claude Code is having a moment right now but if you've been using tools like Cursor you might be wondering is Claude Code actually better well I've been using Cloud Code inside of Cursor for the last couple of weeks and I've become kind of obsessed with these two questions one is on the interface is a terminal-based interface with a coding agent like cloud code actually better than the UI that I get with cursor for chatting with an agent and reviewing and approving code changes or is that the wrong comparison to make the second question is about performance so if I already have access to Claude's most powerful models Sonnet and Opus inside of Cursor then do these models perform better if I use them through the Claude code interface H so in this video I'll show you how I'm actually using Cloud Code inside of Cursor and the specific situations where I actually reach for Cloud Code versus the times when I would actually prefer to use the agent that's built into Cursor i'm seeing a pretty strong case for both but what's fascinating to me is the statement that Anthropic is making with their release of Cloud Code what are they telling us with this product where is this going well what they're telling us is that the future of software engineering is not in engineering code it's engineering agents who write the code now a quick reality check have we arrived at this future where we as professional designers developers we're never going to get our hands dirty in the codebase anymore no in fact in this video I'll show you what my real day-to-day looks like where I'm shifting between these different modes where sometimes I'm delegating entire tasks to an agent with clawed code and other times I'm collaborating directly with the agent and accepting code changes line by line with cursor but still we can't deny where the momentum is shifting agentic coding seems to be the future and with claude code we are actually starting to step into that future right now stepping back I mean the pace at which things are changing right now for us as builders it's like nothing I've ever seen before in this industry so I'm sending out a weekly builder briefing it's a free five-minute read to help you and me keep up with today's tools and our craft in this new era of building with AI you can get your briefing at buildermethods.com so let's try to understand cloud code by listening into this interview with its creator Boris Churnney from anthropic and so I see this as the evolution so this is the languages have sort of I think leveled out kind of all the modern languages belong to kind of similar families you know there's like a few big families of languages and they're pretty similar if you kind of if you glance but the experience is now uh really evolving where now you don't have to deal with punch cards or assembly or even code you you deal with prompts and the model figures out the coding part and this is just hugely exciting to me as a as a programmer now if you're not using Claude day-to-day just yet I want to show you just a few of the features that I think are unique and bring something new to the table so starting off of course we're going to look at the orientation of cloud code being a terminalbased interface so here I am in cursor i've got my terminal down here now of course I already have claude installed on my system that's a simple oneline npm command you can get the instructions through cla's website but once you have it installed you can run it in any terminal and that's one of the great things about cloud code is that you can literally take it anywhere or take it into any other environment or IDE code editor whether you're using cursor VS Code Windsurf or even just any terminal client you can run Cloud Code anywhere but what's really cool is that they also have these direct integrations with the popular IDE including cursor and VS Code and Windsurf and all the others so once you have Claude running in the terminal you can run slash IDE to manage your IDE connections and as you can see I already have the cursor integration installed and so what that means is if I'm in a file here you can see here that it knows exactly which file that I'm in also if I highlight some lines it'll show that it it has specifically five lines selected so that goes right into the context window when I'm asking the agent something or instructing the agent on these specific lines of code or we're kind of zeroing in on this file in my codebase you know just like I would if I were in the chat window in cursor now I don't have to actually use it in the actual terminal window it also has this direct placement in cursor up here so if I click that icon I can get rid of the terminal and then kind of use cloud code in its own dedicated window and I've been going back and forth with this and you know I don't know how much I actually like it to have it in its own window like this i think it could just be like the placement on the screen in in the in the layout because it really does sort of like cover up my actual code files that's part of the philosophy behind cloud code is that you're interacting with agents in the terminal and we're starting to sort of take us away from our actual on the ground codebase right but the reality is I still jump back and forth quite often between the code and the cursor agent and my terminal so I actually prefer to stay in here because then I can use my my command T to pop open and close the terminal i can you know fire up additional terminals and then I can use my command L shortcut to open and close the cursor chat that's sort of like the rhythm and the layout of using claw code inside of cursor that that I tend to like now another really interesting feature of course is MCP integration anthropic are the creators of the MCP protocol and just two days ago from when I'm recording this video they announced an official MCP support in cloud code what does that mean well that means that you can integrate all sorts of other tools that you might be using in your development workflow and interact with them right from within cloud code so for me I'm a fan of using Linear for you know my ticket and issue management writing up bugs and planning out the roadmap for my products and so we can use Linear's MCP server integrated with Cloud Code which really improves the workflow of taking the next linear issue and having Claude Code work on it let's take a look at that right now so I've already connected the linear MCP to my Cloud Code and you can check Cloud Code's official documentation on how to do that they've got really great docs by the way which are really helpful so having that installed allows me to literally just type in and say things like "What's the most recent linear issue?" All right so it did find one but that's actually not the one that I want to do but an even faster approach is to literally just go into linear and and grab the the latest issue okay so it found it it actually took a few moments to to get there there and by the way I must say I have found claude code to be a bit slow for these smaller trivial tasks and that's why I think claude code is actually better at just acting as an agent and going off and doing larger tasks we're going to get into that but it did find it and it found all the details that I had already written into this particular issue so we're going to have Claw Code actually work on this and let's take a look at what I'm talking about here so I've been working on the new website for builder methods and if I go down to the bottom there's this contact form which is just name email what's this regarding and a message and I want to actually add another field in between the email and the actual subject line which is going to be a selection dropdown with a few options in terms of like are they inquiring about AI coaching or sponsorship or you know one of my upcoming courses on building with AI i need to know like what the contact message is going to be about so I'm asking it to add that selection dropdown and these are going to be the options for for what that selection will be i also want the agent to update the mailer that I get sent when a contact message is sent so that we get that piece of information included in the mailer so you know there's a couple different files involved it's not a huge task here but let's see it actually work on something and by the way while this is working I want to show you something else so in my in my claw.md file which is very similar to cursor rules it's like the memories for Claude so every time it's going to work on something it can kind of refer to these notes i added some instructions on how to handle linear MCP instructions meaning when you start working on a task that comes from a linear issue I wanted to do a few specific things like changing the status on the issue to in progress and when it's done it's going to drop a comment on the issue and then when also when it's done it's going to change from in progress to in review so just some like organizational stuff but again like I don't need to repeat those instructions they're just built into my claw.md file now this comes up from time to time it asks permission to do something and I want to do shift tab to put autoacc edits on so that's actually going to give it permission for that specific request so I might see that come up again all of those permissions are in clawed settings local and then it's going to sort of just like list out all these different permissions that you grant it as you go along so this is one of those differences in the experience of working with the agent through clawed code where we can actually let it just go and make lots of changes across lots of files i don't have to actually accept each individual file whereas with cursor when it does make edits to multiple files I still need to review and approve like literally click the button to approve every single file edit sometimes that's good and sometimes I prefer something a little bit more hands-off like cloud code so this is all done now by the way I don't know if you actually heard that but in my claw.md file I also added this line to to play the glass sound when a task is done it sort of gets my attention so if I have the agent working on something and I'm working in a different app it'll sort of alert me with that audio sound just like you know cursor has that feature built in as well which I really like so I just added that to the Claude instructions or Claude memories if you will and and that works so this task apparently is finished let's take a look at linear to see what it did there it actually checked off the checklist items that I put into the description in in linear i instructed it to do that it also posted a comment it looks like it's from me but this is actually from Claude Code through the integration and it lists out all the changes that it made here and finally it actually changed the status from in progress to in review exactly what I asked it to do so that looks really good in my project management setup in linear let's actually take a look at the app and we'll take a look at the code and see how it did so yeah here is the select dropdown i do need to fix some styling to make it look more like a selection but looks like it has the options that I want there so that's good all right let's test this out i'm going to select that and make sure that this piece of information is going to be passed along in the mailer that I receive when a contact message is sent okay so here is that mailer and it actually did a good job of just adding it right into the content of the mailer that I received i can review the individual code changes by looking back in the logs here in cloud code now you know I must say like this experience of looking at each individual code change is just not as good in the cloud code terminal interface this is the kind of thing where I really prefer interacting with the cursor interface if I want to have that granular control and reviewing and approving each individual line of code or maybe you know rejecting certain lines of code that's a lot easier to do in the way that cursor lays it out but as we can see here it basically did everything exactly as I wanted it to so Claude clearly is a very capable agent this of course was a pretty simple task but I think it does just as well on much more complex tasks so another feature in cloud code that makes it really powerful are slash commands i mean custom slash commands right so it is a terminal interface and it's all about you know using the the slash key to to get to things now this top one get commit I actually created that as a custom slash command so your custom slash commands live inside your cloud folder inside commands now in this project I've only created one that's for git commit it's a markdown file these are my instructions for how I want a git commit message to be created every time it's a really deep description of exactly how I want commit messages written examples of good commit messages examples of bad commit messages and everything so obviously every time I ask cloud code to commit the changes I'm not going to rewrite all of these instructions so these custom commands are sort of like my collection of saved commands that I can reuse again and again and of course the more we use cloud code the larger we can grow our library of these custom commands so when I use it it's nice because all I really need to do is hit the slash key and it puts your custom commands at the very top so this first one is my command if I had more they would also be at the top and then below that we have the other slash commands that cloud code gives us now these custom commands are also scriptable so you can even pass arguments into these commands and do more powerful things with them again this is what's getting us into engineering the agents as a layer in between us and the actual code so this shift is just really interesting another interesting feature I'm actually not using this quite yet but I'm probably will at some point is the GitHub actions integration with Cloud Code so you can actually set it up so that from within GitHub you can just mention Claude and ask it to review a PR or give it a task and it'll go off and do that via GitHub actions you know I always like to think about what is the meaning behind the tools that we're using to build with AI and right now it's pretty fascinating what's happening with cloud code right so if working with cloud code in the terminal takes us one step away from working with the code directly interacting with cloud code via GitHub actions takes us like another step away from the terminal that it we're now literally talking about like a remote agent that we are delegating tasks to it's going off and doing them and completing them it's not even happening in our local environment again it's this theme of we are moving toward a world of agents doing the work for us and our work as designers and developers and architects is to engineer the agents themselves now let's get back to my question about the interface and the performance of cloud code i thought this part of the interview with Boris was interesting you know like if you're used to handwriting code I think now the industry is shifting to a place where you're orchestrating agents that write your code and it's more about reviewing code than handwriting code and yeah I think people have to kind of deal with this transition and I think as a programmer it's incredibly exciting because you can do so much more so much faster right and there's still some stuff we I I'll have to dip down and kind of handwrite code but now I kind of dread it now obviously I agree that directionally this is where we're going away from handwriting code and toward agentic coding but there are still certain situations where I prefer to be more hands-on even if I'm still working with AI i would say that roughly 80% of the code in my projects is now written by AI and not me actually typing it that the way that I interact with AI agents is different depending on the task you know there are still lots of situations where I just prefer cursors interface for actively collaborating with the AI to design and craft and complete a feature and for me this is especially true for front-end design tasks where I'm writing Tailwind CSS I'm checking the browser I'm tweaking I'm making things pixel perfect i also like to think of cursors agent interface as like a pair programmer where we are zooming in on on a particular area of logic and I can accept or reject individual lines of code until we get the system just right you know cloud code doesn't really do a good job of having that level of granularity or direct collaboration on a feature but there are other situations where cloud code really shines like when I have a large task involves lots of different files in my codebase a lot of leg work cloud code is just really capable and really good at methodically executing tasks like that as long as they are well planned and well scoped and I'm going to get to planning mode in cloud code in just a second now in my last video on this channel I showed you Cursor's new background agents feature which has a very similar use case as cloud code because in both cases we're essentially delegating a task for the agent to work on elsewhere while we do other things that's how I think of these background agents in cursor or you know claude code in its ideal use case it's an agent that can take on a big task while I do something else and this is where I think that claude code really has an edge when it comes to performance because when I think of cloud code as an actual autonomous agent it just seems to be more successful and confident with these larger tasks whereas the background features in cursor is still useful it seems to perform better in smaller bug fixes and tightly scoped background tasks really good for knocking out your backlog as I talked about in the other video so I do think that cloud code is highly capable but it's only as capable as the plan that you're asking it to execute so let's talk about plan mode in cloud code and to understand why plan mode is now a core part of the cloud code product let's hear what Boris had to say about that i'd say the biggest thing that I've seen power users start to do in and out of anthropic is ask quad to make a plan before it starts to code and something people will do sometimes when they first start using quad code is they'll be like "Hey write this really big complicated feature." And then they get frustrated when it doesn't do it the way that they imagined it in their mind and a really good way to kind of align what you want to do with what Claude wants to do is ask it to make a plan and run it by you first so just a few days ago now the team at Anthropic announced plan mode as a core feature in cloud code so you can get to it with shift tab and you sort of toggle between auto accept edits on that's like when you're actually working on a task or you get into plan mode here so let's go ahead and actually plan out the next feature that I want to build and I'll show you what we're working on here so on builder methods I'm planning to release some courses on building with AI i call them just in time AI learning courses now these four courses are in like pre-launch mode so if you click on any one of them we're going to get to this like waiting list form and each one of them has a different waiting list that anyone can sign up for so when someone fills out their email and they join the waiting list currently they just get to this page which just says you're in but since these are upcoming courses I want to actually collect some feedback and understand what the top questions are so that I can really make these courses as useful and helpful as possible right so I like to have these survey forms but I want to make sure that I have a survey for all four of the courses and understanding which course each individual subscriber is interested in and having those questions be specific to the course so you know we're going to need a little bit of coordination between figuring out which course the person has raised their hand to say that they're interested in and then sending in the survey response for each of the four different courses right so I wrote up this feature in this linear issue so this was just me writing this i could have used AI to help me craft this it's sort of just me getting like the raw like functionality out of my head and this is sort of the description i could just copy and paste this into cloud code but since I do have the linear MCP integration all hooked up I'm going to go ahead and just do this i'm going to grab that linear issue right there please plan this issue now while it's working on that plan I want to tell you about the think hard keyword right so what I could have done in that prompt was I could have asked it "Please review this issue and plan it out and think hard." That's that's a key word that's going to trigger Claude Code to spend a little bit extra time thinking and analyzing and coming up with a more reasoned response you can even say think harder or you can even say ultrathink on this task and each one of those sort of ratchets up the level of resources and time that it's going to spend on the planning all that being said I'm currently using the Claude Opus model which is sort of like the best of the best of course it's kind of expensive but I am on the Max plan which is probably the best deal right now for working with Claude Code so I'm sort of just waiting for it to you know analyze and think through this plan and here we go so here is Claude's plan again like the planning phase is a core part of the process it's a best practice and so the team at Anthropic designed it in as a first class feature you could see with the borders like they're really emphasizing like this is how you can be successful with cloud code is if you actually spend the time and sort of frontload the effort in the planning phase and cloud code does a really good job of supporting that workflow so here is the plan that they came up with now I do want to sort of like clarify the plan here with Claude before we move forward because it's sort of describing some things that are already in place and I want to make sure that it's going to focus its effort on the back end on implementing these forms okay so I'm responding to Claude here i'm saying you've described the design and interface as we already have please propose your plan for implementing the survey form couple more details i told it to reference the linear issue and I'm asking it to think hard this time let's see what happens okay so here is Claude's updated plan for implementation you know it did a really good job of numbering this out with bullet points and I reviewed everything here and it all looks exactly how I would want it to be implemented so let's go ahead and proceed okay so Claude is working on the task that I gave it and it's making lots of changes i love how Claude makes its own to-do list and it sort of crosses off the items as it goes now while Claude is cooking I I think this is a good time to talk about something that I just find a little bit challenging right now when I'm working with cloud code and that's multitasking right so you know cloud code really works best as like an autonomous agent where it can go off and do these large tasks on its own and I could open up a second and third and fourth instance of Cloud Code in different terminal tabs but they would all be on the same git branch and of course that means that we would risk running into code conflicts unless I have the tasks really working on totally separate parts of the codebase which is kind of rare it's also kind of frustrating that like while it's working like I can't really go and work on other parts of the codebase again for that same reason i might run into conflicts now there is a workflow to do proper multitasking using git work trees it's a little bit more complicated i'm still kind of working out the workflow for that myself once I do I'll probably post another video here on the channel to really cover multitasking in cloud code because I think that that would be a really powerful use case especially given the fact that it it is so strong as an autonomous agent rather than like an active agent the way the cursor is okay looks like it's all done and it gave me all the information about what has been implemented and let's refresh the page and test it out so I'm going to go down to courses and let's go ahead and subscribe to a course okay so here we are and I can see the new survey form it sort of added some some extra styling around it that I didn't quite ask for so I think I might start to update my clawed memory files to give it just more specific instructions to really dial in exactly what I want it to do and what I don't want it to do you know and really that's the process of working with these agents is like working with them and finding the areas where you just want to course correct it and build it into the systems build it into the instructions whether it's cursor rules or the claude MD file and that's what it means to engineer our agents right but overall this is looking good let's actually go ahead and test it out and it also did a good job of passing along the email address from the first step that's something that I did not have in place before it added in that logic so let's go ahead and send okay so we've got this thank you page again it's it's the front-end details that I'm going to need to sort of tweak but it has all the content that I wanted it to have so all of that is actually working really well but you know there's one thing that I just noticed about this implementation i submitted the survey form but where did that submission go with my contact form I actually get an email sent to me with that survey information and so I see this so Claude actually added this comment for the survey submit action in my controller it says like here you would typically save survey data to your database for now we'll just redirect to the success page so like it didn't really complete the feature but that's not a fault of cloud code that's my fault i did not include that detail in my plan and I sort of overlooked it when we were reviewing the plan i just sort of thought that it would assume that we need to send the the survey response the same way that we already do with our contact form but we can't really assume those details these models whether it's Claude Code or any other agent that you're working with they're extremely capable at carrying out the tasks that they're given but you do need to be really explicit about exactly what you wanted to do so we just can't assume it's going to understand these details naturally thought that was a really interesting teachable moment so I hope you're starting to see a pattern here right the idea of hashing out a plan before we ask the AI to build a feature or execute a task or multiple tasks in a large project that's quickly becoming the industry best practice no matter which tool you're using whether you're using cloud code which does an excellent job at that sort of workflow or using cursor or windsurf or any of these other ones to build out the plan ahead of time and execute the plan step by step so developing PRDS or project requirements documents is the way that professionals are building with AI today and that's the topic of my other video on how vibe coding goes pro so after you hit subscribe on this channel go ahead and watch that one next i'll see you there