, and code. And for
11:03 this feature, you're going to need co-work, which you can also use hotkeys
11:08 to bump to. In my case, command 2. Now, once you're done, head down to this
11:11 dispatch feature, and it'll say something like co-working on the go. In
11:15 order to get this started, it walks you through sort of the functionality, which
11:18 is that your phone is like a walkie-talkie that can communicate with
11:22 Claude on your computer. You just send Claude a message and it'll work on tasks
11:26 locally using your computer. Um, the one caveat to this is you do need to
11:28 Setting Up Claude Dispatch
11:30 remember to keep your computer awake. Uh, this is a big thing that a lot of
11:33 people were doing with OpenClaw. They would purchase specific computers like
11:37 Mac minis or whatever, plug them in, and just have them working 24/7. This is the
11:41 exact same thing. you just have, you know, your computer where you want to do
11:44 these tasks, awake, and then Claude dispatch can take care of the rest of it
11:48 for you. Anyway, all you do is you click get started and then you have to pair
11:52 with the Claude mobile app. So, I've already downloaded the Claude mobile
11:54 app. All I'm going to do is open this puppy up. It looks like I'm already
11:58 signed in to my phone. So, I have the Claude desktop app open right over here.
12:03 It has a chat window um ready to to talk to Sonnet 4.6. Next up, it'll ask, "Hey, can I give
12:08 Claude access to your files?" So, it'll ask me when I click finish, should I
12:12 keep this computer awake, which you're going to want to click yes to.
12:15 Otherwise, dispatch is going to run into some problems. Um, give it the ability
12:19 to use Chrome. And then ensure that all connectors are on. Okay. Then just click
12:24 finish setup. It'll ask to access files in your desktop and documents and so on
12:27 and so forth. You're going to have to click this button a bunch of times. And
12:32 now you have the ability to communicate with your phone and Claude. Now, on my
12:38 phone's end, all I need to do is in my Claude app, scroll down to where it says
12:42 dispatch, click the button, and then I'll get something that looks like this.
12:46 And as you can see, the UX and stuff like that is very similar to what we
12:49 just did. So, all I need to do is click pair with your desktop, and then it'll
12:53 connect automatically with the conversation that's in this window. So
12:57 much so that my hands are off the computer now. All I'm going to do is
13:01 type hey. And I want you guys to see how this immediately pops up on my computer
13:05 as well. And so I'm actually having a conversation with Claude uh coowork
13:08 through dispatch right now. It's fully managed for me. And then I obviously
13:11 have like a contiguous conversation history which is really sweet. And
13:14 here's where I could start by I don't know running some skills. So I'll say
13:20 run the lead scraper skill. Get me some leads for I don't know HVAC in Texas.
13:26 Let's just do 100 to start. I'm now going to send it the message and it's
13:30 going to go through the process of connecting to the skill which is in my
13:34 co-work instance. I'm finding all the session information and stuff like that
13:38 and then actually launching it. And then I will have all of the information and
13:42 everything I need um popping up on my phone as well. And you can see when I
13:46 actually kick off a skill execution, I create a new tab here that is prepended
13:51 with the words dispatch. In this case, it's scrape HVAC leads Texas. That
13:55 actually opens this up. And then you can see the progress bar on the top rightand
13:58 corner. Just cuz this is a pre-established skill that I, you know,
14:01 I put together a while back. um it'll go through the debugging and everything
14:04 else it needs to and you know I'm I'm not actually going to see a message back
14:08 like on my phone until this thing is done and it delivers the results in this
14:11 way. Essentially what it's doing is it's like spinning up new co-work instances
14:16 new co-work conversations almost as like sub agents going doing the task and then
14:20 reporting back the results of said task to me. Also, hypothetically, if you were
14:23 using something like cloud code and you wanted this oneshot skill run
14:27 functionality from your phone, a really simple and easy way to do it is you
14:30 basically just get all your skills from whatever your IDE or your cloud code
14:34 instantiation is storing them. So, in my case, it's right over here. And then, in
14:38 my case, I just duplicated them to this new folder, which is cloud/skills. And
14:41 then I just said, hey, I want you to add them all to co-work. It then goes
14:45 through, sends you a couple of brief little requests. Then after all this
14:48 stuff is said and done, um you just get a big list of skills that you can just
14:51 copy over to Cloud Co-work. Uh and this is going to allow it to work with
14:54 dispatch really easily. So I'm just going to click copy to my skills.
14:57 Actually, this one I think is already part of it. I just did this as a demo.
15:00 So this one should be good. That's actually going to like add auto research
15:04 diagrams directly over to the system. The next one, face variants for
15:07 instance, and so on and so forth. Then you'll get this copied to your skills
15:09 message at the top and now you can access it in dispatch really easily.
15:12 Okay, hopefully you guys appreciated this video. Um dispatch is really cool.
15:15 Conclusion and CTA
15:16 This is honestly a stand-in replacement for 80 to 90% of all the features that
15:19 Open Claw and other extraordinarily insecure implementations of autonomous
15:23 decentralized agents uh provide for you. So, I'm using this thing all the time
15:27 now. And it's not just because I'm like I love Big Daddy Anthropic, although I
15:31 do. Um it's because I just think this is a much safer and better solution than
15:34 something like uh you know what most most other people on on Twitter and X
15:37 are probably showing you. If you guys like this sort of video, please sub to
15:40 the channel. uh around 70% of you guys aren't for whatever reason. So, you'd be
15:43 doing me a big solid like and then comment down below with whatever the
15:46 heck you want me to make videos on. Um I really make videos based off of audience
15:49 feedback at this point. So, if you guys have anything in particular you want me
15:53 to record or cover, let me know. I'm more than happy to do it for you. Have a
15:55 lovely rest of the day. Catch all y'all