Claude Code Will 10x Your Productivity

Key Takeaways

Welcome to this video where we're going to take a look at Cloud Code. Cloud Code is a very powerful AI tool that runs in your terminal. It promises a huge leap forward in developer productivity and can change the way we think about building software. By this point, you've probably used ChatgPT or Cursor for AI assisted coding already. So, if you're like me when I first heard of Cloud Code, you might think it's just a CLI version of an AI chat, but that would be a huge understatement. Cloud code is a generalpurpose AI agent that runs in your terminal. It can read and modify files for you in your project directory and you can also easily integrate it with external APIs like GitHub or Slack. This basically gives it the ability to act as an autonomous agent and do complex things for you, things that would normally take hours or days. Here's an example I ran earlier with Cloud Code implementing a complete feature across several files. um it created a plan and a checklist of tasks that it needed to do and then it did them one by one. And if I go to my GitHub page, I can actually see the pull request it's created. Um all I have to do is review and decide if I want to merge that. And this all happened with a single command. So it's not just answering questions or changing uh one part of the code at a time. It's actually doing this job in multiple steps, planning, making changes, testing it. um and then creating the git commit and pull request to complete the feature. In this video, I'll show you how you can install cloud code, how to configure it, and how you can use it at a basic level. Then, we'll go through its three essential features that you need to learn first before you can really leverage its full potential. And of course, I'll be supporting this with real examples along the way. From what I've seen so far, I think this is the real deal. And I don't mean cloud code specifically, but rather the idea of uh general purpose agents to plan and work on your project for you. I think that this will still take an incredible amount of technical knowledge and skills to leverage effectively. Um but it's going to be a different set of skills from traditional software engineering. And the exciting thing is that I don't think that anyone has it quite figured out yet. So anyway, let's get started. First of all, what is claude code and how is it different from other AI tools you might have used like cursor? Well, unlike chat GPT where you just have a back and forth conversation or cursor where you collaborate on specific parts of your code, uh cloud code takes it a step further. It can do much longer, larger complex task uh that has multiple steps and execute them in sequence, figuring out what to do next until it gets the job done. Here's an example of a typical workflow. You can give cloud code an instruction like implement a new user profile endpoint to see usage limits and then validate these with unit test and finally create a pull request. Now let me actually run that in my terminal for you. We'll see what happens. You could see that it starts by creating a plan um a checklist of all the things it has to do and then it starts reading the files throughout the project to figure out where it needs to make these changes. And I don't even have to interact with it or do anything. I can just watch it work while I go and grab a coffee. Uh eventually it gets to a point where it's actually changing the code and you can browse and see all the changes that it's making here. You don't have to approve them, but you can also ask it to slow down uh and ask for your approval if you wanted. Now, here is an example where it's actually stopped uh to ask me for a decision. Uh this is because it's doing something new. It's trying to run a Python test and I haven't told it that it could do that before in this session. So, it's going to ask me and I can choose uh I can let it proceed or I can tell it to uh don't ask me again for similar commands in the future. Now, of course, it's not going to go perfect the first time. Uh in this case, when it ran a test, there was a failure. Uh so instead of failing and just stopping and asking me for the next instruction, it actually analyzes this failure and then makes another plan on how to fix it. And you could see here um it's identified the fix is that it needs to patch the dependency properly uh for fast API. And overall this entire workflow took about 5 minutes and most of it was just me watching it work. Um but now it's done. And then when I go back to the GitHub for this project, I could see that it's created a feature branch ready to be reviewed and merged with the project. So as you can see, this development experience is really quite different uh from using chat GPT in the browser or from using something like cursor. You can still use it alongside cursor or VS code. So it's not really a replacement for an IDE. Uh you can think of it more like another software engineer that can work with you in parallel and in the background. It's pretty easy to get cloud code set up. You will need NodeJS 18 or higher. So make sure you have that installed first. Then you can just run this command to install it. Uh and this will work if you are on Mac or Linux. Cloud code doesn't natively run on Windows though. Uh but you can use it through WSL. It will work but it will probably have some limitations. Once it's installed, you can run Claude in your terminal and it'll walk you through authentication. You have a few options here, but I strongly recommend using a Claude Pro or a Claude Max subscription instead of API credits. That's because it can burn through tokens really quickly when it's working on complex task. Uh, and with API billing, you could accidentally end up with a huge bill. With a subscription, at least you're charged a flat monthly cost. And once you're authenticated, uh the workflow is that if you want to use it for a project, you navigate to that project folder first. If you have a Python environment or a node environment, you can enter that environment in your terminal as well. And then just run this cloud command to start the agent with that project directory and with that execution environment. So let's go ahead and take a look at that here. Um here I'm in my project, which is a to-do list. Uh let's start the cloud app. And the first thing it will ask you is if you trust the files in this folder. So I'm going to trust it. And once you're in uh your interface will look a little bit like this. You will have a command prompt here where you can type any instruction you want. And there's also a couple of tips for getting started. And it will also offer to create this claw.md file for you which is what you will use to provide instructions for that particular project. Um there's a couple of shortcuts as well. If you press question mark, you'll see that there's a couple of shortcuts here for different things that you can do. For example, if we press slash, you'll see different commands you could use. And if we press at, you can see that we can refer to specific files within this directory. And to exit, we just press Ctrl + C twice. Now, with the tool set up, let's walk through some practical examples to see what Cloud Code can actually do. I'm going to start it up again in one of my project folders and we'll use it to do some simple tasks and slowly build up to a more complex workflow. First, let's have Cloud Code analyze our project to understand what we're working with. So, this is the command I'm going to run. And you can see it's scanning through our entire codebase, reading all the files. And uh it's giving us a complete overview of the projects, the structure, and all the interesting files uh that it needs to know about. Now, I think that this is pretty useful if you're joining a new project or you're coming back to an old project after a while. Now, instead of just having this explanation pop up in the terminal, what if I wanted to save it to a file in my project? Here I'm going to ask it to write this entire description and save it into a file called project description. The first time it needs to do something new. In this case, creating a file in my project, it's going to ask me for permission. I can choose to grant that permission once or I can choose to grant it throughout the rest of the session. But if I don't like the instruction, I can also choose to decline and tell it to do something different. In this case, I'm just going to grant it the permission. So you can see cloud code has full access to your file system. It can read the files. It can also create new ones and modify existing ones. Uh this is pretty different from just chatting with AI because it's actually working within your project. Now let's take this a step further. I'm going to ask it to read this project description for context and then use it to research and brainstorm 10 new feature ideas for this project. Uh and here I'm using the at symbol and you could see that it will let me reference any file or directory uh in the prompt. And here's my complete prompt for the next instruction. I wanted to use this existing file that we just created. Um, I wanted to brainstorm a list of 10 simple new features, uh, using the web for inspiration and then save all of those ideas into a separate file. Now, if I wanted to see what Claude plans to do, uh, instead of it actually going ahead and doing all of that, I can actually switch between planning mode and execution mode. Uh, and the hotkey for that is shift tab. So, if you see here, I'm pressing shift tab. It's now switched to planning mode. And now when I run this, it will think about what to do, but it won't actually do it. Of course, this is a pretty simple instruction, and I'm confident it's going to do the right thing. So, I'm just going to switch it back to the default mode. And then I'm going to run this. And again, we can just watch while it's doing its job. Here, it's following our instruction to search the web uh for ideas and inspiration as well, which is what we told it to. Now, I didn't have to set up or install anything special for this web search. Cloud Code is able to do that by default. This is really useful if you needed to look up API instructions or documentations. Um, and here I use the bash command to uh see the output of this file I created. And you can see that it's given us 10 simple examples for features that we can implement. Now those three steps we just did, analyzing the project, documenting it and then coming up with new features, um it's already quite a complex workflow with several steps and we had to sit there and guide it through it manually. Um and also imagine that if you go to a new project and you wanted to do the same workflow again, uh you'd have to provide it step-by-step instructions. So there's actually a better way to do this. If you find a workflow that works really well and you wanted to use it again, uh you can turn it into a custom command. I've actually set up this command called project analysis that does exactly all of those steps we just walked through. Um and it will show up when we type slash here. Uh so these are some of the default commands and at the top this project analysis command is our custom one. And now it's able to run that entire workflow uh and produce these two files for us without me having to intervene at all. And this was just from one command in a completely different project. Now that we've seen some end to end examples of what Cloud Code can do, let's take a look at these three essential features. First, we'll take a look at how you can configure Cloud Code. And this is really important because your configuration will be tailored specifically to you and your projects. Uh, next I'll show you how to set up a custom workflow like the one you just saw. And finally, I'll show you how to connect it to external tools like GitHub or Linear. To configure Claude Code, there is a special file that Claude will look for, and it's called claude.md. Here is an example of what that could look like. By creating a cloud MD file in your project root, you give cloud code instructions for your project, including coding style, architecture decisions, and best practices or conventions that you'd like it to use. It will follow this guidance really closely uh for whatever task you tell it to do. And this is just a normal markdown file. There's no specific format or structure. Just write this as if it were a set of instructions for a human to read. And if you're using cloud code for the first time in a new project, you could actually use this /init command uh and it will generate a new cloud MD file for you. Here it's going to try to reverse engineer your project to figure out what set of instructions work really well for this project. And here is the contents of the file that it's generated for me for this project. It has details like what commands to run and what architectural patterns it's seen in the project as well. Now, if you wanted to configure cloud code globally, you can also put this cloud MD file into your home directory like in this example here. This will make it apply to all of your cloud sessions. Now, exactly what kind of things should go into this file? Well, that's completely up to you. I recommend just start with it empty and then whenever you run into a behavior that you wanted to change, just add that into the file. For example, if you are using this to build a Python project and you run into a problem where it's trying to test your app by just running the app, if you didn't want it to do that, you could instead add a rule uh saying that instead of trying to run the script, uh write and run unit tests instead. Now, moving on, you remember that SL project analysis command that I showed you earlier? Cloud Code gives you the ability to create custom slash commands for repetitive task. You can turn any multi-step workflow into a single command uh that you can reuse across multiple projects. Uh and to do this, you start by creating a command folder. You can either make this project specific uh by putting them in.cloud/comands or again make these commands global by putting them into your home directory. Then it's really simple. Any markdown file you put into that command folder becomes a slash command that you can run anywhere. So now let's create that project analysis workflow we used earlier as a global user command. Within that file, I have just put the set of instructions in plain text uh for claw to use whenever I run the command. You could read through these if you want, but it's pretty much the same set of instructions that we used in the earlier part of the video. And so now when we go into claude in any project at all um we will see that command pop up and we could use it uh by just typing slash project analysis. Now another thing worth knowing is that these custom slash commands don't have to be just a static prompt either. You can make them take arguments and the syntax looks like this. All you have to do is just use this dollar sign arguments keyword inside your command. uh and then you can pass that in when you actually type the slash command in your cloud session. I really like this feature because it really helps you to build uh your own library of useful tools and automation workflows. So uh when you're using cloud code, I want you to think about what kind of things you do repeatedly like for example code reviews, uh feature implementation or testing and see if you can turn them into custom commands. The third essential feature of cloud code and probably the most powerful is its ability to connect to external development tools through something called MCP model context protocol. Now if you're not familiar with MCP, it's basically just an AI adapter for LLMs. If you want to learn more, I have a short video about it which I'll link in the description. Basically, instead of being limited to just reading and writing files within your project, Cloud Code can use MCP to interact with other important systems like databases, um, cloud providers, GitHub, Slack, or basically anything else that is accessible via API. Now, let's look at a practical example by connecting to Linear, which is a popular project management tool. On this MCP directory page, I could see a bunch of MCP servers that are available uh right now to connect with cloud code. And here are the documentations and endpoints. So this is the endpoint that we need. Uh this is a remote server which we want to connect to. So I'm just going to go ahead and copy that. Once we have the MCP endpoint, all we have to do is use this command in our terminal to connect it with cloud code. You could add this MCP server locally per project or you could use a scoping flag to add this globally for your user. And for MCP servers that need authentication like this one, there'll usually be a separate step when we boot up Cloud Code asking us to authenticate in the browser. Once an MCP server is connected, you can type /mcp uh to see all of your servers. And you can click enter to look into the details. And then if we wanted to see all of the tools or the functions available, we can click to view tools. If you haven't used Linear before, it's basically a project management tool. So you could see that it's got all of these functions for us to interact with our project. Now that it's connected, we don't have to do anything special. Uh Cloud Code should figure out automatically when it should be using these tools. Here, I'm going to type a command for it to show me all the issues related to this task management project. And once it's done, it's found these 14 open issues for me uh from my linear dashboard. And I can verify that by popping over to my linear dashboard here as well uh and seeing these issues. And now I'm starting to see a really cool uh workflow integration where you're just managing your project and your issues through whatever app you're normally using like linear in this case. Um and cloud code can sit in the middle and actually read those issues and execute them on the project. So for example this issue here creating a quick capture component. If we click into that, there's a couple of instructions um so that for our task application uh we can kind of use a hotkey to add a task really quickly. Let's see if cloud code can actually pull this ticket down and implement it in our project uh with one command. So now I'm asking it to implement uh this task which is to create that quick capture component. And once again it's making a plan on how to approach this uh and taking off these action items one by one. Now, usually when you ask it to do a larger task like this that has multiple steps, it could take a while. So, normally you could go and grab a coffee or even just kind of put this into the background and do something else while you wait for it to do its job. Now, this took about 2 minutes. It's gone ahead and updated the entire project for me. And now I can pop over to the local version of the site uh and test the feature. So, if I press command K, uh I'll see that the quick capture window pops up. And this is a completely new feature implemented by cloud code. Now I can close the loop on the workflow and also ask it to update the status and the progress of that ticket on my linear server. And when it's done, we should be able to go back to our linear app uh and see that the ticket has been updated on our dashboard. We look into that. We'll see that it's been changed to in progress here and that the description and comments have been updated with what's changed as well. You will probably need to have a really solid review uh or a quality control workflow to make sure that it's on the right track, but at this point you could probably already see uh how much of a productivity boost this can give you. Now, before we get carried away with using this, uh, let's take a quick look at what model and what pricing Cloud Code actually uses under the hood. By default, Cloud Code uses Sonnet 4.0, which I think is actually good enough for most use cases, especially if you structure your project well. Uh, and you can set it to use Opus model as well, but that will cost about five times as much in terms of tokens. And in terms of cost, there is no free tier, and there's currently two ways to pay for this. You can do it by using your API key or by being subscribed to the pro or the max plans. And if you choose to use the API key, you will be build by token usage. I strongly recommend not doing this because uh you can easily use a lot of tokens without realizing it and then get stuck with a huge bill. Generally, I think it's much better for most people to just stick with one of the paid plans so that you have fixed usage and predictable monthly costs. The pro plan is $17 a month and I recommend starting with that. Uh then if you run into limits, you can consider upgrading it to the max plan which is $100. Using cloud code so far has made me think differently about how we build software. Instead of writing code line by line, uh we're kind of moving to designing systems and workflows for AI agents to use. As I said at the beginning of the video, even though it looks like this makes things a lot easier, I still think that it will require a lot of knowledge and skills to be able to use effectively. And I think there'll be a big gap between those who have figured it out and those who haven't. I'm going to be figuring out a lot of this stuff for myself as well. So, subscribe to this channel if you want to be a part of that journey and learn with me. I'll see you in the next one.