This Made Claude Code Way Smarter (Copy Me)
By Beau Johnson · · 18 min · 736 views
About this video
Transcript
Read the full transcript
0:00If you're trying to use Cloud Code to build out your startup idea, then this video is going to be for you because we're going to be going over the five lessons that I learned 30 days in using Cloud Code and best practices. So, let's jump right into it and kind of give you a quick overview on where we're coming from. We are working on EasyFlip.ai. This is an AI listing tool that is going to help resellers list items on marketplaces like eBay, Poshmark, and Merkari using AI automations. So, this is how far we've gotten 30 days in. We've redesigned the landing page. All this was built out in Cloud Code. And I'm going to be walking you through the process on the things that I learned, the mistakes that I made to help build a good foundation for you as you go along your building journey. We have a complete dashboard integration connected with the eway API. We are so close to
0:45launching this product after 30 days of working on it. Now, we're just kind of tying up the loose ends. And I just want to give you some context because 30 days ago, I knew nothing about coding. I didn't know what a bash command was. I didn't know what MCP meant. I didn't even know how to write a single line of code. And you shouldn't because the things that we're going to be working on and the skills that you need to work on are things prompt engineering and context engineering and learning how to use clawed code effectively. And I'm going to be going over the mistakes on what I did and you hopefully you can learn from them. So going over these five mistakes hopefully brings you a little bit more clarity. Start with proper project structure. What does that even mean? The problem. Jumping straight
1:30into cloud code without setup leads to configuration headaches. The solution is writing a proper workflow. What does that even mean? Generating a proper product requirement document. And I'm going to tell you how I went about this. All right. So, I'm using cloud code within cursor and you can see here I have a PRD markdown file. This is necessary because it gives Claude code instructions and it gives it more context on what you're trying to build. So, this is our PRD. It's very lengthy. You don't need to type this out yourself. Here's what I suggest you do. You use Claude to help generate your product requirement document based on the idea that, and here we go. We have a product requirements document based on
2:16the easyflip.ai idea. Amazing. goes over the product vision, strategic alignment, problem solving like this the problem that this app solves within the niche. Uh resources, problem statement and opportunity, quantified impact, market evidence, opportunity analysis. So, this is just going to give the vision for your project to Claude to be able to build a foundation to build your project. This is key. I did this way late in the process and I learned my lesson because not only was I getting errors, bugs, hallucinations, I had to go back and redo everything, y'all. I had to restart over just a week ago and that's why I'm making this video because I don't want you to be in the same position. This is a very tedious technical process and I'm not used to it and I learned the hard way. And maybe you got gray hair right now. Maybe
3:02you've aged 10 years and just a couple days pulling your hair out trying to figure it out. You don't need to necessarily learn how to code, but you need to learn how to communicate to get proper outputs. And I was filling up the context window. I was chatting back back and forth with cloud code which may work for some people. It just didn't work for me. I needed to get good a few skills. Creating good effective prompts and learning to communicate effectively for the task that I want. So using cloud code and implementing your PRD is key. Once you get this done, you're able to have claude markdown file based off of the product requirements doc. And the cloud markdown file is more of like a project overview. This is going to give you like the key features that you're trying to build out, maybe the steps and the time horizons you're trying to get
3:48those features built out in. Um, goes over um key PRD requirements identified. And you could just be straight up with cloud code and just say, "Hey, I put in my PRD. Now create a claude markdown file based on that information." And it did it. Uh, very cool. Don't think that you need to type this out manually. Use the tools that you have available and rely on. I say just rely on them. And so that once that is done, you have your project set up. This is something I didn't do when I first started and I regret it. And if you go back and watch where we started, we started this whole idea on Bolt.new, which is great. I love Bolt. It's a great platform to build, you know, your startup idea. But where I w was and where I wanted to be, the project just got too complicated, got
4:34too big for Bolt. and we moved to cloud code so we can have a little bit more control over you know what we're trying to integrate when you start implementing APIs and databases and authentication and um payments it just gets a little bit trickier so I wanted a little more control take my life back I highly recommend doing what we just talked about in step one which is creating proper project structure so PRD cloud markdown file a real example was just like I said I wasted so much time uh messing around because we didn't have the foundation set up properly. Then we transferred to claude code middle of the project. I have this in a playlist you can go back and watch and it's just raw and real and I make a lot of mistakes. I just don't want you to go through that. I've learned more in the last 30 days
5:19than in the last years combined. Guys, this has been amazing. It's challenged me in ways that I haven't been challenged before back to when we first started our first business 5 years ago. And I love learning and I just want to help you fasttrack your process so you don't make mistakes. Use Claude Desktop for better prompts is number two. The discovery was Claude isn't just for writing code. It's incredible for crafting better prompts. Think of it this way. Claude is going to be your prompt engineer. It's going to be your context engineer. It's going to be how you're able to break things down so you can understand. Maybe you understand it already. I didn't 30 days ago. I needed explanations that needed to be broken down so a 5-year-old can understand. And that's what I used Claude for. I have
6:04Claude open on this window here and I have Cloud Code open on my monitor. So, I'm able to just go back and forth between Claude and Claude Code to make sure that when I send a prompt for fixing a bug or an issue, it's effective. It's efficient. It portrays what I'm trying to accomplish rather than me just saying figure it out. If you just say figure it out, you're going to go through this constant circle of hallucinations, maybe even breaking a few things. That's where I don't want you to mess up. So, using Claude for better prompts and I could just show you right here, I have a full running conversation with Claude. I have a project called easyflip.ai. This project within Claude helps create a database of what we're trying to figure out, what we're trying to solve, all the problems that are going on within this build.
6:50This is necessary and a must if you have no experience, y'all, because it's so overwhelming with the apps that are coming out, all the innovation happening in the space. getting some clarity by using claude in conjunction with claude code is going to be to your benefit and it's going to help you learn these skills faster which is the goal. So creating a project within claude based around your project is going to be I say a must because there's no way I would have gotten this clarity without for example just for an example of what we've been dealing with today. Oh, let's just go. Something we were struggling with today. I just tell Claude more errors specifically with the taxonomy scope when analyzing a steel jacket men's 2XL. It puts it in the kids clothing category under shoes, which doesn't make any sense. I have copied
7:35the console code. Clearly, this is a men's 2XL jacket, which is what it says in the title. So, I don't know why it got it all wrong, but we got an error. And here it is. And I copied the console code and gave it to Claude. And it totally breaks it down and reads exactly what's going on. Then we're able to debug the issue within cloud and then we can track down what exactly is happening during the token exchange during the API calls during the vision analysis and then we can correct it accordingly. If you have no experience doing this like I do, it's going to be tough. Like I have what am I looking? So doing this gives you just a different viewpoint and Claude can break it down for you so you can understand. This has been the key for me to learn this quick. Nothing else. No courses, no mentors, no guides
8:20is just and this is just how I learn. Maybe you learn differently but I learn by doing and by having things broken down at a very simple level. so I can understand the concepts, the principles, the ideas, and why we're doing what we're doing. This is important. Don't just be oneshotting things and not knowing how things come together and how they're working, especially if you want to do this in the long run. This is the first app we've ever even built, y'all. Like I said, 30 days ago, we were not doing any. So, this has been key, and that's just an example. The log shows several issues that needed addressing. It goes through the server error, the front-end JavaScript error, the taxonomy miscatategorization, um the cell scopes were misaligned in the API, the front-end data structure was incorrect, and it gave me a prompt. Um, so it says tell cloud code to debug these issues in order of priority. I
9:07take this and I just drop it right into cloud code and we execute. I do have to say I run in plan mode. I don't know if you see that down there before I just hit enter on the prompt and have it execute. I have it plan the execution. Once it's planned it out, I can copy the plan, give it back to Claude and say, "What do you think of this plan?" Then I have before I even start running the prompt that Claude initially gave to Claude Code, I can see what is Claude Code doing. This has saved hours of headaches and frustration. We've been able to troubleshoot more errors and bugs than ever before quickly. Not because, you know, I'm getting smarter or better, you know, maybe incrementally day by day, but it's because I'm using the tools more effectively. And this is just the workflow that I'm hoping can save you hours of frustration. I don't
9:54want you to give up. I don't want you to quit on your idea. And that's why we're going through this step by step. So using Claude for better prompts. Claude isn't just for writing code. It is your personal assistant assistant to accomplishing the dreams that you want to accomplish. So, how it works? The iterative feedback loop helps you refine your requests. That's it. And I just walked you through that process. Business impact. More precise prompts equals faster development equals a quicker path to you making money. Practical tip. Use cloud to help you write better prompts for your AI for other AI tools. I mean, this has multiple use cases, but this is just how I'm using it specifically in cloud code. Number three is leverage existing libraries instead of building from scratch. Oh my goodness. And we started on React for building this out. Then we added on Superbase. We added on Stripe.
10:41We added on Netlfi and all that work. We were just building these things out individually. When you're just starting out, you probably don't know how to build this out. Neither did I. So the mistake was trying to build eBay authentication integration from scratch. Trying to build it all from scratch. And then I found this repository on GitHub. It is a TypeScript library that solved all the OOTH issues immediately. Having proper documentation that is current is key to your project success. Can I get an amen? Holy moly. Praise Jesus. This is what I'm talking about. The people people have probably already built what you're trying to build and it's probably on a repository somewhere on GitHub. Just saying. No need to build things out and reinvent the wheel here. Shout out hent, my boy Hint, whoever you are. This gave us proper current documentation to at least set a foundation on connecting
11:27all these APIs from our back end. So, I implemented this into our project and it changed everything. Also too, keep in mind cloud code isn't amazing at searching the internet. So it's pulled documentation from 2018 didn't work. So by implementing this into our project it really helped fasttrack the process and we had spent days just trying to figure that. So that helped key insight always research existing solutions before custom development. So this is something that I I just didn't know. I didn't have the knowledge. I didn't know that someone had built this already that I could have put into our project. Time saved. This saved weeks of ooth debugging avoiding avoided by using mature. So maybe you're trying to build a feature within your and you're having a hard time. Check if there's a repository that you can implement. Do your research first cuz it could also mess things. Ask Claude. Number four is
12:13plan your tech stack before you code. The evolution. So we started with Bolt.new which Bolt's great. I think you could build probably 70 80% of the app ideas that you have on Bolt. And it's gotten a lot better just in the last month. So a month ago though, we were filling up the context window. We were burning through tokens, hundreds of dollars worth, and we didn't get the end result we want. Then we got frustrated and we made the switch to claw code. So we learned each platform has its strengths. Choose based on your end goal. What is the thing you're trying to build? Is it complex? Is it a walking app? Is it a habit tracking app? Is it a task manager app? Is it Amazon? What are you trying to build? You need to figure that out first. So cloud code excels at API integrations, database management, and AI agent workflows. And I think it works even better by using cloud in conjunction with the cloud code. So
12:58strategic thinking. Who is your target market? Who are you building for? If it's rather simple idea, it can be built on bolt, lovable, replet. If you want more complexity, use clock. So the thing here for me, and this has always been, not always, but just within building businesses over the last year, um just becoming a lifelong learner, you know, learn as much as you can and be willing to fail. So making mistakes seems bad initially. You don't want to make mistakes, especially when it comes to programming or creating something that's complex because sometimes you may not know how to solve those mistakes and errors that you end up making and it may end up trashing your project. But these mistakes help you grow. The skills that matter the most here are context engineering and prompt engineering. Without being able to communicate effectively to these LLMs, you'll be left with AI sloped and context rot and age 10. So that is key here. What is
13:45context rot? Just want to cover it briefly. is that the models that you use when you fill up the context window end up degrading over time. You can see here not as efficient once you get into several back and forths with your cloud code. It's going to just get worse. For example, understanding the context window took me weeks to figure out what that even meant. Your output's just going to get worse the more that you go back and forth with cloud code. So that's why I'm using Claude to be effective in my prompt and have it understand what is the problem. create a prompt that I can give to claude code so it can solve that problem. These LLMs are going to outprompt you. And this is why this is even just a hack I learned a couple months ago is that tell the LLM to create the prompt for you to either prompt itself or create a prompt for you
14:30know cloud code or whatever you're trying to accomplish. You're going to notice the prompts are way more elaborate, way more detailed, much better than you could do in that amount of time especially. So context Windows. Okay. So yeah, 200,000 token capacity and it's rolling out to a million. I just upgraded to the $200 plan on Claude Code and I've never got I've never been put on time. So check that out, understand it. Don't just ask Claude Code, how's it going? What you doing today? You want to make sure every time you're putting in a prompt Claude code based around your project. You want it to be efficient and effective and you want it to be optim. So that was really helpful learning about what that even meant. And that was the answer to why I was getting bad output, plain and simple. Another little extra here, let's just call this number six. Um, Ras Mike
15:15here on YouTube, his name's Michael. Um, put out this Nex.js starter kit uh about a week ago. And this changed the entire project. This changed the trajectory. This changed every single thing about our project because we were building this thing from scratch using React, building our own database, building our own authentication, connecting the APIs oursel, building in the payment structure, and when you don't know what you're doing like I did, um it's difficult. So, Nex.js starter kit um built the foundation for us. It has authentication using better off. It has polar for payments. It uses um Neon for database, Cloudflare R2 for uh image storage. Changed everything. It made our project flow. And I redid this. I
16:01completely redid everything a week. Like I I was pulling my hair out. We had thousands of lines of code trailing off in areas that did not matter and it was just making Claude code bogged down, not giving us the answers that we wanted. So we started over from scratch. Of course, I consulted with Claude and we implemented this starter pack. I'm going to link it down below because I want this to help you build your project. And this has got everything, y'all. The Nex.js SAS starter kit. Thank you, Michael, by the way, if you ever watch this. Uh, authentication and user management. It uses better off. It has the Google OOTH already. I mean, we did all this by oursel and it was so frustrating. This is already plug-and-play. Polar for payments, AI integration with Open AI, modern UI and UX using Tailwind, Shad CN. It's amazing, y'all. Database, storage, neon,
16:48drizzle, cloudflare. So good analytics. Oh my goodness, guys. And we were able to build out a beautiful landing page just like this in a beautiful dashboard where everything is ready to go. If you want to get listed on eBay, you drag and drop your photos there. You assign the SKS, you generate the listings, you publish to eBay. All this was done in just a few days. So, I just say this to you that maybe you're building your project right now and you're going through a little bit of frustration like I was nearly giving up multiple times. Don't do that. You maybe just need to get better at using the tools. You don't need to go out and be a professional programmer overnight, but you do need to understand the concepts and the principles behind the things that you're doing to be an effective person who builds. So, that's why I wanted to come on here and make this video. If you guys have any questions or maybe feedback,
17:33let me know. We're going to keep the videos coming. We're on day 30 and we're going to be shipping this thing here out in the next few weeks.
Build this with us, live
Shipping Skool is a free community with 4 live build sessions a week. Bring your idea.
Join Shipping Skool — Free





