People Who Hype Cursor Usually Lack Technical Skills
en.smallyu.net
en.smallyu.net
I do think we are in a transitional period now. Eventually all editors will have the same agentic capability. Thats why editor agnostic tools like Claude code and aider are much more exciting to me.
EDIT: Oh, and as a sibling poster mentioned: A huge number of security vulnerabilities -- except now they can be purposefully injected by just posting random subtly-wrong code on the interweb. Not that you couldn't do that before, but reach would be much more limited unless you got your 'seemingly correct' code posted on SO.
One thing is certain though - it's an amazing time for security professionals. We already have careless developers without AI, but now? Oh boy oh boy.
It does matter.
Not sure that this is true. cursor agent mode is different from cursor code completion and cursor code completion is a legitimately novel model I believe.
I was surprised to learn this, but they made some interesting choices (like using sparse mixture-of-experts models for their tab completion model, to get high throughput/low latency).
Originally i think they used frontier models for their chat feature, but I believe theyve recently replaced that with something custom for their agent feature.
One thing people don't realize is there's also randomness in the answers. Also, some of the editors allow you to tweak temperature and others do not. This is why I found Roo Code extension to be better.
https://www.coplay.dev/blog/a-brief-history-of-cursor-s-tab-...
People who hype autocomplete usually lack technical skills.
People who hype memory-safe languages usually lack technical skills.
People who hype compilers usually lack technical skills.
There was some wordplay, which is just attacking the current technology that makes ttechnology more attainable by more people. However with LLMs, there is one major worry - in that it encourages de-skilling and getting addicted to having a LLM think for us. Now, if you can run your own LLMs, youre resilient to that. But the really bad side is when the LLM companies put pricetags to use the 'think-for-you' machine. And that represents a great de-skilling and anti-critical-thought.
I'm not saying "don't use LLMs". I am saying to run them yourselves, and learn how to work with them as an interactive encyclopedia, and also not to let them have core control over intellectual thought.
Not sure what you mean. Using a local LLM or a cloud one can get you addicted in the same way?
For example, you can run multiple LLMs to understand each outputs.
You can issue system messages (commands) to do specific actions, like ignore arbitrary moralities, or to process first and second derivative results.
By running and commanding an LLM locally, you become a actual tool-user, rather than a service user at the whim of whatever the company (ChatGPT, etc) wishes.
What kind of hardware is needed to get something useful running locally?
Cursor/windsurf/roo/kilo/zed are about smoothing over rough edges in actually getting work done with agenetic models. And somewhat surprisingly, the details matter a lot in unlocking value.
It could very well be that Cursor isn’t very helpful. It could also be the case that the person prompting is not providing enough context for the problem at hand.
It’s impossible to tell without the full chat history.
Knowing the correct pronunciation means you watch videos or have a social group that discusses it, that doesn't mean you are necessarily competent like someone who reads primary sources.
So you know... experience in the industry.
But I'm not even talking about the weird names for things either.
It’s often a very intellectually dishonest industry.
It’s always amazing such smurt people like you always seem to forget your own statistical irrelevance.
“Oh no the sound forms don’t come out just right.” What a giant foot going to squish us? Minus the anxiety of being smiting you’re projecting some bizarre obligation to your lived experience.
Why don’t you say things in the old Latin language? Yeah that’s right we move on from written and spoken traditions, just need to properly apply physical statistics to the real engineering. The stupid babble is in a YT video is far removed from actually building a bridge or another context where the physics matters more than the Anglicanized circumlocution.
Just a bit miffed at the usage here, given Claude is a GPT model, but OpenAI names their models in such a way that it's strange to say "ChatGPT model" to specify that category.
After using Cursor heavily the past few weeks I agree with the authors points. The ability to work outside of Cursor/AI is paramount within small software teams because you will periodically run into things it can't do - or worse it will lead you in a direction that wastes a lot of developer time.
Cursor will get better at this over time, and the underlying model, but the executive vision of this is absolutely broken and at this point I can only laugh at the problems this generation of startups will inevitably go through when they realize their teams no longer have the expertise to solve things in more traditional manners.
A little too sure of itself, incautious and overly generalising.
> In fact, Cursor’s code completion isn’t much better than GitHub Copilot’s. They both use the same underlying models
The difference is in the tooling around the models: codebase indexing, docs, MCP servers, rules, linter error feedback, and agents that automatically incorporate all of those other things together. If you don't use all that, then the models will only reach a fraction of their potential usefulness.
I agree that Cursor is overhyped by some, but it sounds like the author hasn't given it a fair chance.
But here's the thing. You CAN make these tools do cool things, but you're going to be fiddling around a lot to do so.
...so are they really that productive? That depends. A newer engineer or if it's a new language or framework or library to a seasoned engineer? Absolutely. It shortens the whole RTFM loop. But in some cases screwing around with these tools is slowing people down.
Use them responsibly and for educational purposes. They're handy. They can help with productivity or hurt.