@devinhedge
I appreciate you putting this video together because I’ve been curious about cursor for sometime. I’m an experienced developer. As in I actually remember punchcard and mainframe computers. You explained a lot of your complaints and quirks that you’re running into using cursor and large language models to write code. From everything you said, it sounds like you may have a blind spot that I find so many developers that I mentor have. All of the problems that you’ve explained you were running into can be attributed topoor object, oriented design and not using clean code principles. I hope this helps you find a new area to explore and grow your skills further. I have found that these new tools amplify the need for test first development, small modular, functions and classes, another principal, such as using good design patterns. Maybe this comment can keep all the excitement rolling while at the same time reinforcing the need for first understanding the principles of software engineering. I’m finding so many developers using these tools and they don’t even understand how a large language model chat session even works. Some examples are : size of context, window, effect of referencing documents in the size of the context window, token limits, how attaching the document to the chat reduces the number of tokens available for inference by the large language model, the need to use a RAG or CAG and went to use which one. “Fundamentals. Fundamentals. Fundamentals.” Should be our mantra. All the best.
@Adeltraut
But VSCode has all those features also in extensions. I have also exactly the same cons for the VSCode extension Github Copilot Chat. Why specifically is Cursor better?
@dennist.9376
Network engineer here. I used this for a project to pull all sorts of CDP info from switches as well as gathering info for the remote device interface. Then all the data is dropped into a neatly sorted Excel file with different sheets. Huge timesaver and a really powerful tool. I have very little Python experience and have played around with Ansible several years back.
@DigitalAdamTech
Cursor AI can't even count the number of characters for the content it writes. The AI tells me the recommended character count for a title and description tag for a website, and then writes the content based on the copy of the webpage. I'm thinking "this is awesome!" I check it with external tools, and it's over the character limit on what the AI recommends. I ask the AI to check how many characters it just wrote. It give me an answer, and it's wrong. I tell the AI that, and it says I'm correct, and rewrites it. After rewriting it, I check it again. It's still over the recommended amount it recommended. Again I ask the AI to tell me how many character it just rewrote. It tells me a number that is incorrect. I let the AI know it's wrote, and the AI say it's mistaken once again, and rewrites it to the correct character count. Some days I'm impressed with AI, and others I'm thinking how stupid it truly is.
@mfpears
I think the real lesson is that VS Code had all kinds of intricate plans of the features they were rolling out and then AI came along and they sort of accommodated it but didn't change much what they were doing. So cursor comes along and now finally we're getting some obvious low-hanging fruit improvements in VS Code like AI generated commit messages. This seemed so obvious to me. I just can't believe it's taken them two years to figure this out. But I'm sure not even cursor is super great either. I mean the first thing I thought of that I would want to do is select a range of code and press a hot key and use my voice to tell it to refactor it a certain way. Is that possible? It feels like it wouldn't be possible. Does cursor provide APIs to hook into its features?
@ShaharHarshuv
When you ask for a change in one file, I would expect it to take other files into account, but it doesn't seem like would do that. Copilot does learn from your other files, so it can use patterns and apis that are unique to your project
@k.ryashwanthreddy3265
well ,as of now, all those features mentioned in the video can be done in vscode with copilot extension.
@ChadzMedia
Comparison to github copilot please?
@mrssk.online
Great video. I love your honesty!
@mxz2024
you cannot use this in most companies, at least in germany etc., because of data privacy! if you usr Chatgpt etc. it sutomaticslly has access to your whole repositiry..the sourcecode is property of the customer, not of open AI. major data privacy issue or let s say, it s not allowed to just hand over customer property for free to some company
@thekinggamer21
This definitely feels sponsored
@rdemorais
7:01 here is the most important thing developers must understand about using AI to aid software engineering
@ev.c6
This tool works great for smaller projects, but when it comes to bigger, more complex ones, there’s no way to make it efficient. The editor has to upload all the project files to the AI backend to get the context, which ends up being pretty expensive. It makes you wonder if using a locally running AI model built specifically for code analysis would be a better option.
@Nostalgiamusic-h1z
"Do not use this app under any circumstances. There is absolutely no support, and in case of any issues, no one responds. The application keeps disconnecting. The AI lacks accuracy and messes up all your files. If you want to lose your project, then by all means, go ahead and use it."
@edwardferry8247
I used it, it was absolutely dreadful, couldn’t build a basic API call app without loads of errors then couldn’t fix the errors ending up hallucinating nonsense.. shocking stuff.
@DannyFavela
Does the CMD+i "create a project" multi-file feature result in standard boilerplate in the project? Or (like most consumer-facing LLMs) does it come up with something different than the last time you might have asked?
@vimallama5981
I think same kind of job can be done in vscode with copilot
@imakethesites3048
Thanks for this
@Rhythm.Rhapsody
what is the difference between this and having copilot on VScode
@smokinglife8980