The Truth About Coding With Cursor AI

Key Takeaways

I've been creating web apps and Native mobile apps completely using curser with absolutely zero coding knowledge for the past few months here's what I learned about coding with cursor as a complete beginner so first things first you have to understand how these AI models work at their core when you prompt with them so these llm models can't reason or think like a human being the way llm models answer prompts is by referencing information on the internet so all of the responses generated from these AI models have been answered before by some someone somewhere on the internet so basically these llm models have memorized the entire internet so what this all means is that AI is great at regurgitating information but don't expect it to have any deep logical reasoning capabilities especially for issues that have had no previous public documentation before now this may only be an issue if you're trying to create a highly specialized Enterprise solution in a very Niche industry with limited data or just simply some revolutionary algorithm that has never been implemented before luckily though most successful SAS applications are not that groundbreaking most if not all the functions you would ever need have been solved before either directly or indirectly so pretty standard things like user authentication chat functions automations just general core business functions have all been implemented before in fact looking at the landscape right now most projects that you see launch daily on product hunt or trending on X are all just basically chat gbt rappers dressed up in their own application now diving into the actual development process you'll need to First choose a language as well as a framework for the type of app that you're going to build even though llms are pretty competent in every framework I would just stick with nextjs for web apps and react native for mobile apps both of them use JavaScript so there will be some carryover nextjs is basically the most popular modern framework to build web apps it has massive support from other web tools and allows you to handle both the front end and backend development for even more Simplicity react native is more like the same thing but for Native mobile apps created for either iOS or Android and then when you're done with all your development you can use versell to deploy your app in under 60 seconds this is hands down the most easiest and capable workflow for using cursor to build entire web or mobile applications next the key thing that most people don't realize when building with cursor especially as a beginner is that you need to have some basic understanding of coding Concepts and no this doesn't mean you need to be able to know how to code a full stack application from scratch but you should know at a minimum the file structure of your application and how each file relates to your app this helps you know what are the right questions to ask AI now ai and llm models in general are very competent at building out component files but it's up to you to understand when to create new files when to tweak existing files and how the files all relate to one another it's very common for cursor to end up duplicating functions or even entire files which can end up breaking your entire code so it's up to you to be mindful especially as your codebase grows since AI won't always know how your app is structured so you can think of this basic necessary understanding in more of a macro sense you should know what each section of lines do in your code and if you don't know just literally copy and paste that line of code into cursor and have it explain it to you keep doing this as you build so you have a mental map of how your app is structured so that when you do encounter errors you know exactly where to look and pinpoint the issue I also found that you don't need to worry as much diving deeper into the code and what I mean by this is that cloud is pretty good at writing correct Syntax for any language that you're in and then within cursor you can always provide the latest documentation for any framework that you're using so that the llm model has the latest and greatest when it's generating AI code for you the only times that there could be errors is if there's any recent changes within the commands or functions of the framework that you're working in and this could happen with nextjs given that it's a pretty modern framework that's constantly updating so again I would just always make sure to update cursor with the latest documentation for any framework that you're using inside your application next AI will get stuck and it's up to you to have enough patience to work past these choke points when building out your application and if you do have a limited coding background there are going to be times in your application build which will really have your head scratching as to what you should do next and this typically happens from either asking the wrong prompts or not providing enough context within the prompt you're trying to create when this inevitably happens when you try to implement a new feature or function it's important to really just step back and understand one what it is you're trying to do within this specific goal and what context you should be providing to the llm model to get past this point so I would first just send over the entire file structure of your codebase and then send over any relevant files to the llm at a minimum so that they understand how your project is structured and from there they can provide you information on which files should be modified which files should be created and they can also just audit the code that you currently have to let you know hey this function is currently being duplicated or this shouldn't be here now ai generated code often requires human debugging and when this happens you need to provide the full airor logs from both the terminal as well as the web console when you run your application and if these don't provide enough details to the AI model you need to include more context when you try to solve the error a way to go about this is to have aii ADD comprehensive error handling and logging to your files so that it provides more detailed error messages from there you can then Circle back and provide the llm with even more context and then finally solving your problem so this basically creates a feedback loop that feeds into itself for increasingly more detailed information now all of this can be done solo but to really have the most efficient AI coding experience especially as a beginner we have an active community of AI Builders as well as expert developers on call to answer questions about AI generated code and even debug your projects for you when AI get stuck within our program at AI code Pathways another convenience feature that I don't see too many people using is cursor rules these are preset rules that you can apply to the llm model that you're using so that it remembers a template every single time you prompt so if you're creating a nextjs project with typescript and app router make sure to say so in the cursor Rules by saying something like like you're an expert in typescript and app router for an xjs project in this way you don't need to keep repeating this context to the AI model and lastly the larger and more complex your app grows the harder it will be to generate code directly from AI now most tutorials that you'll see online showing you how to use cursor AI typically just goes over how to create a simple landing page or how to create a very simple mobile app such as identifying this plant with a screenshot there's never any real complexity or even somewhat of a backend for registered users or a save database and the reason for this isn't because AI isn't capable it's because creating complex applications completely using AI takes weeks if not months not something that can be condensed down to a 10-minute edited video or an hour long live stream anybody can just go into VZ create a simple landing page and then bring that into cursor AI to truly build a robust full stack application with complex business logic as well as full backend capabilities you need to take your time building each feature slowly and truly understanding every single aspect of your project and learning along the way you will need to begin developing skills on how to manage large code bases and like I mentioned before really understand how every component in your project relates to one another at its core you can think of generating AI code as being a coordinator so you have to first make sure that all of the components work by themselves and then from there you have to make sure that they work well with the other components inside the project project to finish up and make one complete package you never have to write the code yourself or get into the nitty-gritty but as a project coordinator you have to be well aware of all the functions in your app and where everything is located so that when someone asks you how to change a feature you know exactly which file to modify or which files relate to that specific feature cser is an amazing and truly groundbreaking tool and given the right context it can do just about anything but it's always important to just remember that the tool is only a smart as the user behind it and if you're interested in more weekly videos covering the ins and outs of cursory AI as well as coding with AI make sure to like comment and subscribe for more information on all of these AI tools in the future [Music]