Claude Code: The Future of Coding?

--- type: transcript date: 2025-07-05 youtube_id: L9qCRED--go --- # Transcript: Claude Code: The Future of Coding? In this video, I'm going to be going over Claude Code. I've used this over the past 30 days or so, and I'm ready to share my thoughts. I think it's probably one of the best coding agents that is out there. It's quite a bit different than what you're probably used to if you're using something like Cursor or Windsurf or one of those other AI assisted that are out there. So, in this video, I wanted to just touch on why I think Cloud Code is as impressive as it is and also just touch on where we're potentially going as an industry. First up, I just want to pull out a Google Trends chart. I just wanted to touch on the history of some major AI coding tools. GitHub copilot. So this came out over four years ago, believe it or not. But it wasn't until after Chat GPT where we really started to see GitHub Copilot gain some significant momentum. And this was right around the time when we had things like GPT4 come out and models just started to get increasingly better at being able to code and actually being useful in what we use day-to-day. Fast forward to late last year, shortly after Curser announced their series A in the summer of last year, they had a ton of momentum. Everyone was talking about them on social media. There was a ton of creativity in terms of how people were leveraging the tools. Also paired with what the Cursor team was shipping in terms of new capabilities. They launched something called Composer at the time, Cursor Agent. And that momentum has really just continued on. So right now they're a multi-billion dollar company. It's just an incredible story. the cursor story. But what's interesting as of recently is Claude Code. And why is this important? Why is Claude Code so different? I have a few different ideas on this. There are a handful of things that I want to touch on. First up, Claude Code came out in February, but one thing you'll notice is if we look on this chart in February, we had a little bit of a bump here, but it wasn't until Claude 4 came out when people really realized, "Holy this is incredible in terms of its capabilities." myself included and in particular Claude 4 Opus. Now the one thing that I do want to touch on what it really solidified was how important metrics are like Swebench as well as terminal bench. These are effectively agentic coding benchmarks. And one of the interesting things within this and I think it actually fell under the radar a little bit but there were some very large companies most notably rakutin who was able to run claude independently for 7 hours with sustained performance. And when I saw that I was like okay this can't be real. What are they doing for 7 hours to just kick off cla code for that long to be able to do that? And after using it I have an idea. I've been able to get it to run for about 15 minutes, 20 minutes, 25 minutes if I'm explicitly trying to test how long it will go. But it is really impressive once you give it instructions and you give it the proper permissions to run what it needs to run for the terminal commands and all of that, it will just keep going. It will just continue on. You can instruct it to test things, build things, and it really is like a smart developer. It's great at actually writing code, yes, for sure. But actually knowing when to take action, when to stop, when to test, when to invoke the proper tools, that is where these models really shine within a tool like Cloud Code. Next up, in terms of what I think they got right with this was the pricing. They came out with this max plan, which is similar to something like Chat GPT's $200 a month plan, but the difference with this is it actually comes with higher usage to Claude Code. you're going to be able to have 5x the outputs or 20x the outputs depending on the max tier for using cloud code in particular. I've been using the $200 a month max plan. And this is the first tool that I've noticed that it's actually a step function above what I was able to previously produce with leveraging cursor. And that's my benchmark with these types of tools in terms of how sticky they are is what tool actually allows me to accomplish what I want without running into major bugs and all of those types of things. And that's the thing with cloud code is it isn't trying to be a ton of different things. It's actually designed and deliberately built for an environment where we don't necessarily know what the UX is going to look like for leveraging these LLMs within coding applications. First up in terms of getting started, all that you need to get started with this is you can run the installation command. Once you've gone through the installation steps, you can select whether it's your API key or you can log in to your cloud account to be able to leverage this. The other thing that I think they really nailed with this is the form factor. You're able to use this within any terminal. You could spin this up on a Linux instance. You can use it within iTerm or the terminal that's built into the Mac. Now, I want to show you how this works. So, first what I'm going to do is I'm going to create a new Nex.js application here and then I'm just going to go through all of the commands. Initialize cla within here. Do you trust the files within this folder? I'm going to say yes. Now, within here, the first thing that I do want to point out, you do have slash commands. There's a number of helpful things within here, whether it's hooks or looking at costs or adding in an MCP server. There's just a bunch that you can configure within there. Now, the one thing that I do want to note is I actually don't use too many slash commands. And the way that I generally use it, I don't use any sort of task manager or to-do list feature that is outside what's built in. I generally come into the tool with what I want to accomplish and then I just iterate on what I need to do and then I hop out once I'm done whatever task that I had in mind. Now, in terms of how I like to leverage Cloud Code, I like to have my development server in a separate tab here so I can go and move it around and do what I need to do. Typically, I'd have these on one screen and then I'd have what I'm working on the other screen. So, if I take a look here, we have our Nex.js development server and then on the right hand side, we have cloud code. So, just to demonstrate this, I'm going to say replace the homepage with a beautiful hero that reads developers digest. And this is just a simple query, but just to demonstrate how this works, one thing that's really powerful with how cloud code is built is it has a really effective way on how it can actually traverse and find the relevant bits of code of what you're asking for. And the way that this is set up is it's actually quite a bit different than the semantic chunking methods that we see within some other editors, whether it's cursor or some other tools that are out there. I did see an interview from the creator of Claude Code and one of the things that he mentioned was actually by leveraging things like Grep and relying on the LLM to write out the commands to search and traverse your codebase with things like reax. I think taking a different approach than what was the industry standard in terms of how to actually traverse a codebase. It does seem like it is quite an effective strategy. One thing that I do want to know with this and these are going to be the things that you use quite a bit within cloud code and this is going to be shift tab. This is going to allow you to pivot from the manual mode where you have to approve everything, the auto mode as well as the planning mode. If I just turn this on here, what I often do when I'm leveraging cloud code is I will just shift between those three modes. If it's more of a high stakes change and I want to see everything that it's doing, I'll put it within the manual mode. If it's a little bit more of a haphazard thing that I'm asking for or if I have high confidence that it will be able to accomplish it, I can turn on that auto mode. And if I don't have a lot of confidence, that's when I turn on the planning mode. One of the things with the planning mode is you'll find it's really effective if it already has context within the conversation. If it's already done a number of changes and it has the contextual awareness on what you've done that's related to what you're about to do, if you turn on that planning mode, it can be really helpful to just think through the process of what it needs to do next. Here is the change that we asked for. Now I'm going to ask for something a little bit more involved. I want to create a header footer as well as a contact and blog page. I want this to be in a glass morphism theme. Now if you send in a complex query like that, what we'll see claude generate is it will think through what it needs to do, but it will actually generate a to-do list. And this built-in to-do feature is honestly quite killer because what we'll see is it will go through step by step that list in that order. We're going to see that it's going to build out the header. It's going to build out the footer. I'm going to give it permission to create the components directory. And now we'll see it work through the to-dos. And why it is nice to have the separate terminal with your separate server is we'll be able to actually watch these updates in real time as they happen. We see it's working through. It's created the header for us. We've created the footer. And now it's going to update the layout to actually include the header and footer that it created. And the one thing with this that I really quite like, especially if you give it a little bit more instructions at the time of when you prompt it, is it will actually go in and test things and build things and it can actually reflect on things about whether it's the proper approach or if you want to say something like, "Hey, improve this or I have this idea. Maybe you can guide me." Use it as almost like a rubber duck. That's something that I really like with tools like this. This is what we ask for. We have our header. We have our hero area in this glassmorphism theme. We have this footer. And if I take a look at our header here and I click over to blog, I can see we have this blog page. If I click over to contact, I have this contact form. Now, the one thing to know with this is what I find is if you just ask it without specifying these styles. Obviously, this was just a really simple implementation. But the thing that's really impressive with this, and this is the one tool that I've never used before, it can actually get further without you having to look at code than tools like cursor or obviously GitHub copilot. Now, obviously that's just a really simple example, but what you can do with this is we can just iterate from here. So, I'm going to say I want to add an app router API that returns hello world on click of a button in the hero area. Here's another simple request, but mind you, it does not need to be simple. This is a very sophisticated model. You can ask it to do quite involved things and it probably will do quite well. But again, if we take a look at this, it creates a to-do list. We see that it wrote out a simple hello world route. We're going to go ahead and add this button where we call that endpoint. And then ultimately, we have the state update for when we call it and all of that good stuff. And then it describes everything that we have here is it will generate a to-do list. But you can exit out at any point if you don't like the direction that it's going. Or alternatively, if you wanted to plan, if I click shift tab, you can put it into plan mode. Say if it's something hard like I want to add some crazy difficult thing. Let's brainstorm ideas. And one thing that I really like about this is all that I need to know in terms of the different key bindings and commands for actually leveraging this is just shift tab. You can use those slash commands and what have you, but I find the manual mode, the auto mode, as well as the plan mode, that honestly can probably get you through most of what you want to do, like 90%. You can definitely try out some of the cool features that are built in like hooks and MCPs and all of that good stuff. That is all definitely great stuff to check out, but in terms of the core functionality, just shift tabbing through what you want to do can get you quite far. Now I wanted to show a couple charts that I saw that I thought were quite interesting from the creator of Cloud Code, Boris Churnney. We basically talked about programming, how it went from punch cards in the ' 50s all the way through to these lower level languages like forran or cobalt and basic and then we get C and C++. Eventually we have JavaScript and Python and Java and now today we have things like TypeScript, Rust, Swift, so on and so forth. Now the interesting thing is where are we going to go in terms of what we actually write for code and the trajectory is undeniable that increasingly so over the coming years we're going to be writing more and more code with natural language just like you saw me demonstrate in this video and the other thing to know with this is in terms of actually where we write code there was Ed and Vim and Emacs and eventually Visual Basic Eclipse Sublime Neoim Copilot and then we had Cursor Devon and then Cloud Code. Now you could look at some charts like this and say okay hey this is maybe a little bit self- serving but I think directionally these are very correct like maybe cloud code isn't going to be the tool that we leverage for actually writing code of the future maybe it's going to be cursor or maybe it is going to be Devon or some other tool but the thing in common with all of these types of tools is the big difference with all of these is obviously they leverage AI but the key way in terms of how we actually increasingly write code is with natural language And I think it's pretty undeniable that more and more of our code is going to be generated from natural language over the coming years. But otherwise, that's pretty much it for this video. I just wanted to do a little bit of a different look at Claude Code. I can do some more demonstrations in terms of how you can leverage Claude Code if you're interested in that sort of thing. But otherwise, if you found this video useful, please comment, share, and subscribe. Otherwise, until the next