I just ended my Cursor subscription today, and upgraded from Claude's Pro to Max plan to reduce my Claude Code costs. They now include a healthy Claude Code allowance in the Max plans.
If one has already set up Claude Code with metered API use, one toggles between plans using the /login command. Once to start using Max, then whenever one hits a five hour rate limit and wants to keep working.
I've tried many platforms. I kept Cursor long after Windsurf, but Claude Code is a clear winner, as most people report who don't bristle at the cost.
When Cursor or Windsurf forks VS Code, they have a reason. Their chat panes always felt like periscopes; one has better control over Claude Code in a terminal, and this frees up one's choice of editor. I now use Sublime Text, fast and lean.
Signal-boosting this -- I use Claude Code too, and it's beautiful: all the benefits of typing long-form thoughts, ideas, strategy, combined with direct access (for the llm api) to the codebase (no uploading/downloading), and Anthropic's promise of not training on your inputs or outputs.
Cursor is not about vibe coding. Vibe coding means you don't care about the AI's code output as long as it works. Cursor is all about efficiently reviewing the AI-proposed changes and hitting Tab only when you approve them. Much of the editing process is hitting Esc because the proposed change is not good.
I know this is a meta point but I'm pretty sure vibe coding is just an X meme that means whatever the poster intends. I'm not sure you can say vibe coding does or doesn't care about relative quality
Yeah, I'm afraid "vibe coding" is a term that quickly lost its meaning because everyone was using it to mean different things.
Some people use it to mean using AI for writing code in general. I've preferred for it to mean when someone who doesn't know how to code uses AI to write code and doesn't understand the output.
Any IDE based editor feels like a stopgap to me. We may not be there yet, but I feel that in the future a "vibe coder" isn't even going to look at much code at all. Much of what developers who are relying on Cursor, Windmill, Replit, etc etc are doing is performative as it relates to code. There is just a lot of copy/pasting of console errors and asking for things one way or another.
Casual or "vibe" coding is all about the output. Doesn't work? Roll back. Works well? Keep going. Feeling gutsy? Single shot.
Cursor isnt for vibe coding. I use it. I ask the AI to do something I know how to do but it can do it faster. I check the changes to make sure everything looks good.
Vibe coding is just a prototyping tool / "dev influencer" gimmick. No one serious is using Cursor for vibe coding, nor will anyone serious ever vibe code. It's for AI assisted development-- in other words, a more powerful intellisense.
I felt the same way for a while, but I am really not so sure now. Cursor is definitely drawing on the influencer/growth well to drive some portion of these #s.
It's a lot easier and more scaleable to get 1000 people "vibe coding" than it is to get 10 experienced engineers using you for autocomplete.
WARNING for Cursor users:
Cursor is currently stuck using an outdated snapshot of the VSCode Marketplace, meaning several extensions within Cursor remain affected by high-severity CVEs that have already been patched upstream in VSCode. As a result, Cursor users unknowingly remain vulnerable to known security issues. This issue has been acknowledged but remains unresolved: https://github.com/getcursor/cursor/issues/1602#issuecomment...
Given Cursor's rising popularity, users should be aware of this gap in security updates. Until the Cursor team resolves the marketplace sync issue, caution is advised when using certain extensions.
Yeah, this is what I don't get. How is Cursor that much better than Aider? When you have a great, free, open source tool, why would an investor pump money into Cursor?
I'm not sure that Andrej envisioned this when he tweeted out his Vibe Coding take what seems like forever ago now.
I applaud anybody who jumps into Cursor (or other AI Assisted Coding Tools) to build a new product. I think that a way to express ideas is awesome, and allowing for these ideas to materialize is valuable for society and users will determine what is valuable/usable.
However, it's well documented that the expression of these tools is limited. I think that the bet here is that LLMs will continue to get better and better, paving the way for these tools to become more valuable: which I haven't been convinced with yet.
At it's core you can list out the primary functionality of an AI Assisted Coding platform and how these components interact. Their prompts have been dumped, and the tools have been replicated, plus the big LLM providers are in this space as well and understand more nuances around the models and how they interact with the different components.
$9B seems bonkers, but time will tell. There are a few outcomes here: pop, life changes incredibly, or this is the stagnation period that seems to happen with AI/ML. LLMs have changed the way I work already, the question is "what is next". I am hoping that I am ahead of others on the Hype cycle, but only time will tell (from heavy use of AI tools).
> I think that the bet here is that LLMs will continue to get better and better
I don't think so. I think the way we use the same LLMs will continue to get better. Cursor is built on essentially the exact same LLM models as VSCode/Github Copilot, yet Cursor managed to wring a lot more usefulness out of them.
I think it's still early days in understanding how to use LLMs as a foundational technology to build out other products, and improving the models isn't all that necessary. In my view.
I think it's a combination of both, the LLM's today for coding are just average containing a lot of pre-2024 knowledge. The vibe tools are getting around some of the shortcomings and increased token limits which is great, but up to date current knowledge can't rely on llm.txt doc updates as context and expect reasonable code generation. Give me some monthly updated topic related LLM's to use (coding, content writing, history), I don't need the entire world all the time.
Good for them. IMO they should ditch the editor though. I see no reason that they should tie themselves to one editor. It seems like a waste of time. If Claude code let me use me subscription I'd be off cursor pretty quick.
Asking a programmer to ditch their editor is just wild. A lot of this thread seems to be full of non-programmers trying to figure out why Cursor exists lmao
The value for me in cursor is the tab feature not the chat. I use the chat to generate code maybe once per week but the tab autocompleting saves me probably 30 minutes per day.
Are different coding agents better at different languages? Like if I’m trying to write in python vs Golang vs PHP vs C#, am I going to want a different agent for each? Or is one agent going to be more or less consistent among all languages?
Just because (perhaps) a majority of SWEs are using it responsibly as a tool, doesn't mean that a likely fairly wide swath of newcomers aren't jumping on the vibe coding wave.
No I don’t. I just created an account for the first time. I just wanted to point out augment since nobody had mentioned it and it seemed relevant. Roo code is another very good alternative that’s better than cursor but it requires a bit more manual work to get good results. I’ve spent a lot of time trying out all of these tools, augment is the best one IMO but was also the last one I discovered. I figured sharing it here might spare others the searching and frustration I went through with cursor/windsurf/roo/pear/trae and so on.
Augment does not have a model picker. It uses Claude 3.7 right now. The context engine is the magic sauce. It’s miles ahead of all the other tools, almost always gets it right where others fail.
This is similar to the sort of hype which happened with Clubhouse app ($4B valuation, then the users stopped signing up.), Hopin ($9BN until the pandemic ended) and finally Inflection AI ($4B and almost no-one uses it after the initial AI hype).
One of biggest risks is Microsoft who can further lower prices with Copilot (and they can afford to do that for years) for longer and rapidly copy Cursor just like they destroyed Slack with Teams.
There really is no lock-in case for Cursor (unless they acquire something else) as users can easily cancel and switch back to VS Code and Cursor can lose that ARR very quickly and the cost is the entire company.
For Microsoft? Costs them nothing.
This $9B valuation is peak euphoria and this is the best time for Cursor to sell as they are getting very greedy after rejecting a buyout from OpenAI (twice).
+1, they have no most, and we clearly see that big actors are willing to run at loss to get market shares.
At this point I guess people are looking for an exit, and the last investors will own an empty shell
>> I know it works as intended. Just curios why Microsoft decided to enforce this all of a sudden.
Because they can. Also...
>> Did GitHub Copilot just give up on playing fair with Cursor, admitting it's winning?
The game wasn't 'fair' to begin with and it was rigged for Microsoft to win anyway. (Cursor being based on VS Code).
If you're competing against Microsoft, expect them to race you to zero for years (extinguish) at close to no cost for them.
It is all for Cursor to lose if they continue as they are and competitors like Microsoft catch up (and they will do so very quickly whilst lowering prices).
I want to agree, but whatever they are doing is working incredibly well so far. The results I get from cursor outweigh what I can get from Copilot by orders of magnitude. Maybe long term there's no moat, but you'd think if there were no moat now the folks at Microsoft would be able to compete.
This is something that I think Copilot is working on fixing actively. I think that they were focused on Agent mode previously, so put less effort into the autocomplete functionality but they are hearing more and more that this is somewhere they need to focus.
I don't think user base is a moat, at least not for tooling like this. Switching costs is practically zero, as evidence by how quickly Cursor came up and ate Copilot's lunch. Presumably all of Cursor's users switched away from VS Code or another editor. How did that moat go for them?
That’s not a moat when it comes to tooling like this, I can and will cancel my sub when I try something objectively better, there is zero lock in, even if I make a poor choice I can come back.
This is not a moat, the fact there are a bunch of companies hot on their heels proves this too.
Considering how hostile Microsoft is being to forks - even benign ones like VS Codium - it's probably worth it to explore switching to another IDE anyways.
Why are so many people conflating Cursor's main value proposition with "vibe coding"? AI assisted development is not "vibe coding", which is just a silly gimmick to churn out half-assed insecure garbage quickly.
Edit: oh, it's because the article erroneously claims it's a "vibe coding app." Yikes.
But it (at least one of the bubbles) did pop. Tons of companies who got gargantuan raises and valuations in 2020/2021 are essentially in the position where they won't hit those valuations again for years, if ever.
That must be tough for them. But everybody talking about it was referring to "pop" of 2001, when most of the companies didn't just hit lower valuations, they went out of business.
Zero-interest rate policy began in 2008 and ended in 2022. Bubbles can last a long time when borrowing money has been free for the VC class to throw into AI, alongside food delivery, crypto, office rentals and taxicabs under the guise of 'technology'.
California just became the 4th largest economy largely due to tech, and currently ranks 42nd among states in terms of job growth in Q1 2025. The Pop won't be a collapse of capital it'll be in continued decline for the 95% as tech oligarchs continue to facilitate the transfer of wealth to the very top.
The entire workflow for "AI coding agents" boils down to:
1. You write a prompt
2. The agent wraps it in a system prompt and sends it to the LLM
3. The LLM sends back a response
4. The agent performs specific actions based on that response (editing files, creating new ones, etc.)
I don't see why anyone would ditch their current (non-AI) IDE for Cursor just to get this functionality (especially if you're getting hit with a monthly subscription fee on top of it.)
P.S. I maintain a VS Code extension that does the 4 steps above as a baseline[1]
1) Popularity. While there are plenty of die-on-a-hill users for ____ app, there are just as many people who will step away to try something and find they like it. Lots of devs use VScode, but its only been around for 10 years. Some people still swear by Notepad++
2) Demand from on-high: When the non-tech boss shows up and says "Everyone use this now". I don't know how much this happens, but it does happen. Technical dictates from someone who shouldn't be making the decision, probably for a non-technical reason.
3) I hesitate to bring this one up, but here we go: People don't know any better. There is a new generation of developers coming up who are leaning hard into vibe coding. And just when I was young, there are plenty of seasoned developers crying out about it's validity. The new generation will pick their own tools - in part to distance themselves from the current generation.
Cursor has consistently felt faster and easier to use with better inline auto-complete and faster large edits in chat than VSCode ever did. The way suggestions and chat is shown is just a bit easier to read and more elegantly presented.
After trying a lot of them, the need for a real ide is even stronger. Every single tool I've tried creates bugs, unrequired code, mistakes, hallucinations. Currently playing with Junie and Augment, and CoPilot holds up surprisingly well, not sure why people are so eager to ditch it.
I've been very happy with VSCode + Gemini 2.5 (a recentish integration). I will re-evaluate Cursor again but I can't imagine they're going to be able to beat Microsoft to the punch.
Why do people pay for water bottles when it comes out of the sink for free? Why do people use Dropbox when you could just mount a filesystem with curlftpfs? Why do people pay for Docusign or Postman or Duolingo?
I doubt it. Cursor is an IDE with an ai co-pilot deeply integrated into it. They've literally changed the paradigm of software development by making ai-assisted coding feasible. The vibe-coding mention is reductive imo.
Another way of looking at it:
Maker of "pricey electronic typewriter", Apple hits $9B valuation (FT 1984)
I mean I tried C# integration and Cursor does not even fix all compilation errors before reporting it has "completed the task". Feels like that's the most basic integration you can have beyond reviewing diffs.
That's a fair question and point. I'm not die-hard Cursor fan. Use what works best for you, but I'm just more so commenting that the vibe-coding part totally minimizes what the offering is.
This is like saying cars are for doing skids and speeding illegally, then telling the person who drives for a living if they “don’t have a very inflated idea about what a car is”.
Vibe coding is letting the AI take the wheel for every decision, not verifying output, progress above all. Of course it’s possible to use it in a more subtle collaborative capacity with heavy oversight.
On the off chance there was any doubt still... Alas, seeing a bubble provides almost no good information except that you know something bad will happen sometime.
Except that there is one problem and that is Microsoft GitHub who has a much more massive platform (220m+ users), distribution and ecosystem lock-in which Cursor does not have.
Yeah, good things companies only use the bare-bones git functionality of GitHub /s. Forget things like GitHub Actions, Issues, Projects, complicated security settings, etc. etc.
I've been at a company that migrated from GitHub to GitLab and it was a substantial undertaking, and the company was a very small new startup - it would have been many orders of magnitude more difficult for a larger company with multiple dev teams to move.
I am looking from perspective of vibe coding task for some regular dev. If he has access to github from his current IDE, nothing prevents Cursor from having the same level of access.
It’s how a lot of code is being generated by owners of small startups with minimal engineering engagement. PMs are producing full walking skeletons of designs that used to take a week to polish to that fidelity.
Sure--but it's also the most widely used IDE for integrated AI assistance to normal software engineers. It's a "vibe coding" app in the same way that a washing machine is a "sock cleaning machine." I mean, yes, it does that, but that's a small part of its designed and in-practice usage.
If you're using Cursor (as I do), but find it tends to struggle once your project reaches a certain size, and you're ok with something CLI-based, you might find the open source CLI coding agent I've built interesting: https://github.com/plandex-ai/plandex
- It can handle up to 2M tokens of context directly, and can index/work with/chat with projects up to 20M tokens (1M+ lines). Here's an example of chatting with with SQLite codebase to learn about how transactions are implemented: https://plandex.ai/_next/static/media/plandex-sqlite.0ee6cb2...
- All changes are committed to a version-controlled sandbox by default, preventing the problem of stray changes that you don't notice being left behind in the project.
- Being terminal-based allows for more seamless and powerful execution control and automated debugging. Here's an example of automatically debugging a browser app (via redirection of console logs/errors to the terminal): https://www.youtube.com/watch?v=g-_76U_nK0Y&embeds_referring...
I've used it. Honestly it's the same place as goose and aider ... it tries to DIY its own solutions too often as opposed to using existing robust libraries, writes inflexible code where simple piped commands would do, and basically Katamari Damacies the problem into a complicated sprawl that doesn't work.
I think the reason you tend to see these same kinds of problems across tools is that you are running up against the reasoning limitations of the underlying models. I see my goal as trying to push that horizon out as far possible, and I think Plandex pushes it significantly farther than most other tools, but you do of course still run into those limitations.
That said, I think even setting aside improvements to the underlying models, there's a lot of potential to improve on these issues. I think they're all basically addressable at current capability levels, though they are difficult problems. It's what I'm personally most excited to work on.
Honestly, I think it simply becomes the wrong modality at a certain level of engagement with any problem.
When human developing a project successfully you engage in different collective behaviors throughout the process. That's why the scrum/agile stuff that tries to normalize it doesn't actually work very well.
On computers, relationships are workflows and you need to do a dance of fluid relationships to use AI effectively throughout the execution process - otherwise that's why you either abandon it after you get to some point or you feel like you're just wasting time.
If you don't see any similarities, I'm not really sure what to say. I built it specifically to improve the workflow of what has come to be labelled 'vibe coding', and I believe it's more effective for this purpose than Cursor... though I find that Cursor is excellent for IDE auto-completion.
It's MIT-licensed and can be used for free, so yes I do mention it when relevant conversations come up, because I think can be useful to people, and I think that is well within the spirit of HN, which is supposed to be a maker community. Since I'm an HN addict, I read/post a lot, and so I notice when these topics come up.
https://archive.ph/yQlLu
I just ended my Cursor subscription today, and upgraded from Claude's Pro to Max plan to reduce my Claude Code costs. They now include a healthy Claude Code allowance in the Max plans.
If one has already set up Claude Code with metered API use, one toggles between plans using the /login command. Once to start using Max, then whenever one hits a five hour rate limit and wants to keep working.
I've tried many platforms. I kept Cursor long after Windsurf, but Claude Code is a clear winner, as most people report who don't bristle at the cost.
When Cursor or Windsurf forks VS Code, they have a reason. Their chat panes always felt like periscopes; one has better control over Claude Code in a terminal, and this frees up one's choice of editor. I now use Sublime Text, fast and lean.
Do you also use aider? How does it compare to Claude code? The concept seems to be similar, a stand-alone cli agent.
Signal-boosting this -- I use Claude Code too, and it's beautiful: all the benefits of typing long-form thoughts, ideas, strategy, combined with direct access (for the llm api) to the codebase (no uploading/downloading), and Anthropic's promise of not training on your inputs or outputs.
Codex with o3 worked much better for my projects than Claude code. Plus it’s open source so I can switch models easily.
Claude code is actually great. I get to use any tools/editor I want, and get the AI agent workflow.
Right, it decouples your IDE choice from the "Code Assistant" choice, giving you massive flexibility.
How does this compare with using openrouter + OSS? I'm currently averaging $20/month with copious usage (and no limits, obviously).
Cursor is not about vibe coding. Vibe coding means you don't care about the AI's code output as long as it works. Cursor is all about efficiently reviewing the AI-proposed changes and hitting Tab only when you approve them. Much of the editing process is hitting Esc because the proposed change is not good.
I know this is a meta point but I'm pretty sure vibe coding is just an X meme that means whatever the poster intends. I'm not sure you can say vibe coding does or doesn't care about relative quality
Yeah, I'm afraid "vibe coding" is a term that quickly lost its meaning because everyone was using it to mean different things.
Some people use it to mean using AI for writing code in general. I've preferred for it to mean when someone who doesn't know how to code uses AI to write code and doesn't understand the output.
Any IDE based editor feels like a stopgap to me. We may not be there yet, but I feel that in the future a "vibe coder" isn't even going to look at much code at all. Much of what developers who are relying on Cursor, Windmill, Replit, etc etc are doing is performative as it relates to code. There is just a lot of copy/pasting of console errors and asking for things one way or another.
Casual or "vibe" coding is all about the output. Doesn't work? Roll back. Works well? Keep going. Feeling gutsy? Single shot.
Cursor isnt for vibe coding. I use it. I ask the AI to do something I know how to do but it can do it faster. I check the changes to make sure everything looks good.
Vibe coding is just a prototyping tool / "dev influencer" gimmick. No one serious is using Cursor for vibe coding, nor will anyone serious ever vibe code. It's for AI assisted development-- in other words, a more powerful intellisense.
I felt the same way for a while, but I am really not so sure now. Cursor is definitely drawing on the influencer/growth well to drive some portion of these #s.
It's a lot easier and more scaleable to get 1000 people "vibe coding" than it is to get 10 experienced engineers using you for autocomplete.
The people who outsource their jobs to mechanical Turks will be using it until their employer(s) find out.
Have you tried Bolt.new? You can vibe code with it.
WARNING for Cursor users: Cursor is currently stuck using an outdated snapshot of the VSCode Marketplace, meaning several extensions within Cursor remain affected by high-severity CVEs that have already been patched upstream in VSCode. As a result, Cursor users unknowingly remain vulnerable to known security issues. This issue has been acknowledged but remains unresolved: https://github.com/getcursor/cursor/issues/1602#issuecomment...
Given Cursor's rising popularity, users should be aware of this gap in security updates. Until the Cursor team resolves the marketplace sync issue, caution is advised when using certain extensions.
I've flagged it here, apologies for the repost: https://news.ycombinator.com/item?id=43609572
Is there a list of extensions to avoid?
Cursor may be $9B but Aider is priceless
Yeah, this is what I don't get. How is Cursor that much better than Aider? When you have a great, free, open source tool, why would an investor pump money into Cursor?
EE monies?
I'm not sure that Andrej envisioned this when he tweeted out his Vibe Coding take what seems like forever ago now.
I applaud anybody who jumps into Cursor (or other AI Assisted Coding Tools) to build a new product. I think that a way to express ideas is awesome, and allowing for these ideas to materialize is valuable for society and users will determine what is valuable/usable.
However, it's well documented that the expression of these tools is limited. I think that the bet here is that LLMs will continue to get better and better, paving the way for these tools to become more valuable: which I haven't been convinced with yet.
At it's core you can list out the primary functionality of an AI Assisted Coding platform and how these components interact. Their prompts have been dumped, and the tools have been replicated, plus the big LLM providers are in this space as well and understand more nuances around the models and how they interact with the different components.
$9B seems bonkers, but time will tell. There are a few outcomes here: pop, life changes incredibly, or this is the stagnation period that seems to happen with AI/ML. LLMs have changed the way I work already, the question is "what is next". I am hoping that I am ahead of others on the Hype cycle, but only time will tell (from heavy use of AI tools).
> I think that the bet here is that LLMs will continue to get better and better
I don't think so. I think the way we use the same LLMs will continue to get better. Cursor is built on essentially the exact same LLM models as VSCode/Github Copilot, yet Cursor managed to wring a lot more usefulness out of them.
I think it's still early days in understanding how to use LLMs as a foundational technology to build out other products, and improving the models isn't all that necessary. In my view.
I think it's a combination of both, the LLM's today for coding are just average containing a lot of pre-2024 knowledge. The vibe tools are getting around some of the shortcomings and increased token limits which is great, but up to date current knowledge can't rely on llm.txt doc updates as context and expect reasonable code generation. Give me some monthly updated topic related LLM's to use (coding, content writing, history), I don't need the entire world all the time.
I use both and I'm amazed at how poor the VSCode proposal is compared to Cursor.
Didn’t they reject $20B from OpenAI? This seems like a 50% haircut.
Good for them. IMO they should ditch the editor though. I see no reason that they should tie themselves to one editor. It seems like a waste of time. If Claude code let me use me subscription I'd be off cursor pretty quick.
Please do not ditch the editor! Cursor's integrated IDE with easy changes I can read is the whole reason I like it the best!
Claude Max can be used with Claude Code
Yea just learned that today but it's quite expensive imo.
Cursor editor autocomplete (cursor tab) is the one feature I use constantly. Everything else is more selective for my type of work.
Asking a programmer to ditch their editor is just wild. A lot of this thread seems to be full of non-programmers trying to figure out why Cursor exists lmao
easier sell/setup
> Makes it an easier sell to devs, "you don't have to 'change' editors" and it makes setup easier if you happen to be using VSC
But tying it to an editor (including VSCode) means "you have to change editors".
I don't use VSCode, so any solution requiring it is a no-go.
When we have aider[1], which works with any editor/IDE, I just don't see the value in trying Cursor, et al.
[1] https://aider.chat/
The value for me in cursor is the tab feature not the chat. I use the chat to generate code maybe once per week but the tab autocompleting saves me probably 30 minutes per day.
Are different coding agents better at different languages? Like if I’m trying to write in python vs Golang vs PHP vs C#, am I going to want a different agent for each? Or is one agent going to be more or less consistent among all languages?
really, prompting and workflow style is 80%.
AI is like any other program, good output can't come from bad input.
Calling Cursor a vibe coding app is just wild. The journalists and influencers are so out of touch with the software engineering world.
There's plenty of people using Cursor this way though:
https://youtu.be/opB25teOxYQ
Just because (perhaps) a majority of SWEs are using it responsibly as a tool, doesn't mean that a likely fairly wide swath of newcomers aren't jumping on the vibe coding wave.
Does this fundraising only make sense if cursor is planning to jump into the infrastructure side too? To do application hosting layer like Replit.
$200M ARR so that’s a 45x multiple. Wow.
Why not go public?
Because there's plenty of private equity lined up and dumb enough to buy at a 45x multiple.
Cursor isn’t very good. It aggressively limits context I assume to save money. Augment runs circles around it.
Do you have any relationship to/with Augment?
No I don’t. I just created an account for the first time. I just wanted to point out augment since nobody had mentioned it and it seemed relevant. Roo code is another very good alternative that’s better than cursor but it requires a bit more manual work to get good results. I’ve spent a lot of time trying out all of these tools, augment is the best one IMO but was also the last one I discovered. I figured sharing it here might spare others the searching and frustration I went through with cursor/windsurf/roo/pear/trae and so on.
His only posts are promoting it
Which models are available on Augment right now? Which do you use?
https://www.augmentcode.com/blog/ai-model-pickers-are-a-desi...
Augment does not have a model picker. It uses Claude 3.7 right now. The context engine is the magic sauce. It’s miles ahead of all the other tools, almost always gets it right where others fail.
Also noticed it. It often forgets recent changes then proceeds to revert them without being asked to touch that part of code.
This is similar to the sort of hype which happened with Clubhouse app ($4B valuation, then the users stopped signing up.), Hopin ($9BN until the pandemic ended) and finally Inflection AI ($4B and almost no-one uses it after the initial AI hype).
One of biggest risks is Microsoft who can further lower prices with Copilot (and they can afford to do that for years) for longer and rapidly copy Cursor just like they destroyed Slack with Teams.
There really is no lock-in case for Cursor (unless they acquire something else) as users can easily cancel and switch back to VS Code and Cursor can lose that ARR very quickly and the cost is the entire company.
For Microsoft? Costs them nothing.
This $9B valuation is peak euphoria and this is the best time for Cursor to sell as they are getting very greedy after rejecting a buyout from OpenAI (twice).
+1, they have no most, and we clearly see that big actors are willing to run at loss to get market shares. At this point I guess people are looking for an exit, and the last investors will own an empty shell
Well said. There are so many AI assist tools (windsurf, cline/roocode, trae, etc) and it’s so easy to switch between them.
This is crack smoking. This company is going to get wiped out by GitHub Copilot in short order.
Currently, GH co-pilot sucks. At least the plan we have at work...
And/or wiped out by VS Code plugins like Cline and Kilo which are free.
Probably not — I know of companies dropping GH Copilot because of low internal usage compared to cursor.
it's vscode with chatgpt
Quite a valuation for a VSCode fork. They have no moat.
There are already VS Code extensions (Cline, Roo Code and Kilo Code) that do the same + are much better IMO.
Disclaimer: I'm one of the maintainers of Kilo.
It's a rented castle, too.
Indeed.
"DID YOU JUST BAN CURSOR?" https://github.com/microsoft/vscode-dotnettools/issues/1909
> The error seems very clear to me. Dev Kit is licensed only for use with VS Code, Cursor is not VS Code, ergo it is not licensed to use it.
>
> Not to mention that only VS Code can use the official plugin registry in the first place. Everything working as intended.
Exactly. From: [0]
>> I know it works as intended. Just curios why Microsoft decided to enforce this all of a sudden.
Because they can. Also...
>> Did GitHub Copilot just give up on playing fair with Cursor, admitting it's winning?
The game wasn't 'fair' to begin with and it was rigged for Microsoft to win anyway. (Cursor being based on VS Code).
If you're competing against Microsoft, expect them to race you to zero for years (extinguish) at close to no cost for them.
It is all for Cursor to lose if they continue as they are and competitors like Microsoft catch up (and they will do so very quickly whilst lowering prices).
[0] https://github.com/microsoft/vscode-dotnettools/issues/1909#...
Ah, good old Microsoft.
More like a squatted castle.
I want to agree, but whatever they are doing is working incredibly well so far. The results I get from cursor outweigh what I can get from Copilot by orders of magnitude. Maybe long term there's no moat, but you'd think if there were no moat now the folks at Microsoft would be able to compete.
This is something that I think Copilot is working on fixing actively. I think that they were focused on Agent mode previously, so put less effort into the autocomplete functionality but they are hearing more and more that this is somewhere they need to focus.
Their user base is the moat. Same as any network effect business.
If all your peers are using a thing, it's really hard to convince entire industry to switch even if something better is available.
I don't think user base is a moat, at least not for tooling like this. Switching costs is practically zero, as evidence by how quickly Cursor came up and ate Copilot's lunch. Presumably all of Cursor's users switched away from VS Code or another editor. How did that moat go for them?
That’s not a moat when it comes to tooling like this, I can and will cancel my sub when I try something objectively better, there is zero lock in, even if I make a poor choice I can come back.
This is not a moat, the fact there are a bunch of companies hot on their heels proves this too.
Early adopters are the most fickle version of a user moat I can imagine.
We're talking about an IDE, switching is pretty frictionless.
Well, switching IDEs can actually be hard due to all the integrations. But this is VScode to VSCode …
Network effect? What network?
No need to change your IDE to AI/vibe code.
There are already a few good VS Code extensions like Cline and Kilo Code which do 80/20 of the job.
the proper timeline is
cline -> roo -> kilo
There's also things like goose, plandex, and aider.
The real problem is what to with all those bugs written by the AI, however you choose to vibe them.
That's what my latest effort, https://github.com/kristopolous/llmehelp is trying to address.
Using Cursor is a lot easier than Roo Code or Cline. It has a much better UI and it's faster for streaming tool uses, and I believe it's cheaper.
Considering how hostile Microsoft is being to forks - even benign ones like VS Codium - it's probably worth it to explore switching to another IDE anyways.
Even Copilot has caught up with its chat and agent editing features.
Why are so many people conflating Cursor's main value proposition with "vibe coding"? AI assisted development is not "vibe coding", which is just a silly gimmick to churn out half-assed insecure garbage quickly.
Edit: oh, it's because the article erroneously claims it's a "vibe coding app." Yikes.
I have a paint-rapido pointed at a keywords list with JDVance in a dumpster at the center.
Kilo is not "Vibe" coding, nor does it need to steal "Intellectual" "property" at the end of it's 5 hours shift.
I am about done with VSCode and it's claims of usefulness. I started long before the internet and PCs.
It may come back to be relevant "Here Comes Another Bubble v1.1 - The Richter Scales" but some lingo needs update (but other parts are still spot on).
https://www.youtube.com/watch?v=I6IQ_FOCE6I
I've been on HN since around 2010 (I think) and I have heard "we're in a bubble and it's going to pop" this whole time.
But it (at least one of the bubbles) did pop. Tons of companies who got gargantuan raises and valuations in 2020/2021 are essentially in the position where they won't hit those valuations again for years, if ever.
That must be tough for them. But everybody talking about it was referring to "pop" of 2001, when most of the companies didn't just hit lower valuations, they went out of business.
Zero-interest rate policy began in 2008 and ended in 2022. Bubbles can last a long time when borrowing money has been free for the VC class to throw into AI, alongside food delivery, crypto, office rentals and taxicabs under the guise of 'technology'.
California just became the 4th largest economy largely due to tech, and currently ranks 42nd among states in terms of job growth in Q1 2025. The Pop won't be a collapse of capital it'll be in continued decline for the 95% as tech oligarchs continue to facilitate the transfer of wealth to the very top.
[1]: https://www.mercurynews.com/2025/05/03/economy-jobs-layoff-w...
Why do we need a new IDE for "AI/vibe coding"?
The entire workflow for "AI coding agents" boils down to:
1. You write a prompt
2. The agent wraps it in a system prompt and sends it to the LLM
3. The LLM sends back a response
4. The agent performs specific actions based on that response (editing files, creating new ones, etc.)
I don't see why anyone would ditch their current (non-AI) IDE for Cursor just to get this functionality (especially if you're getting hit with a monthly subscription fee on top of it.)
P.S. I maintain a VS Code extension that does the 4 steps above as a baseline[1]
[1] https://github.com/Kilo-Org/kilocode
It’s vibes investing to match the vibes coding
I think there are 2 or 3 reasons:
1) Popularity. While there are plenty of die-on-a-hill users for ____ app, there are just as many people who will step away to try something and find they like it. Lots of devs use VScode, but its only been around for 10 years. Some people still swear by Notepad++
2) Demand from on-high: When the non-tech boss shows up and says "Everyone use this now". I don't know how much this happens, but it does happen. Technical dictates from someone who shouldn't be making the decision, probably for a non-technical reason.
3) I hesitate to bring this one up, but here we go: People don't know any better. There is a new generation of developers coming up who are leaning hard into vibe coding. And just when I was young, there are plenty of seasoned developers crying out about it's validity. The new generation will pick their own tools - in part to distance themselves from the current generation.
> P.S. I maintain a VS Code extension that does the 4 steps above as a baseline[1]
Still needs an IDE!
Why should I use kilo instead of aider?
https://aider.chat/
I'm "old school", I do these steps manually.
(It actually helps more, IMO).
Honestly main reason is the UI and speed.
Cursor has consistently felt faster and easier to use with better inline auto-complete and faster large edits in chat than VSCode ever did. The way suggestions and chat is shown is just a bit easier to read and more elegantly presented.
These things make a big difference.
After trying a lot of them, the need for a real ide is even stronger. Every single tool I've tried creates bugs, unrequired code, mistakes, hallucinations. Currently playing with Junie and Augment, and CoPilot holds up surprisingly well, not sure why people are so eager to ditch it.
I've been very happy with VSCode + Gemini 2.5 (a recentish integration). I will re-evaluate Cursor again but I can't imagine they're going to be able to beat Microsoft to the punch.
There is literally no need for that. Roo or Kilo (extensions for VSCode) are open-source multi-agent plugins that don't require a whole new IDE.
The flat monthly fee charged by Cursor is attractive to many businesses and individuals.
Why do people pay for water bottles when it comes out of the sink for free? Why do people use Dropbox when you could just mount a filesystem with curlftpfs? Why do people pay for Docusign or Postman or Duolingo?
Wow, AI 'vibe coding' app? What a reductive statement on what Cursor is.
Seems about right to me
Are you sure you don't have a very inflated idea about what Cursor is?
I doubt it. Cursor is an IDE with an ai co-pilot deeply integrated into it. They've literally changed the paradigm of software development by making ai-assisted coding feasible. The vibe-coding mention is reductive imo.
Another way of looking at it: Maker of "pricey electronic typewriter", Apple hits $9B valuation (FT 1984)
pretty reductive
Are they different from Copilot though?
I mean I tried C# integration and Cursor does not even fix all compilation errors before reporting it has "completed the task". Feels like that's the most basic integration you can have beyond reviewing diffs.
That's a fair question and point. I'm not die-hard Cursor fan. Use what works best for you, but I'm just more so commenting that the vibe-coding part totally minimizes what the offering is.
This is like saying cars are for doing skids and speeding illegally, then telling the person who drives for a living if they “don’t have a very inflated idea about what a car is”.
Vibe coding is letting the AI take the wheel for every decision, not verifying output, progress above all. Of course it’s possible to use it in a more subtle collaborative capacity with heavy oversight.
9B, no moat, it’s bubble time!
On the off chance there was any doubt still... Alas, seeing a bubble provides almost no good information except that you know something bad will happen sometime.
That sometime will be when reality sinks in regarding revenue, but by then some of the original investors have made their money and split.
they have userbase, revenue, super-fast growth, leading market position and capable team.
Yup, but good unit economics are what matters in the long run.
Except that there is one problem and that is Microsoft GitHub who has a much more massive platform (220m+ users), distribution and ecosystem lock-in which Cursor does not have.
Things can change very quickly in 6 - 12 months.
there is no lock in on github, since communication happens through git protocol, so no external tool/client can be locked out.
Yeah, good things companies only use the bare-bones git functionality of GitHub /s. Forget things like GitHub Actions, Issues, Projects, complicated security settings, etc. etc.
I've been at a company that migrated from GitHub to GitLab and it was a substantial undertaking, and the company was a very small new startup - it would have been many orders of magnitude more difficult for a larger company with multiple dev teams to move.
I am looking from perspective of vibe coding task for some regular dev. If he has access to github from his current IDE, nothing prevents Cursor from having the same level of access.
Better sell fast, then enjoy life.
I mean... it's a little disingenuous to call it a "vibe coding" app.
It’s how a lot of code is being generated by owners of small startups with minimal engineering engagement. PMs are producing full walking skeletons of designs that used to take a week to polish to that fidelity.
Sure--but it's also the most widely used IDE for integrated AI assistance to normal software engineers. It's a "vibe coding" app in the same way that a washing machine is a "sock cleaning machine." I mean, yes, it does that, but that's a small part of its designed and in-practice usage.
> but it's also the most widely used IDE for integrated AI assistance to normal software engineers.
Is it? I'd be surprised if GitHub Copilot didn't have more paid users.
If you're using Cursor (as I do), but find it tends to struggle once your project reaches a certain size, and you're ok with something CLI-based, you might find the open source CLI coding agent I've built interesting: https://github.com/plandex-ai/plandex
- It can handle up to 2M tokens of context directly, and can index/work with/chat with projects up to 20M tokens (1M+ lines). Here's an example of chatting with with SQLite codebase to learn about how transactions are implemented: https://plandex.ai/_next/static/media/plandex-sqlite.0ee6cb2...
- All changes are committed to a version-controlled sandbox by default, preventing the problem of stray changes that you don't notice being left behind in the project.
- Being terminal-based allows for more seamless and powerful execution control and automated debugging. Here's an example of automatically debugging a browser app (via redirection of console logs/errors to the terminal): https://www.youtube.com/watch?v=g-_76U_nK0Y&embeds_referring...
I've used it. Honestly it's the same place as goose and aider ... it tries to DIY its own solutions too often as opposed to using existing robust libraries, writes inflexible code where simple piped commands would do, and basically Katamari Damacies the problem into a complicated sprawl that doesn't work.
Thanks, I appreciate the feedback.
I think the reason you tend to see these same kinds of problems across tools is that you are running up against the reasoning limitations of the underlying models. I see my goal as trying to push that horizon out as far possible, and I think Plandex pushes it significantly farther than most other tools, but you do of course still run into those limitations.
That said, I think even setting aside improvements to the underlying models, there's a lot of potential to improve on these issues. I think they're all basically addressable at current capability levels, though they are difficult problems. It's what I'm personally most excited to work on.
Honestly, I think it simply becomes the wrong modality at a certain level of engagement with any problem.
When human developing a project successfully you engage in different collective behaviors throughout the process. That's why the scrum/agile stuff that tries to normalize it doesn't actually work very well.
On computers, relationships are workflows and you need to do a dance of fluid relationships to use AI effectively throughout the execution process - otherwise that's why you either abandon it after you get to some point or you feel like you're just wasting time.
You're right. I think the solution essentially boils down to modeling those other modalities and incorporating them into the agent.
I see too much self promotion over this tool all over HN and it is not anywhere near the same as Cursor?
I don't see any similarities other than it uses AI and that is about it.
If you don't see any similarities, I'm not really sure what to say. I built it specifically to improve the workflow of what has come to be labelled 'vibe coding', and I believe it's more effective for this purpose than Cursor... though I find that Cursor is excellent for IDE auto-completion.
It's MIT-licensed and can be used for free, so yes I do mention it when relevant conversations come up, because I think can be useful to people, and I think that is well within the spirit of HN, which is supposed to be a maker community. Since I'm an HN addict, I read/post a lot, and so I notice when these topics come up.