How do you ensure that cursor follows it's rules file consistently? Mine just decides to ignore it from time to time. Also, how do you ensure that cursor ignores files or doesn't change/break current functions?
Hi - it uses a context window so I’d follow the rule of thumb of using a thread per coding session. If you notice it slowing down or chugging, start a new one. That problem goes away over time just like the bandwidth problem did in the early days of the internet. You can create a .cursorignore file and add files just like you would with a .gitignore. You still need to read the code it’s producing. You’re always going to want to provide context for any relevant changes / files needed. There’s also a command you can run to pack your entire repo into a formatted text file that’s easily consumed by an LLM that creates a detailed readme of file tree/dependencies etc. Making a video about that soon!
🎯 Key points for quick navigation: 00:00:00 *💻 Introduction to Using Cursor for Coding Projects* - Overview of the speaker’s coding projects using Cursor, - Mention of previous projects like a stats website and a complex health data app, - Transition to explaining how to optimize Cursor settings for better productivity. 00:01:22 *⚙️ Cursor Updates and Rules Setup* - Using Homebrew to keep Cursor updated efficiently, - Importance of setting up pre-prompts and version control with Cursor rules, - Explanation of how to manage and customize Cursor directories based on tasks. 00:03:03 *📂 Indexing Codebase and Custom Docs* - Importance of regularly reindexing your codebase, especially after large changes, - Custom documentation setup for APIs and frameworks like Google Calendar and Supabase, - Tips for organizing and indexing documents in Cursor for better reference during development. 00:06:26 *🔍 Fine-Tuning Cursor Settings and Performance* - Disabling "search the web" feature for more explicit control over prompts, - Turning off unnecessary features like chat streams and adjusting scroll settings, - Importance of thoroughly reviewing generated code and maintaining a clean workspace by deleting chats and cache. Made with HARPA AI
I think a big thing that people dont realize is how good these tools are for learning. I've been coding for a while and between reviewing code that AI generates to make sure it makes sense, and asking for explanations on things i wouldn't have done or dont understand, ive actually reinforced older knowledge and learned quite a few new tricks!
Agree! I usually tell it to also comment things heavily when there's something I think I might forget why it was done in the future. I could see how a junior engineer might make a mess, but as an experienced dev, it just makes me able to iterate so much faster.
Totally agree. Some helpful prompts I use to learn: 1) What are other ways we could write this code? What are the pros and cons of those approaches? 2) Explain this code from the perspective of a <your language> programmer. What things are easier and what is harder? 3) Are there ways of refactoring this code to be more modular, readable, and robust? 4) What are common mistakes people make when writing code like this? Show examples
I agree, this was such an interesting POD! I recommend watching Todd Howard and John Carmack on various podcasts. Can you recommend more Podcasts for Coders please. Very interesting video, I'll sub for more (: I think you may be using to many Hard Cuts IMO. I recommend using a free noise gate, And turn up the BGM / or add atmospheric audio that Flows, rather than get cut. This will make your audio and video more professional, and it masks all the Hard Cuts ;)
Yes this pod was fantastic. Love Carmack. Yea I will make more vids like this Sounds good. I honestly was trying out CapCut to do all the cuts with ai -- recently switched to davinci so should be better :-) just uploaded btw, not podcast video tho. Thx for the comment!
TL;DW: 1. Follow & Binge WebDevCody, Theo, Pieter Levels, DHH, Primeagen 2. Find problems you want to solve with tech, then solve them. 3. Search GitHub for example projects you can read 4. Go bang your head against the editor to solve your problem, ask Google & GPT's billions of questions 5. Trick yourself to make it FUN!!!
BetterTouchTool is king. Assign all of your apps to keybindings. Use Karabiner to make your caps lock key a hyperkey. I can't live without it. Also, turn on the "less motion" feature in the system settings so it's faster to switch apps.
Composer doesn’t have side bar- that’s the AI panel. For AI panel you just do cmd+L. Or shift cmd+L to add a block to a thread. For instructor you have floating window with Cmd+I and full screen with Shift+Cmd+I
I wonder if you added up all the time spent researching, installing, configuring and learning to use all these "productivity" tools if it would actually outweigh the time if you just instead got stuff done. When I see stuff like this it's almost like a procrastinators dream. Wasting time by being "more productive" instead of actually... just being productive. Random thought I just had watching this. Never thought there would be a tool for just rearranging windows (takes literally seconds).
Takes maybe 2x as long to setup as it did for us to comment to each other then you get the benefits forever. If you use a computer all day you should be a power user
There is however a corner of the internet where people just pontificate about productivity, and don’t actually ship meaningful work. They come up with new stuff bc that’s their job. Avoid that like the plague
I use the same python file tree on my own project folder, as well as uploading key files to project knowledge to generate Readmes on my own project. I found it helps orient the context of Claude better too. (Relative file references etc). Ditto on cursor
@@parkerrex use python to extract my vscode project or workspace folder structure and feed the result back into Claude Project Knowledge. So Claude understands structure
Working on a x0.1 problem, you can sure x10 that to become a x1 engineer. You have never seen a real x10 engineer. You will be mind-blown, not only by his speed but more specifically extreme deep knowledge, very high reasoning capabilities and very high accuracy. I'll suggest to watch "George Hotz" and admire a real x10 engineer. But anyway, I don't think you are serious or seriously think that shiet is good, It's a waste of a time for any problem bigger than 0.1x difficulty. I, personally, find an LLM useful when I'm learning something new, or discovering, I ask questions I would ask someone familiar with that topic. In that case, yes it's useful and makes me learn faster, know the best practices, how others do it etc.... all the knowledge needed. But to apply it in a real app it's too garbage because there is ABSOLUTELY 0 reasoning behind. An LLM by definition is a "generative model" no reasoning is involved at any step. At the same time, something like Github Copilot is usefull sometimes (using it since the Beta) Because it's much simpler, and doesn't claim to code everything for you, it focuses on your cursor, which in many cases pretty good, but even that it's not great, because you find yourself reviewing code and adjusting it and make sure it respect the code base and everything else. What you said about having a markdown file in the future and an LLM will just do everything, yes like you said "in the future" but not anytime soon. It needs SO MUCH more like x1000000 more powerful to get there, researchers need to find other things than the "Attention layer" which is used in all LLM's. I suggest to watch some "Yann LeCun" podcasts the guy who invented convolutional neural networks. You are surely great at doing business, and I would suggest doing the same to research a little more on LLM's and tech.
Not sure if this meant to be a dunk or not but I appreciate the comment regardless. You're spot on about copilots being tools, not crutches. I'm all about that Hotz wisdom too - guy's a legend. My video title aside (still learning the RU-vid ropes), I'm genuinely excited about how LLMs are changing coding. They're not replacing solid skills, but they're definitely smoothing out those initial learning curve for new libraries, picking up on new codebases etc.. They reduce time to productivity and delete the bottom quartile of tasks. Developers will spend brain cycles on more challenging concepts. This means more value to the world. They make great guesses based on what they’re trained on. Anthropic is fantastic at react/typescript/shadcn. If you’re not using cursor, you are stuck in the dark ages. GitHub Copilot is eons behind. I’ve also been the vim elitist who says you need to raw dog code. I get that ego thing, but once you’re past that you realize these tools help you and customers don’t care about your ego. I'm only 2 years into using LLM's. We need to keep pushing the boundaries - who knows what we'll be coding with LLMs in another two years?
@@parkerrex Not mean to be a dunk. I just don’t like inaccuracies I guess and false hope. Many self proclaimed swe I see on twitter talk BS about all sort of AI tools and they genuinely believe in what their saying, being hit by a hard Dunning-Kruger effect without noticing. You seem to have knowledge and some experience in writing software from the very little code you showed, the way you navigate and the specifics you talked about. I use LLM’s everyday every time, in reality the most difficult thing is to find a solution, not to write it down. That’s why it’s useless. For example in your video you already knew everything you needed to make the google oauth work, so if the LLM fails you’ll notice it immediately, and a bad software engineer will just get stuck and have no idea where to go and what to do. What I mean is that all this « time saving » thing is BS because most of the time is spent thinking and planning and finding the right solutions. Once you understand it’s simple to write. If you don’t understand, even with the best LLM you’ll notice be able to make it because you’ll get stuck at some point and can’t make progress and no one can debug the garbage code that was generated or just rewrite from scratch which is generally the fastest solution. Anyway, I have a lot more to say about this lol 😂 I suggest following ThePrimeagen and Theo. They are both great and real. Not fake as 99% of indie hackers who sell products to their audience
Agree -- they're just tutors at the end of the day. I'm working with composer more and more. I think the solution bit can also be helped with LLM's. Eg. I took Elon's 5-step algorithm for breaking down problems and use it as a text snippet for breaking down solutions. ``` md You are a 100x programmer specializing in typescript, next.js app router, shadcn, supabase for google auth with calendar perms and db queries and mutations with next server actions postgreSQL, AI, HONOjs for apis, and react. Your job is to write requirements to solve a customer problem we're trying to solve. Your answer to the question must strictly follow decision-making algorithm: **Decision Making Algorithm** 1 Write the requirements. Question the requirements. Make the requirements you just wrote less dumb. Requirements **are always dumb** to some degree. Reduce the number of requirements while still solving the problem. No matter how smart these requirements seem, they are still dumb to some degree. You have to start here otherwise you get the perfect answer to the wrong question. 2 Try to delete whatever the "step" is or process step. You will come up with a plan of steps in the form of requirements to solve the problem we come up with. You will then need to choose which ones to delete because they are not all required. If you're not adding something back in later, we're not deleting enough. It's OKAY to add things in later, we do NOT want to be overly conservative. 3 Optimize and simplify the solution. Warning: The most common mistake of a smart engineer is to optimize something that should NOT exist. **The problem to solve:** [ insert your problem statement ] ``` You have to give it context though and the proper snippets... Theyre great - fell down the rabbit hole and did Prime's entire vim setup haha.
and yeah - have 7 years of experience designing/ launching products as pm & and UX designer, light FE programming. Last 2 years typescript stack & some native iOS. Plenty to learn but thats the beauty of programming. Infinite rabbit holes pair well with infinite curiosity!
Not sure if you are sponsored or have any work relations with Cursor, but this sound and feels like a badly wrapped ad. Especially with the background music.
No affiliation -- just a legit product. I get why you'd think that since most YT vids are shilling stuff for kickbacks/rev share... But nah not this one. Give it a shot sometime I bet you'll be surprised.
yeah let it rip. Put in a system prompt -- promise you wont go back to vscode. Delete your chats as you go or itll eat up your cache and slow down. LMK if you have any questions its a game changer for me - hence me yapping.
Hi ! What is great pitch for the product manager for digital product domain job for interview point of view. Project case story discussion about digital product.
Excellent video! However, I would like to know if you know about interaction with smart watches. I would like to know if there is a connection between the app and the watch. Thanks in advance!😊