really good enough, and the second one is that like the scaffolding, the thing on top of the model, wasn't good enough. And when we initially launched Claude Code, the very, very first versions late last year, I think it was still using Sonnet 3. 5. This wasn't even 3. 6, or whatever we call this thing...there was a few engineers sitting there; and one of them was Robert, and there was a couple of other engineers. And I just walked in, and I saw Claude Code on their screen the first time. And like I just gave this to them the day before and they're already using it. And it was just the craziest
Поиск по транскрипциям
wanna talk a little bit on the model front. So, up until just recently, Claude Code was powered mainly by Claude 3. 7 Sonnet. So now with the Claude 4 models powering Claude Code under the hood, what has this unlocked and where do you think we're headed? - Yeah, maybe a couple months or so, before the models came...just remember my jaw dropping at how much more capable it feels. So, I think there's all these new use cases that are unlocked. when you're using Claude Code synchronously in a terminal, I think one of the big changes is Claude is just much better at holding your instructions. And so you tell it to do something
Importantly here, we should say that this is not something that Claude Code would just do. Like if you or I just prompted Claude Code right now to do. There's been some jailbreaking or something going on here, right? - That's right. - Yeah, so in this case, the actor found the right prompt and put it in the right...like bludgeon the AI into just like continuing, continuing on, putting words in its mouth, all sorts of things like that. So that's what they're doing to Claude Code. - Yeah, actually, in this case, they were doing role play. So they were pretending to be a, your average security person doing network penetration testing to make sure that
because when the agent fails, it's very hard to get it back on track. So, what Anthropic also show us here is this clawed code feature. So Claude code is basically Antropics version of a coding assistant and it basically you know it's just a tool that lives within your coding tool. So you've got your terminal...code reviews or bug reports and it will reply to feedback fix errors and suggest code changes. So you just have to in run install GitHub app inside of Claude code to activate it. So this is crazy because I don't think people realize the shift that is occurring here because this means that Claude is now moving beyond
exact same thing. Like you're managing units in a video game. So someone's like, "Okay, when is someone going to build that interface for managing claude code? " I built a SAS product. It failed. Everyone is loving it but not signing up. Should I start building a new project? Well, if everyone's loving it, why would...have some videos on my channel around uh iOS apps if you're interested. Let's see. We have 25 votes and 84% are for the new Claude Code interface. All right, let's get into it. Should we just It sounds like that's the winner to me. We're going to build a gamified Claude code interface
they're going for the same feel where they have crunching, moseying, unraveling and so forth. So yeah, it feels very familiar in a way. Now, I think that Claude Code Web, which is like Codex CLI Web, is significantly better for tasks that require only one or two things, like very small changes in the sense that...complete, let's go through the brand new sandbox mode that they have available. Hey, so as a short aside, after popular demand, I made a brand new Claude Code class that's available. There will be a link down below and I basically cover every single Claude Code feature that's available and I'm continuing to update
other chat here, you can see we've got the progress and it's ticked off in the top right. Somebody asked, is this not just Claude code with a UI? So, actually claude code with UI came out last week inside the desktop app for Claude. So, you can use Claude now, right, on the desktop. And then...code over here, you can use claude code, right? But the problem is that inside claw code it's not really that useful for I don't know what you call them but people who don't use AI normally right it's not that useful whereas for example inside co-work this basically is a lot more user friendly
start new chat, it will know everything it added in past sessions. This is all automatically loaded in. So, if you're building an app right now with Claude Code and you're not using GitHub, make sure you go to github. com, create a new repository. It's all for free. And then tell Claude to commit the code...make commits to GitHub, back up your code, and it'll know every time moving forward all the changes it made in GitHub. This again is going to make Claude code so much more powerful because it will remember all its changes from the past. Use GitHub. Use it a ton. Constantly commit your code to GitHub. It'll make
recommend? What are some best practices or ways for me to get started? - Yeah, so super-tactically, actually the number one thing that we recommend right now is the Claude Code SDK. And what's really, really interesting about the Claude Code SDK is we essentially built an agent harness an agentic harness around the model to run that loop...that really great starting place right out-of-the-box. - Right, I feel like this is a pretty common misconception, at least when I talk to developers about the Claude Code SDK. So I'm not building a coding application. Why would I wanna use this? But you can kind of remove the coding-specific parts, - Yeah, I mean
happen faster. And I basically don't know if the premise here is correct and I want to know what changes have happened so far. So I can tell Claude Code, hey, can you spawn up two forked subagents? One of them should make a Mermaid diagram of all the changes that we made so far and another one should...light theme instead, press enter. And that would go over to that fork with that message queued. So this kind of feels like agent teams in Claude Code, if you are aware of that. So our verification fork is done and it passed this information back into main session. And this information is much richer and like more nuanced because
next one is remote control. Right? So remote control is basically a new feature that bridges your local claude code terminal sessions with claude AI code, the claude iOS app and the claude Android app. Right? So your session keeps running on your machine the entire time and the web and mobile interface is just a window into your local
wake up, I have code to review the moment I wake up. So my app is always improving. You need to be doing this as well with your Claude code subscription. You get this Claude for web app, go to claude. ai, go to the Claude code section, spin up agents right before...your computer and b you have agents in the cloud building tasks for you on the go. This will multiply the amount of work you can get done with Claude code. Which brings
particular file with the name dot Claude dot local dot MD, right? And this is only for your personal purpose, right? You want to give more context to the Claude code, you know? So I will go ahead and create Claude dot local dot MD file, you know, with more information. I don't want to share it that with...project level, this you can actually share with the entire team, you know? So that team will also be, you know, when they are using Claude code, some basic context information will be available. And then you can also save this in a global dot Claude dot MD file, which applies to every project that you create in this particular
released Claude Code Desktop. And this is so sick, too. Inside the Clawed Desktop app, which is out on Mac and Windows, you now have this new button right here, which switches you to the coding section where you now can use Claude Code right on your desktop. It does it both locally and on the web. So, you have...well as working locally on your computer. I've been testing this app all day and it is excellent. It is the new way I'm going to use Claude Code. I'm not going to use Visual Studio Code anymore. be using Cursor with the CLI in it or just the CLI by itself. I think this
edge though because everything else I'd say is weaker, but the fact that it was able to get the price done in V1 is pretty good. So in Claude Code, I'm going to say that seems to work. What are next steps? I'm going to assume it's pulling in real data is the next step...going to say that works. The UI works, the pricing works, but there's no AI data to see what it does from there. Okay, so Claude Code shows the updated to-do list, which I love. I love that Claude code works this way. It works in kind of a to-do fashion where it builds and maintains
claude code for web you need to have your code hosted on GitHub which is a code repository. So if you haven't yet sign up for GitHub go to github. com create a free account. Then we can do new repository. We're going to call this a journaling app. You can make it public or private...going to do is I'm going to say, can you upload this code to and then put in the address for our repo and hit enter. And now Claude Code, like the good employee it is, is going to take our code and put it on that new repository. I'm just going
then dole it out, okay, and delegate subchunks of that task to different models. And so what's occurring here is in this hypothetical example, we're using claude code to be our manager. We then give it some task like hey make me a SAS app that does X Y and Z. And then what it's doing...files for this course you can find in the top link in the description below. What I'm going to do is go back to Claude Code and open up a new session. And then I'm going to select this folder that I've actually already created for this purpose called multiplatform orchestration. Now, as mentioned, you guys will
this is exactly how I built the trading dashboard on just a Sunday afternoon. And it starts with plan mode. So, I go into Claude Code, and I'll show on my screen, but you go into plan mode, where Claude doesn't accept any edits. It's not making any changes to scripts or files. You're just brainstorming...have that plan. It's going to be just an. md file, and then you're going to go into build mode, where you're going to give Claude Code that plan, accept edits, and it's going to build the skeleton of the dashboard. Then the third one, the third step, is where it gets fun. This is where
Okay, so it seems Anthropic have been pushing updates like crazy to Claude Code because we have yet another big update. Meanwhile, the OpenAI Codex team have been taking a nap because we haven't seen something particularly interesting or a big update for the last few weeks. But anyways, we're going to be going through this update...Claude Code, which includes Claude Code Web and also a brand new sandbox mode as well. Now firstly with Claude Code
just dropped my prompt inside of there. Now, one really quick thing that I should mention is like if you ever get a lot of text inside of your Claude code instead of having to backspace everything manually, okay, you can hit escape twice and it will just clear everything for you. All right, that's another little...then say published, all right? So, I might have different actual names for a project, but this is the idea. So, let's call this like how to use Claude code. All right? And then we'll just click save. That'd be a video idea. Close that. And then let's just see here. How to use Claude code