I keep switching away from and back to Cursor (mainly due to frontier models explicitly fighting their apply model, the first few times it’s funny to see the LLM itself write “this is frustrating” but I digress).
And every time I find it having diverged further from VSCode compatibility.
This wouldn’t be so bad if it was an intentional design choice but it seems more that Microsoft is starting to push them out? Like MS Dev Containers plugin is still recommended by some leftover internal, but if you install it you get pushed to a flow that auto uninstalls it and installs Remote Containers by Anysphere (which works differently and lacks support for some features). And I end up rebuilding my Dev Container once more… I also noticed recent extensions such as the Postgres one from MS also doesn’t exist.
techpression · 24m ago
I fully expect MS to change the VS Code license in the not so far future to make applications like Cursor not possible. Forking might be a thing initially but will quickly fade since without the backing of MS the ecosystem around it will die.
evo_9 · 4m ago
This is why I’m using Zed now, and Claude Code. I like to keep Zed pretty minimal and I’m slowly weening off of Cursor in favor of Claude Code when I need it
ai_assisted_dev · 11m ago
I still don't understand how cursor is making any money at all. I spend so much time inside cursor, that I am spending 10-20$ per day on additional requests. Now if I connec model provider APIs to windsurf, I'd be spending upwards of 100$ due to amount of tokens I use through the API per day. And if I connect my own API key to Cursor, I immediately get rate limited for any request, because I go well above 50 per minute. And I did try claude code, but its just not on par with my experience with Cursor.
I could probably go much lower, and find a model that is dirt cheap but takes a while; but right now the cutting edge (for my own work) is Claude 4 (non-max / non-thinking). To me it feels like Cursor must be hemorrhaging money. The thing that works for me is that I am able to justify those costs working on my own services, that has some customers, and each added feature gives me almost immediate return on investment. But to me it feels like the current rates that cursor charges are not rooted in reality.
Quickly checking Cursor for the past 4 day period:
Requests: 1049
Lines of Agent Edits: 301k
Tabs accepted: 84
BudaDude · 7h ago
I love Cursor, but it feels like a ticking time bomb with extensions not being updated at the same rate as VSCode.
Also another issue I am starting to see is the lack of shared MCP servers. If I have VSCode, Cursor, and Claude open, each one is running its own instance of the MCP server. You can imagine that with a dozen or so MCP's, the memory footprint becomes quite large for no reason.
nojs · 5h ago
It’s more of a ticking time bomb because it relies so heavily on upstream model providers who all have competing products, particularly Claude Code.
ramoz · 1h ago
I think about this daily. More devs are starting to pick up on Claude Code. The initial “not an IDE!” scare is usually diminished within the initial session.
I don’t think the future of agentic software development is in an IDE. Claude Code gives me power to orchestrate - the UX has nothing to do with terminal; it just turns out an agent that lives on the OS and in the filesystem is a powerful thing.
Anthropic can and will evolve Claude Code at a pace cursor cannot evolve IDE abstractions. And then yea - they are designing the perfect wrapper because they are also designing the model.
It… sure is something. I’m still thinking about if it’s horrible, visionary, or both.
ramoz · 1h ago
Wow yea, definitely resonates thanks for sharing.
This is why I think the future may be less about “agents” and more about “intelligent infrastructure”
I don’t want to chat. I want to point a wand and make things happen.
Aeolun · 5h ago
Until Claude Code becomes manageable price wise, I don’t think Cursor really sees them as their competition. I can burn the whole cursor subscription price in a single day with Claude Code.
travbrack · 1h ago
I'm surprised nobody is mentioning how cheap copilot pro is. $20 and you get all you can eat inference without using your own api key for the models on vs code agent mode.
nextos · 52m ago
I've found Emacs plus gptel very pleasant to use. And following the ethos of Emacs, it is backend agnostic and very malleable.
Besides, if you want something inexpensive, using Gemini 2.0 Flash as a backend is completely free. Google provides an API key at no cost.
mogili · 1h ago
I use with the Max plan ($100 per mo) and its well worth the money and only hit the limit once so far.
vunderba · 3h ago
Yeah, if you're a heavy user of Claude code, you pretty much need to use it with a Max subscription rather than a BYOK approach. But that starts at $100 / month so it's a significant bump from cursor.
unmole · 2h ago
> need to use it with a Max subscription rather than a BYOK
Why is that?
steveklabnik · 1h ago
Max is far cheaper.
myth_drannon · 4h ago
They just announced Claude Code will come with the pro subscription ($20)
falcor84 · 4h ago
Interesting, but the limit is really low:
> Pro ($20/month): Average users can send approximately 45 messages with Claude every 5 hours, OR send approximately 10-40 prompts with Claude Code every 5 hours. [0]
it's a MCP server with WASM plugin system, packaged, signed & published via OCI registry.
manorek · 1h ago
I don't understand. Is this meant to run locally? Because I tried to deploy my agent using GitHub MCP server to K8s. I can't ask my agent to run docker command in a pod.
Cursor and other VSCode forks connect to Open VSX [1] for extensions. Barring some of the Microsoft extensions, I've found that pretty much all the extensions I use are available and kept up to date on Open VSX. Cursor seems to have enough funding to support their own variants of the Microsoft extensions, like Python and C++.
The one issue I've run into is that the VSCode version Cursor uses is several months old, so we're stuck using older extensions until they update.
This is good user feedback. If Cursor is "Claude + VSCode", why do you need the other 2 open?
TechDebtDevin · 5h ago
Just run it on http
vFunct · 5h ago
this is the way.
STDIO MCP is really just a quick hack.
mmasu · 6h ago
why would you open 3 IDEs all at once :-)
curiousElf · 1h ago
I wish they had proper support for multi root repos (even though the last update promised better support, it was just a line in the release notes with no docs - which seems to be their usual change management style).
Its so painful - the model never knows the directory in which it is supposed to be and goes on a wild goose chase of searching in the wrong repo. I have to keep guiding it to the right repo. Anyone here has had success with such a setup?
SkyPuncher · 51m ago
I have a cursor rule that tells it about directories. Basically just X is UI, Y is BE, Z is auth.
Keep it short. It's enough for it to realize it needs to navigate directories.
feldstein · 19m ago
Dev here, Can you give me more details about whats going on here? Screenshots or request ids (or both) would be best
You can email me directly at feldstein at anysphere.co
jameslk · 1h ago
Cursor wishlist item for any PM listening:
When reviewing the changes made from agent mode, I don’t know why the model made the change or whether the model even made the change versus a tool call making the change. It’s a pain to go fish out the reason from a long response the model gives in the chat.
Example: I recently asked a model to set up shadcn for a project, but while trying to debug why things looked pretty broken, I had sift through a bunch of changes that looked like nasty hallucinations and separate those from actual command line changes that came from shadcn's CLI the model called. Ended up having to just do things the old fashioned way to set things up, reading the fine manual and using my brain (I almost forgot I had one)
It would be nice if above every line of code, there’s a clear indication of whether it came from a model and why the model made the change. Like a code comment, but without littering the code with actual comments
namanyayg · 4m ago
Was just discussing this a friend today
Hand written code needs to be distinguishable and considered at a higher priority for future code generation context
jadbox · 7h ago
VSCode with extensions Copilot [autocomplete] + CLINE [AI chat] + FOAM [obsidian-esk markdown support] is goat. There's no way a closed-source alternative to going to compete with this.
AstroBen · 7h ago
Have you given the alternatives a genuine test? My experience with Aider (never used Cline) is that it's nowhere near as good
CuriouslyC · 5h ago
The benefit of Aider is that you can configure a very involved static analysis toolchain to edits which directly triggers new edits in response, and everything is a git commit so it's easy to revert bad edits quickly. I have used both and I find Aider provides more control and produces code faster due to leaner prompts (it's also easier to run multiple Aider instances than Cursor instances), while Cursor has a prettier interface, and I do like being able to see diffs live in files (though I almost never spend the time reading them to accept/reject). I imagine if you don't spend any time configuring Aider cursor would probably seem far better.
AstroBen · 5h ago
what are the most useful changes you've made to the configuration? This could be it - I haven't played with that a whole lot
jprokay13 · 5m ago
Create a file like conventions.md in the root of your repository with specific commands for common tasks: running tests, linters, formatters, adding packages
Set this as part of the files it reads on startup. Then ask aider to look at your codebase and add to it :)
Aider has a lot of slash commands to familiarize yourself with. Ask and web are crucial commands to get the most out of it.
My recommendation to anyone is to use ask the most then tell it to “implement what we discussed” when it looks good.
Hope that helps
CuriouslyC · 3h ago
The biggest thing is to set it to autofix lint/test issues, then to set up a really good lint/test config. Also, I find that Aider's default system prompt setup is a little less preconfigured out of the box than Cursor's, so it helps to have detailed styleguide/ai rules documents that are automatically added to the chat. I usually configure my projects to add README.md, STYLEGUIDE.md (how to structure/format code) and AIRULES.md (workflow stuff, for instance being socratic with the user when requirements aren't clear or the prompt contains ambiguity, general software engineering principles/priorities, etc).
tunesmith · 6h ago
Are you saying Aider isn't as good as Cursor, or that Cursor isn't as good as Aider?
AstroBen · 6h ago
aider isn't as good
jadbox · 5h ago
In my limited tests, Aider and CLINE are very similar, but it's really hit/miss depending on the specific task.
eikenberry · 5h ago
Doesn't this rely completely on the AI it is using and not the client?
mogili · 1h ago
I used Cline and Claude Code extensively for a project. Claude code is much better.
esafak · 4h ago
Why not? What does it do that Copilot Agent or Junie can't? All the competitors have a similar UX and the same selection of models.
ramoz · 1h ago
Cline copying features of Claude Code seems like sustainable competition.
kurtis_reed · 6h ago
What does "is goat" mean?
bluetidepro · 6h ago
Goat is slang for “(the) greatest of all time.”
kshacker · 5h ago
I was today years old minus maybe a few months when I learned this, and I had seen it referenced so many times.
SR2Z · 5h ago
This one is not particularly new. I wanna say that GOAT was new in the early 90s.
kshacker · 4h ago
I was not commenting on the vintage of this, I was commenting on my ignorance which is where the parent or grandparent comment started from. Sorry if it did not come out right
copilot autocomplete? my experience with it has been very delusional, cursor prediction in cursor(bad naming let's be honest) is simply unmatched
hu3 · 6h ago
Copilot in VSCode has autocomplete and also something they call "next edit".
In my experience, next edit is a significant net positive.
It fixes my typos and predicts next things I want to do in other lines of the same file.
For example, if I fix a variable scope from a loop, it automatically scans for similar mistakes nearby and suggests. Editing multiple array values is also intuitive. It will also learn and suggest formatting prefences and other things such as API changes.
Sure, sometimes it suggests things I don't want but on average it is productive to me.
Martinussen · 6h ago
Cursor also does this.
attentive · 5h ago
Did you change CompletionModel to 'gpt-4o-copilot'? - it may be the default now, provided you keep copilot extension updated.
this is remarkably lowkey for a 1.0 of a big product.
where is the splashy overproduced video? where is the promises of agi? where is the "we are just getting started" / "we cant wait to see what you'll build"? how do i know what to think if you aren't going to tell me what to think?
We tried to make the video lowkey-ish! Appreciate any feedback if came off differently.
thegrim33 · 1h ago
It's an IDE but they don't bother to list on their website what languages you can use the IDE to develop for? I feel like I'm going crazy here. How can they not bother to mention that on their website / marketing?
Vilian · 1h ago
Vs code also don't advertise what language it's for, isn't the same case here?
Salgat · 21m ago
On VS Code's front page it explicitly lists 12 languages and that it "supports almost every major programming language. Several ship in the box, like JavaScript, TypeScript, CSS, and HTML, but extensions for others can be found in the VS Code Marketplace."
ecb_penguin · 32m ago
They support every language through their plugin architecture.
"Cursor works with any programming language. We’ve explicitely worked to improve the performance of our custom models — Tab included — on important but less popular languages like Rust, C++, and CUDA."
You should be good as long as you use one of the mainstream languages: Python, JS, Java, C++, Golang, Ruby, Rust etc
benoittravers · 1h ago
These just look like small updates. Doesn’t warrant a full version upgrade.
smcleod · 4h ago
I don't really see why people still use Cursor over tools like Cline / Roo Code. I'm guessing it's as they clearly have a larger viral marketing department than engineers, as the application itself doesn't perform nearly as well, requires you to have another IDE installed and their subscriptions nerf the models context sizes etc...
jen729w · 1h ago
Remember, not everyone is a pro dev. I'm certainly not: and in that context I find Cursor to be incredibly simple and useful.
It's just like VSC, which I was using, but it has these magical abilities. I was using it within a minute of downloading it. Unlike Cline, I guess, whatever that is.
SparkyMcUnicorn · 20m ago
Install Roo Code (or Cline) in VS Code, plug in your API key(s), and start using them. It's that simple.
demosthanos · 4h ago
Because at some point we have to stop riding the treadmill and just pick a tool and use it to make stuff. Cursor was the first to really arrive at a useful agent mode. It does everything I need it to and more. It's not worth it to me to keep hopping to new tools every time a new one becomes the hyped up hot thing.
Like it or not, we're hitting the slope of enlightenment and some of us are ready to be done with the churn for a while.
smcleod · 2h ago
Cline was agent based from day one and doesn't try to do copilot style tab completes at all. It's been our go to agentic coding app across the majority of our large clients since mid-late 2024. Cursor has been trying to play catch up but has not delivered us the same results.
demosthanos · 1h ago
> doesn't try to do copilot style tab completes at all
Which is another reason why I'll stick with Cursor. Cursor's tab complete can barely be described as Copilot-style, it's nearly a different paradigm and it's what actually got me to pay for it in the first place. I only tried agent mode because it was included with the bundle.
> from day one
July 5, 2024 if going by initial commit. So, yes, technically before Cursor, but Cursor was building agent mode before Cline drew any attention in the mainstream. Cline's first interest on HN dates back to January.
I'll concede that it appears Cline did get to agents first, but it's still a new development in terms of actually drawing interest.
ecb_penguin · 30m ago
> I don't really see why people still use Cursor over tools like Cline / Roo Code
Because we're developers with things to build and we don't have time to play with every AI tool backed by the same LLM.
Glyptodon · 4h ago
Because my company decided to pay for everyone's cursor and I don't have the bandwidth to spend my time constantly evaluating what's better and pitching it?
h2782 · 4h ago
I couldn't possibly disagree with you more that Cline is better than Cursor. Cursor's success isn't because of "a larger viral marketing department"; it's because they made superior software and service.
ramoz · 1h ago
A lot of power in social influence. Especially with the younger generations who remix that influence - compound spread of mindshare. Cursor is all over social media.
BoorishBears · 1h ago
That's silly. Cursor has the best autocomplete experience, period, and some people prefer that to agent-style interactions.
There's still a ton of low hanging fruit that other Copilot-style autocomplete products don't seem to be picking up, like using clipboard contents, identifying the next place in the file to jump to, etc.
I primarily save time coding with AI with autocomplete, followed by chat, with agentic flows a very distant 3rd, so Cursor is a legitimately better product for me.
ramoz · 1h ago
It’s not silly at all. There is a lot of hyper activity going in terms of social influence.
I didn’t say cursor has poor UX.
I tab too. And use agent for cheaper work I don’t care too much about. That said, the best autocomplete is arguably evolving and cursor does not own that.
BoorishBears · 51m ago
Someone said "I don't really see why people still use Cursor over tools like Cline / Roo Code"
And your answer is "A lot of power in social influence.", which is a bit silly when autocomplete is the first form of AI assistance a critical mass of people found intuitive + helpful and Cursor has the best implementation of it... meanwhile Cline/Roo Code don't provide it.
ramoz · 26m ago
You don’t get it - autocomplete is evolving from keyboard clicks to prompts. Tab-ing is not as effective as agentic coding.
Your beloved cursor will go all in on this front, less and less priority on focused cursors in the editor.
mntruell · 4h ago
> their subscriptions nerf the models context sizes etc
You can use the full-context if you prefer that cost/speed tradeoff! Just have to turn on Max Mode.
Cline is great for many users, but a bit of a different product. Lots of Cursor's value come from custom models that run in the background (e.g. Tab, models that gather context, etc.).
nfRfqX5n · 4h ago
the tab feature is really good
helloplanets · 7h ago
Is there a good chart somewhere, to compare all the current AI coding CLIs / IDEs / extensions with one another?
I don't think it's being kept up to date. I believe for the IDEs, it requires manual testing to get the numbers. Since things change so quickly, it's mostly just a historical artifact. Hopefully some future version is automated.
tptacek · 6h ago
I don't think you really want to boil this down to a number; there's a whole lot of feature and workflow differences to capture:
* BYO model or not
* CLI, UI, VSC-plugin or web
* async/sync
* MCP support
* context size
* indexed or live grep-style search
There's probably like 10 more.
wiradikusuma · 3h ago
Never heard SWE-agent until now, and seems to beat Aider (the tool I use) consistently.
Does anyone know if it's GitHub-only or can it be used as a CLI (i.e., Aider replacement)?
mirkodrummer · 6h ago
I keep getting back and forth between Cursor and Zed, but Cursor autocomplete and next cursor prediction are still the best in class between all the competitors, I don't use chats and agents, yet I feel very productive and fast. I sometimes go back to Copilot too just to see how is it going but it has been very delusional so far regarding code suggestions. The only thing I hate about Cursor is the overwrite of some of the shortcuts of vscode, I remapped some and learned some new, and that vim mode plugin is a bit buggy. This and the fact that performance compared with Zed is shit and that's why I go back to Zed sometimes, I'm pondering the idea of just using both and keep them both open
RollingRo11 · 4h ago
> I'm pondering the idea of just using both and keep them both open
Do it. I've started editing with Zed and just keeping Cursor/Intellij open on the side. (Cursor b/c of the the free student plan, Intellij for school assignments).
I feel spoiled by the performance, especially on promotion displays. I've started noticing some dropped frames in Cursor and measured an avg of 45-60 fps in Intellij (which is somewhat expected for such a huge IDE). I basically exclusively write in Zed, and do everything else in their respective apps.
Glyptodon · 4h ago
I think not using agent chat is kind of a missing forest for the trees sort of thing.
That said, I do continue to think that agents are in this weird zone where it's more natural to want to interact through ticketing layer, but you kind of want to editor layer for the final 5%.
asar · 7h ago
Cursor recently lost me as a customer. Too many updates that disturb my workflow and productivity, no easy way to roll back versions, super sparse changelogs, lots of magic in context building, really untransparent pricing on max mode. I recently made the switch to Claude Code on the Max plan and I couldn't be happier. The only real thing I'm missing is the diff view across files, but I assume it's just a matter of time until that's properly implemented in Zed or VSCode.
lopatin · 7h ago
I feel unstoppable with Claude Code Max. I never thought I'd pay $200 per month for any developer tool, yet here we are, and I also couldn't be happier with it.
ed_mercer · 3h ago
Can you elaborate? How is it better than Cursor?
remixff2400 · 51m ago
I just started with it, so still getting my feet wet, but it's been better than any other tool at really grokking my codebase and understanding my intent. The workflow feels better than a strict IDE integration, but it does get pricey really quickly, and you pretty much need at least the $100 Max subscription.
Luckily, it should be coming with the regular $20 Pro subscription in the near future, so it should be easier to demo and get a feel for it without having to jump in all the way.
koakuma-chan · 1h ago
Try it
deadbabe · 6h ago
Would you pay $400?
esafak · 5h ago
Don't give people ideas.
deadbabe · 1h ago
Other professions pay a lot for their tools, and developers are loaded with cash.
1ucky · 7h ago
Since last week it’s possible to use Claude Code in the VSCode terminal where it now automatically installs a plugin to display the diffs.
h2782 · 4h ago
The current max pricing is actually as transparent as it has ever been: It's 20% more to use Max than the APIs directly. I am not sure if your feedback is outdated/based on a previous version of reality?
nojs · 5h ago
> The only real thing I'm missing is the diff view across files
You can commit checkpoints prior to each major prompt and use any IDE’s builtin visual diff versus last commit. Then just rebase when the task is done
punkpeye · 4h ago
Exciting! Particularly the emphasis on smoother MCP integrations.
Will be adding the Add to cursor button to Glama later today today.
If anyone from Cursor is reading this, we are rolling out MCP server usage analytics where we aggregate (anonymous) usage data across several providers. Would be amazing to include Cursor (reach me at frank@glama.ai). The data will be used to help the community discover the most used (and therefore useful) clients and servers.
DidYaWipe · 1h ago
Is what?
philip1209 · 7h ago
There has to be some kind of joke in here about how long it takes people to declare a 1.0 release. "1 million ARR? Not yet. 10 million ARR? Not yet. 300 million ARR? Maybe soon."
esafak · 5h ago
"A million dollars isn't cool. You know what's cool? A billion."
theappsecguy · 5h ago
Cursor had been atrocious. Building on top of an already crappy IDE, you’d hope that they are at least keeping up with VSCode improvements and updates. But they are far behind and instead keep slapping on more garbage.
The agent stuff is largely useless. The tab prediction go nuts every few seconds completely disrupting flow.
Aeolun · 5h ago
Huh, I have the opposite experience. The only tab complete worth anything is in Cursor.
Glyptodon · 4h ago
I agree it has good completes, but it also hallucinates completes for blank lines or in the middle of typing a one word change and that can be irritating. Still better than plain VS code autocomplete though.
girvo · 5h ago
> The tab prediction go nuts every few seconds completely disrupting flow.
This is my main gripe with it, too. It's still been semi-useful at least for some analysis and examination of our code-base, but editing and autocomplete I've not found super useful yet.
ukblewis · 6h ago
I can’t be the only one for which the Python support in Cursor has been absolutely garbage the past week. I’m super disappointed with Cursor. I wanted to love it
And every time I find it having diverged further from VSCode compatibility.
This wouldn’t be so bad if it was an intentional design choice but it seems more that Microsoft is starting to push them out? Like MS Dev Containers plugin is still recommended by some leftover internal, but if you install it you get pushed to a flow that auto uninstalls it and installs Remote Containers by Anysphere (which works differently and lacks support for some features). And I end up rebuilding my Dev Container once more… I also noticed recent extensions such as the Postgres one from MS also doesn’t exist.
I could probably go much lower, and find a model that is dirt cheap but takes a while; but right now the cutting edge (for my own work) is Claude 4 (non-max / non-thinking). To me it feels like Cursor must be hemorrhaging money. The thing that works for me is that I am able to justify those costs working on my own services, that has some customers, and each added feature gives me almost immediate return on investment. But to me it feels like the current rates that cursor charges are not rooted in reality.
Quickly checking Cursor for the past 4 day period:
Requests: 1049
Lines of Agent Edits: 301k
Tabs accepted: 84
Also another issue I am starting to see is the lack of shared MCP servers. If I have VSCode, Cursor, and Claude open, each one is running its own instance of the MCP server. You can imagine that with a dozen or so MCP's, the memory footprint becomes quite large for no reason.
I don’t think the future of agentic software development is in an IDE. Claude Code gives me power to orchestrate - the UX has nothing to do with terminal; it just turns out an agent that lives on the OS and in the filesystem is a powerful thing.
Anthropic can and will evolve Claude Code at a pace cursor cannot evolve IDE abstractions. And then yea - they are designing the perfect wrapper because they are also designing the model.
Long bet is Claude Code becomes more of an OS.
It… sure is something. I’m still thinking about if it’s horrible, visionary, or both.
This is why I think the future may be less about “agents” and more about “intelligent infrastructure”
I don’t want to chat. I want to point a wand and make things happen.
Besides, if you want something inexpensive, using Gemini 2.0 Flash as a backend is completely free. Google provides an API key at no cost.
Why is that?
> Pro ($20/month): Average users can send approximately 45 messages with Claude every 5 hours, OR send approximately 10-40 prompts with Claude Code every 5 hours. [0]
I can probably run through that in 5 minutes.
[0] https://support.anthropic.com/en/articles/11145838-using-cla...
https://github.com/tuananh/hyper-mcp
it's a MCP server with WASM plugin system, packaged, signed & published via OCI registry.
i wrote a MCP with plugin system where you only need to run 1 instance and add plugins via config file.
https://github.com/tuananh/hyper-mcp
The one issue I've run into is that the VSCode version Cursor uses is several months old, so we're stuck using older extensions until they update.
[1] https://open-vsx.org/
This is good user feedback. If Cursor is "Claude + VSCode", why do you need the other 2 open?
STDIO MCP is really just a quick hack.
Its so painful - the model never knows the directory in which it is supposed to be and goes on a wild goose chase of searching in the wrong repo. I have to keep guiding it to the right repo. Anyone here has had success with such a setup?
Keep it short. It's enough for it to realize it needs to navigate directories.
When reviewing the changes made from agent mode, I don’t know why the model made the change or whether the model even made the change versus a tool call making the change. It’s a pain to go fish out the reason from a long response the model gives in the chat.
Example: I recently asked a model to set up shadcn for a project, but while trying to debug why things looked pretty broken, I had sift through a bunch of changes that looked like nasty hallucinations and separate those from actual command line changes that came from shadcn's CLI the model called. Ended up having to just do things the old fashioned way to set things up, reading the fine manual and using my brain (I almost forgot I had one)
It would be nice if above every line of code, there’s a clear indication of whether it came from a model and why the model made the change. Like a code comment, but without littering the code with actual comments
Hand written code needs to be distinguishable and considered at a higher priority for future code generation context
My recommendation to anyone is to use ask the most then tell it to “implement what we discussed” when it looks good.
Hope that helps
In my experience, next edit is a significant net positive.
It fixes my typos and predicts next things I want to do in other lines of the same file.
For example, if I fix a variable scope from a loop, it automatically scans for similar mistakes nearby and suggests. Editing multiple array values is also intuitive. It will also learn and suggest formatting prefences and other things such as API changes.
Sure, sometimes it suggests things I don't want but on average it is productive to me.
where is the splashy overproduced video? where is the promises of agi? where is the "we are just getting started" / "we cant wait to see what you'll build"? how do i know what to think if you aren't going to tell me what to think?
edit: oh haha https://x.com/cursor_ai/status/1930358111677886677
"Cursor works with any programming language. We’ve explicitely worked to improve the performance of our custom models — Tab included — on important but less popular languages like Rust, C++, and CUDA."
Hundreds of languages supported: https://code.visualstudio.com/docs/languages/overview
It's just like VSC, which I was using, but it has these magical abilities. I was using it within a minute of downloading it. Unlike Cline, I guess, whatever that is.
Like it or not, we're hitting the slope of enlightenment and some of us are ready to be done with the churn for a while.
Which is another reason why I'll stick with Cursor. Cursor's tab complete can barely be described as Copilot-style, it's nearly a different paradigm and it's what actually got me to pay for it in the first place. I only tried agent mode because it was included with the bundle.
> from day one
July 5, 2024 if going by initial commit. So, yes, technically before Cursor, but Cursor was building agent mode before Cline drew any attention in the mainstream. Cline's first interest on HN dates back to January.
I'll concede that it appears Cline did get to agents first, but it's still a new development in terms of actually drawing interest.
Because we're developers with things to build and we don't have time to play with every AI tool backed by the same LLM.
There's still a ton of low hanging fruit that other Copilot-style autocomplete products don't seem to be picking up, like using clipboard contents, identifying the next place in the file to jump to, etc.
I primarily save time coding with AI with autocomplete, followed by chat, with agentic flows a very distant 3rd, so Cursor is a legitimately better product for me.
I didn’t say cursor has poor UX.
I tab too. And use agent for cheaper work I don’t care too much about. That said, the best autocomplete is arguably evolving and cursor does not own that.
And your answer is "A lot of power in social influence.", which is a bit silly when autocomplete is the first form of AI assistance a critical mass of people found intuitive + helpful and Cursor has the best implementation of it... meanwhile Cline/Roo Code don't provide it.
Your beloved cursor will go all in on this front, less and less priority on focused cursors in the editor.
You can use the full-context if you prefer that cost/speed tradeoff! Just have to turn on Max Mode.
Cline is great for many users, but a bit of a different product. Lots of Cursor's value come from custom models that run in the background (e.g. Tab, models that gather context, etc.).
* BYO model or not
* CLI, UI, VSC-plugin or web
* async/sync
* MCP support
* context size
* indexed or live grep-style search
There's probably like 10 more.
Does anyone know if it's GitHub-only or can it be used as a CLI (i.e., Aider replacement)?
Do it. I've started editing with Zed and just keeping Cursor/Intellij open on the side. (Cursor b/c of the the free student plan, Intellij for school assignments).
I feel spoiled by the performance, especially on promotion displays. I've started noticing some dropped frames in Cursor and measured an avg of 45-60 fps in Intellij (which is somewhat expected for such a huge IDE). I basically exclusively write in Zed, and do everything else in their respective apps.
That said, I do continue to think that agents are in this weird zone where it's more natural to want to interact through ticketing layer, but you kind of want to editor layer for the final 5%.
Luckily, it should be coming with the regular $20 Pro subscription in the near future, so it should be easier to demo and get a feel for it without having to jump in all the way.
You can commit checkpoints prior to each major prompt and use any IDE’s builtin visual diff versus last commit. Then just rebase when the task is done
Will be adding the Add to cursor button to Glama later today today.
https://glama.ai/mcp/servers
If anyone from Cursor is reading this, we are rolling out MCP server usage analytics where we aggregate (anonymous) usage data across several providers. Would be amazing to include Cursor (reach me at frank@glama.ai). The data will be used to help the community discover the most used (and therefore useful) clients and servers.
The agent stuff is largely useless. The tab prediction go nuts every few seconds completely disrupting flow.
This is my main gripe with it, too. It's still been semi-useful at least for some analysis and examination of our code-base, but editing and autocomplete I've not found super useful yet.