IMO other than the Microsoft IP issue, I think the biggest thing that has shifted since this acquisition was first in the works is Claude Code has absolutely exploded. Forking an IDE and all the expense that comes with that feels like a waste of effort, considering the number of free/open source CLI agentic tools that are out there.
Let's review the current state of things:
- Terminal CLI agents are several orders of magnitude less $$$ to develop than forking an entire IDE.
- CC is dead simple to onboard (use whatever IDE you're using now, with a simple extension for some UX improvements).
- Anthropic is free to aggressively undercut their own API margins (and middlemen like Cursor) in exchange for more predictable subscription revenue + training data access.
What does Cursor/Windsurf offer over VS Code + CC?
- Tab completion model (Cursor's remaining moat)
- Some UI niceties like "add selection to chat", and etc.
Personally I think this is a harbinger of where things are going. Cursor was fastest to $900M ARR and IMO will be fastest back down again.
Agreed on everything. Just to add, not only anthropic is offering CC at like a 500% loss, they restricted sonnet/opus 4 access to windsurf, and jacked up their enterprise deal to Cursor. The increase in price was so big that it forced cursor to make that disastrous downgrade to their plans.
I think only way Cursor and other UX wrappers still win is if on device models or at least open source models catch up in the next 2 years. Then i can see a big push for UX if models are truly a commodity. But as long as claude is much better then yes they hold all the cards. (And don't have a bigger company to have a civil war with like openai)
Seems like the survival strategy for cursor would be to develop their own frontier coding model. Maybe they can leverage the data from their still somewhat significant lead in the space to make a solid effort.
- Forking VSCode is very easy; you can do it in 1 hour.
- Anthropic doesn't use the inputs for training.
- Cursor doesn't have $900M ARR. That was the raise. Their ARR is ~$500m [1].
- Claude Code already support the niceties, including "add selection to chat", accessing IDE's realtime warnings and errors (built-in tool 'ideDiagnostics'), and using IDE's native diff viewer for reviewing the edits.
Their base is $20/mth. That would equal 3.75M people paying a sub to Cursor.
If literally everyone is on their $200/mth plan, then that would be 375K paid users.
There’s 50M VS Code + VS users (May 2025). [1] 7% of all VS Code users having switched to Cursor does not match my personal circle of developers. 0.7% . . . Maybe? But, that would be if everyone using Cursor were paying $200/month.
Seems impossibly high, especially given the number of other AI subscription options as well.
The $20/month cursor sub is heavily limited though, for basic casual usage that's fine but you VERY soon run into its limits when working at any speed.
I also just prefer CC's UX. I've tried to make myself use Copilot and Roo and I just couldn't. The extra mental overhead and UI context-switching took me out of the flow. And tab completion has never felt valuable to me.
But the chat UX is so simple it doesn't take up any extra brain-cycles. It's easier to alt-tab to and from; it feels like slacking a coworker. I can have one or more terminal windows open with agents I'm managing, and still monitor/intervene in my editor as they work. Fits much nicer with my brain, and accelerates my flow instead of disrupting it
There's something starkly different for me about not having to think about exactly what context to feed to the tool, which text to highlight or tabs to open, which predefined agent to select, which IDE button to press
Just formulate my concepts and intent and then express those in words. If I need to be more precise in my words then I will be, but I stay in a concepts + words headspace. That's very important for conserving my own mental context window
Claude Code is just proving that coding agents can be successful. The interface isn’t magic, it just fits the model and integrates with a system in all the right ways. The Anthropic team for that product is very small comparatively (their most prolific contributor is Claude), and I think it’s more of a technology proof than a core competency - it’s a great API $ business lever, but there’s no reason for them to try and win the “agentic coding UI” market. Unless Generative AI flops everywhere else, these markets will continue to emerge and need focus. The Windsurf kerfuffle is further proof that OpenAI doesn’t see the market as must-win for a frontier model shop.
And so I’d say this isn’t a harbinger of the death of Cursor, instead proof that there’s a future in the market they were just recently winning.
I was being hyperbolic saying their ARR will go to zero. That's obviously not the case, but the point is that CC has revealed their real product was not "agentic coding UI", it was "insanely cheap tokens". I have no doubt they will continue to see success, but their future right now looks closer to being a competitor to free/open tools like cline/roo code, as well as the CLI entrants, not a standalone $500M ARR juggarnaut. They have no horse in the race in the token market, they're a middleman.
They either need to create their own model and compete on cost, or hope that token costs come down dramatically so as to be too cheap to meter.
Digging in here more... why would you say it isn't in Anthropic's interest to win the "agentic coding UI" market?
My mental model is that these foundation model companies will need to invest in and win in a significant number of the app layer markets in order to realize enough revenue to drive returns. And if coding / agentic coding is one of the top X use cases for tokens at the app layer, seems logical that they'd want to be a winner in this market.
Is your view that these companies will be content to win at the model layer and be agnostic as to the app layer?
Cursor see it coming - it's why they're moving to the web and mobile[0]
The bigger issue is the advantage Anthropic, Google and OpenAI have in developing and deploying their own models. It wasn't that long ago that Cursor was reading 50 lines of code at a time to save on token costs. Anthropic just came out and yolo'd the context window because they could afford to, and it blew everything else away.
Cursor could release a cli tomorrow but it wouldn't help them compete when Anthropic and Google can always be multiples cheaper
> Anthropic just came out and yolo'd the context window because they could afford to
I don’t think this is true at all. The reason CC is so good is that they’re very deliberate about what goes in the context. CC often spends ages reading 5 LOC snippets, but afterwards it only has relevant stuff in context.
I think this is an interesting and cool direction for Cursor to be going in and I don't doubt something like this is the future. But I have my doubts whether it will save them in the short/medium term:
- AI is not good enough yet to abandon the traditional IDE experierence if you're doing anything non-trivial. Hard finding use cases for this right now.
- There's no moat here. There are already a dozen "Claude Code UI" OSS projects with similar basic functionality.
Strictly speaking about large, complex, sprawling codebases, I don't think you can beat the experience that an IDE + coding agent brings with a terminal-based coding agent.
Auto-regressive nature of these things mean that errors accumulate, and IDEs are well placed to give that observability to the human, than a coding agent. I can course correct more easily in an IDE with clear diffs, coding navigation, than following a terminal timeline.
You can view and navigate the diffs made by the terminal agent in your IDE in realtime, just like Cursor, as well as commit, revert, etc. That’s really all the “integration” you need.
Some excellent points. On “add selection to chat”, I just want to add that the Claude Code VS code extension automatically passes the current selection to the model. :)
I am genuinely curious if any Cursor or Windsurf users who have also tried Claude Code could speak to why they prefer the IDE-fork tools? I’ve only ever used Claude Code myself - what am I missing?
Cursor's tab completion model is legitimately fantastic and for many people is worth the entire $20 subscription. Lint fixes or syntax-level refactors are guessed and executed instantly with TAB with close to 100% accuracy. This is their final moat IMO, if Copilot manages to bring their tab completion up to near parity, very little reason to use Cursor.
I'd like to ask the opposite question: why do people prefer command line tools? I tried both and I prefer working in IDE. The main reason is that I don't trust the LLMs too much and I like to see and potentially quickly edit the changes they make. With an IDE, I can iterate much faster than with the command line tool.
I haven't tried Claude Code VS Code extension. Did anyone replaced Cursor with this setup?
I can roll back to different checkpoints with Cursor easily. Maybe CC has it but the fact that I haven’t found it after using it daily is an example of Cursor having a better UX for me.
I like using Claude Code through Roo Code (vscode extension). I find it easier to work with text using a mouse, vscode diff viewer etc. I guess if you're very good at vim shortcuts etc you can use that in Claude Code instead of selecting text with a mouse. Claude Code has a vscode extension too so I feel that using Claude Code through vscode just adds a better UI.
IIRC problem is that VS Code does not allow extensions to create custom UI in the panels areas except for WebViews(?). It makes for not a great experience. Plus Cursor does a lot with background indexing to make their tab completion model really good - more than would be possible with the extensions APIs available.
Not if you want custom UI. There are a lot of things you can do in extension land (continue, cline, roocode, kilocode, etc. are good examples) but there are some things you can't.
One thing I would have thought would be really cool to try is to integrate it at the LSP level, and use all that good stuff, but apparently people trying (I think there was a company from .il trying) either went closed or didn't release anything note worthy...
When the Copilot extension needs a new VS Code feature it gets added, but it isn't available to third party extensions until months later... Err, years later... well, whenever Microsoft feels like it.
So an extension will never be able to compete with Copilot.
I use Augment extensively and find it superior to cursor in every way - and operates as an extension. It has a really handy task planning interface and meta prompt refinement feature and the costs are remarkably low. The quality of output implantation is higher IMO and I don’t have to do a lot of model selection and don’t get Max model bill explosions. If there’s something Cursor provided that Augment doesn’t via extension it was not functionally useful enough to notice.
I've really struggled with using the extensions - their UI/UX is worse, they're much more limited in what they can do and they're much more unstable (in IntelliJ at least).
Then again writing mostly kotlin I cannot get along with the VS Code forks as they're just not that great outside of typescript projects in my experience.
I tend to prompt in cursor/windsurf and refactor in IntelliJ which is okay but a bit of a pain.
One competitor to Claude Code that I don't hear much about is Jetbrains Junie. From my experience, the code it generates is as good as CC, and if you've purchased a Jetbrains license you probably have some amount of free Junie every month.
I'll fill in some context. I think the value of Cursor as an IDE is probably somewhat ephemeral. It's mostly combating Microsoft's ambitions to keep other players busy and box them out of the market. Agents gain a lot of value from model context protocol and there's an amazingly short list of clients that fully support the protocol, but the VSCode Chat window is one of them: https://modelcontextprotocol.io/clients
I actually do prefer the view that having the agent built into an IDE brings me but I'll be damned if I'm forced to use CoPilot/OpenAI. Second to that, the agent does have access to a lot more contextual tools by being built into the editor like focused linting errors and test failures. Of course that demands your development environment is setup correctly and could be replicatable with Claude Code to some extent.
I never got the valuation. I (and many others) have built open source agent plugins that are pretty much just as good, in our free time (check out magenta nvim btw, I think it turned out neat!)
> with a simple extension for some UX improvements
What are the UX improvements?
I was using the Pycharm plugin and didn’t notice any actual integration.
I had problems with pycharm’s terminal—not least of which was default 5k line scroll back which while easy to change was worst part of CC for me at first.
I finally jumped to using iterm and then using pycharm separately to do code review, visual git workflows, some run config etc.
But the actual value of Pycharm—-and I’ve been a real booster of that IDE has shrank due to CC and moving out of the built in terminal is a threat to usage of the product for me.
If the plugin offered some big value I might stick with it but I’m not sure what they could even do.
#1 improvement for VS Code users is giving the agent MCP tools to get diagnostics from the editor LSPs. Saves a tremendous amount of time having the agent run and rerun linting commands.
Does anyone have a comparison between this and OpenAI Codex? I find OpenAI's thing really good actually (vastly better workflow that Windsurf). Maybe I am missing out however.
Codex CLI is very bad, it often struggles to even find the file and goes on a rampage inside the home directory trying to find the file and commenting on random folders. Using o3/o4-mini in Aider is decent though.
> What does Cursor/Windsurf offer over VS Code + CC?
Cursor's @Docs is still unparalleled and no MCP server for documentation fetching even comes close. That is the only reason why I still use Cursor, sometimes I have esoteric packages that must be used in my code and other IDEs will simply hallucinate due to not having such a robust docs feature, if any, which is useless to me, and I believe Claude Code also falls into that bucket.
> Cursor's @Docs is still unparalleled and no MCP server for documentation
I strongly disagree. It will put the wrong doc snippets into context 99% of the time. If the docs are slightly long then forget it, it’ll be even worse.
just curious because I'm inexperienced with all the latest tools here
> - Tab completion model (Cursor's remaining moat)
What is that? I have Gemini Code Assist installed in VSCode and I'm getting tab completion. (yes, LLM based tab completion)
Which, as an aside I find useful when it works but also often extremely confusing to read. Like say in C++ I type
int myVar = 123
The editor might show
int myVar = 123;
And it's nearly impossible to tell that I didn't enter that `;` so I move on to the next line instead of pressing tab only to find the `;` wasn't really there. That's also probably an easy example. Literally it feels like 1 of 6 lines I type I can't tell what is actually in the file and what is being suggested. Any tips? Maybe I just need to set some special background color for text being suggested.
and PS: that tiny example is not an example of a great tab completion. A better one is when I start editing 1 of 10 similar lines, I edit the first one, it sees the pattern and auto does the other 9. Can also do the "type a comment and it fills in the code" thing. Just trying to be clear I'm getting LLM tab completion and not using Cursor
I use Windsurf so I remain in the driver's seat. Using AI coding tools too much feels like brain rot where I can't think sharply anymore. Having auto complete guess my next edit as I'm typing is great because I still retain all the control over the code base. There's never any blocks of code that I can't be bothered to look at, because I wrote everything still.
I often use the same setup. Qwen 2.5 coder is very good on its own, but my Emacs setup doesn’t also use web search when that would be appropriate. I have separately been experimenting with the Perplexity Sonar APIs that combine models and search, but I don’t have that integrated with my Emacs and Qwen setup - and that automatic integration would be very difficult to do well! If I could ‘automatically’ use a local Qwen, or other model, and fall back to using a paid service like Perplexity or Gemini grounding APIs just when needed that would be fine indeed.
I am thinking about a new setup as I write this: in Emacs, I explicitly choose a local Ollama model or a paid API like Gemini or OpenAI, so I should just make calling Perplexity Sonar APIs another manual choice. (Currently I only use Perplexity from Python scripts.)
If I owned a company, I would frequently evaluate privacy and security aspects of using commercial APIs. Using Ollama solves that.
Windsurf big claim to fame was that you could run their model in airgap and they said they did not train on GPL code. This was an option available for Enterprise customers until they took it away recently to prevent self hosting
I agree it has a good chance of catching up, but the difference in quality is pretty noticeable today. I'd much rather stick with vscode, because I hate all the subtle ways Cursor changes the UI; like taking over the keyboard shortcut for clearing the scrollback in the terminal. But I find it's pretty hard to use Copilot's tab completion after using Cursor for a while.
I think CC is just far more useful; I use it for literally everything and without MCP (except puppeteer sometimes) as it just writes python/bash scripts to do that far better than all those hacked together MCP garbage bins. It controls my computer & writes code. It made me better as well as now I actually write code, including GUI/web apps, that's are always fully scriptable. It helps me, but it definitely helps CC; it can just interrogate/test everything I make without puppeteer (or other web browser control, which is always brittle as hell).
CC would explode even further if they had official Team/Enterprise plan (likely in the work, Claude Code Waffle flag), and worked on Windows without WSL (supposedly pretty easy to fix, they just didn't bother). Cursor learnt the % of Windows user was really high when they started looking, even before they really supported it.
They're likely artificially holding it back either because its a loss leader they want to use a very specific way, or because they're planning the next big boom/launch (maybe with a new model to build hype?).
Cursor's multi-file tab completion and multi-file diff experience are worth $20 easily IMO.
I truly do not understand people's affinity for a CLI interface for coding agents. Scriptability I understand, but surely we could agree that CC with Cursor's UX would be superior to CC's terminal alone, right? That's why CC is pushing IDE integration -- they're just not there yet.
I can't stand the UX, or VS Code's UX in general. I vastly prefer having CC open in a terminal alongside neovim. CC is fully capable of opening diffs in neovim or otherwise completely controlling neovim by talking to its socket.
I strongly agree with you.
I’m more of a CLI guy, and Claude Code just works. Most good projects have a CLI anyway (gcloud, GitHub CLI, Vercel, etc.). I prefer CLI vs MCP’s.
I’m on the $200 plan, and it’s absolutely worth it (never thought I’d say this for a CLI app).
I don’t see how there will be any money to be made in this industry once these models are quantized and all local. It’s going to be one of the most painful bubble deflations we have ever seen and the biggest success of open source in our lifetimes.
The forked IDE thing I don't understand either, but...
During the evaluation at a previous job, we found that windsurf is waaaay better than anything else. They were expensive (to train on our source code directly) but the solution the offered was outperforming others.
A lot of engineers underestimate the learning curve required to jump from IDE to terminal. Multiple generations of engineers were raised on IDEs. It's really hard to break that mental model.
Claude Code is totally different paradigm. You don't edit your files directly so there is no tab autocomplete. It's a chat session.
There are IDE integrations where you can run it in a terminal session while perusing the files through your IDE, but it's not powering any autocomplete there AFAIK.
As far as I can tell, terminal agents are inferior to hosted agents in sandboxed/imaged environments when it comes to concurrent execution and far inferior to assisted ide in terms of UX so what exactly is the point?. The "UI niceties" is the whole point of using cursor and somehow, everyone else sucks at it.
Done. Now you have a SOTA agentic AI with pretty forgiving usage limits up and running immediately. This is why it's capturing developer mindshare. The simplicity of getting up and going with it is a selling point.
You’re missing the point tho. The point of the cli agent is that it’s a building block to put this thing everywhere. Look at CCs GitHub plugin, it’s great
It's another Character.ai situation [0]. Unfortunate for any employees who aren't founders or researchers, as they don't get any payout or a nice new job from this exit structure. In fact they lose their whole time invested at the company.
What a harsh time to work for an AI startup as a rank and file employee! I wonder how the founders justify going along with it inside their mind.
Edit: Thank you @jonny_eh for the clarification. I can't imagine it feels awesome being a leftover but at least you vested out. "Take the money and leave" is still a bit raw when the founders and researchers are now getting the initial payout + generous Google RSU's.
Character.ai reached out to me for an opportunity, but they've already been carved up.
I think it's great that the rank and file got some of their equity cash-out (based on the other comment), but I imagine it isn't an attractive prospect as a start-up to join at this point.
I just ignored the recruiter. I can't imagine their would be a second liquidity event.
> Unfortunate for any employees who aren't founders or researchers, as they don't get any payout or a nice new job from this exit structure. In fact they lose their whole time invested at the company.
Windsurf’s value didn’t go to $0 overnight. The company will continue and their equity is likely still worth a decent amount wherever the company ends up.
Obviously a disappointing outcome for the people who thought life changing money was right around the corner, but they didn’t lose everything.
Just like with Character I'm assuming the employees get something. Whatever nonsense "licensing" fee Google is paying to not cause an antitrust investigation should be paid out straight to employees
The general character of capital markets is to pay as little as possible. Otherwise you lose out to those who are more ruthless. It is plausible that Windsurf employees really are getting very little value for their work. We need to see details of the deal.
Isn't there not some contractual agreement between the VCs and the founders? (I understand that a non-compete might not apply [in CA], but taking VC money is a little different that simply getting hired).
Were I a Windsurf investor, I'd be pissed right now and calling my lawyer.
I went from Emacs to VS Code, then to Cursor, next to Claude Code, which is so good that I feel like I am having half a dozen junior devs at my fingertips, 24/7.
Since Claude Code is cli based, I reviewed my cli toolset: Migrated from iTerm2 to Ghostty and Tmux, from Cursor to NeoVim (my God is it good!).
Just had a 14h workday with this tooling. It’s so good that I complete the work of weeks and months within days! Absolutely beast.
At this point I am thinking IDEs do not reflect the changing reality of software development. They are designed for navigating project folders, writing / changing files. But I don’t review files that much anymore. I rather write prompts, watch Claude Code create a plan, implement it, even write meaningful commit messages.
Yes I can navigate the project with neovim, yes I can make commits in git and in lazygit, but my task is best spent in designing, planning, prompting, reviewing and testing.
I work at a company with over 700 employees. And there are tons of use cases where a simple CRUD app is sufficient. Or where glue code needs to be written / changed for legacy systems. Or where an OS system like Camunda is deployed and needs to be configured, workflows developed, etc
The reality of companies out there is much simpler than the challenges of a startup that needs to build systems that are state of the art, scale for millions of users, etc There are companies out there that make millions, in areas you‘ve never heard of, and their core business does not depend on software development best practices.
In our company we have an IT team with the median age of fifty, team members who never have developed software, just maintain systems, delegate hard work to expensive consultants.
Now in that setting someone coming from a startup background is like someone coming from the future. I feel like a wizard who can solve problems in days, instead of weeks or months waiting for a consultant to solve.
I review PRs/commits, not files. Given the right cage to lock the agent inside, and guardrails built around, and conventions and guidelines, and agentic flows so it can pull in what's needed.. the need to look at every line and file during implementation is significantly lessened. So then I review the final output (which is a unit of work/task wrapped in a PR).
yes vibecoding is addicting like that. but if you are not reviewing any code and simply vibing then
in my expreience you'll eventually get stuck in "its still not working" loops beause you have no other context or insight to provide it other than that. Then you have either accept what you have or throw the whole thing out and/or actually read the code . kind of rules out last option because code is now just too far gone with too many special cases hardcoded because AI sucks at abstraction or real software engineering.
At work we’re encouraged to use AI, so I do. For me the one thing that works well is using it to write one off scripts that do stuff and would be a chore to write.
Usually in 2-3 prompts I can get a python or shell script that reads some file list somewhere, reads some json/csv elsewhere. Combines it in various ways and spits out some output to be ingested by some other pipeline.
I just test this code if it works it’s good.
Never in my life would I put this in a critical system though. When I review these files they are full of tiny errors that would blow up in spectacular manner if the input was slightly off somewhere.
It’s good for what it is. But I’m honestly afraid of production code being vibe coded by these tools.
Ghostty is gpu accelerated. It’s super fast, and tmux in it is a joy to use. That combined with NeoVim gives me an increadibly smooth dev experience in the terminal, something I had never have with iTerm2 and emacs.
Haha :) I lived inside Emacs, used orgmode for everything, have written tons of Elisp, used org-roam as my second brain, used vanilla Emacs shortcuts instead of Evil (with a special keyboard settup using Karabiner Elements), did even my googling from Emacs, used emacs calc instead of my calculator, but in the end I spent more time tinkering my Emacs setup than doing real work. Emacs was a lifestyle. At some point I realized: Unix and the terminal are what Emacs try to be: It tries to be a one-stop shop offering you everything: Surfing the web, writing emails, word processor, calculator, planner, terminal. Unix and the terminal offer me all of that. Plus any scripting language. Why miss all the beautiful apps, just to be an Emacs zealot? The editor in emacs is just one usecase. Neovim does it just as well, if not better.
But relax, noone is taking your Emacs from you :) I still like it, but am not a disciple anymore ;)
Well you can't risk Claude quitting overnight. It forgets everything it did the day before and now you have to start over ... must ... finish ... tonight ... within ... context ... window.
Cursor (and Garry Tan’s X post) has shown us that the VC money is propping up these companies astounding growth, the only way for them to become profitable is to increase the cost per a request, which means they need to innovate like crazy.
The moat is paper thin.
GitHub has open sourced copilot.
The open source community is working hard on their own projects.
No doubt Cursor is moving fast to create amazing innovations, but if the competition only focuses on thin wrappers they are not worth the billion dollar valuations.
I love watching this space as it is moving extremely fast.
There is no moat. If you’re a true believer that strong agents are around the corner, then all of these add on companies will be obsolete in a few years. The first company to strong agents can trivially rebuild Cursor or Windsurf.
I think the recent Grok release and considering xAI was relatively late to the game shows that the only moat to training giant models is how many GPUs you can buy. ChatGPT was earth-shattering and it took less than two years for multiple credible competitors to match or exceed them. Making these models profitable is proving extremely difficult in the face of so much competition and such unsustainable expectations being set. Google seems to be most likely to sustain themselves through this melee. Them and the Chinese companies.
Cursor just committed mass consumer fraud at worst, and at best pissed off all their best customers. I feel really sorry for those who invested at a 9bb valuation.
I seriously cannot keep up. I fell a bit behind and now I feel I need a primer to know who owns/acquired/developed all these additional things surrounding the ai space
This whole situation feels shockingly close to the Meta/Scale situation, where founders and specific employees were plucked out, and effectively gutted any future prospects for the company.
At least in the Scale case there seemed to be some form of payout to employees and equity holders, but this takes it a whole lot further by just throwing out all other employees.
There is supposed to be the concept that “all common stock is the same”. These fake-acquisitions completely undermine that.
Yep, if investors and early employees keep getting left out in the cold while execs get a soft landing at Big Tech, it's going to shake a lot of trust in the startup game
I don’t think anyone trusts any tech company much these days. It’s been a steep decline in the past 5 years, from arbitrary mandates to the constant talk about firing everyone and hiring an AI. Even as an investor, it’s hard to trust that the “honor system” that once existed is still in play.
So Google, Meta, and Microsoft will just hollow out the best AI startups of their talent instead of buying them - out of fear of monopoly lawsuits I'm assuming?
Nice plan I guess. Kind of obvious to spot though.
I actually don't know if there's much that can be done unless there's some non-competes in those employees' contracts which are usually not very enforceable outside of finance iirc.
I’m honestly just surprised that the CEO and co-founder decided to walk away from the company and leave behind all these employees he was leading. Especially considering many of them probably joined for lower pay, hoping for a big upside.
“Buying the startup” just means handing over megabucks to do-nothing investors. If Google isn’t buying any product or technology, why should investors get a talent fee?
Do nothing investors who enabled the company to reach this point? Employees who chose lower salaries in expectation of shares being worth something? Come on now.
This is overly reductionist. The are plenty of laws that work well.
Any time I hear someone talk about more or less regulation, instead of talking about better or worse regulation, I suspect they are ideologists and trying to shift the narrative, or else they would be able to criticise based on actual merit.
There are many AI startups and we are just in the beginning of learning how to use them. There will be some stupid company like those you’ve listed that figures out a way to use AI that is far better than any other implementation, and Google, Meta, and Microsoft may go the way of Yahoo and AOL, but we’ll see
I never knew anyone who used Windsurf. These AI acquisitions have been unbelievable(in a bad way). WIX acquired some garbage Lovable.dev clone for 80 million. I think many of us are waiting for this bubble to pop(economy will likely pop too)
It was barely better than Cursor and they got shafted by Anthropic because of the takeover announcement so nobody really used it anymore because let's face it - Claude Sonnet is just the best coding model. Design-wise the chat panel and autocomplete integration was a bit nicer than in Cursor but not by much. Subscription for Windsurf was/is also 5$ cheaper.
i don't think it was better than or comparable to cursor at all. except for the month prior to the OpenAI Acquisition news where some minor influencers on X were calling it better.
Everyone has a niche, Windsurf is the only large provider if you are a Jetbrains shop.
There are some alternatives like continue.dev or Jetbrains own AI offering but no Cursor or Claude Code ( Sonnet 3.7/4) you can get through Jetbrains plugin or others, but Anthropic does not provide support same with cursor.
Check out sweep. Completely unaffiliated, their only offering is the jetbrains plugin so it gets a lot more focus than windsurf. Only downside is that Claude code is still a better agent, but at least its tab complete is some of the best
Windsurf was also used by enterprises because of their on-prem plan. They gutted that after OpenAI acquisition was announced and since then I am sure none of those enterprises that used it will switch to their cloud offering and look for other venues.
I used it for about six weeks in the spring, at which point I tried cursor and found out that it was far better for my use case (backend Go): better and faster autocomplete that can work across files, less buggy UI, and it actually follows your rules.
I’m not bothered by this news — it’d be pretty shortsighted to assume that any of these disposable AI assistants will continue to exist forever.
Base44 is absolutely not garbage. I’ve tried it and can say it’s as good or better of a vibe-builder than Lovable or Bolt. Have you benchmarked it against the competition or can you otherwise substantiate the “garbage” claim? FWIW I do know one amazing engineer using Windsurf
Readit News
Let's review the current state of things:
- Terminal CLI agents are several orders of magnitude less $$$ to develop than forking an entire IDE.
- CC is dead simple to onboard (use whatever IDE you're using now, with a simple extension for some UX improvements).
- Anthropic is free to aggressively undercut their own API margins (and middlemen like Cursor) in exchange for more predictable subscription revenue + training data access.
What does Cursor/Windsurf offer over VS Code + CC?
- Tab completion model (Cursor's remaining moat)
- Some UI niceties like "add selection to chat", and etc.
Personally I think this is a harbinger of where things are going. Cursor was fastest to $900M ARR and IMO will be fastest back down again.
I think only way Cursor and other UX wrappers still win is if on device models or at least open source models catch up in the next 2 years. Then i can see a big push for UX if models are truly a commodity. But as long as claude is much better then yes they hold all the cards. (And don't have a bigger company to have a civil war with like openai)
Do you have a citation for this?
It might be at a loss, but I don’t think it is that extravagant.
- Anthropic doesn't use the inputs for training.
- Cursor doesn't have $900M ARR. That was the raise. Their ARR is ~$500m [1].
- Claude Code already support the niceties, including "add selection to chat", accessing IDE's realtime warnings and errors (built-in tool 'ideDiagnostics'), and using IDE's native diff viewer for reviewing the edits.
[1] https://techcrunch.com/2025/06/05/cursors-anysphere-nabs-9-9...
Their base is $20/mth. That would equal 3.75M people paying a sub to Cursor.
If literally everyone is on their $200/mth plan, then that would be 375K paid users.
There’s 50M VS Code + VS users (May 2025). [1] 7% of all VS Code users having switched to Cursor does not match my personal circle of developers. 0.7% . . . Maybe? But, that would be if everyone using Cursor were paying $200/month.
Seems impossibly high, especially given the number of other AI subscription options as well.
[1] https://devblogs.microsoft.com/blog/celebrating-50-million-d...
Last disclosed revenue from Cursor was $500mil. https://www.bloomberg.com/news/articles/2025-06-05/anysphere...
But the chat UX is so simple it doesn't take up any extra brain-cycles. It's easier to alt-tab to and from; it feels like slacking a coworker. I can have one or more terminal windows open with agents I'm managing, and still monitor/intervene in my editor as they work. Fits much nicer with my brain, and accelerates my flow instead of disrupting it
There's something starkly different for me about not having to think about exactly what context to feed to the tool, which text to highlight or tabs to open, which predefined agent to select, which IDE button to press
Just formulate my concepts and intent and then express those in words. If I need to be more precise in my words then I will be, but I stay in a concepts + words headspace. That's very important for conserving my own mental context window
And so I’d say this isn’t a harbinger of the death of Cursor, instead proof that there’s a future in the market they were just recently winning.
They either need to create their own model and compete on cost, or hope that token costs come down dramatically so as to be too cheap to meter.
My mental model is that these foundation model companies will need to invest in and win in a significant number of the app layer markets in order to realize enough revenue to drive returns. And if coding / agentic coding is one of the top X use cases for tokens at the app layer, seems logical that they'd want to be a winner in this market.
Is your view that these companies will be content to win at the model layer and be agnostic as to the app layer?
The bigger issue is the advantage Anthropic, Google and OpenAI have in developing and deploying their own models. It wasn't that long ago that Cursor was reading 50 lines of code at a time to save on token costs. Anthropic just came out and yolo'd the context window because they could afford to, and it blew everything else away.
Cursor could release a cli tomorrow but it wouldn't help them compete when Anthropic and Google can always be multiples cheaper
[0] https://cursor.com/blog/agent-web
I don’t think this is true at all. The reason CC is so good is that they’re very deliberate about what goes in the context. CC often spends ages reading 5 LOC snippets, but afterwards it only has relevant stuff in context.
- AI is not good enough yet to abandon the traditional IDE experierence if you're doing anything non-trivial. Hard finding use cases for this right now.
- There's no moat here. There are already a dozen "Claude Code UI" OSS projects with similar basic functionality.
Auto-regressive nature of these things mean that errors accumulate, and IDEs are well placed to give that observability to the human, than a coding agent. I can course correct more easily in an IDE with clear diffs, coding navigation, than following a terminal timeline.
CC has some integration with VSC it is not all or nothing.
I resisted moving from Roo in VS Code to CC for this reason, and then tried it for a day, and didn't go back.
I am genuinely curious if any Cursor or Windsurf users who have also tried Claude Code could speak to why they prefer the IDE-fork tools? I’ve only ever used Claude Code myself - what am I missing?
I haven't tried Claude Code VS Code extension. Did anyone replaced Cursor with this setup?
Occasionally they lose their connection to the terminal in VSCode, but I’ve got no other integration complaints.
And I really prefer the bring-your-own-key model as opposed to letting the IDE be my middleman.
Not if you want custom UI. There are a lot of things you can do in extension land (continue, cline, roocode, kilocode, etc. are good examples) but there are some things you can't.
One thing I would have thought would be really cool to try is to integrate it at the LSP level, and use all that good stuff, but apparently people trying (I think there was a company from .il trying) either went closed or didn't release anything note worthy...
So an extension will never be able to compete with Copilot.
Then again writing mostly kotlin I cannot get along with the VS Code forks as they're just not that great outside of typescript projects in my experience.
I tend to prompt in cursor/windsurf and refactor in IntelliJ which is okay but a bit of a pain.
I actually do prefer the view that having the agent built into an IDE brings me but I'll be damned if I'm forced to use CoPilot/OpenAI. Second to that, the agent does have access to a lot more contextual tools by being built into the editor like focused linting errors and test failures. Of course that demands your development environment is setup correctly and could be replicatable with Claude Code to some extent.
What are the UX improvements?
I was using the Pycharm plugin and didn’t notice any actual integration.
I had problems with pycharm’s terminal—not least of which was default 5k line scroll back which while easy to change was worst part of CC for me at first.
I finally jumped to using iterm and then using pycharm separately to do code review, visual git workflows, some run config etc.
But the actual value of Pycharm—-and I’ve been a real booster of that IDE has shrank due to CC and moving out of the built in terminal is a threat to usage of the product for me.
If the plugin offered some big value I might stick with it but I’m not sure what they could even do.
Does anyone have a comparison between this and OpenAI Codex? I find OpenAI's thing really good actually (vastly better workflow that Windsurf). Maybe I am missing out however.
Cursor's @Docs is still unparalleled and no MCP server for documentation fetching even comes close. That is the only reason why I still use Cursor, sometimes I have esoteric packages that must be used in my code and other IDEs will simply hallucinate due to not having such a robust docs feature, if any, which is useless to me, and I believe Claude Code also falls into that bucket.
I strongly disagree. It will put the wrong doc snippets into context 99% of the time. If the docs are slightly long then forget it, it’ll be even worse.
I never use it because of this.
> - Tab completion model (Cursor's remaining moat)
What is that? I have Gemini Code Assist installed in VSCode and I'm getting tab completion. (yes, LLM based tab completion)
Which, as an aside I find useful when it works but also often extremely confusing to read. Like say in C++ I type
The editor might show And it's nearly impossible to tell that I didn't enter that `;` so I move on to the next line instead of pressing tab only to find the `;` wasn't really there. That's also probably an easy example. Literally it feels like 1 of 6 lines I type I can't tell what is actually in the file and what is being suggested. Any tips? Maybe I just need to set some special background color for text being suggested.and PS: that tiny example is not an example of a great tab completion. A better one is when I start editing 1 of 10 similar lines, I edit the first one, it sees the pattern and auto does the other 9. Can also do the "type a comment and it fills in the code" thing. Just trying to be clear I'm getting LLM tab completion and not using Cursor
I get all AI or none, so it’s always obvious what’s happening.
Completions are OK, but I did not enjoy the feeling of both us having a hand on the wheel and trying to type at the same time.
My local ollama + continue + Qwen 2.5 coder gives good tab completion with minimal latency; how much better is Cursor’s tab completion model?
I’m still weary of letting LLM edit my code so my local setup gives me sufficient assistance with tab completion and occasional chat.
I am thinking about a new setup as I write this: in Emacs, I explicitly choose a local Ollama model or a paid API like Gemini or OpenAI, so I should just make calling Perplexity Sonar APIs another manual choice. (Currently I only use Perplexity from Python scripts.)
If I owned a company, I would frequently evaluate privacy and security aspects of using commercial APIs. Using Ollama solves that.
Plus recently launched OpenCode, open source CC is gaining traction fast.
There was always very little moat in the model wrapper.
The main value of CC is the free tool built by people who understand all the internals of their own models.
VSCode & CoPilot now offer it.
Is it as good? Maybe not.
But they are really working hard over there at Copilot and seem to be catching up.
I get an Edu license for Copilot, so just ditched Cursor!
They're likely artificially holding it back either because its a loss leader they want to use a very specific way, or because they're planning the next big boom/launch (maybe with a new model to build hype?).
I truly do not understand people's affinity for a CLI interface for coding agents. Scriptability I understand, but surely we could agree that CC with Cursor's UX would be superior to CC's terminal alone, right? That's why CC is pushing IDE integration -- they're just not there yet.
I can't stand the UX, or VS Code's UX in general. I vastly prefer having CC open in a terminal alongside neovim. CC is fully capable of opening diffs in neovim or otherwise completely controlling neovim by talking to its socket.
During the evaluation at a previous job, we found that windsurf is waaaay better than anything else. They were expensive (to train on our source code directly) but the solution the offered was outperforming others.
agentic tool + anthropic subsidized pricing.
Second part is why it has "exploded"
A lot of devs are not superstar devs.
They don't want a terminal tool, or anything they have to configure.
A IDE you can just download and 'it just works' has value. And there are companies that will pay.
That said, the creator of Claude Code jumped to Cursor so they must see a there there.
Claude Code - Agentic/Autonomous coding usecases.
Both have their own place in programming, though there are overlaps.
gemini cli is very expensive.
There are IDE integrations where you can run it in a terminal session while perusing the files through your IDE, but it's not powering any autocomplete there AFAIK.
Dead Comment
- > curl -fsSL http://claude.ai/install.sh | bash
- > claude
- > OAuth to your Anthropic account
Done. Now you have a SOTA agentic AI with pretty forgiving usage limits up and running immediately. This is why it's capturing developer mindshare. The simplicity of getting up and going with it is a selling point.
I doubt MS has ever made $900M off of VS Code.
What a harsh time to work for an AI startup as a rank and file employee! I wonder how the founders justify going along with it inside their mind.
[0] Character.ai CEO Noam Shazeer Returns to Google https://news.ycombinator.com/item?id=41141112 - 11 months ago (87 comments)
Edit: Thank you @jonny_eh for the clarification. I can't imagine it feels awesome being a leftover but at least you vested out. "Take the money and leave" is still a bit raw when the founders and researchers are now getting the initial payout + generous Google RSU's.
Hopefully Windsurf employees are treated well here.
Note: I worked at Character until recently.
$2.4 billion.
Character.ai reached out to me for an opportunity, but they've already been carved up.
I think it's great that the rank and file got some of their equity cash-out (based on the other comment), but I imagine it isn't an attractive prospect as a start-up to join at this point.
I just ignored the recruiter. I can't imagine their would be a second liquidity event.
Source: I was in GDM when character was acquired.
Otherwise why not merge all of engineering into ElGoog?
Windsurf’s value didn’t go to $0 overnight. The company will continue and their equity is likely still worth a decent amount wherever the company ends up.
Obviously a disappointing outcome for the people who thought life changing money was right around the corner, but they didn’t lose everything.
Edit: the people downvoting this clearly can't read, I made the exact same point as jonny_eh.
High interest rates make VC funding more expensive and now bigtech can swoop in, poach all the necessary staff and deprive investors of an exit.
What is the point any more?
Were I a Windsurf investor, I'd be pissed right now and calling my lawyer.
Since Claude Code is cli based, I reviewed my cli toolset: Migrated from iTerm2 to Ghostty and Tmux, from Cursor to NeoVim (my God is it good!).
Just had a 14h workday with this tooling. It’s so good that I complete the work of weeks and months within days! Absolutely beast.
At this point I am thinking IDEs do not reflect the changing reality of software development. They are designed for navigating project folders, writing / changing files. But I don’t review files that much anymore. I rather write prompts, watch Claude Code create a plan, implement it, even write meaningful commit messages.
Yes I can navigate the project with neovim, yes I can make commits in git and in lazygit, but my task is best spent in designing, planning, prompting, reviewing and testing.
The reality of companies out there is much simpler than the challenges of a startup that needs to build systems that are state of the art, scale for millions of users, etc There are companies out there that make millions, in areas you‘ve never heard of, and their core business does not depend on software development best practices.
In our company we have an IT team with the median age of fifty, team members who never have developed software, just maintain systems, delegate hard work to expensive consultants.
Now in that setting someone coming from a startup background is like someone coming from the future. I feel like a wizard who can solve problems in days, instead of weeks or months waiting for a consultant to solve.
Say no more.
You don't review the code? Just test it works?
Usually in 2-3 prompts I can get a python or shell script that reads some file list somewhere, reads some json/csv elsewhere. Combines it in various ways and spits out some output to be ingested by some other pipeline.
I just test this code if it works it’s good.
Never in my life would I put this in a critical system though. When I review these files they are full of tiny errors that would blow up in spectacular manner if the input was slightly off somewhere.
It’s good for what it is. But I’m honestly afraid of production code being vibe coded by these tools.
they don't review files anymore though.
But relax, noone is taking your Emacs from you :) I still like it, but am not a disciple anymore ;)
and yet you're pulling 14 hour workdays..
Deleted Comment
Would love to hear more.
The moat is paper thin.
GitHub has open sourced copilot.
The open source community is working hard on their own projects.
No doubt Cursor is moving fast to create amazing innovations, but if the competition only focuses on thin wrappers they are not worth the billion dollar valuations.
I love watching this space as it is moving extremely fast.
Deleted Comment
Because they didn't do their jobs properly?
What happened?
The thesis is that once you’re paying $200 a month, you’re beholden and won’t pay and compare it with anything else.
At least in the Scale case there seemed to be some form of payout to employees and equity holders, but this takes it a whole lot further by just throwing out all other employees.
There is supposed to be the concept that “all common stock is the same”. These fake-acquisitions completely undermine that.
Nice plan I guess. Kind of obvious to spot though.
Maybe there’s more to the story.
Gentle reminder that more startups die by suicide than homicide, and that an early-stage startup is a total crapshoot.
You sweet summer child.
Any time I hear someone talk about more or less regulation, instead of talking about better or worse regulation, I suspect they are ideologists and trying to shift the narrative, or else they would be able to criticise based on actual merit.
if it was better it would have survived.
There are some alternatives like continue.dev or Jetbrains own AI offering but no Cursor or Claude Code ( Sonnet 3.7/4) you can get through Jetbrains plugin or others, but Anthropic does not provide support same with cursor.
Jetbrains Junie is supposedly the same thing but no Rider and that's my current project so didn't get into that yet.
Windsurf was just disappointingly bad in intellij (like any other plugin I've tried so far)
I am still a paid subscriber but most of my usage is claude code now becaue Windsurf does not Sonnet 4 included in their plan.
I’m not bothered by this news — it’d be pretty shortsighted to assume that any of these disposable AI assistants will continue to exist forever.
...which no one talks about anymore. Okay I guess you have a point.
Dead Comment