I use Cursor daily - here's how I avoid the garbage parts

https://www.nickcraux.com/blog/cursor-tips

by striat • 1 year ago

250 250 1 year ago

kevingadd• 1 year ago

Like mine will keep forgetting about nullish coallescing (??) in JS, and even after I fix it up it will revert my change in its future changes. So of course I put that rule in and it won't happen again.

I'm surprised that this sort of pattern - you fix a bug and the AI undoes your fix - is common enough for the author to call it out. I would have assumed the model wouldn't be aggressively editing existing working code like that.

worldsayshi2• 1 year ago

Yeah I have seen this a bunch of times as well. Especially with deprecated function calls. It generates a bunch of code. I get deprecation warnings. I fix them. Copilot fixes them back. I have to explicitly point out that I made the change for it to not reintroduce the deprecations.

I guess that while code that compiles is easier to train for but code with warnings less so?

I remember there are other examples of changes that I have to tell the AI I made to not have it change it back again, but can't remember any specific examples.

Aeolun2• 1 year ago

It’s due to a problem with Cursor not updating the state of the files that have been manually edited since the last time they were used in the chat, so it’ll thing the fix is not there and blindly output code that doesn’t have it. The ‘apply’ model is dumb, so it just overwrites the corrected version with the wrong one.

I think the changelog said they fixed it in 0.46, but that’s clearly not the case.

oefrha• 1 year ago

Yep I asked about this exact problem the other day: https://news.ycombinator.com/item?id=43308153 Having something like “always read the current version of the file before suggesting edits” in Cursor rules doesn’t help, the current file is only read by the agent sometimes. Guess no one has a reliable solution yet.

siquick• 1 year ago

Cursor in agent mode + Sonnet 3.7 love nothing better than rewriting half your codebase to fix one small bug in a component.

I've stopped using agent unless its for a POC where I just want to test an assumption. Applying each step takes a bit more time but means less rogue behaviour and better long term results IME.

worldsayshi2• 1 year ago

love nothing better than rewriting half your codebase to fix one small bug in a component

Relatable though.

krets• 1 year ago

Sounds like a human colleague of mine

WA• 1 year ago

If you stopped using agent mode, why use Cursor at all and not a simple plugin for VSCode? Or is there something else that Cursor can do, but a VSCode plugin can't?

askonomm• 1 year ago

So, if I liked being a manager more than a developer, I'd use Cursor, and lean in entirely on AI?

lonelyasacloud• 1 year ago

Yes; it can be used in agentic mode and along with the joys it also has a few of the frustrations that would be familiar if have managed human devs.

TingPing• 1 year ago

If you don’t understand what it outputs then it’s just random garbage.

jillesvangurp• 1 year ago

The UX of tools like these is largely constrained by how good they are with constructing a complete context of what you are trying to do. Micromanaging context can be frustrating.

I played with aider a few days ago. Pretty frustrating experience. It kept telling me to "add files" that are in the damn directory that I opened it in. "Add them yourself" was my response. Didn't work; it couldn't do it somehow. Probably once you dial that in, it starts working better. But I had a rough time with it creating commits with broken code, not picking up manual file changes, etc. It all felt a bit flaky and brittle. Half the problem seems to be simple cache coherence issues and me having to tell it things that it should be figuring out by itself.

The model quality seems less important than the plumbing to get the full context to the AI. And since large context windows are expensive, a lot of these tools are cutting corners all the time.

I think that's a short term problem. Not cutting those corners is valuable enough that a logical end state is tools that don't do that that cost a bit more. Just load the whole project. Yes it will make every question cost 2-3$ or something like that. That's expensive now but if it drops by 20x we won't care.

Basically large models that support huge context windows of millions/tens of millions of tokens cost something like the price of a small car and use a lot of energy. That's OK. Lots of people own small cars. Because they are kind of useful. AIs that have a complete, detailed context of all your code, requirements, intentions, etc. will be able to do a much better job that one that has to guess all of that from a few lines of text. That would be useful. And valuable to a lot of people.

Nvidia is rich because they have insane margins on their GPUs. They cost a fraction of what they sell them for. That means that price will crash over time. So, I'm optimistic that a lot of this stuff will improve rapidly.

jampekka• 1 year ago

Nvidia is rich because they have insane margins on their GPUs. They cost a fraction of what they sell them for. That means that price will crash over time. So, I'm optimistic that a lot of this stuff will improve rapidly.

OTOH currently the LLM companies are probably taking a financial loss with each token. Wouldn't be surprised if the price doesn't even cover the electricity used in some cases.

Also e.g. Gemini already runs on Google's custom hardware, skipping the Nvidia tax.

_heimdall2• 1 year ago

Nvidia is rich because they have insane margins on their GPUs. They cost a fraction of what they sell them for. That means that price will crash over time. So, I'm optimistic that a lot of this stuff will improve rapidly.

That still leaves us with an ungodly amount of resources used both to build the GPUs and to run them for a few years before having to replace them with even more GPUs.

Its pretty amazing to me how quickly the big tech companies pivoted from making promises to "go green" to buying as many GPUs as possible to burn through entire powerplants worth of electricity.

_1tem2• 1 year ago

Try Claude Code. It figures out context by itself. I’m having a lot of success with it for a few days now, whereas I never caught on with Cursor due to the context problem.

_--__--__• 1 year ago

I have not tried Claude Code, but besides the model lock-in the number one complaint I have heard is that it consistently over provides context leading to high token usage.

mbeex• 1 year ago

aider [...] It kept telling me to "add files" that are in the damn directory that I opened it in.

That's intentional, and I like it. It limits the context dynamically to what is necessary (of course it makes mistakes). You can also add files with placeholders and in a number of other ways. but most of the time I let Aider decide. It has a repomap (https://aider.chat/docs/repomap.html), gradually building up knowledge and makes proposals based on this and other information it gathered also with token costs and out-of-context-window in mind.

As for manual changes: aider is opinionated regarding the role of Git in your workflow. At first glance, this repels some people and some stick to this opinion. For others, it is exactly one of the advantages, especially in combination with the shell-like nature of the tool. But the standard Git handling can still be overridden. For me personally, the default behavior becomes more and more smooth and second nature. And the whole thing is scriptable, I only begin to use the possibilities.

In general: Tools have to be learned, impatient one-shot attempts are simply not enough anymore.

cyprx• 1 year ago

I had been using Cursor for a month until a day when my house got no internet, then i realized that i started forgetting how to write code properly

ant6n3• 1 year ago

Just get a Mac Studio with 512GB RAM and run a local model when the internet is down.

_puk2• 1 year ago

Back up your $20 a month subscription with a $2000 Mac Studio for those days your internet is down.

Peak HN.

jjude2• 1 year ago

Which local model would you recommend that comes close to cursor in response quality? I have tried deepseek, mistral, and few others. None comes close to quality of cursor. I keep coming back to it.

pknerd• 1 year ago

Can one run cursor with local LLMs only?

eadmund• 1 year ago

But then I’d be using a Mac, and that would slow my development down and be generally miserable.

SCdF• 1 year ago

... to make *completely* sure that they forgot how to program?

risyachka• 1 year ago

I had the exact same experience, pretty sure this happens in most cases, people just don’t realize it

vrnvu• 1 year ago

Me too. I completely forgot the standard library and basic syntax of my daily language. wow. I went back to VSCode and use Cursor for the AI model to ask questions.

laborcontract7• 1 year ago

Cursor's current business model produces a fundamental conflict between the well-being of the user and the financial well-being of the company. We're starting to see these cracks form as LLM providers are relying on scaling through inference-time compute.

Cursor has been trying to do things to reduce the costs of inference, especially through context pruning. For instance, if you "attach" files to a conversation, Cursor no longer stuffs the code from those files into the prompt. Instead, it'll run function calls to open those files and read bits and pieces of the code until the model feels it has enough information. This seems like a perfectly reasonable strategy until you realize you cannot do the same thing with reasoning models, if you're limiting the reasoning to just the initial prompt.

If you prune out context from the initial prompt, instead of reasoning on richer context, the llm reasons only on the prompt itself (w/ no access to the attached files). After the thinking process, Cursor runs function calls to retrieve more context, which entirely defeats the point of "thinking" and induces the model to create incoherent plans and speculative edits in its thinking process, thus explaining Claude's bizarre over-editing behavior. I suspect this is why so many Cursor users are complaining about Claude 3.7.

On top of this, Cursor has every incentive to keep the thinking effort for both o3-mini and Claude 3.7 to the very minimum so as to reduce server load.

Cursor is being hailed as one of the greatest SAAS growth stories but their $20/mo all-you-can-eat business model puts them in such a bad place.

rafaelmn• 1 year ago

Cursor has been trying to do things to reduce the costs of inference, especially through context pruning. For instance, if you "attach" files to a conversation, Cursor no longer stuffs the code from those files into the prompt. Instead, it'll run function calls to open those files and read bits and pieces of the code until the model feels it has enough information. While that seems like a perfectly reasonable strategy, it starts to fall apart when integrating reasoning models.

In general I feel like this was always the reason automatic context detection could not be good in fixed fee subscription models - providers need to constrain the context to stay profitable. I also saw that things like Claude Code happily chew through your codebase, and bank account, since they are charging by token - so they have the opposite incentive.

Roritharr2• 1 year ago

This is only surface-level deep. Cursor already has Quotas for their paid plans and Usage-based Pricing for their larger models, which I run into and fall over to their usage based model every month.

Imo most of their incentive on context-pruning comes not just from reducing the token amount, but from the perception that you only have to find "the right way"tm to build that context window automatically, to get to coding panacea. They just aren't there yet.

laborcontract7• 1 year ago

If you’re going to pay on the margin, why not use those incremental dollars running the same requests on cline? I’m assuming cost is the deciding factor here because, quality-wise, plugging directly into provider apis with cline always does a much better job for me.

NitpickLawyer• 1 year ago

This seems like a perfectly reasonable strategy until you realize you cannot do the same thing with reasoning models, if you're limiting the reasoning to just the initial prompt.

Keep in mind that what we call "reasoning" models today are the first iteration. There's no fundamental reason why you can't do what you stated. It's not done now, but it can be done.

There's nothing stoping you from running "tinking" in "chunks" of 1-2 paragraphs, doing some search, and adding more context (maybe from pre-reasoned cache) and continuing the reasoning from there.

There's also work being done on think - summarise - think - summarise - etc. And on various "RAG"-like thinking.

MrBuddyCasino2• 1 year ago

Cursor has been trying to do things to reduce the costs of inference, especially through context pruning.

You can also use cline with gemini-2.0-flash, which supports a huge context window. Cline will send it the full context and not prune via RAG, which helps.

laborcontract7• 1 year ago

I love cline but i’ve never tried the gemini models with it. I’ll give it a shot tonight, thanks for the tip!

greyman• 1 year ago

Or you can also use Gemini Code Assist extension for VS Code, which is basically free, but so far, the code it wrote almost never worked for me. So far I use only Claude 3.7 or Grok in chat mode. Almost no model, as of today, is good at coding.

laborcontract7• 1 year ago

I've just tried gemini-2.0-flash, this is an incredible model that's great for making edits. I haven't tried any heavy lifting with it yet but It's replaced Claude for a lot of my edits. It's also great at agentic stuff too!

IanCal2• 1 year ago

Instead, it'll run function calls to open those files and read bits and pieces of the code until the model feels it has enough information. This seems like a perfectly reasonable strategy until you realize you cannot do the same thing with reasoning models, if you're limiting the reasoning to just the initial prompt.

There's nothing about this that conflicts with reasoning models, I'm not sure what you mean here.

laborcontract7• 1 year ago

what i mean is that their implementation (thinking only on the first response) renders zero benefit because it doesn’t see the code itself. They run multiple function calls to analyze your codebase in increments. If they ran the thinking model on the output of those function calls, then performance would be great but, so far, this is not what they are doing (yet). It also dramatically increases the cost of running the same operation.

namaria• 1 year ago

Reflecting on your comment I realized that using a huge amount of GPUs is akin to an Turing machine approaching infinite speed. So I think the promise of LLMs writing code is basically saying: if we add a huge number of reading/writing heads with unbounded number of rules, we can solve decideability. Because what is the ability to generate arbitrarily complex code if not solving the halting problem? Maybe there's a more elegant or logical way to postulate this, or maybe I'm just confused or plain wrong, but it seems to me that it is impossible to generate a program that is guaranteed to terminate unless you can solve decideability. And throwing GPUs at a huge tape is just saying that the tape approaches infinite size and the Turing machine approaches infinite speed...

Or put another way, isn't the promise of software that is capable to generate any software given a natural language description in finite time basically assuming P=NP? Because unless the time can be guaranteed to be finite, throwing GPU farms and memory at this most general problem (isn't the promise of using software to generating arbitrary software the same as the promise that any possible problem can be solved in polynomial time?) is not guaranteed to solve it in finite time.

sandbach• 1 year ago

I think you're right, but what company's business model doesn't produce a conflict between the user's well-being and the company's finances?

divan3• 1 year ago

How does the current state of Cursor agentic workflow compare to Windsurf Editor?

I've been using Windsurf since it was released, and back then, it was so ahead of Cursor it's not even funny. Windsurf feels like it's trained on good programming practices (check usage of the function in other parts of the project for consistency, double checking for errors after changes made, etc). It's also surprisingly fast (it can "search" the 5k files codebase in, like, 2 seconds. It even asked me once to copy and paste output from Chrome DevTools because it suspected that my interpretation of the result was not accurate (and it was right).

The only thing I truly wish is to have the same experience with locally running models. Perhaps Mac Studio 512GB will deliver :)

jjude2• 1 year ago

I too liked windsurf better than cursor until ...

I asked it to refactor an authenticatedfetch block of code. It went on a loop exhausting 15 credits (https://bsky.app/profile/jjude.com/post/3ljuhrxs3442k).

throwaway3141554• 1 year ago

How much does a credit cost in USD?

divan3• 1 year ago

To be honest, I switched from Cursor to Windsurf precisely because of how much less of a credits it uses. Even using daily, I couldn't even remotely hit the limits of the credits in Windsurf. Well, initially they didn't even show how many credits I'm using :), now it's more visible, but still for 10$ per month I still can't hit the limits and I'm not restricting myself (not abusing either).

blainm• 1 year ago

I've found tools like Cursor useful for prototyping and MVP development. However, as the codebase grows, they struggle. It's likely due to larger files or an increased number of them filling up the context window, leading to coherence issues. What once gave you a speed boost now starts to work against you. In such cases, manually selecting relevant files or snippets from them yields better results, but at that point it's not much different from using the web interface to something like Claude.

Semaphor• 1 year ago

I had that same experience with Claude Code. I tried to do a 95% "Idle Development RPG" approach to developing a music release organization software. At the beginning, I was really impressed, but with more and more complexity, it becomes increasingly incoherent, forgetting about approaches and patterns used elsewhere and reinventing the wheel, often badly.

blitzar• 1 year ago

Or the context not being large enough for all the obscure functions and files to go into the context. I am too basic to have dug deep enough, but a simple (automatic) documentation context for the entire project would certainly improve things for me.

turnsout2• 1 year ago

Agreed. One useful tip is to have Cursor break up large files into smaller files. For some reason, the model doesn't do this naturally. I've had several Cursor experiments grow into 3000+ line files because it just keeps adding.

Once the codebase is reasonably structured, it's much better at picking which files it needs to read in.

yard2010• 1 year ago

How can I stop Cursor from sending .env files with secrets as plain text? Nothing I tried from the docs works.

M4v3R• 1 year ago

This is a huge issue that was already raised on their forums and it's very surprising they didn't address it yet.

[0] https://forum.cursor.com/t/environment-secrets-and-code-secu...

timrichard3• 1 year ago

I have been adding .env files to .cursorignore so far.

I can see from that thread that the approach hasn’t been perfect, but it seems that the last two releases have tried to address that :

“0.46.x : .cursorignore now blocks files from being added in chat or sent up for tab completions, in addition to ignoring them from indexing.”

mexicocitinluez• 1 year ago

lol "move fast and break stuff....like really, really break stuff. i mean, break it so bad you'll probably cause people to lose their jobs and livelihoods"

torginus• 1 year ago

I have been a religious Cursor + Sonnet user for like past half a year, and maybe I'm an idiot, but I don't like this agentic workflow at all.

What worked for me is having it generate functions, classes, ranging from tens of lines of code to low hundreds. That way I could quickly interate on its output and check if its actually what I wanted.

It created a prompt-check-prompt iterative workflow where I could make progress quite fast and be reasonably certain of getting what I wanted. Sometimes it required fiddling with manually including files in the context, but that was a sacrifice I was willing to make and if I messed up, I could quickly try again.

With these agentic workflows, and thinking models I'm at a loss.

To take advantage of them, you need very long and detailed prompts, they take a long time to generate and drop huge chunks of code on your head. What it generates is usually wrong due to the combination of sloppy or ambiguous requirements by me, model weaknesses, and agent issues. So I need to take a good chunk of time to actually understand what it made, and fix it.

The iteration time is longer, I have less control over what it's doing, which means I spend many minutes of crafting elaborate prompts, reading the convoluted and large output, figuring out what's wrong with it, either fixing it by hand, or modifying my prompt, rinse and repeat.

TLDR: Agents and reasoning models generate 10x as much code, that you have to spend 10x time reviewing and 10x as much time crafting a good prompt.

In theory it would come out as a wash, in practice, it's worse since the super-productive tight AI iteration cycle is gone.

Overall I haven't found these thinking models to be that good for coding, other than the initial project setup and scaffolding.

timrichard3• 1 year ago

I think you’re absolutely right and I’ve come to the same conclusion and workflow.

I work on one file at a time in Ask mode, not Composer/Agent. Review every change, and insist on revisions for anything that seems off. Stay in control of the process, and write manually whenever it would be quicker. I won’t accept code I don’t understand, so when exploring new domains I’ll go back with as many questions as necessary to get into the details.

I think Cursor started off this way as a productivity tool for developers, but a lot of Composer/Agent features were added along the way as it became very popular with Vibe Coders. There are inherent risks with non-coders copypasting a load of code they don’t understand, so I see this use case as okay for disposable software, or perhaps UI concept prototypes. But for things that matter and need to be maintained, I think your approach is spot on.

_heimdall2• 1 year ago

Have you found that this still saves you time overall? Or do you spent a similar amount of time acting as a code reviewer rather than coding it yourself?

theshrike792• 1 year ago

Do you have any Cursor rules defined? Those tend to control its habit of trying to go off the rails and solve 42 problems at once instead of just the one.

Amekedl• 1 year ago

Compounding the opinions of other commentors, I feel that using Cursor is a bad idea. It's a closed source SaaS, and with these components involved, service quality can do wild swings on a daily basis, not something I'm particularly keen of.

rco87863• 1 year ago

This is true of every single service provider outside of fully OSS solutions, which are a teeny tiny fraction of the world's service providers.

turnsout2• 1 year ago

There's always Aider with local models!

Aurornis• 1 year ago

AI tools aren’t all or nothing. You use them when it makes sense and you go back to regular coding when they don’t (or when they’re unavailable).

You’re also not limited to a single tool. You can switch to different tools and even have multiple editors open at the same time.

walthamstow7• 1 year ago

Eng leadership at my place are pushing Cursor pretty hard. It's great for banging out small tickets and improving the product incrementally kaizen-style, but it falls down with anything heavy.

I think it's weakening junior engineers' reasoning and coding abilities as they become reliant on it without having lived for long, or at all, in the before times. I think may be doing the same to me too.

Personally, and quietly, I have a major concern about the conflict of interest of Cursor deciding which files to add to context then charging you for the size of the context.

As with so many products, it's cheap to start with, you become dependent on it, then one day it's not cheap and you're fucked.

rco87863• 1 year ago

I’ve been a paying cursor user for 4-5 months now and feeling the same. A lot more mistakes leaking into my PRs. I feel a lot faster but there’s been a noticeable decrease in the quality of my work.

Obviously I could just better review my own code, but that’s proving easier said than done to the point where I’m considering going back to vanilla Code.

ljm2• 1 year ago

Same result - I tried it for a while out of curiosity but the improvements were a false economy: time saved in one PR is time lost to unplanned work afterwards. And it is hard to spot the mistakes because they can be quite subtle, especially if you've got it generating boilerplate or mocks in your tests.

Makes you look more efficient but it doesn't make you more effective. At best you're just taking extra time to verify the LLM didn't make shit up, often by... well, looking at the docs or the source.. which is what you'd do writing hand-crafted code lol.

I'm switching back to emacs and looking at other ways I can integrate AI capabilities without losing my mental acuity.

DanHulton• 1 year ago

There's this concept in aviation of "ahead of or behind the plane". When you're ahead of the plane, you understand completely what it's doing and why, and you're literally thinking in front of it, like "in 30 minutes we have to switch to this channel, confirm new heading with ATC" and so forth. When you're behind the plane, it has done something expected and you are literally thinking behind it, like "why did it make that noise back there, and what does that mean for us?"

I think about coding assistants like this as well. When I'm "ahead of the code," I know what I intend to write, why I'm writing it that way, etc. I have an intimate knowledge of both the problem space and the solution space I'm working in. But when I use a coding assistant, I feel like I'm "behind the code" - the same feeling I get when I'm reviewing a PR. I may understand the problem space pretty well, but I have to basically pick up the pieced of the solution presented to me, turn them over a bunch, try to identify why the solution is shaped this way, if it actually solves the problem, if it has any issues large or small, etc.

It's an entirely different way of thinking, and one where I'm a lot less confident of the actual output. It's definitely less engaging, and so I feel like I'm way less "in tune" with the solution, and so less certain that the problem is solved, completely, and without issues. And because it's less engaging, it takes more effort to work like this, and I get tired quicker, and get tempted to just give up and accept the suggestions without proper review.

I feel like these tools were built without any sort of analysis if they _were_ actually an improvement on the software development process as a whole. It was just assumed they must be, since they seemed to make the coding part much quicker.

Aeolun2• 1 year ago

I feel like this is also related to cursor getting worse, not better, over the past few months.

theshrike792• 1 year ago

What do you consider "heavy"? Is it optimising an algorithm or "rewrite this whole codebase in <a different language>"?

walthamstow7• 1 year ago

Refactoring a critical function that is very large and complex. Yeah, maybe it shouldn't be so large and so complex, but it is and that's the codebase we have. I'm sure many other companies do too.

malux85• 1 year ago

Then only use it for the small tasks? There's one button you have to click to turn it off.

KronisLV• 1 year ago

Personally, and quietly, I have a major concern about the conflict of interest of Cursor deciding which files to add to context then charging you for the size of the context.
As with so many products, it's cheap to start with, you become dependent on it, then one day it's not cheap and you're fucked.

If it gets too expensive, then I guess the alternative becomes using something like Continue.dev or Cline with one of the providers like Scaleway that you can rent GPUs from or that have managed inference… either that, or having a pair of L4 cards in a closet somewhere (or a fancy Mac, or anything else with a decent amount of memory).

Whereas if there are no well priced options anywhere (e.g. the upfront investment for a company to buy their own GPUs to run with Ollama or something else), then that just means that running LLM based systems nowadays is economically infeasible for many.

onion2k5• 1 year ago

I think it's weakening junior engineers' reasoning and coding abilities as they become reliant on it without having lived for long, or at all, in the before times.

As someone old enough to have built websites in Notepad.exe it's totally reasonable that I ask my teams to turn off syntax highlighting, brace matching, and refactoring tools in VSCode. I didn't have them when I started, so they shouldn't use them today. Modern IDE features are just making them lazy.

walthamstow7• 1 year ago

Is this supposed to be funny?

moron4hire4• 1 year ago

Not all change is progress.

Change comes with pros and cons. The pros need to outweigh the cons (and probably significantly so) for change to be considered progress.

Syntax highlighting has the pro of making code faster to visually parse for most people at the expense of some CPU cycles and a 10 second setting change for people for whom color variations are problematic. It doesn't take anything away. It's purely additive.

AI code generation tools provide a dubious boost to short term productivity at the expense of extra work in the medium term and skill atrophy in the long term.

My junior developers think I don't know they are using AI coding tools. I discovered it about 2 months into them doing it, and I've been tracking their productivity both before and after. In one case, one might be committing to the repository slightly more frequently. But in all cases, they still aren't completing assignments on time. Or at all. Even basic things have to be rewritten because they aren't suitable for purpose. And in our pair programming sessions, I see them frozen up now, where they weren't before they started using the tools. I can even see them habitually attempt to use the AI, but then remember I'm sitting with them, and halt.

I tried to use AI code generation once to fill in some ASP.NET Core boilerplate for setting up authentication. Should be basic stuff. Should be 3 or 4 lines of code. I've done it before, but I forgot the exact lines and had been told AI was good for this kind of lazy recall of common tasks. It gave me a stub that had a comment inside, "implement authentication here". Tried to coax the AI into doing what I wanted and easily spent 10x more time than it would have taken to look up the documentation. And it still wasn't done. I haven't touched AI code gen since.

So IDK. I'm very skeptical of the claims that AI is writing significant amounts of working code for people, or that it at all rivals even a moderately smart junior developer (say nothing of actually experienced senior). I think what's really happening is that people are spending a lot of time spinning the roulette wheel, always betting on 00, and then crowing they're a genius when it finally lands.

timothygold4• 1 year ago

I think your mistaking programmer productivity with A.I that generates all the code for you allowing you to switch off your brain completely. prompt engineering code is not the same skill as programming and being good at it does not mean you actually understand how code or software works.

hakunin• 1 year ago

Yes. It is reasonable to ask them to turn off these things temporarily as an exercise, if you think it will help junior engineers write better code.

All teaching works this way. Would it be totally reasonable to have junior pilots only practice with autopilot?

hansvm• 1 year ago

I interview people every day who don't recognize glaring flaws despite the syntax highlighting and other linting in their favorite environment telling them something is definitely very wrong. Much like with LLMs, I'm firmly in the camp of (1) master working without it, (2) master working with it, (3) integrate it into your workflow in the way that suites you best.

If your team can't recognize bullshit without syntax highlighting, they're going to struggle even when it's turned on. Once they've mastered the underlying principles, syntax highlighting will make them much more effective.

kamaal9• 1 year ago

>but it falls down with anything heavy.

If you are using LLMs to write anything more than a if(else)/for block at a time you are doing it wrong.

>I think it's weakening junior engineers' reasoning and coding abilities as they become reliant on it without having lived for long, or at all, in the before times.

When I first started work, my employer didn't provide internet access to employees, their argument would always be how would you code if there was no internet connection, out there in the real world? , As it turns out they were not only worried about the wrong problem, but the got the whole paradigm about this new world wrong.

In short it was not worth building anything at all in a world internet doesn't exist.

>then one day it's not cheap ...

Again you are worried about the wrong thing, your worry should not be what happens when its no longer cheaper, but when it, as a matter of fact gets cheaper. Which it will.

bluefirebrand5• 1 year ago

If you are using LLMs to write anything more than a if(else)/for block at a time you are doing it wrong

Then what value are they actually adding?

If this is all they are capable of, surely you could just write this code yourself much faster than trying to describe what you want to the LLM in natural language?

I cannot imagine any decently competent programmer gaining productivity from these tools if this is how limited they still are

Why are people so bullish on them?

bitwize• 1 year ago

In a few years, if you mention you just straight up code with an editor or IDE, there's going to be a hackernews who's going to ask you about the size of your project at work (LOC, dev team headcount) because you can't really be working on something big and important without AI.

evo_9• 1 year ago

Has anyone tried limiting Cursor or Cline, etc, to a higher level role such as analysis and outlining proposed changes, and then coding those changes yourself with minimal LLM interaction? Aka, ask to define / outline only a high level set of changes, but do no actual changes to any file; then proceed through the outlined work, limiting Cursor to roughing out the work and hand writing the actual critical bits? That’s the approach I’ve been taking, a sort of best of both worlds that greatly speeds me up without taking the hands 100% off the wheel.

real0mar• 1 year ago

Yeah. Reasoning models like r1 tend to be good for architecting changes and less optimal for actually writing code. So this allows the best of both worlds.

vineyardmike• 1 year ago

This seems like the worst of both worlds. The human still has to do the "boring" work of writing out all the boiler plate stuff, but now there is a machine telling the human what to do. Oh, and the machine is famously not great at big question type stuff while being being much more performant at churning out boilerplate.

tgdude• 1 year ago

I find the opposite. I tend to think through the problem myself, give cursor/claude my understanding, guide it through a few mistakes it makes, have it leave files at 80% good enough as it codes and gets stuck, and then spend the next 20 min or so cleaning up the changes and fixing the few wire up spots it got wrong.

Often I will decompose the problem into smaller subproblems and feed those to cursor one by one slowly building up the solution. That works for big tickets.

For me the time saving and force multiplier isn't necessarily in the problem solving, I can do that faster and better in most cases, but the raw act of writing code? It does that way faster than me.

dr_kiszonka• 1 year ago

I think it's weakening junior engineers' reasoning and coding abilities as they become reliant on it without having lived for long, or at all, in the before times.

I have a feeling there will be a serious shortage of strong "old-school" seniors in a few years. If students and juniors are reliant on AI, and we recruit seniors from our juniors, who will companies turn to when AI gets stuck?

mwgalloway2• 1 year ago

I’m looking forward to when the job market recovers, but I’m not looking forward to the prospect of a significant amount of future demand being in the realm of having to scale and maintain the AI slop code that’s being generated now.

HenryBemis2• 1 year ago

I feel that (I can feel it in my own skin/life) that those 'oldies' with "ok" SME skills and "great" business acumen will be the ones harnessing the hordes of 'prompt engineers'.

It is like they will be the 'new nerds' and we will have the 'street-smarts'.

mbil• 1 year ago

I wrote about this recently. I agree, though I think it might take a little longer. There will be a deskilling where the easy problems are solved by AI, so new grad workers won’t have gradual experience to grow into seasoned experts.

arkh2• 1 year ago

I think it's weakening junior engineers' reasoning and coding abilities as they become reliant on it without having lived for long, or at all, in the before times.

Same as StackOverflow, same as Google, same as Wikipedia for students.

The problem is not using the tools, it's what you do with the result. There will always be lazy people who will just use this result and not think anymore about it. And then there will always be people who will use those results as a springboard to what to look for in the documentation of whatever tool / language they just discovered thanks to Cursor.

You want to hire from the second category of people.

m_fayer• 1 year ago

Quantity has a quality all it’s own.

That is to say, as we strive for better and better tools along a single axis, at some point the social dynamics shift markedly, even though it’s just “more of the same”. Digital distribution was just better distribution, but it changed the very nature of journalism, music, and others. Writing on computers changed writing. And the list goes on.

“This is just the next thing in a long line of things” is how we technologists escape the disquieting notion that we are running more and more wild social experiments.

addicted• 1 year ago

Same as StackOverflow, same as Google, same as Wikipedia for students.

Is it though?

If I used stack overflow, for example, I still needed to understand the code well enough to translate it to my specific codebase, changing variable names at the very least.

breakfastduck• 1 year ago

This is not the same. Because SO, google etc require actual research, introspection, prompting AI does not.

firecall• 1 year ago

As with so many products, it's cheap to start with, you become dependent on it, then one day it's not cheap and you're fucked.

Exactly!

As you learn a codebase, you and it become better together.

It can take a while, but the investment in understanding the stack you are working with does pay off.

Trying to shortcut junior devs to productivity is the plot to a cyberpunk horror movie!

Torkel• 1 year ago

I saw this post from a professor this morning:

https://x.com/lxeagle17/status/1899979401460371758

Students are not asking questions anymore.

Small assignments work well!

But then the big test comes and scores are at an all time low.

HenryBemis2• 1 year ago

Yes it is exactly what we think is causing it. Students use LLMs to get 10/10 in assignments so they don't learn 'the thing' and they tank on the big-closed-books-non-LLM-exams.

BUT (apologies for the caps), back in the day we didn't have calculators and now we do. And perhaps the next phase in academia is "solve this problem with the LLM of your choice - you can only use the free versions of LLAMA vX.Y, ChatGTP vA.B, etc. - no paid subscriptions allowed" (in the same spirit that for some exams you can use the simple calculator and not the scientific one). Because if they don't do it, they (academia/universities) will lose/bleed out even more credibility/students/participation.

The world is changing. Some 'parts' are lagging. Academia is 5 years behind (companies paying them for projects help though), Politicians are 10-15 years behind (because the 'donors' (aka bribers) prefer a wild-wild-west for a few years before rights are protected. (Case and point writers/actors applying a lot of pressure when they realized that politicians won't do anything until cornered)

belter• 1 year ago

It seems these days every other post on HN is about Cursor.

Is Cursor ultimately not using GPT-4, GPT-4o, Claude 3.5/3.7 Sonnet and so on? If some of the auto completion and agent features might be nice, is this not too much push for what is essentially just another UI plugin?

rafram• 1 year ago

They have their own next edit prediction model, and for me personally that’s the killer feature. I don’t like the VS Code platform very much, but it’s just magical to be able to press tab-tab-tab and watch Cursor jump between lines in the file and make the edits that I was clearly about to make.

smokel• 1 year ago

Don't underestimate how much effort it takes to apply a technique (LLMs) to a certain domain.

Creating a good user interface, processing the code into appropriate embeddings for useful matching, and marketing are just three things that take a lot of effort to get right.

The fragmented landscape of alternatives is not attractive at all.

One thing I'm quite worried about, though, is that Cursor is a fork of VS Code. This will never be maintainable in the long run, in particular if Microsoft does not want them to continue.

siva72• 1 year ago

Has anyone else noticed how sharp the quality of composer and chat output has dropped since their newest release? I had a good opinion about them but they seem to have dropped the ball

brushfoot• 1 year ago

Cursor deciding which files to add to context then charging you for the size of the context

Copilot is unlimited for a flat fee [1]. I've been happy with it.

They also have a free tier now, but it's limited to 50 chats a month, which I'd burn through pretty quickly.

1: https://docs.github.com/en/copilot/about-github-copilot/subs...

lolinder• 1 year ago

Personally, and quietly, I have a major concern about the conflict of interest of Cursor deciding which files to add to context then charging you for the size of the context.

Can you elaborate on what you're referring to? I don't used Cursor extensively, but I do pay for it, and it was a flat fee annual subscription for unlimited requests, with the "fast" queue being capped at a set number of requests per month with no reference to their size.

Claude Code does work the way you say, since you provide it your Anthropic API key. But I have not seen Cursor charging for context or completion tokens, which is actually the reason why I'm paying for it instead of using something like Aider.

https://www.cursor.com/pricing

j452• 1 year ago

Heavier architecture / systems design stuff is sometimes easier to do with a differently configured cursor, or claude.

The key I find is experience doing that kind of stuff to begin with, vs domain experience as well, vs little to none and wanting to learn the ropes.

quotz2• 1 year ago

Cline is much better

cheema33• 1 year ago

Why not list the reasons why you like it over Cursor?

quotz2• 1 year ago

I was in a hurry and hoped others would continue the thread. Cline is open source, pay as you go, using your own Openrouter API keys. You can use any model you like. Its more expensive than Cursor of course, depending on usage and models.

r_singh• 1 year ago

Just use Cline, it beats Cursor hollow — saves me like hours per day

mrlowlevel• 1 year ago

Do any of these tools use the rich information from the AST to pull in context? Coupled with semantic search for entry points into the AST, it feels like you could do a lot…

zarathustreal• 1 year ago

Don’t they all do this? Surely they’re not just doing naive text, n-gram, regex, embeddings, etc, right?

marxplank• 1 year ago

they are just a language model generating text which happens to be code, ASTs are not used afaik

hannah_creator• 1 year ago

Very useful!!!

factsaresacred• 1 year ago

Too bad they removed the ability to use Chat (rebranded as Ask) with your own API keys in version 0.47. Now every feature requires a subscription.

Natural for Cursor to nudge users towards their paid plans, but why provide the ability to use your own API keys in the first place if you're going to make them useless later?

siva72• 1 year ago

Oh no didn't realise that. I had a high opinion of this team but it starts changing now...

2sk21• 1 year ago

I read this point in the article with bafflement:

"Learn when a problem is best solved manually."

Sure, but how? This is like the vacuous advice for investors: buy low and sell high

dkersten• 1 year ago

By trying things and seeing what it’s good and bad at. For example, I no longer let it make data modelling decisions (both for client local data and database schemas), because it had a habit of coding itself into holes it had trouble getting back out of, eg duplicating data that it then has difficulty keeping in sync, where a better model from the start might have been a more normalised structure.

But I came to this conclusion by first letting it try to do everything and observing where it fell down.

ookblah• 1 year ago

parts of the article are spot on. after the magic has worn off i find it's best to literally treat it like another person. would you blindly merge code from someone else or huge swaths of features? no. i have to review every single piece of code, because later on when there's a bug or new feature you have to have that understanding.

another huge thing for me has been to scaffold a complex feature just to see what it would do. just start out with literal garbage and an idea and as long as it works you can start to see if something is going to pan out or not. then tear it down and do it again with those new assumptions you learned. keep doing it until you have a clear direction.

or sometimes my brain just needs to take a break and i'll work on boilerplate stuff that i've been meaning to do or small refactors.

ThinkBeat• 1 year ago

What programming languages do you primarily use ? I feel that knowing what programming languages a llm is best at is valuable but often not directly apparent.

flipgimble• 1 year ago

Cursor overwrites the “code” command line shortcut/alias that’s normally set by VS Code. It does this on every update with no setting to disable this behavior. There are numbers of forum threads asking about manual solutions. This seems like a deliberately anti-user feature meant to get their usage numbers up at all costs. This small thing makes me not trust the decision making process at Cursor won’t sell me out as a user.

throwaway3141554• 1 year ago

This is the primary reason I uninstalled Cursor and subsequently realized that, hey, VS Code has most of these features now.

What in the hell were they thinking?!

rhodescolossus• 1 year ago

I've tried Cursor a couple of times but my complain is always the same: why forking VS Code when all this functionality could just be an extension, same as Copilot does?

Some VSCode extensions don't work, you need to redo all your configuration, add all your workspaces... and the gain vs Copilot is not that high

frereubu• 1 year ago

and the gain vs Copilot is not that high

I think that's (at least part of) your answer. More friction to move back from an entirely separate app rather than disabling an extension.

andrewl-hn• 1 year ago

why forking VS Code when all this functionality could just be an extension, same as Copilot does?

Have you programmed extensions for VSCode before? While it seems like a fairy extensible system overall, the editor component in particular is very restrictive. You can add text (that's what extensions like ErrorLens and GitLens are doing), inlay hints, and on-hover popup overlays (those can only trigger on words, and not on punctuation). What Cursor does: the automatic diff-like views of AI suggestions with graphic outlines, floating buttons, and whatnot right on top of the text editing view - is not possible in vanilla VSCode.

This was originally driven by necessity of tighter control over editor performance. In its early days VSCode was competing with Atom - another extensible JS-powered editor from GitHub, and while Atom had an early lead due to larger extensions catalog VSCode ultimately won the race because they manged to maintain lower latency of their text editor component. Nowadays they still don't want to introduce extra extension points to it, because newer faster editors pop out all the time, too.

satvikpendem5• 1 year ago

Now the Atom (and Electron, hence the name) creators are working Zed, which is even faster than VSCode, albeit not as extensible.

DiabloD35• 1 year ago

I'm sorry, but isn't Cursor just an editor? Maybe an editor shouldn't actually have garbage parts to avoid?

Why not just use an editor that is focused on coding, and then just not use an LLM at all? Less fighting the tooling, more getting your job done with less long term landmines.

There are a lot of editors, and many of them even have native or semi-native LLM support now. Pick one.

Edit: Also, side note, why are so many people running their LLMs in the cloud? All the cutting edge models are open weight licensed, and run locally. You don't need to depend on some corporation that will inevitably rug-pull you.

Like, a 7900XTX runs you about $1000. You probably already own a GPU that cost more in your gaming rig.

krapht• 1 year ago

Edit: Also, side note, why are so many people running their LLMs in the cloud? All the cutting edge models are open weight licensed, and run locally. You don't need to depend on some corporation that will inevitably rug-pull you.

???

Deepseek R1 doesn't run locally unless you program on a dual socket server with 1 TB of RAM. Or enough cash to have a cabinet of GPUs. The trend for state-of-the-art LLMs is to get bigger over time, not smaller.

Look, I've played with llava and llama locally too, but the benchmarked performance is nowhere near what you can get from the larger cloud providers who can serve hundred-million+ parameter models without quantization.

DiabloD35• 1 year ago

You wouldn't use full fledged R1 for coding. There are distilled models using R1 for coding that get you most of the way there. R1 also doesn't take 1TB of RAM, go use read Unsloth's writeup on how to reduce model size without reducing quality (they got it to fit into 131GB): https://unsloth.ai/blog/deepseekr1-dynamic tl;dr parameter count is where the statistical model lives or dies, not weight precision; you can't blindly shrink every weight, and tooling is learning how to not butcher models.

Also, performance between cloud-ran models and models I've ran locally with llama.cpp seem to be actually pretty similar. Are you sure your model didn't fit into your VRAM, or something else may have been misconfigured? Not fitting into VRAM slows everything to a halt. All the coder models that are worth looking at fit into 24GB cards in their full sized variants with the right quantization.

zild3d• 1 year ago

All the cutting edge models are open weight licensed, and run locally.

No? from https://lmarena.ai/ coding:

...

Rank* ... Model ... Score ... Org ... License

1 ... Grok-3-Preview-02-24 ... 1414 ... xAI ... Proprietary

1 ... GPT-4.5-Preview ... 1413 ... OpenAI ... Proprietary

3 ... Gemini-2.0-Pro-Exp-02-05 ... 1378 ... Google ... Proprietary

3 ... o3-mini-high ... 1369 ... OpenAI ... Proprietary

3 ... DeepSeek-R1 ... 1369 ... DeepSeek ... MIT

3 ... ChatGPT-4o-latest (2025-01-29) ... 1367 ... OpenAI ... Proprietary

3 ... Gemini-2.0-Flash-Thinking-Exp ... 1366 ... Google ... Proprietary

3 ... o1-2024-12-17 ... 1359 ... OpenAI ... Proprietary

3 ... o3-mini ... 1353 ... OpenAI ... Proprietary

4 ... o1-preview ... 1355 ... OpenAI ... Proprietary

4 ... Gemini-2.0-Flash-001 ... 1354 ... Google ... Proprietary

4 ... o1-mini ... 1353 ... OpenAI ... Proprietary

4 ... Claude 3.7 Sonnet ... 1350 ... Anthropic ... Proprietary

DiabloD35• 1 year ago

Yes, I'm aware of various rankings. Try all of those models on something that isn't commonly used on a benchmark, and you'll notice that a lot of the proprietary models have trouble actually producing statistically relevant results.

The only one that I've come across that makes me think LLMs will maybe be useful someday is Deepseek R1 and the redistillations based on it.

I've seen HN's fascination with OpenAI's products, and I can't understand why. Even O1 and O3, they're always too little too late, somebody else already is doing something better and throwing it into a HF repo. Must be the Silicon Valley RDF at work.

stared3• 1 year ago

Other useful things I've discovered:

- Push for DRY principles ("make code concise," "ensure good design").

- Swap models strategically; sometimes it's beneficial to design with one model and implement with another. For example, use DeepSeek R1 for planning and Claude 3.5 (or 3.7) for execution. GPT-4.5 excels at solving complex problems that other models struggle with, but it's expensive. - Insist on proper typing; clear, well-typed code improves autocompletion and static analysis.

- Certain models, particularly Claude 3.7, overly favor nested conditionals and defensive programming. They frequently introduce nullable arguments or union types unnecessarily. To mitigate this, keep function signatures as simple and clean as possible, and validate inputs once at the entry point rather than repeatedly in deeper layers.

- Emphasize proper exception handling. Some models (again, notably Claude 3.7) have a habit of wrapping everything in extensive try/catch blocks, resulting in nested and hard-to-debug code reminiscent of legacy JavaScript, where undefined values silently pass through multiple abstraction layers. Allowing code to fail explicitly is a blessing for debugging purposes; masking errors is like replacing a fuse with a nail.

stared3• 1 year ago

Some additional thoughts on GPT-4.5: it provides BFK-9k experience - eats e̶n̶e̶r̶g̶y̶ ̶c̶e̶l̶l̶s̶ budget ($2 per call!) like there is no tomorrow, but removes bugs with a blast.

In my experience, the gap between Claude 3.7 and GPT-4.5 is substantial. Claude 3.7 behaves like an overzealous intern on stimulants. It delivers results but often includes unwanted code changes, resulting in spaghetti code with deeply nested conditionals and redundant null checks. Although initial results might appear functional, the resulting technical debt makes subsequent modifications increasingly difficult, often leaving the codebase in disarray. GPT-4.5 behaves more like a mid-level developer, thoughtfully applying good programming patterns.

Unfortunately, the cost difference is significant. For practical purposes, I typically combine models. GPT-4.5 is generally reserved for planning, complex bug fixes, and code refinement or refactoring.

In my experience, GPT-4.5 consistently outperforms thinking models like o1. Occasionally, I'll use o3-mini or DeepSeek R1, but GPT-4.5 tends to be noticeably superior (at least, on average). Of course, effectiveness depends heavily on prompts and specific problems. GPT-4.5 often possesses direct knowledge about particular libraries (even without web searching), whereas o3-mini frequently struggles without additional context.

kikimora2• 1 year ago

Wouldn’t it be easier instead of juggling with models and their quirks to just write the code the old way?

pestkranker• 1 year ago

Is there an equivalent to cursorrules and copilot-instructions for the Jetbrains IDEs (Rider) + GitHub Copilot extension?

DaveMcMartin• 1 year ago

For those of you who, like me, use Neovim, you can achieve "cursor at home" by using a plugin like Avante.nvim or CodeCompanion. You can configure it to suit your preferences.

Just sharing this because I think some might find it useful.

gregwebs• 1 year ago

AI blows me away when asked to write greenfield code. It can get a complex task using hundreds of lines of code right on the first try or perhaps it needs a second try on the prompt and an additional tweak of the output code.

As things move from prototype to production ready the productivity starts to become a wash for me.

AI doesn’t do a good job organizing the code and keeping it DRY. Then it’s not easy for it to make those refactorings later. AI is good at writing code that isn’t inherently defective but if there is complexity in the code it will introduce bugs in its changes.

I use Continue for small additions and tab completions and Claude for large changes. The tab completions are a small productivity boost.

Nice to see these tips- I will start experimenting with prompts to produce better code.

flippyhead• 1 year ago

Note that the latest update (0.47.x) made this useful change:

Rules: Allow nested .cursor/rules directories and improved UX to make it clearer when rules are being applied.

This has made things a lot easier in my monorepos.

dimgl• 1 year ago

The new Cursor update (0.47) is cursed. They got rid of codebase searching (WTF?) and the agent is noticeably worse, even when using Sonnet 3.5.

I'm really shocked, actually. This might push me to look at competitors.

HugoDias• 1 year ago

I saw this post on the first page a few minutes ago (published 5 hours ago), but it quickly dropped to the 5th page. Given its comments and points, that seems odd. I had to search to find it again. Any idea why?

timothygold4• 1 year ago

I tried cursor for a day or two and then asked for a refund... here's why:

* It has terrible support for Elixir (my fav language) because the models are only really trained on python.

* Terrible clunky interface... it would be nice if you didn't have to click around, do modifier ctrl + Y stuff ALL the time.

* The code generated is still riddled with errors or naff (apart from boiler plate)... so I am still * prompt engineering * the crap out of it.. which I'm good at but I can prompt engineer using phind.com...

* The fact that the code is largely broken first time and they still haven't really fixed the context window problem means you have to copy paste error codes back into it.. defeating the purpose of an in integrated IDE imo.

* The free demo mode stops working after generating one function... if I had been given more time to evaluate it fully I would never have signed up. I signed up to see if it was any good.. which it isn't.

TheAnkurTyagi• 1 year ago

nice also you can use project-specific structure and markdown files to ensure the AI organizes content correctly for your use case. we are using it on 800k lines of golang and it works well. https://getstream.io/blog/cursor-ai-large-projects/

Jayakumark• 1 year ago

Will you be able to share those ai_*.md files ?

timothygold4• 1 year ago

I made a quick prototype to demonstrate what I think A.I code assistance should be..

https://github.com/hibernatus-hacker/ai-hedgehog

This is a simple code assistant that doesn't get in your way and makes sure you are coding (not losing your ability to program).

You configure a replicate API token from replicate... install the tool and point it at your code base.

When you save a file it asks the LLM for advise and feedback on the file as a "senior developer".

Run this along side your favorite editor to get feedback from an LLM as your working on (open source code nothing you don't want third parties to see).

You are still programming and using your brain but you have some feedback when you save files.

The feedback is less computationally expensive or fraught with difficulty than actually getting code from LLM's so it should work with much less powerful models.

It would be nice if there was a search built in so it could search for useful documentation for you.

9dev• 1 year ago

It sounds very interesting, I would suggest adding some kind of demo or video on the readme; I think i know how it works, but it would be nice to see Hedgehog in action.

I’m going to try it tomorrow!

trash_cat• 1 year ago

And then at the top of the file, just write some text about what the project is about. If you have a particular file structure and way of organising code that is great to put in as well.

By asking the AI to generate a context.md file, you get an automatically structured overview of the project, including its purpose, file organization, and key components. This makes it easier to onboard new contributors, including other LLMs.

ZeroTalent• 1 year ago

Cursor is not SOTA. It's just popular. Look into alternatives. CLine, Augment Code, etc.

mattwad• 1 year ago

Everyone posting should be considerate of your language/stack. It's pretty likely that Cursor doesn't work equally for every language. I'm working on a Next.js/Typescript/Solidity monorepo with multiple apps and packages and it handles pretty much anything I throw at it. I know I can squeeze more out because I have only been really using it heavily for the past month or so.

DeathArrow• 1 year ago

Apart from the fact that it chews fast requests like there's no tomorrow I dislike how it does changes I didn't ask to. And if I ask to undo what it did without being asked, it goes on and beaks more code.

In my test application I had a service which checked the cache, then asked the repository if no data is in cache, then uses external APIs to fetch some data, combine it and update the DB and the cache.

I asked Cursor to change using DateTime type to using Unix timestamp. It did the changes but it also removed cache checks and calling external APIs, so my web app relied just on the data in DB. When asked to add back what it removed, it broke functionality in other parts of the application.

And that is with a small simple app.

austin-cheney• 1 year ago

With so many caveats, so many exceptions, so many rules, so much manual validation, and so little trust why bother? It sounds like an employee pending termination.

atoav• 1 year ago

I have been using AI coding for a while now and the issues I keep having is the following:

- LLM keeps forgetting/omitting parts of the code

- LLM keeps changing unrelated parts of the code

- LLM does not output correctly typed code (with Rust this can feel like throwing mud at a wall and see what sticks, in the end you're faster on your own)

- LLM flip-flops back and forth between two equally wrong answers when asked about a particularly (from the perspective of the LLM) to answer problem

In the end the main thing any AI coding tool will have to solve, is how to get the human in front of the LLM to trust that the output does what it does without breaking other things.

But of course LLMs are already crazy good at whst they do. I just wonder how people who have no idea what they are doing will be able to handle that power.

fifticon• 1 year ago

I've had colleagues like that since the 1990's :-) :-( But yes, we would like AIs to be better than Kevin..

hakaneskici• 1 year ago

What is your mental model when coding with Cursor?

When I code with with AI assistance; I "think" differently and noticed that I have more memory bandwidth to think about the big picture rather than the details.

With AI assistance, I can keep the entire program logic in my head; otherwise I have to do expensive context switching between the main components of the program/system.

How are you "thinking" when typing prompts vs typing actual code?

kobe_bryant• 1 year ago

I was trying to figure out what he does and his website proudly states at the very top “ No templates, no no-code, no AI slop - just great sites built to grow.”. interesting!

eleumik• 1 year ago

How much is the price of a kilogram of code in the us ?

I use Cursor daily - here's how I avoid the garbage parts

Related Stories