Cursor – The AI-First Code Editor | Hacker News Reader

Cursor – The AI-First Code Editor

cursor.sh

40 pointsrobenkleene57 comments

I was mildly interested in this, until i saw the "testimonials"

> If you're still using VS Code, you're falling behind.

Click through to their twitter bio

> CEO @HyperWriteAI, @OthersideAI - I make AIs do the impossible.

Okay, right. Sure. There's potential for interesting products in this space, but seeing all these blue check AI hype men do their 'thoughtleading' is an instant red flag for me. It's the exact same crypto bro bullshit all over again.

> If you're still using VS Code, you're falling behind.

Except this is still VS Code based...........................................

> If you're still using VS Code, you're falling behind.

For a moment I thought that what would come after that sentence would be "... Use Jetbrains IDEs instead"

That's what I was thinking. Jetbrain's autocompleting

    widgets.foreach

    for (const widget of widgets) {
    }

Still remains a next level feature for me. No AI suggestions have been able to top that.

(One of the Cursor devs here)

Appreciate the feedback! Can see how this can come across as hype-y.

We're a small team of all software engineers and I think we have a thing or two to learn about branding -- kinda just picked out some of our favorite tweets

  import foo from 'foo'

  > add comments

  // imports
  import foo from 'foo'

oh thanks, now I know that these lines that start from "import" are actually imports

I just tilted on the same thing. Worst comments ever!

The other part of the same example isn't any better. "Collection name".

The twitter embeds are wild too: "If you're still using VS Code, you're falling behind."

It's like crypto bros but for software development.

The current hype cycle indeed reminds me a lot of cryptocurrency. I think there is more substance to generative AI, but the dynamic of wading through all the hype trying to figure out what is and what isn't substantive is giving me deja vu.

Fair but I don't think Cursor is one of those. I'm not using it myself but I've heard a lot of praise. What they're innovating on is the UX where the diff gets integrated into your code in real-time. It's like watching someone fix or change your code per your instructions, instead of using ChatGPT where you basically copypaste it after it's done.

Isn't that what just the 'native' Copilot is?

I think what differentiates it is what they call "Command K": https://cursor.sh/features#cmd-k

It basically gives you a Github style diff/PR right in the code, whereas Copilot just autocompletes.

The NFT grifters have simply moved on to AI. I wonder what shiny object will infatuate them next?

I don't think that's entirely fair. There's a product here. I think they're over selling it quite a bit, but it's a real tangible product. This at least has the possibility to be useful.

The “add error handling” example is baffling. I’m not even sure if it’s valid JS to put an export statement in a try/catch block?

Edit: No, it's not valid javascript https://i.imgur.com/9g28y3l.png

I'm curious how this compares pricing-wise to Github Co-pilot for personal use:

Co-Pilot:

$10/month or $100/year (save $20 if you buy the yearly). Unlimited requests, no throttling

It uses OpenAI’s Codex model, a descendant of GPT-3[1]. They do seem to be working on the next iteration of co-pilot which will be GPT-4 powered, code named "Co-Pilot X"[1,2]

Cursor:

- Basic plan: FREE, but with 50 GPT-4 uses (I assume total because it doesn't specify per month), and 200 GPT-3.5 uses per month

- Pro plan: $20/month 500 "fast" GPT-4 uses per month, unlimited "slow" GPT-4 and unlimited everything else.

If GPT-4 proves to much better than Co-Pilot's version of GPT-3 then maybe Cursor might be worth it. Wonder how long that 500 would last for a typical dev.

sources:

1: https://github.blog/2023-03-22-github-copilot-x-the-ai-power...

2: https://github.com/features/preview/copilot-x

https://github.com/features/copilot#pricing

https://cursor.sh/pricing

Not to play into the ever-present trope of HN users talking about the presentation of the website rather than the tool it's advertising, but this site is extremely unresponsive and janky to navigate. I hope that form-over-function experience isn't indicative of what I can expect from the Cursor editor.

Yeah it feels so js heavy, it’s horrible to use.

Cursor's major innovation is in the UX of integrating AI-assisted tooling into your editor.

- Inline chat about code blocks

- Inline instructions on modifying/creating code

- Import of documentation to the local project for better suggestions and referencing

You highlight the code you're interested in and hit CMD+K to have it do modifications for you.

This kills the copy + paste of code from your editor to ChatGPT and back, which beyond being a pain also loses all the context of where the code is in your project and is also a step ahead of Co-Pilot which is more just super intelligent code completion.

In general, I've found a great deal of dev productivity improvements in using it as a coding assistant where I'm still orchestrating and guiding the work and using the AI assistance to very rapidly blow through boilerplate tasks, reformatting things, double checking if I'm doing something right, etc.

Isn't the inline chat feature the same as Ctrl+I (https://code.visualstudio.com/assets/updates/1_79/copilot-ed...) in GitHub Copilot? I find it really useful, although the main page of GitHub Copilot does not appear to mention this feature.

I think that's "Copilot Chat" which I also used and at time (couple months ago) was only in beta and you needed Visual Studio Code Insiders release to use it.

Alternatively, if you just want to integrate Copilot into Neovim and get on with your day, I recently discovered that the latest LazyVim makes this option as an extra out of the box.

I actually discovered this while working on Shell Bling Ubuntu, which is a couple of easy scripts to get you a bunch of modern command line tools nice and configured in one go, but you can just scroll down to "Add Copilot to Neovim" to see. It's refreshingly user friendly for NV configs.

https://github.com/hiAndrewQuinn/shell-bling-ubuntu

I've been using Cursor quite a bit on my two current personal projects in Python and C++ respectively, and I think it's quite useful:

I haven't written a unit test by hand since I started using Cursor. Because it has the full codebase as context, it can easily generate unit tests. I would say the code is correct around 50% of the time on the first try? Otherwise, you can quickly submit a follow-up prompt to fix stuff. Obviously, it's still GPT-4 in the background, so hallucinations still happen.

I tried adding a simple feature to my interpreted custom programming language using the chat feature, giving the full codebase as context: The programming language already supported indexing into arrays, but not strings. I asked in chat what I need to change to make this feature work. It generated code and showed me in which files I have to add or modify existing code. Very impressive. Again, the code wasn't perfect, I had to do some minor adjustments like method names, but it understood my code, gave the right files to modify, and also proposed code that was almost correct. There is a beta feature that applies the proposed code changes to the codebase, but that did not work yet when I tried it (I am using an OpenAI API key, maybe it's related to that).

Overall, I think it is very helpful, especially when I am tired from $dayjob, but still want to do some coding in the evening. For me, it's much easier to formulate what I want to code in human language, instead of writing the code "by hand".

Caveat: Even if you use the OpenAI API key, ALL your code is sent to the cursor.so servers for prompt generation, indexing etc., even with "local mode" on. "Local mode" just changes a flag that cursor won't save code on their servers, but it's still sent there for processing. I assume this is a deal-breaker for most companies for now.

It’s a fork of VS Code? I guess it’s tougher to monetize an extension?

edit: <ignoreme>Also why is it Mac only if VS Code supports all three OS?</ignoreme>

It appears to have windows support also?

It does. It prompts me to download for Windows.

Indeed I can't see the point of shipping the "AI first" feature as a whole editor. It seems to me that it should be a feature of any editor, or else I'd like to be convinced by arguments less marketing oriented than those I see on the home page.

It's likely that they changed the UI in a way that stock VSCode does not allow. Extensions can't modify everything.

On the website: "Cursor is a fork of VSCode"

I'm aware. It was meant to convey a tone of confusion - not as a literal question.

Easier to monetize but more effort spent in keeping up with upstream. Good for extracting money from dumb VCs, bad for users.

I see download options for windows, linux, and mac.

Seems they did some user agent detection. Whoops! Shallow criticism on my end.

Fair call I think, just because I'm on platform A should not preclude me from knowing about cross-platform availability.

And you are not. When I visit the page I see "download for Linux" with an element immediately below for "other".

My only criticism is "other" could be called out more.

(One of the Cursor devs here)

We forked VSCode because we wanted to change the UI in ways Microsoft wouldn't allow (e.g. our Cmd-K isn't possible as a VSC plugin; Microsoft is building their own version and isn't letting plugins use it)

Really pleasant experience learning a bit more python and implementing some small projects with the help of cursor.sh.

I love the way it will just type out things, you can ask it to do thing you obviously know how to do, and it will spit out readable usable code without the typing. Remove all comments, work on code, ask it to put back comments, just silly time saving things but it all adds up, you can feed it fresh docs straight out of github and it does a fair job of interpreting them, there are a lot of things it can do, seems short-sighted perhaps to focus on what it cannot.

Pleased to see the recent '$8 million in seed funding led by OpenAI’s Startup Fund', I'm not experienced enough with any IDE to judge their integrations, however, a lot of the criticism here seems to be a side track and not actually user experiences, their UX is under-development, to me, it looks cleanly implemented and well thought out for a 'young' product.

I hate the pricing schemes emerging for these tools and I hope it gets fixed.

“500 fast GPT-4 calls” & “Unlimited slow GPT-4 uses”. Wait. What is slow GPT4? Seems “your requests may be queued behind others”.

Luckily, you can bring your own OpenAI key it seems but it’s a lot of granularity to ponder for a tool that’s supposed to make AI seamless.

1. I don’t think it’s too hard to understand BYOK or n number of tokens. I’ll have to talk to the market at large but personally, pretty clear.

2. Maybe you can suggest an alternative?

The alternate is what the competition does: $10/month. No quotas.

https://github.com/features/copilot#pricing

What is a "GPT-4 call" anyway? What counts as a call in an AI-first editor? Every autocomplete suggestion eats away at my quota?

Fixed price as copilot has? Who is going to pay token for coding? I have 100s of recommends in copilot per day. Has to be fixed.

> Build software faster in an editor designed for pair-programming with AI

Isn't it known that what takes the most time (in software engineering) is to understand a given business problem, decompose it in manageable chunks that can be worked in parallel, then provide a solution for each chunk, and then maintain such a solution? Using the IDE to write the actual code, sure, it needs to be done, but it's rather the part that takes the least amount of time. So, if any their slogan "Build software faster in an editor designed for pair-programming with AI" is rather misleading. I actually spend more time reading code with my IDE than writing code.

> Trusted by tens of thousands of engineers

How do these large corporation find softwares like cursor.sh?

Also curious about this and questioning its validity…

It’s an easy trick: they just check the domains of the email addresses of people who created an account and display the logos of the high profile ones.

That's pretty fraudulent. Unless you signed a B2B contract, you can't claim thath.

Well, they can; it’s just illegal.

Worst case they are going to get a legal email from those companies to remove the logos.

When I worked at one of the large startup company, we all were submitting various websites to our legal department to remove our logo.

I own B2B business, just never felt confident to ask other businesses if that is ok to show theirs logos, so don’t show any. Probably a miss of opportunity.

Ask a friend to use it once on a free trial. They work for Uber. Logo won. Rinse/repeat.

I heard of cursor on the grapevine before it was posted here and I am not someone who stays on top of this stuff so they are doing a bit of marketing I reckon.

Same, too, for sites that say "trusted by students a ${PRESTIGIOUS_INSTITUTIONS}". Overuse of that little advertising trick has absolutely zeroed out whatever value it had.

The UX of Cursor is the best I’ve experienced so far. GitHub Copilot Chat has been released in beta, but the experience isn’t as good. The context and the real-time code diff make a huge difference.

Cursor would be more promising if it wasn't just a VS Code fork that lags behind upstream VSCode and has broken extensions as a result. It has other minor bugs and glitchiness as well, and its main selling point atm, that the LLM gets your entire codebase as context, also isn't reliable with an even moderately-sized codebase.

1. How does CoPilot integration work together with ChatGPT in this editor - or is there no support for copilot?

2. Is there intent to support Code Llama?

For that matter, what serious contenders to ChatGPT4 are there for code completion, which of these are self hosted, and are there any integrations to existing IDEs?

Once again, an AI generator fails majestically when you ask a simple question. Don't these companies actually test their own products, or do they rely on the unit-tests generated by this product?

"Keep your code private

With privacy mode, we store none of your code in our servers or logs." But they then say they still use other peoples servers though. This copy text needs sorting out!

Vi keybindings?

cursor sh or so?

Reply on news.ycombinator.com