
To learn for free on Brilliant, go to https://brilliant.org/DevelopersDigest/ . You’ll also get 20% off an annual premium subscription TOOLS I USE → Wispr Flow (voice-to-text): https://ref.wisprflow.ai/ddai → Screen Studio (screen recording): https://screen.studio/?aff=Jv2k0 LEARNING RESOURCES → Scrimba: https://scrimba.com/?via=developersdigest FOLLOW ME → X/Twitter: https://x.com/dev__digest In this video, we delve into Anthropic's new release, Claude Sonnet 4.5, highlighting its ability to maintain focus on complex multi-step tasks for over 30 hours. We showcase the capabilities of Claude Code within the terminal by building a Next.js application complete with a homepage, blog, and various functionalities in parallel. Additionally, we push the limits by creating classic arcade games like Pong, Connect Four, Snake, and more, all with minimal prompts. Finally, the video features a sponsorship message from Brilliant.org, emphasizing their innovative learning approach. 00:00 Introduction to Claude Sonnet 4.5 00:16 Exploring Claude Code in VS Code and Terminal 01:06 Building a Next.js Application with Claude Code 03:15 Enhancing the Application with Additional Features 03:49 Testing the Application and Observations 05:56 Creating a Games Page with Claude Code 06:47 Sponsor Message: Brilliant.org 07:51 Reviewing the Generated Games 10:55 Final Thoughts and Conclusion This video is sponsored by Brilliant
Weekly deep dives on AI agents, coding tools, and building with LLMs - delivered to your inbox.
Free forever. No spam.
Subscribe FreeNew tutorials, open-source projects, and deep dives on coding agents - delivered weekly.
--- type: transcript date: 2025-10-03 youtube_id: U9bjOBOU7Nc --- # Transcript: Anthropic Sonnet 4.5 in Claude Code in 10 Minutes So, just this week, Anthropic released Claude Sonnet 4.5 with some really impressive results. They claimed that they've observed it maintaining focus for more than 30 hours on complex multi-step tasks. In this video, I wanted to do a quick look at what it looks like within Claude Code specifically. So, if you haven't used Cloud Code before, effectively what it allows you to do is to have a very effective autonomous agent. And the nice thing with Claude Code is you have a couple different options. So they have a brand new VS code extension where it's a very similar sort of look and feel to something like the cursor agent or something like GitHub copilot where you have the panel on the right hand side here and you can effectively have the agent and interact with it with natural language. Alternatively there is the option where you can use this within your terminal which I think a lot of people myself included do prefer as an option. Now, additionally, there is a Cloud Agents SDK, which is pretty interesting because it's not actually intended to be just used for code. They did rename this from the Claude Code SDK to the Cloud Agent SDK, where there's a ton of interesting use cases in terms of how you can leverage this. So, in this video, what I'm going to do is I'm going to be showing you Cloud Code within the terminal. We're going to try and build out a quick app. We're going to see how it looks, and we're just going to dive in and see what the new model's like. Okay. So, first up, I'm going to open up my terminal here. And I'm going to make a new directory. We're going to call this claude 4.5. The first thing that I'm going to ask is I'm going to say create a Nex.js application. And specifically, I want the homepage to read developers digest. Let's include a header as well as a footer. And let's include a blog page with some markdown example blog posts. Next, what I'm going to ask for is I'm going to actually want to spin off some different sub aents to have this go through and break down the task. So I'm going to ask specifically cloud code to instead of running through this in a sequential manner is I'm going to say I want to run as many of these component generations in parallel. What cloud code is going to do is we're going to go ahead and run the command to spin up that fresh Nex.js template. So within here we see that it's creating a folder called developers digest. We're going to be using TypeScript, Tailwind, ESLint, so on and so forth. We could obviously specify all of this as well. So once it's done that, what it's going to do is it's going to begin to create all of the components and pages in parallel. So the cool thing with cloud code is what it allows you to do is you can actually have your request like you see here, write out all of those different components at the same time. Okay, so here we're starting to see all of the different components come in. And now it's going to create the homepage and blog in parallel. This is just one little trick that you can use to speed up the process. Now there definitely are still some benefits to run through things linearly in some cases. Now we see it's going through it's creating the post. It's creating the blog and now it needs to install some relevant packages to actually read the markdown files. We're going to install gray matter for parsing the markdown. It's gone ahead. It's created our nextjs application like we saw it create the relevant components for what we had asked for the header footer. We also have the homepage where it reads developers digest as the main heading. We have the blog post to display the markdown. It installed the relevant packages along the way and then we also seated it with three different blog posts. Here we go. So here is the starting point here. So within here we actually don't have clickable elements for the blog post. Okay. So now since I didn't specify how I want it to look, what I'm going to ask for is I'm going to say I want to make sure that this looks beautiful. So let's use a neo brutalist theme. And additionally, let's fill out the homepage. Let's have a complete SAS landing page. So things like a pricing section, a FAQ section. Let's have a really rich footer as well with some Lauram Ipsum type of text. And overall, let's add in much more content specifically to the homepage. And I want to resolve the blog post so I can click through each page and be able to have some beautiful interactive code blocks as a part of the examples as well. Now, the one interesting thing with Claude Sonnet 4.5 is they did mention within the announcement that they have observed it running for upwards of 30 hours on some tasks. You can imagine if you get creative enough with your instructions for cloud code and you let it run off autonomously, it could potentially do things like complicated refactors or if you have a very detailed spec that maybe requires iteration along the way. It could potentially run for several hours. Now, I've only ever got it to really run for something like 20 30 minutes at most. And that was really just to try and have it explicitly run on something of a complex task. So like generate a ton of different applications. s autonomously and just to see how long it would actually run without intervention. But I'm definitely curious if you have used it and have had it run for hours before. Let me know in the comments below. I'm interested in hearing people have used this. Here we go. So here is what it has generated for us. So within here we have a much better looking site. We have a much better example on the homepage. And again with just two prompts, this is what we were able to get. We have a functional FAQ section. We have some relevant placeholder information. We have some call to actions. And then we also have a really rich footer as well. Now if I go over to our blog post. So previously we weren't able to interact with these. But now if I go and I click through to one of these. So we have the modern CSS techniques. And again with just one prompt we even have these beautiful renditions of these little code blocks with the syntax highlighting which is something that I didn't explicitly ask for but it's nice to see that it did also include that. We have the modern CSS techniques. We have the CSS properties. Then within here, if I just go through the actual content of here, we actually have relevant coherent content within at least this post. And if I just go through TypeScript best practices, we have the starting point some of these different things that we could have within our application. And at this point, since it is a static site, if you just wanted to deploy this at Verscell, you could just go ahead and run Verscell, set up the deployment, and in just a number of seconds, you would have this as a live site. So now I want to push the limits a little further. I'm going to say I want to create a games page. And specifically, what I want to have is I want to start with Asteroids and have through the years some of the most popular games all the way through to the year 2000. Specifically, what I want you to do is I want you to first create the page where all of the different games will live and then in parallel create each respective page for each of those games. Now, the one thing that I do want you to be mindful of is I want to make sure that they all have the neo brutalist theme. And finally, I want to make sure that we have at least 10 games of varying complexity. Okay, so I'm going to go ahead and send that off. And then as soon as it's done, we'll circle back and take a look at what it has generated for us. Here we go. It's created the games page for us. We have 10 classic arcade games of varying complexity. And then all of the games include the full neo brutalist style. Before I move on, I want to thank today's sponsor, brilliant.org. or Brilliant takes a fresh approach to learning that perfectly complements these AI developments. Imagine a fitness center for your mind where every lesson makes you mentally stronger. Instead of sitting back, you're diving in tackling challenging problems with crystal clear explanations. Research shows this approach is six times more effective than conventional learning methods. Their latest course on large language models especially caught my attention, offering a comprehensive yet digestible exploration of AI technology. You don't need to be a coding wizard or an academic scholar to benefit from Brilliant. It's designed for anyone eager to enhance their analytical thinking and broaden their understanding of modern concepts. Whether you're on a daily commute, lunch break, or evening downtime, Brilliant transforms spare moments into valuable learning opportunities. To learn for free on Brilliant, go to brilliant.org/developers digest. Scan the QR code on screen or click the link in the description. And Brilliant's also given our viewers 20% off an annual premium subscription which gives you unlimited daily access to everything on Brilliant. So without further ado, let's go ahead and take a look at what it has generated for us. We'll go ahead and we'll refresh our page here. I'll go and I'll click on games. And then within games, let's first take a look at Pong. Here is our game of Pong. So we have the simulated opponent here. And if I go and I hit the ball, I can see that we have the proper collision detection and all of that. And if I go and I deliberately try and miss this shot, I can see the scoring works as well. We have pong functional. Next up is connect four. So if I just go ahead and test this out, if I get four in a row, I can see that the game has ended. So we do have that functionality. We also had the reset button there. Next up, we have the game of snake. Now, the one thing we'll just have to disable the events of the keyboard presses here where we have the layout shifting, but we overall have the functionality working where we're adding to our snake as we collect the different pieces. Here is the classic game of Breakout. And again, it's matched the styles that we have, the overall aesthetic quite well. And we see the collision detection and all of the scoring work on this one. Now, if I go over to asteroids, I again I have the same issue with the event handlers. So that would be able to be resolved probably within just one prompt here. But if I'm able to shoot all of these different asteroids here, I can see that they begin to dissipate in size. We have that same aesthetic. All of it seems to be functioning. Now for Pac-Man, this is one I've tested on some other models before and often times, more often than not, it doesn't work. Let's go ahead and see how well this does. Within here, we see the Pac-Man not turning. We also don't have any simulated aliens within this. In terms of actually collecting the pellets, we do see that it collects them at the end of each of the rows, but it's not actually collecting them as we go by. So, there are some bugs in this one, but again, for just one prompt for asking for 10 different games, it does seem to be doing quite well. Next up within here, we have Missile Command where we're trying to protect our six different cities. This actually is quite difficult and again, it has the same sort of look and feel. We had that collision detection and we also had the scoring all work. Here is our game of Frogger. So we sort of have this 8bit game and then within here I cannot really tell what is supposed to be roads as well as cars. So within here it looks like that dark blue is actually what causes a collision. But I'm able to go within here and then if I get enough collisions, we have the lives that decrement and then we also have the final score here. I have tried Frogger with other models before and that is something that I haven't had success with. And then here we go. So here is Tetris. So I'm able to use my keyboard. Again, we just have to disable the event handlers on the keyboard. So it's scrolling down the page for all of these. But otherwise, it does actually work. So again, just to circle back to what we have here, we have our main homepage that we were able to generate with Oneshot. We have the blog that is functional with the coding blocks. And then we also have all of these different games within here. Honestly, in terms of the functionality with just three prompts as well as this agentic coding style, I didn't even have to write a terminal command to instruct it to create a Nex.js application. Overall, I'm curious your thoughts. Obviously, this is a relatively simple demonstration. It was able to do quite well, and mind you, a lot of people aren't going to be generating these arcade games, but just to do a little bit of an initial test with the model. I'm curious your thoughts. Let me know what you think or what you've built with this latest model from Sonnet 4.5. If you found this video useful, please like, comment, share, and subscribe.
Technical content at the intersection of AI and development. Building with AI agents, Claude Code, and modern dev tools - then showing you exactly how it works.