guys can run everything that you need from the convenience of your device. So, as some of you guys will know, I'm a big fan of using claude code to do a lot of economic knowledge work for me. This video is not trying to say that co-work is better than cloud code, but it is a different...annoying. It's often pretty fragile and then it takes a fair amount of setup. So, Claude Co-work solves all this for us completely out of the box. Okay, so totally hands-off demo. I have co-work set up with cloud code dispatch. It's right over here. And uh I just want to scrape 200 agency owners
Поиск по транскрипциям
much more done because I can have the servers on the same screen as Cloud Code. I can have multiple terminals open without my computer slowing down, be able to go back and forth. So now I can go in here, give Claude another prompt for what to work on while continuing to work on my other app on Creator...branches and work trees. Again, if you want a full breakdown video on that workflow, let me know down in the replies below as well. I have tried every Claude Code workflow on the planet. As of this date that this video is posted, this is my favorite workflow ever. I get a bunch of complaints when I show
turns your core code into an agency um, using 61 specialists and is free to set up. So, I'm going to show you exactly how it works. It's a free open source project on GitHub that just exploded to over 12,000 stars. It could completely change how you use tools like, for example, Claude, Cursor, and Gemini...gives claw code a bunch of AI specialists that they can start using. Now, if you want to get started with this, you can just copy this command into claude code. So, let's get started with
Claude Code just released agent swarm, and this is honestly one of the biggest updates I've seen in AI automation this year. We're talking about multiple Claude agents working together simultaneously on complex tasks. Not just one agent doing one thing, a whole team of AI agents collaborating like actual employees. Now, before I dive into exactly
little bit more closely so for example to use a concrete example anthropic is really big like their CEO is saying well he said last year coding 90% of code is going to be written by AI tools and quite honestly I thought that was full of crap but practically we are at that point so why is anthropic still...actually hiring experts. They hire, they aqua hired Jared from Bun, the very competent developer, or you could say he is now the adult in the room working on Claude Code, making uh maybe cleaning up a bit of the mess that was made when you go fast and break things or nowadays when you vibe code, which you know
models is that you should try avoid using the word think when you have thinking mode off in Claude Code, for example, especially when using Opus 4. 5 because that's particularly sensitive to a word. So basically, if you use a word think with a Claude model and you don't have thinking mode on and you say something
right now. And uh yes, this project, yeah, I'm I am actively developing it. It's just me. I'm work well, it's me and Claude Code, which I've been using a bunch. Um but I have a background in software development somewhat, but um yeah, I've been using cloud code a lot and uh yeah
this in a couple of different ways. One is if you write forward slashplugin, this is going to give you a list of different plugins that are supported by claude code. So if I go into the search bar, I can type in Figma. And you can see here this allows you to uh integrate with the Figma...ahead and open up the browser asking me if I would like to give Claude access to my Figma. All right. And now let's go back. Okay. So, now that we've got the Figma MCP server set up with Cloud Code, it's time for me to actually start bringing my designs over. I'm going to start
like all of anything that I would do to touch the keyboard I'm now passing it on to some type of agent hardness um whether it's claude code or it's codeex or any of these tools and so my job suddenly turns into like I have ideas I pass them on to cloud code and then
coding and not opus 46. I'm to save tokens on uh opus because it's so expensive. I codeex is good. Codeex is perfectly fine. I like claude code better but codeex is perfectly fine to save money. I love your I love you brow. You are the best. I assume you mean bro. I assume you didn
model to create you a private GitHub repo and then push your env. After all that's done, it's literally as easy as just going into your cloud code and just saying, "Hey, clone this into our workspace. " So, you just give it the link of the workspace that you set up. Um, and then you just send...folder and then open this up. Let that be the parent folder. Uh, and now you have everything that you need. Um, you know, assuming you download and install claude code on the user. So, hey, tell me about this workspace. And uh it's just going to go through get a highle overview presumably from the agents claude
light tasks for so much cheaper. I spend thousands of dollars a month on the claude API because it powers basically all the apps I build out. This is going to reduce my cost so much using the API. And then finally is clawed code execution. I very specifically say execution here. This is where the secret to using Haiku...notifications right now. So many more cool products and announcements coming out very shortly here I'm going to cover. But let's talk about using Haiku and Claude Code because this is the secret. This is where you're going to get the most sauce. So I'm in Visual
code or um one of these other ones before I started to play around with open claw. And I would also uh have all the processes built out in claude code first. So, for example, that executive assistant over the next uh two weeks, I might build out a bunch of skills like the Sebastian refer skill, um like
chain them so you can tell it to run one skill and then run another skill if needs be, but you can't really run them in parallel. Whilst code is able to run many tools in parallel and skills are essentially tools, you will quickly fill up the context window and then it will basically be unusable...code review of subagent, you want a lot of deep context and you want it to forget that previous context and not to have it to complete the main Claude code agent conversation. And another pretty useful way of using them is to have like a problem solver sub agent. So when you're really stuck on a pesky
time. Additionally, it can really help to ask Clot Code to create a plan before you start building the app, especially if you're trying to create something larger. Another tip would be is that you could duplicate this invoice generator and set a second terminal window, say Claude, and work on two versions of the feature at once...with the one that works better for you. And finally, I want to add that this also really works well if you have some old projects that you want Claude Code to explain to you or if you work as a developer and you need to get familiar with a brand new project. Well, you just open up Cloud Code
said was, "Hey Claude, I want you to take this YouTube video and repurpose it into a LinkedIn X and Instagram post. " Then I dropped in the link to the YouTube video and shot it off. Not only did it create all these assets, but it also found bugs in its own code...fixed those. And then we have this folder over here called drafts. And if I open it up, you can see that we have building beautiful websites with claude code, which is the video I gave it. And then in here we have Instagram with our actual post text and five visuals. We've got LinkedIn with our post text
Claude models? - So that's excellent question. The canonical standard way of doing this is to install the MCP SDK, set up your own loop, like you mentioned with Claude Code and handle connecting to any MCP server that you need to get connected to. But it's essentially your responsibility as a software developer to put together
your specific needs without paying per API call or hitting rate limits. Now, I know what some of you are thinking. Is this really as good as Claude or GPT? For coding, yes, the benchmarks prove it. M2. 1 is purpose-built for coding and agentic tasks. It's trying to be the best coding assistant possible. And at that
great about this newer model? Honestly, from what I've seen so far, Claud is still beating it by a long way. The outputs, the speed, Claude has nearly finished coding the second task, whereas AI Studio is still on the first one and struggling. So, from what I can see, you get a much better output from Claude versus
Okay, so I'll be going over the Claude code updates of last week. Firstly, you want to make sure you're on the right version, so you need to be on 0. 88 or greater. So you can write claude --version, and you can see it over here, I'm on 0. 89. And you can write claude update