I've been running Claude Code in my Cursor IDE for a while now via extension. I like the setup, and I direct Claude on one task at a time, while still having full access to my code (and nice completions via Cursor). I still spend time tweaking, etc. before committing. I have zero interest in these new "swarms of agents" they are trying to force on us from every direction. I can barely keep straight my code working on one feature at a time. AI has greatly helped me speed that up, but working serially has resulted in the best quality for me. I'll likely drop Cursor for good now and switch back to vanilla VsCode with CC.
> have zero interest in these new "swarms of agents" they are trying to force on us from every direction.
Good for you! Personally waiting for one agent to do something while I shove my thumb up my butt just waiting around for it to generate code that I'll have to fix anyway is peak opposite of flow state, so I've eagerly adopted agents (how much free will I had in that decision is for philosophers to decide) so there's just more going on so I don't get bored. (Cue the inevitable accusations of me astroturfing or that this was written by AI. Ima delve into that one and tell there was not. Not unless you count me having stonks in the US stock market as being paid off by Big AI.)
I have personally found that I cannot context switch between thinking deeply about two separate problems and workstreams without a significant cognitive context-switching cost. If it's context-switching between things that don't require super-deep thought, it's definitely doable, but I'm still way more mentally burnt-out after an hour or two of essentially speed-running review of small PRs from a bunch of different sources.
Curious to know more about your work:
Are your agents working on tangential problems? If so, how do you ensure you're still thinking at a sufficient level of depth and capacity about each problem each agent is working on?
Or are they working on different threads of the same problem? If so, how do you keep them from stepping on each other's toes? People mention git worktrees, but that doesn't solve the conflict problem for multiple agents touching the same areas of functionality (i.e. you just move the conflict problem to the PR merge stage)
It's easier when I have 10 simple problems as a part of one larger initiative/project. Think like "we had these 10 minor bugs/tweaks we wanted to make after a demo review". I can keep that straight. A bunch of agents working in parallel makes me notably faster there though actually reviewing all the output is still the bottleneck.
It's basically impossible when I'm working on multiple separate tasks that each require a lot of mental context. Two separate projects/products my team owns, two really hard technical problems, etc. This has been true before and after AI - big mental context switches are really expensive and people can't multitask despite how good we are at convincing ourselves we can.
I expect a lot of folks experience here depends heavily on how much of their work is the former vs the later. I also expect that there's a lot of feeling busy while not actually moving much faster.
Yes, also doesn't work for me. If the changes are simple, it is fine but if the changes are complex and there isn't a clear guideline then there is no AI that is good enough or even close to it. Gives you a few days of feeling productive and then weeks of trying to tidy up the mess.
Also, I have noticed, strangely, that Claude is noticeably less compliant than GPT. If you ask a question, it will answer and then try to immediately make changes (which may not be related). If you say something isn't working, it will challenge you and it was tested (it wasn't). For a company that is seems to focus so much on ethics, they have produced an LLM that displays a clear disregard for users (perhaps that isn't a surprise). Either way, it is a very bad model for "agent swarm" style coding. I have been through this extensively but it will write bad code that doesn't work in a subtle way, it will tell that it works and that the issues relate to the way you are using the program, and then it will do the same thing five minutes later.
The tooling in this area is very good. The problem is that the AI cannot be trusted to write complex code. Imo, the future is something like Cerbaras Code that offers a speed up for single-threaded work. In most cases, I am just being lazy...I know what I want to write, I don't need the AI to do it, and I am seeing that I am faster if I just single-thread it.
Only counterpoint to this is that swarms are good for long-running admin, housekeeping, etc. Nowhere near what has been promised but not terrible.
I tried swarms as well, but I came back to you as well. It’s not worth it even th e small worse description double-checking, fine-tuning is not worth the effort the worse code will cost me in the future. Also when I don’t know about it.
It's not that difficult. You get it to work on one deep problem, then another does more trivial bug fixes/optimizations, etc. Maybe in another you're architecting the next complex feature, another fixes tests, etc etc
For my work I’ve never found myself sitting around with nothing to do because there’s always so much review of the generated code that needs to be done
The only way I can imagine needing to run multiple agents in parallel for code gen is if I’m just not reviewing the output. I’ve done some throwaway projects where I can work like that, but I’ve reviewed so much LLM generated code that there is no way I’m going to be having LLMs generate code and just merge it with a quick review on projects that matter. I treat it like pair programming where my pair programmer doesn’t care when I throw away their work
> Personally waiting for one agent to do something while I shove my thumb up my butt just waiting around for it to generate code that I'll have to fix anyway
I spend that time watching it think and then contemplating the problem further since often, as deep and elaborate as my prompts are, I've forgotten something. I suspect it might be different if you are building something like a CRUD app, but if you are building a very complicated piece of software, context switching to a new topic while it is working is pretty tough. It is pretty fast anyway and can write the amount of code I would normally write in half a day in like 15 minutes.
In my workflow, it's totally interactive: Give the LLM some instructions, wait very briefly, look at code diff #1, correct/fix it before approving it, look at code diff #2, correct/fix it before approving it, sometimes hitting ESC and stopping the show because the agent needs to be course corrected... It's an active fight. No way I'm going to just "pre-approve all" and walk away to get coffee. The LLMs are not ready for that yet.
I don't know how you'd manage a "swarm" of agents without pre-approving them all. When one has a diff, do you review it, and then another one comes in with an unrelated diff, and you context switch and approve that, then a third one comes in with a tool use it wants to do... That sounds absolutely exhausting.
I find it puzzling whenever someone claims to reach "flow" or "zen state" when using these tools. Reviewing and testing code, constantly switching contexts, juggling model contexts, coming up with prompt incantations to coax the model into the right direction, etc., is so mentally taxing and full of interruptions and micromanagement that it's practically impossible to achieve any sort of "flow" or "zen state".
This is in no way comparable to the "flow" state that programmers sometimes achieve, which is reached when the person has a clear mental model of the program, understands all relevant context and APIs, and is able to easily translate their thoughts and program requirements into functional code. The reason why interrupting someone in this state is so disruptive is because it can take quite a while to reach it again.
Working with LLMs is the complete opposite of this.
Thank you so much. These comments let me believe in my sanity in an over-hyped world.
I see how people think its more productive, but honestly I iterate on my code like 10-15 times before it goes into production, to make sure it logs the right things, it communicates intent clearly, the types are shared and defined where they should be. It’s stored in the right folder and so on.
Whilst the laziness to just pass it to CC is there I feel more productive writing it on my own, because I go in small iterations. Especially when I need to test stuff.
Let’s say I have to build an automated workflow and for step 1 alone I need to test error handling, max concurrency, set up idempotency, proper logging. Proper intent communication to my future self. Once I’m done I never have to worry about this specific code again (ok some error can be tricky to be fair), but often this function is just practically my thought and whenever i need it. This only works with good variable naming and also good spacing of a function. Nobody really talks about it, but if a very unimportant part takes a lot of space in a service it should be probably refactored into a smaller service.
The goal is to have a function that I probably never have to look again and if I have to do it answers me as fast as possible all the questions my future self would ask when he’s forgotten what decisions needed to be made or how the external parts are working. When it breaks I know what went wrong and when I run it in an orchestration I have the right amount of feedback.
As others I could go very long about that and I’m aware of the other side of the coin overengineering, but I just feel that having solid composable units is just actually enabling to later build features and functionality that might be moat.
Slow, flaky units aren’t less likely to become an asset..
And even if I let AI draft the initial flow, honestly the review will never be as good as the step by step stuff I built.
I have to say AI is great to improve you as a developer to double check you, to answer (broad questions), before it gets to detailed and you need to experiment or read docs. Helps to cover all the basics
I used to have a pro-cursor subscription, but it was way too expensive because I'd always hit my limit. I realized I could just use claude code + the free version of cursor for autocomplete and it worked even better. At this point, I'm not understanding the value that cursor is bringing. A souped up claude code? All I have to do is wait a few months and anything useful will be in claude code or codex or whatever.
Man, I wish they'd keep the old philosophy of letting the developer drive and the agent assist.
I feel like this design direction is leaning more towards a chat interface as a first class citizen and the code itself as a secondary concern.
I really don't like that.
Even when I'm using AI agents to write code, I still find myself spending most of my time reading and reasoning about code. Showing me little snippets of my repo in a chat window and changes made by the agent in a PR type visual does not help with this. If anything, it makes it more confusing to keep the context of the code in my head.
It's why I use Cursor over Claude Code, I still want to _code_ not just vibe my way through tickets.
My guess would be this is less driven by product philosophy, more driven by trying to maximise chances of a return on a very large amount of funding in an incredibly tough market up against formidable, absurdly well-funded competitors.
It's a very tough spot they're in. They have a great product in the code-first philosophy, but it may turn out it's too small a market where the margins will just be competed away to zero by open source, leaving only opportunity for the first-party model companies essentially.
They've obviously had a go at being a first-party model company to address this, but that didn't work.
I think the next best chance they see is going in the vibe-first direction and trying to claim a segment of that market, which they're obviously betting could be significantly bigger. It's faster changing and (a bit) newer and so the scope of opportunity is more unknown. There's maybe more chances to carve out success there, though honestly I think the likeliest outcome is it just ends up the same way.
Since the beginning people have been saying that Cursor only had a certain window of time to capitalise on. While everyone was scrambling to figure out how to build tools to take advantage of AI in coding, they were one of the fastest and best and made a superb product that has been hugely influential. But this might be what it looks like to see that window starting to close for them.
As these products mature people are going to see more of this stuff. These are the contours of the market. The technology is incredible but it’s still subservient to the economics of building products.
It’s the “why can’t Facebook just show me a chronological feed of people I follow”. Because it’s not in their interests to do so.
Yeah, and just like all social media platforms adopted short form video sooner or later they are going to give in to what consumers pay for (in attention or money). Right now it's anyone's guess what that might be in the context of software development.
It's a very tough spot they put themselves into. If the goal wasn't to get filthy rich quick it would probably be possible to make a good product without that tough spot.
You know, it’s stuff like this making me think maybe the anti capitalists have a point.
A company makes a popular product customers like, but to satisfy the VCs the company must make a product the customers don’t like but could make the VCs more money.
Not sure this is the “invisible hand” Adam Smith had in mind.
According to the comment I replied to you have to make the product VCs think will make VCs the most money, even if that’s at odds with what your customers are telling you they want.
It is more of a point that it may require leaving your old customers behind and disappointing them in order for you to find the customers you can provide the most value to.
Anti-capitalist here: Our point is actually the same point as the one Anti-feudalists had. The consumer hostility observed under capitalism is simply a corollary.
To be fair feudalism (the row-farming kind) kind of collapsed because people found better deals with the rise of trade and mercantilism and such. It wasn’t anything anybody needed to make points over.
IDEs seem headed in the same direction. Seriously, watching Codex rip apart binaries in parallel and Claude go from nothing to app in one prompt, I’m pretty sure there’s no need for me to look at any code. I’m fine using tools that just emit machine code if that’s more efficient.
It is easier to imagine the end of the world then the end of feudalism.
I kind of like the story of how Malthus had his theory of societal collapse because he couldn’t imagine a better system then mercantilism. That societies would rather collapse then to end their colonial monopolies.
I see a similar theory today with around depopulation, that as society gets older and relatively fewer working age people there are, that society would rather collapse then we find a better system then Capitalism.
(I think I agree), Georgist here, Our point is also that these rent-seeking abilities (something which even the creator of capitalism famously hated the land-lords quite a lot)
This is the issue with something like Cursor and VC's funding because I feel like these private equities essentially seek rent in their own way by hollowing out the products from within, to maximize profits without doing efforts in a manner very similar to Rent-seeking, and most large companies also feel like a rent-seeking on the monopoly that they establish (like google or facebook)
I have made someone who was communist/socialist agree to georgism and I have had someone who was extremely capitalist agree to georgism, and to be honest, whether it be georgism or anti capitalism or socialism, I think that the world just wants a system where a person is treated with dignity within the economic cogs.
My opinion is that as long as we can all agree on the last premise about dignity for individuals within the economic cogs, we can all have meaningful conversations to make that a premise, hopefully a reality.
(I feel like the people who might deny dignity to people within this particular context, have either a bias/incentive to not look towards the problem, or are uninformed, or lack the energy to fight towards change within the system, and more importantly the _hope_ that the future can be better)
I am not hopeful about the current political systems (even around the whole-world at times), I feel like there should be more information and decentralization within politics.
Essentially, politics really just feel unaccountable to me, your vote really stops mattering to politicians if/when money starts talking. But technically, this system can be broken through with enough votes.
I really hope for a future where politics and politicians feel accountable and genuine, maybe even someone from down the street who we can have some chats with to actually know them.
Ironically or unironically, just as how the landlords pushed against Georgism/George within really making political difference, The same is happening right now as well where Online landlord monopolies dictate how people interpret and vote by using their algorithms/influence.
Politics like many other problems feel like a chicken and egg problem, like things work until they don't and things don't work until they do. At a more individual level, stepping outside of most algorithms and the reason why I joined hackernews is for doing something like this, myself.
I have been on the fence if I think composer is useful, but the speed argument is one I hadn’t really considered. I use cursor with Opus almost exclusively but the other day I tried using OpenCode locally with a 6-bit quantized version of Qwen 3.5 and holy crap the speed and latency were mind blowing. Even if not quite as sharp as big boi Opus and the gang.
Now you’ve got me thinking I should give composer another go because speed can be pretty darn great for more generic, basic, tasks.
> They've obviously had a go at being a first-party model company to address this, but that didn't work.
I thought there was an entire initiative to build their own coding model and the fine tunes of in Composer 1.5 and Composer 2 were just buying them time and training data
Yeah, this model where you don't get an editor anymore feels like a step backwards. I don't want to give up LSPs, being able to step into/rename functions and stuff like that. I should still be the one in control of the code - the agent is the assistant, not me.
This is why Zed's direction felt pretty strong to me. Unfortunately their agentic features are kind of stagnating and the ACP extensions are riddled with issues.
I actually run a custom fork of Zed based on their master branch because of how stagnated the built-in agent is. Master branch Zed agent did get sub-agents, parallel threads, better thread management, and worktrees though, and I implemented agent skills and the ability to select which model to use for sub-agents for it. And with those features, I'm fairly satisfied.
It's still a very nice and fast editor, and you can just switch off those AI features. They're still releasing features and fixes for the non-AI parts.
No but I have now. It’s hard to tell from that few seconds but it doesn’t look like it’s really putting the developer in the driving seat, just providing a minimal escape hatch for manual edits.
Agreed completely on this (as a heavy daily user of Cursor). It's been the perfect in-between of coding by hand (never again!) and strictly "vibe coding" for me. Being able to keep my eyes on all the changes in a "traditional" IDE view helps me maintain a mental model of how my systems work.
I'm hoping in this new UI in v3 I can still get that experience (maybe it's just hidden behind a toggle somewhere for power users / not shown off in the marketing materials).
I'm an engineer at Cursor, can try to clarify questions here.
> I wish they'd keep the old philosophy of letting the developer drive and the agent assist. Even when I'm using AI agents to write code, I still find myself spending most of my time reading and reasoning about code.
We very much still believe this, which is why even in this new interface, you can still view/edit files, do remote SSH, go to definition and use LSPs, etc. It's hard to drive and ship real changes without those things in our opinion, even as agents continue to get better at writing code.
> I'm hoping in this new UI in v3 I can still get that experience (maybe it's just hidden behind a toggle somewhere for power users / not shown off in the marketing materials).
This new interface is a separate window, so if you prefer the Cursor 2 style, that continues to exist (and is also getting better).
That's good to hear, I might have jumped a little too quickly in my opinion. It's a bit of a Pavlovian response at this point seeing a product I very much love embrace a giant chat window as a UX redesign haha.
I would love to see more features on the roadmap that are more aligned with users like us that really embrace the Cursor 2 style with the code itself being the focal point. I'm sure there's a lot you can do there to help preserve code mental models when working with agents that don't hide the code behind a chat interface.
Reading diffs is an inescapable skill, needed for evaluating any kind of PR. This just makes it more interactive.
I just use Copilot with VS Code, but my flow is to just ask Claude to make a change across whatever files it needs to touch, then either accept the changes, edit the changes directly, or clarify whatever was different from my expectations.
Reading diffs is central to how I work with these agents.
As a Cursor user who hasn't tried Claude Code yet, am I missing anything? I seem (sometimes) exceptionally productive in it and it's working for me. To my understanding, Claude Code is all terminal, but something like an IDE seems like the better interface to me: I want to see the file system, etc. It seems Cursor doesn't have the mindshare relative to Claude in public discussion spaces.
Claude Code is where you move up one abstraction layer. Almost everyone using it productively has spend a lot of time working on their harness, ensuring that everything is planned out and structured such that all that is left is really type in the code. This typically works without error. Before that, you interact a lot via Claude Code in whatever abstraction you feel is right.
That's basically it. You can review changes afterwards, but that's not the main point of Claude Code. It's a different workflow. It's built on the premise: given a tight and verifiable plan, AI will execute the actual coding correctly.
This will work, mostly, if you use the very best models with a very good and very specific harness.
Cursor, same as Copilot, has been used by people who are basically pair programming with the AI. So, on abstraction down.
I have no idea what is better, or faster. I suspect it depends at least on the problem, the AI, and the person.
> Cursor, same as Copilot, has been used by people who are basically pair programming with the AI. So, on abstraction down.
This is not really true anymore.
Cursor has better cloud agents than Claude. The multi-agent experience is better, the worktree management is better. Tagging specific code or files in chat is better.
It's hard for me to express the level of pain and frustration I feel going from Cursor to Claude / Conductor+Claude / Claude Extension for VS Code, Claude in Zed, etc.
Really hoping Claude puts more energy into Cowork as a competitor for Cursor and Codex.
I think you are still speaking in the lower abstraction in terms of zwaps' provided understanding. "Tagging specific code" or "files" is likely the type of interfacing most Claude Code users are _not_ doing.
Instead they are defining architecture through specs and verification-loops and attempting to one-shot solutions fitting clear tests. On reflection, I personally don't have many prompts with CC referencing files or code directly, rather I speak in specifications I can then track to a given instance of work in review.
This isn't to suggest you can't work at this abstraction in cursor or w/e interface, but the features you suggest are hardly relevant to the divide zwaps is identifying.
It's always funny to see people's reactions to AI because it's the same they would treat junior engineers if nobody was around to raise an eyebrow. I've had a super micromanager who was absolutely insistent on naming variables and whether the open brackets were on the same line or a new line. I've also had people who just gave me the desired functionality and let me figure out the in-between and put in my own creative features, etc with just slight feedback.
We have OG Cursor for the micromanagers (who want to approve/deny every line) and things like Claude Code for those who are less picky about the how, and able to be amazed at what it creates.
I feel like perhaps you haven't used Cursor. I use both CC and Cursor extensively and as far as I can tell there is nothing that the CC agent will do that Cursor won't do just as well (often using Opus as the backend) and at the same time I get the advantage of seeing the changes in a full IDE if I want to. Their new agent-forward UI hides the code if you don't want to see it as much, but I and many others think that it giving me a full, colourful graphical editor to view changes in is a huge advantage.
I'm not telling you to go use cursor, just to help clarify that you can drive both solutions with the exact same approach and skillset and get very similar results - the difference is the UI. I personally like being able to paste screenshots into the agent, etc.
So that sounds like Claude Code is an inferior subset of Cursor. That Cursor can work like Claude Code, but Claude Code is lacking Cursor’s editing capabilities.
If you install the VS Code plugin, it's the same editing functionality. Cursor lacks a lot of the tooling in claude code that makes the experience a lot more... solid.
Claude Code isn't really "all terminal" if you embed that terminal in your IDE. I still use Cursor (for now), but I embed a CC panel via extension. With this launch of Cursor 3, I'll probably get off Cursor for good. I have zero interest in this.
It's good to try Claude Code just so you focus on skills, agents, and CLAUDE.md
Then when you go back to Cursor it will still support all of those things in the settings.
Using Cursor you tend to not think about those as much since Cursor does a lot of it for you as part of the IDE integration. But it's good to refine it your own way.
But for the most part there isn't much difference.
As someone whose work enforced a switch from Cursor to Claude Code, I do keep on top of the code by pairing it with an IDE, tracking/viewing changes etc. There's no real obstacle to using an IDE as you normally would, with Claude Code as a sidecar.
I tried that for a couple weeks and it's no where near as well integrated as Cursor. I hope they get there though because I like Zed.
Zed plus Claude feels more like using isolated browser extensions instead of something part of the browser (unless you pay for Zeds AI thing then the integration is marginally better).
I vibe my way through my ideas. I look at LLM code sometimes to cry and cringe and then I beg LLM to have basic dignity and self respect to write code it shouldn’t be ashamed of. But then I instruct it to do something and it does it with speed I’m never able to achieve, even if the code is ugly. But it works.
Yes, there are so many. As in hand-written code. I don’t take LLM written code for granted and I rewrite is sometimes. I know it’s not perfect. But it’s useful.
Compile code is not perfect also. But who does hand-written assembler anymore? Yes, LLM is another layer, it would be ugly and slower but it’s much faster to use.
In both cases, it's your processes (automated testing, review, manual QA) that is the bulwark against bugs and issues.
With AI, you can set up great processes like having it check every PR against the source code of your dependencies or having it generate tests for what's an intermediate step or ephemeral solution that you would never write tests for if you had to do it yourself.
There's this idea on HN that if you delegate too much to AI, you get worse code. Presumably not appreciating all the code-improving processes you can delegate to AI, particularly processes you were never doing for hand-written code.
There are some critical parts of architecture where sometimes I really do need to see the code and even sometimes put a wall around it and tell the agent they can't touch it.
How would they make money from the tokens then haha? The main revenue driver of these companies is to get people to use more tokens. That’s what they will optimise for. Getting the developers out of the way is the way to do it.
Isn’t Cursor’s business model mostly subscriptions? They’re the ones paying for inference, not the user directly, right? So wouldn’t they be incentivized to minimize token usage per unit of user value, not maximize raw tokens?
Nope. Enterprise you pay for seat to access all of the enterprise features and then you just pay for tokens as you go. Vast majority of their actual revenue comes from enterprise and their revenue is just api pass through to the model providers.
AI labs think they’re building an autonomous replacement for software engineers, while software engineers see these systems as tools to supplement the process of software engineering.
Yeah that's the disconnect though right? Even with the best frontier models, you need to do a lot of system design work, planning, and reviewing before you can let these models run.
These models are infinitely more effective when piloted by a seasoned software engineer and that will always be the case so long as these models require some level of prompting to function.
Better prompts come from more knowledgeable users, and I don't think we can just make a better model to change that.
The idea we're going to completely replace software engineers with agents has always been delusional, so anchoring their roadmap to that future just seems silly from a product design perspective.
It's just frustrating Cursor had a good attitude towards AI coding agents then is seemingly abandoning that for what's likely a play to appease investors who are drunk on AI psychosis.
Edit: This comment might have come off more callous than I intended. I just really love Cursor as a product and don't want to see it get eaten by the "AI is going to replace everything!" crowd.
> AI labs think they’re building an autonomous replacement for software engineers
And management everywhere is convinced that thats what they are paying for. My company is replacing job titles with "builder". Apparently these tools will make builder out of paper pushers hiding in corporate beaurcarcy. I am suddenly same as them now per my company managment.
Imagine you are the top engineer of your company. Everybody wants your attention, many meetings, design sessions, and of-course code reviews.
With Claude Code, I use Gitlab for reviewing code. And then I let Claude pull the comments.
It looks like the new UI has a big focus on multiple agents. While it feels wrong, the more you split up your work into smaller merge requests, the easier it is to review the work.
Chat first is the way to go since you want the agent busy making its code better. Let it first make plans, come up with different ideas, then after coding let it make sure it fully tests that it works. I can keep an agent occupied for over a hour with e2e tests, and it’s only a couple hundred lines of code in the end.
The philosophy still works, you just have to change your view. Instead of trying to work side by side with the agent on every turn (inside of your IDE), instead the agent performs a unit of work and then you review it. You can use your IDE to view the diff, or another diffing tool.
If you've dug in sufficiently on plan mode, then what the agent is executing is not a surprise and shouldn't need input. If it does, the plan was insufficient and/or the context around the request (agents.md, lessons.md, or whatever tools and documents you use ) weren't sufficient.
EDIT: Maybe it doesn't work in cursor, but I continue to use vscode to review diffs and dig in on changes.
Why I harp on owning your stack instead of outsourcing your Ai experience and interface to Big Ai. There are many frameworks that make this much easier today. I chose ADK which is more of a lift, but also works for non-coding use cases.
I agree. I am building www.propelcode.app for this exact reason.
I get the temptation of letting agents do everything. But they create really bad systems still (bad architecture, reimplementation of solved problems etc).
I also get the temptation for beginners and think it’s great that more people are empowered to build software but moving entirely to chat means they won’t learn and level up which in the long run limits their ability.
I could be wrong. And my way of thinking is dying but thankfully I can build the tool I want.
I'm confused how and if Cursor is still relevant since the Claude Code VSCode extension came out.
The biggest downside for me with Cursor was losing access to gated Microsoft extensions like Python and C#. Even when vibing there are times you will still need a debugger or intellisense.
I note in the comments lots of people saying they are moving back and this latest move looks like the final nail in the coffin for Cursor.
1. Cursor is multi-model, meaning you can use at least a dozen different models.
2. Cursor's UI allows you to edit files, and even have the good old auto-complete when editing code.
3. Cursor's VSCode-based IDE is still around! I still love using it daily.
4. Cursor also has a CLI.
5. Perhaps more importantly, Cursor has a Cloud platform product with automations, extremely long-lived agents and lots of other features to dispatch agents to work on different things at the same time.
Cursor was the tool you use to pair program with AI. Where the AI types the code, and you direct it as you go along. This is a workflow where you work in code and you end up with something fundamentally correct to your standards.
Claude Code is the tool you use if you want to move one abstraction layer up - use harness, specs, verifications etc. to nail down the thing such that the only task left is type in the code - a thing AI does well. This is a workflow where the correctness depends on a lot of factors, but the idea is to abstract one level up from code. Fundamentally, it would be successful if you don't need to look at code at all.
I think there is not enough data to conclusively say which of these two concepts is better, even taking into account some trajectory of model development.
I do feel that any reason I have for installing Cursor is that I want to do workflow 1, rather than workflow 2. Cause I have a pretty comprehensive setup of claude code (or opencode, or whatevs) and I think it does everything you list here.
So, as a product engineer, you probably wanna mention why it matters that Cursor UI allows you to edit files with auto-complete.
I hope this comes off as constructive criticism, but I'm confused about what cursor is now.
Cursor is an IDE and an agentic interface and a cli tool and a platform that all work locally and and in the cloud and in the browser and supports dozens of different models.
I don't know how to use the thing anymore, or what the thing actually is.
I'm having the same issue, as a former Cursor user and current Claude Code addict. CC is a very clear mental model. So is "agent in your IDE," like Cursor used to be and Xcode is now. The advantage of my current setup is that it's the terminal and Xcode, just as it has been for over 20 years.
I applaud Cursor for experimenting with design, and seeing if there are better ways of collaborating with agents using a different type of workspace. But at the moment, it's hard to even justify the time spent kicking the tires on something new, closed source and paid.
I would switch to Cursor 3 in a heartbeat if it supported Claude Agent SDK (w/ Claude Max subscription usage) and/or Codex the way that similar tools like Conductor do
And I would happily pay a seat based subscription fee or usage fees for cloud agents etc on top of this
Unfortunately very locked into these heavily subsidized subscription plans right now but I think from a product design and vision standpoint you guys are doing the best work in this space right now
Is there going to be any more development on the frontier of cursor tab completion and features like that (more focused on helping engineer's with llm's for complex tasks) since I feel this is the main reason I dont use claude code or codex. I want to be writing the code, since I want performant, small, codebases that I understand (I am writing eBPF stuff, so agentic coding doesnt work that well)
Computer use in the cloud is the main reason I use them. It's a game changer. It has its own dev env with a browser / shell and can test what it wrote (a bit of a hassle to set it up, but when it's working, wow)
For $20 a month, I can plan and implements a couple features in 4 hours with Claude. Then I have to wait.
For $20 a month, I can plan and implement thousands of features using Composer 2 or Auto with Cursor. The usage limits are insanely higher. Yes, the depth of understanding is not Opus 4.6, but most work doesn't need that. And the work that does need it I pass to Claude.
I can code 8 hours a day using LLMs as my primary driver spending just $40 a month.
You can do this with copilot, for the $40/mo range, AND you get to use opus 4.6 for all of it. Copilot is absurdly cheap if you can make it fit your work profile.
Agree with that, they seem really good limits for daily use on something like Chat GPT Pro $20 account. I'm in the curious situation of using the Codex CLI within Cursor IDE and not really getting value out of my $60 Cursor sub. Plus at every update it seems Cursor seems to break more of their UI in the 'not a cloud agent chat UI' vs the more traditional VSCode sort of layout of code first. I should probably cancel.
Brand recognition. Since "model-is-the-service", various previously-interesting companies become thin API resellers and the moat is between "selling a dollar for fifty cents" and Brand awareness.
I am not saying this in bad faith. Model companies cannot penetrate every niche with the same brand recognition as some other companies you would consider as "API resellers" do.
I kinda quit using it. The tab feature is useful when making minor or mundane changes, but I quite prefer the codex GUI if I am going to be relatively hands off with agents.
Every company I've worked at has still had a few engineers who insist on working exclusively in the CLI with vim/emacs prior to AI. Every other engineer used some flavor of a desktop app ranging from more minimal editors to incredibly complex IDEs. I expect we land back on UIs long term.
I won’t, but it does have a couple features Codex lags, including remote SSH (huge, because the easiest way to sandbox your agent is to put it into a VM), and the ability to kicking things of on your mobile and finishing up on your desktop (again, really nice if you get a good idea out on a walk, or while talking to a colleague.
These are features I am sure Codex will soon have, of course.
Then there is the advantage of multiple models: run a top level agent with an expensive model, that then kicks of other models that are less expensive - you can do this in Claude Code already (I believe), but obviously here you are limited to something like Haiku.
I love Cursor. As a Product Manager who's not really had coding experience, it's been very useful. I'm able to have a browser on the side and make changes easily, and click through exactly what I want to change rather than having the LLM guess which component I'm talking about. Having multiple models has also been great, as well as the MCP integration. Most times I don't need all the MCPs, but I like being able to turn them on or off based on what I'm doing, like JIRA or Grafana.
One of my favorite startups and I genuinely like to keep subscribing to them.
This is a really underwhelming UI for something that is agent-first. It looks like they're mimicking Notion.
The next generation of interfaces are not going to look like an evolution into minimalist text editor v250. This is like people iterating on terminals before building native or web applications.
VS Code wouldn’t have won the mid-2010s editor wars if it was closed source (note that VS Code has not helped MS ramp people up to VS itself). The winner of that war was always going to be an open source editor, it was just Microsoft whose concept won out. Closed source editors like Coda failed to gain traction and even Sublime Text fell eventually.
If MS ever decided to discontinue VS Code or relicense it, there would be blood in the water. I guarantee you there would be multiple compelling competitors in under a year and probably a new open source winner with consolidation in 5.
So to answer your question: they would be forking Atom (which I think would’ve won otherwise).
Sublime Text fell because VS Code was just better, not because it was closed source. I switched from Sublime Text to VS Code, and didn't care one bit how open or close either was.
Not saying there aren't people who care, there are, but they are a small minority.
Zed's not a VSCode clone, and it's fantastic and OSS. They don't really have a business model that I see working though, IMO. I pay them purely because I love the editor, but the editor is free. The AI integration is what you pay for, but I just run claude code in a terminal.
every AI lab have cli for agent coding. you don't need VS Code. if you want coding agent to write code for you just use cli then use any IDE, text editor or whatever you prefer to review, edit or write code.
There was a time that Eclipse was my preferred editor. It was free and it gave cool sexy features that all the cool kids who could afford Visual Studio had, and it worked on Linux!
Nowadays I'm basically a Neovim purist, but I have positive memories of it. I'm kind of afraid to revisit it at this point, though, since everyone hates on it and I suspect I wouldn't like it as much.
Did you consider IntelliJ, even just the community edition?
If not you really should. IntelliJ with Java is one of the best dev experiences I've ever had. I'm a VSCode fan for most other things but for Java I wouldn't even remotely consider using it over IntelliJ if I had the option :-)
I don't think this is the direction where cursor users want to go, they basically free up the market for VSCode and Zed, and won't be able to compete against lab owning their model.
I only used Cursor in short bursts so 20 USD per month was hard to justify. I recently switched to VSCodium + Cline + OpenRouter and I can use any model I want (currently Step 3.5 Flash for "Temu Sonnet"). It scratches the itch very well for me for literal pennies on the dollar.
I should also add: Cline doesn't require any account at all. I just installed the extension and added my OpenRouter API key and that was it.
If you're in the market, OpenCode is quite good and has become my daily driver. You may also consider pi[0], but that's (from what I've heard) more agenty.
The features here don’t seem game changing. The most compelling parts are mostly already available in Claude or Codex or their related apps and services.
The biggest concern is that if you want to use SOTA models I don’t see how they can match what you get with the subscription plans of Anthropic and Open AI, whether your spending $20 or $200 a month.
Even if they could match what you get in terms of token quantity, they are giving their tools away for free for the foreseeable future and Cursor is not.
Cursor died for me when they star putting limits and time waits everywhere even on more expensive plans.
I totally preferred the other way, but at some point , there is boiler plate or organizations you just want done and it does not make sense to put you waiting minutes a time to confirme few refactors.
That literally killed the vibe for cursor to me
I like cursor and its workflow as a tool, but I do wonder whether moving to cloud (I mean for lots of the cool features) will work. Yes we all GET Cursor has to make money. No one is fooled what this is about. It's also fine, the video and screenshot thing is great.
Daily cursor user who's been previewing this a bit while it was in alpha.
I think it's a really solid release, and while cursor seems to have fallen out of the "cool kids club" in the past three months it remains the most practical tool for me doing AI-first work in a large production code base. The new UI works better in a world where agents are doing most of the work and I can hop back into the IDE interface to make changes.
We've set up a linear integration where I can delegate simpler tasks to cloud agents, and the ability to pick that work up in cursor if I need to go back in forth is a real productivity boost. The tighter integration with cloud agents is something I've been hoping for recently.
I appreciate not being tied at the hip to one model provider, and have never loved doing most of my work from the command line. I was on vs code + meta's internal fork of it for years prior to the current AI wave, so that was a pretty natural transition. I'm pretty optimistic on cursor's ability to win in the enterprise space, and think we're going to see open source models + dev tools win with indie devs over things like claude code as costs start getting passed down more and the gap between frontier models and open source gets tighter.
The biggest killer feature Cursor has that so far no one else seems to have is cloud based computer use. It’s such a game changer. You get a walkthrough video instead of just diffs. But as soon as anthropic release it (their computer use is local only, no thanks) I might consider switching though. Mostly due to the subsidized $200 plan.
I left cursor and went back to VS Code b/c the editing experience is basically the same and cursor was adding more and more agentic features which don't appeal to me. I'm a happy Claude Code user and having my code separate from the planning/brainstorming part of the task makes implementing its own step with me driving/writing the code.
I don't understand how this product can be productively useful. It looks like any other AI chat bot, but I remember hearing people speak very positive things about it. What am I missing?
You're missing nothing with this new ui. For me very good autocomplete + stuff than can be automated with an agent on the side while coding on the other was the peak. I want the control, control to activate/disable autocomplete and agents, I don't want to follow an imposed workflow
Me too, I have the bad feeling autocomplete will be sunsetted sooner or later, it clearly isn't the path they're getting into. Also it started to get worse lately, it tries too hard to predict, it wasn't like that some time ago, hopefully you know what I'm saying
No they are not. Tired of this 40 year old terminal setback instead of having real and beautiful GUIs. Its fine for some kind of people but don't think what works for you is acceptable for the other 50% of us.
Oh yeah I am not talking about cursor, maybe what it felt like when it was 1.0. There will be a new breed of GUIs for people who don't touch terminals. Opencode web and devin code review have a little glimmer of that but it will probably look very different.
Cursor is so good for what I do is that I've cancelled my Cursor subscription and went back to VSCode (w/o Copilot) for the diff review and code navigation.
No per-agent auto-worktree? This is the killer feature of Conductor, having to type `/worktree` into every new chat isn't really a resolution. Not even sure what selecting 'Worktree' for a new chat does
"having to type `/worktree` into every new chat isn't really a resolution"
I don't know what you're talking about. My experience with Cursor (before this new v3) is that new Cursor agent tabs / cloud agents already intelligently manage worktrees to prevent conflicts.
Wow, maybe something is wrong with my setup. In Cursor 3, I am clicking "New Agent" at the top left. My root repository is correctly listed on top of the composer, and I clicked the icon to the right of it and selected 'Worktree'. Then, I instruct the model to run `pwd` and tell me it's git status. It's always just on `main` in my root repository. I dug through the settings and couldn't find anything, and after finding this comment[0] on their forums gave up. Would you mind sharing a bit more about your setup/how it works?
Funny how in this space, once a company feels dead, you don’t even check out their release if the video looks decent, it would have to be totally revolutionary.
No it's not very good. But when you run out of Claude tokens it's perfectly fine for small stuff.
Cursor's inline autocomplete is very good though, much better than anything I could reproduce in Zed with various 3rd party "edit" LLMs (although checking google, they announced a new model since I tried it https://zed.dev/blog/zeta2)
Maybe I'm old, but I only recently started using Gemini to assist me in coding. Now it seems everyone is heading to giving agents to do the full-blown coding. I guess if the result code is good, it doesn't matter who's coding (me or AI).
But are they affordable already for developers who don't earn a Silicon Valley salary? Developers in 3rd world countries?
I'm not convinced people who are doing real work on production applications with any sizable user base is writing code through only agents. There's no way to get acceptable code from these models without really knowing your code base well and basically doing all the systems thinking for the model.
Your workflow is probably closer to what most SWEs are actually doing.
You really need to keep them on a tight leash, stop and correct them when they start screwing up, and then the remaining 90% of the work starts after they say their done, where you need to review/refactor/replace a lot of what they produced.
The only way you're going to let an agent go off on its own to one-shot a patch is if your quality bar is merely "the code works."
This, at least for me, has changed in the past six months. Which is the same thing people were saying in the months prior to that, so I will accept some eye rolls. But at least for our pretty large monorepo opus + a lot of engineering work on context got us to a point where a large portion of our engineers are doing most of their work with agents first and a lot of back and forth + smaller hand edits.
Not true. As long as you don't blindly accept their garbage and keep things behind sensible interfaces so you can reimplement if necessary, and have good tests you're fine
What is Cursor doing? They need to relax a little bit. Recently I saw they released "Glass" which WAS here: https://cursor.com/glass, now just redirects to /download.
Is "Cursor 3" == Glass? I get they feel like their identity means they need to constantly be pushing the envelope in terms of agent UX. But they could stand to have like an "experimental" track and a "This is VS Code but with better AI integration" track.
Glass was a codename while the UI was in early alpha with testers. It redirects to download now because there is no special link anymore. It's just part of Cursor 3 itself.
So funny , I remember their talk about re-imagining their editor for the future of agents. They end up copying codex gui lol.
These AI companies are running out of ideas, and are desperate.
I can't imagine investing in companies that are 3 month behind open source alternatives, and their target audience being the most experimental kind there is.
Cursor seems like far worse value than Codex with a ChatGPT subscription. Doesn't equivalent usage of the $200 subscription cost over $1000? I don't understand why people use it when you can just get multiple Pro subscriptions.
so just like how every chat app has to look like slack, every ide has to look like vscode, now every agent workspace has to look like the codex app? codex app, antigravity, and now this all have the exact same UI design...
Thought I'd give it a try and installed the latest version. Application crashes at startup on Linux (Wayland) with: "The window terminated unexpectedly (reason: 'crashed', code: '139')".
Probably yet another instance of developers mostly testing and doing quality assurance on macOS/Windows.
Hey, sorry about that! Some AUR packages share cursor in a way that isn't forward+backwards compatible across releases. We recommend using our official AppImage from https://cursor.com/download Alternatively, please use a different AUR package that doesn't have these issues https://aur.archlinux.org/packages/cursor-nightly-bin
I prefer to avoid AppImages if I can but I gave it a try anyway and it still fails with exactly the same error. What made you think it's just a packaging issue?
I really dislike this push away from augmentation and towards agents. I get that people want to be lazy and just have the LLM do all of their work, but using the AI as an augmentation means you are the driver and can prevent it from making mistakes, and you still have knowledge of the codebase. I think there is so much more we could be doing in the editor with AI, but instead every company just builds a chatbot. Sigh.
Good for you! Personally waiting for one agent to do something while I shove my thumb up my butt just waiting around for it to generate code that I'll have to fix anyway is peak opposite of flow state, so I've eagerly adopted agents (how much free will I had in that decision is for philosophers to decide) so there's just more going on so I don't get bored. (Cue the inevitable accusations of me astroturfing or that this was written by AI. Ima delve into that one and tell there was not. Not unless you count me having stonks in the US stock market as being paid off by Big AI.)
Curious to know more about your work:
Are your agents working on tangential problems? If so, how do you ensure you're still thinking at a sufficient level of depth and capacity about each problem each agent is working on?
Or are they working on different threads of the same problem? If so, how do you keep them from stepping on each other's toes? People mention git worktrees, but that doesn't solve the conflict problem for multiple agents touching the same areas of functionality (i.e. you just move the conflict problem to the PR merge stage)
It's easier when I have 10 simple problems as a part of one larger initiative/project. Think like "we had these 10 minor bugs/tweaks we wanted to make after a demo review". I can keep that straight. A bunch of agents working in parallel makes me notably faster there though actually reviewing all the output is still the bottleneck.
It's basically impossible when I'm working on multiple separate tasks that each require a lot of mental context. Two separate projects/products my team owns, two really hard technical problems, etc. This has been true before and after AI - big mental context switches are really expensive and people can't multitask despite how good we are at convincing ourselves we can.
I expect a lot of folks experience here depends heavily on how much of their work is the former vs the later. I also expect that there's a lot of feeling busy while not actually moving much faster.
Hey don’t say that too loudly, you’ll spook people.
With less snark, this is absolutely true for a lot of the use I’m seeing. It’s notably faster if you’re doing greenfield from scratch work though.
Also, I have noticed, strangely, that Claude is noticeably less compliant than GPT. If you ask a question, it will answer and then try to immediately make changes (which may not be related). If you say something isn't working, it will challenge you and it was tested (it wasn't). For a company that is seems to focus so much on ethics, they have produced an LLM that displays a clear disregard for users (perhaps that isn't a surprise). Either way, it is a very bad model for "agent swarm" style coding. I have been through this extensively but it will write bad code that doesn't work in a subtle way, it will tell that it works and that the issues relate to the way you are using the program, and then it will do the same thing five minutes later.
The tooling in this area is very good. The problem is that the AI cannot be trusted to write complex code. Imo, the future is something like Cerbaras Code that offers a speed up for single-threaded work. In most cases, I am just being lazy...I know what I want to write, I don't need the AI to do it, and I am seeing that I am faster if I just single-thread it.
Only counterpoint to this is that swarms are good for long-running admin, housekeeping, etc. Nowhere near what has been promised but not terrible.
The only way I can imagine needing to run multiple agents in parallel for code gen is if I’m just not reviewing the output. I’ve done some throwaway projects where I can work like that, but I’ve reviewed so much LLM generated code that there is no way I’m going to be having LLMs generate code and just merge it with a quick review on projects that matter. I treat it like pair programming where my pair programmer doesn’t care when I throw away their work
I spend that time watching it think and then contemplating the problem further since often, as deep and elaborate as my prompts are, I've forgotten something. I suspect it might be different if you are building something like a CRUD app, but if you are building a very complicated piece of software, context switching to a new topic while it is working is pretty tough. It is pretty fast anyway and can write the amount of code I would normally write in half a day in like 15 minutes.
I don't know how you'd manage a "swarm" of agents without pre-approving them all. When one has a diff, do you review it, and then another one comes in with an unrelated diff, and you context switch and approve that, then a third one comes in with a tool use it wants to do... That sounds absolutely exhausting.
This is in no way comparable to the "flow" state that programmers sometimes achieve, which is reached when the person has a clear mental model of the program, understands all relevant context and APIs, and is able to easily translate their thoughts and program requirements into functional code. The reason why interrupting someone in this state is so disruptive is because it can take quite a while to reach it again.
Working with LLMs is the complete opposite of this.
I see how people think its more productive, but honestly I iterate on my code like 10-15 times before it goes into production, to make sure it logs the right things, it communicates intent clearly, the types are shared and defined where they should be. It’s stored in the right folder and so on.
Whilst the laziness to just pass it to CC is there I feel more productive writing it on my own, because I go in small iterations. Especially when I need to test stuff.
Let’s say I have to build an automated workflow and for step 1 alone I need to test error handling, max concurrency, set up idempotency, proper logging. Proper intent communication to my future self. Once I’m done I never have to worry about this specific code again (ok some error can be tricky to be fair), but often this function is just practically my thought and whenever i need it. This only works with good variable naming and also good spacing of a function. Nobody really talks about it, but if a very unimportant part takes a lot of space in a service it should be probably refactored into a smaller service.
The goal is to have a function that I probably never have to look again and if I have to do it answers me as fast as possible all the questions my future self would ask when he’s forgotten what decisions needed to be made or how the external parts are working. When it breaks I know what went wrong and when I run it in an orchestration I have the right amount of feedback.
As others I could go very long about that and I’m aware of the other side of the coin overengineering, but I just feel that having solid composable units is just actually enabling to later build features and functionality that might be moat.
Slow, flaky units aren’t less likely to become an asset..
And even if I let AI draft the initial flow, honestly the review will never be as good as the step by step stuff I built.
I have to say AI is great to improve you as a developer to double check you, to answer (broad questions), before it gets to detailed and you need to experiment or read docs. Helps to cover all the basics
I feel like this design direction is leaning more towards a chat interface as a first class citizen and the code itself as a secondary concern.
I really don't like that.
Even when I'm using AI agents to write code, I still find myself spending most of my time reading and reasoning about code. Showing me little snippets of my repo in a chat window and changes made by the agent in a PR type visual does not help with this. If anything, it makes it more confusing to keep the context of the code in my head.
It's why I use Cursor over Claude Code, I still want to _code_ not just vibe my way through tickets.
It's a very tough spot they're in. They have a great product in the code-first philosophy, but it may turn out it's too small a market where the margins will just be competed away to zero by open source, leaving only opportunity for the first-party model companies essentially.
They've obviously had a go at being a first-party model company to address this, but that didn't work.
I think the next best chance they see is going in the vibe-first direction and trying to claim a segment of that market, which they're obviously betting could be significantly bigger. It's faster changing and (a bit) newer and so the scope of opportunity is more unknown. There's maybe more chances to carve out success there, though honestly I think the likeliest outcome is it just ends up the same way.
Since the beginning people have been saying that Cursor only had a certain window of time to capitalise on. While everyone was scrambling to figure out how to build tools to take advantage of AI in coding, they were one of the fastest and best and made a superb product that has been hugely influential. But this might be what it looks like to see that window starting to close for them.
It’s the “why can’t Facebook just show me a chronological feed of people I follow”. Because it’s not in their interests to do so.
It's a very tough spot they put themselves into. If the goal wasn't to get filthy rich quick it would probably be possible to make a good product without that tough spot.
A company makes a popular product customers like, but to satisfy the VCs the company must make a product the customers don’t like but could make the VCs more money.
Not sure this is the “invisible hand” Adam Smith had in mind.
IDEs seem headed in the same direction. Seriously, watching Codex rip apart binaries in parallel and Claude go from nothing to app in one prompt, I’m pretty sure there’s no need for me to look at any code. I’m fine using tools that just emit machine code if that’s more efficient.
I kind of like the story of how Malthus had his theory of societal collapse because he couldn’t imagine a better system then mercantilism. That societies would rather collapse then to end their colonial monopolies.
I see a similar theory today with around depopulation, that as society gets older and relatively fewer working age people there are, that society would rather collapse then we find a better system then Capitalism.
This is the issue with something like Cursor and VC's funding because I feel like these private equities essentially seek rent in their own way by hollowing out the products from within, to maximize profits without doing efforts in a manner very similar to Rent-seeking, and most large companies also feel like a rent-seeking on the monopoly that they establish (like google or facebook)
I have made someone who was communist/socialist agree to georgism and I have had someone who was extremely capitalist agree to georgism, and to be honest, whether it be georgism or anti capitalism or socialism, I think that the world just wants a system where a person is treated with dignity within the economic cogs.
My opinion is that as long as we can all agree on the last premise about dignity for individuals within the economic cogs, we can all have meaningful conversations to make that a premise, hopefully a reality.
(I feel like the people who might deny dignity to people within this particular context, have either a bias/incentive to not look towards the problem, or are uninformed, or lack the energy to fight towards change within the system, and more importantly the _hope_ that the future can be better)
I am not hopeful about the current political systems (even around the whole-world at times), I feel like there should be more information and decentralization within politics.
Essentially, politics really just feel unaccountable to me, your vote really stops mattering to politicians if/when money starts talking. But technically, this system can be broken through with enough votes.
I really hope for a future where politics and politicians feel accountable and genuine, maybe even someone from down the street who we can have some chats with to actually know them.
Ironically or unironically, just as how the landlords pushed against Georgism/George within really making political difference, The same is happening right now as well where Online landlord monopolies dictate how people interpret and vote by using their algorithms/influence.
Politics like many other problems feel like a chicken and egg problem, like things work until they don't and things don't work until they do. At a more individual level, stepping outside of most algorithms and the reason why I joined hackernews is for doing something like this, myself.
Sometimes u need the beef of opus but 80% composer is plenty.
Now you’ve got me thinking I should give composer another go because speed can be pretty darn great for more generic, basic, tasks.
I thought there was an entire initiative to build their own coding model and the fine tunes of in Composer 1.5 and Composer 2 were just buying them time and training data
God forbids you make a great product in a specific niche and are happy with the money flowing.
Nope, has to be more.
This is why Zed's direction felt pretty strong to me. Unfortunately their agentic features are kind of stagnating and the ACP extensions are riddled with issues.
I'm hoping in this new UI in v3 I can still get that experience (maybe it's just hidden behind a toggle somewhere for power users / not shown off in the marketing materials).
> I wish they'd keep the old philosophy of letting the developer drive and the agent assist. Even when I'm using AI agents to write code, I still find myself spending most of my time reading and reasoning about code.
We very much still believe this, which is why even in this new interface, you can still view/edit files, do remote SSH, go to definition and use LSPs, etc. It's hard to drive and ship real changes without those things in our opinion, even as agents continue to get better at writing code.
> I'm hoping in this new UI in v3 I can still get that experience (maybe it's just hidden behind a toggle somewhere for power users / not shown off in the marketing materials).
This new interface is a separate window, so if you prefer the Cursor 2 style, that continues to exist (and is also getting better).
That's good to hear, I might have jumped a little too quickly in my opinion. It's a bit of a Pavlovian response at this point seeing a product I very much love embrace a giant chat window as a UX redesign haha.
I would love to see more features on the roadmap that are more aligned with users like us that really embrace the Cursor 2 style with the code itself being the focal point. I'm sure there's a lot you can do there to help preserve code mental models when working with agents that don't hide the code behind a chat interface.
I dont think there is an inbetween. Its really hard to 'keep an eye' on code by casually reading diffs. Eventually it will become vibe coding.
Software engineers are deluding themselves with spec driven, plans, prds whatever nonsense and thinking its not vibecoding.
Reading diffs is an inescapable skill, needed for evaluating any kind of PR. This just makes it more interactive.
I just use Copilot with VS Code, but my flow is to just ask Claude to make a change across whatever files it needs to touch, then either accept the changes, edit the changes directly, or clarify whatever was different from my expectations.
Reading diffs is central to how I work with these agents.
That's basically it. You can review changes afterwards, but that's not the main point of Claude Code. It's a different workflow. It's built on the premise: given a tight and verifiable plan, AI will execute the actual coding correctly. This will work, mostly, if you use the very best models with a very good and very specific harness.
Cursor, same as Copilot, has been used by people who are basically pair programming with the AI. So, on abstraction down.
I have no idea what is better, or faster. I suspect it depends at least on the problem, the AI, and the person.
This is not really true anymore.
Cursor has better cloud agents than Claude. The multi-agent experience is better, the worktree management is better. Tagging specific code or files in chat is better.
It's hard for me to express the level of pain and frustration I feel going from Cursor to Claude / Conductor+Claude / Claude Extension for VS Code, Claude in Zed, etc.
Really hoping Claude puts more energy into Cowork as a competitor for Cursor and Codex.
Instead they are defining architecture through specs and verification-loops and attempting to one-shot solutions fitting clear tests. On reflection, I personally don't have many prompts with CC referencing files or code directly, rather I speak in specifications I can then track to a given instance of work in review.
This isn't to suggest you can't work at this abstraction in cursor or w/e interface, but the features you suggest are hardly relevant to the divide zwaps is identifying.
We have OG Cursor for the micromanagers (who want to approve/deny every line) and things like Claude Code for those who are less picky about the how, and able to be amazed at what it creates.
I'm not telling you to go use cursor, just to help clarify that you can drive both solutions with the exact same approach and skillset and get very similar results - the difference is the UI. I personally like being able to paste screenshots into the agent, etc.
Then when you go back to Cursor it will still support all of those things in the settings.
Using Cursor you tend to not think about those as much since Cursor does a lot of it for you as part of the IDE integration. But it's good to refine it your own way.
But for the most part there isn't much difference.
Zed plus Claude feels more like using isolated browser extensions instead of something part of the browser (unless you pay for Zeds AI thing then the integration is marginally better).
Compile code is not perfect also. But who does hand-written assembler anymore? Yes, LLM is another layer, it would be ugly and slower but it’s much faster to use.
In both cases, it's your processes (automated testing, review, manual QA) that is the bulwark against bugs and issues.
With AI, you can set up great processes like having it check every PR against the source code of your dependencies or having it generate tests for what's an intermediate step or ephemeral solution that you would never write tests for if you had to do it yourself.
There's this idea on HN that if you delegate too much to AI, you get worse code. Presumably not appreciating all the code-improving processes you can delegate to AI, particularly processes you were never doing for hand-written code.
Now we have 3 ways of coding:
* vim / emacs - full manual
* VSCode / IntelliJ - semi-automatic
* ClaudeCode/Codex/OpenCode/... - fully automated
Cursor can't stay in between
Are you saying they can’t compete with VS Code in the semi-automatic space?
I thought it was primarily a user of Anthropic and OpenAI APIs, so the fewer tokens you use to accomplish a task, the higher their margin.
These models are infinitely more effective when piloted by a seasoned software engineer and that will always be the case so long as these models require some level of prompting to function.
Better prompts come from more knowledgeable users, and I don't think we can just make a better model to change that.
The idea we're going to completely replace software engineers with agents has always been delusional, so anchoring their roadmap to that future just seems silly from a product design perspective.
It's just frustrating Cursor had a good attitude towards AI coding agents then is seemingly abandoning that for what's likely a play to appease investors who are drunk on AI psychosis.
Edit: This comment might have come off more callous than I intended. I just really love Cursor as a product and don't want to see it get eaten by the "AI is going to replace everything!" crowd.
And management everywhere is convinced that thats what they are paying for. My company is replacing job titles with "builder". Apparently these tools will make builder out of paper pushers hiding in corporate beaurcarcy. I am suddenly same as them now per my company managment.
You and I want this. My EMs and HoEs and execs do not. I weep for the future of our industry.
Ignoring the fact that software will just keep getting more and more complex and interconnected... There will always be a new frontier or code and UX
That's because that's exactly where we're headed, and it's fine.
With Claude Code, I use Gitlab for reviewing code. And then I let Claude pull the comments.
It looks like the new UI has a big focus on multiple agents. While it feels wrong, the more you split up your work into smaller merge requests, the easier it is to review the work.
Chat first is the way to go since you want the agent busy making its code better. Let it first make plans, come up with different ideas, then after coding let it make sure it fully tests that it works. I can keep an agent occupied for over a hour with e2e tests, and it’s only a couple hundred lines of code in the end.
If you've dug in sufficiently on plan mode, then what the agent is executing is not a surprise and shouldn't need input. If it does, the plan was insufficient and/or the context around the request (agents.md, lessons.md, or whatever tools and documents you use ) weren't sufficient.
EDIT: Maybe it doesn't work in cursor, but I continue to use vscode to review diffs and dig in on changes.
We needed that jump, there were still floppy disk icons
I get the temptation of letting agents do everything. But they create really bad systems still (bad architecture, reimplementation of solved problems etc).
I also get the temptation for beginners and think it’s great that more people are empowered to build software but moving entirely to chat means they won’t learn and level up which in the long run limits their ability.
I could be wrong. And my way of thinking is dying but thankfully I can build the tool I want.
The biggest downside for me with Cursor was losing access to gated Microsoft extensions like Python and C#. Even when vibing there are times you will still need a debugger or intellisense.
I note in the comments lots of people saying they are moving back and this latest move looks like the final nail in the coffin for Cursor.
2. Cursor's UI allows you to edit files, and even have the good old auto-complete when editing code.
3. Cursor's VSCode-based IDE is still around! I still love using it daily.
4. Cursor also has a CLI.
5. Perhaps more importantly, Cursor has a Cloud platform product with automations, extremely long-lived agents and lots of other features to dispatch agents to work on different things at the same time.
Disclaimer: I'm a product engineer at Cursor!
Cursor was the tool you use to pair program with AI. Where the AI types the code, and you direct it as you go along. This is a workflow where you work in code and you end up with something fundamentally correct to your standards.
Claude Code is the tool you use if you want to move one abstraction layer up - use harness, specs, verifications etc. to nail down the thing such that the only task left is type in the code - a thing AI does well. This is a workflow where the correctness depends on a lot of factors, but the idea is to abstract one level up from code. Fundamentally, it would be successful if you don't need to look at code at all.
I think there is not enough data to conclusively say which of these two concepts is better, even taking into account some trajectory of model development.
I do feel that any reason I have for installing Cursor is that I want to do workflow 1, rather than workflow 2. Cause I have a pretty comprehensive setup of claude code (or opencode, or whatevs) and I think it does everything you list here.
So, as a product engineer, you probably wanna mention why it matters that Cursor UI allows you to edit files with auto-complete.
Cursor is an IDE and an agentic interface and a cli tool and a platform that all work locally and and in the cloud and in the browser and supports dozens of different models.
I don't know how to use the thing anymore, or what the thing actually is.
I applaud Cursor for experimenting with design, and seeing if there are better ways of collaborating with agents using a different type of workspace. But at the moment, it's hard to even justify the time spent kicking the tires on something new, closed source and paid.
And I would happily pay a seat based subscription fee or usage fees for cloud agents etc on top of this
Unfortunately very locked into these heavily subsidized subscription plans right now but I think from a product design and vision standpoint you guys are doing the best work in this space right now
You just add this to your ~/.claude/settings.json:
For $20 a month, I can plan and implement thousands of features using Composer 2 or Auto with Cursor. The usage limits are insanely higher. Yes, the depth of understanding is not Opus 4.6, but most work doesn't need that. And the work that does need it I pass to Claude.
I can code 8 hours a day using LLMs as my primary driver spending just $40 a month.
I am not saying this in bad faith. Model companies cannot penetrate every niche with the same brand recognition as some other companies you would consider as "API resellers" do.
Every company I've worked at has still had a few engineers who insist on working exclusively in the CLI with vim/emacs prior to AI. Every other engineer used some flavor of a desktop app ranging from more minimal editors to incredibly complex IDEs. I expect we land back on UIs long term.
These are features I am sure Codex will soon have, of course.
Then there is the advantage of multiple models: run a top level agent with an expensive model, that then kicks of other models that are less expensive - you can do this in Claude Code already (I believe), but obviously here you are limited to something like Haiku.
One of my favorite startups and I genuinely like to keep subscribing to them.
The next generation of interfaces are not going to look like an evolution into minimalist text editor v250. This is like people iterating on terminals before building native or web applications.
Still curious which ones will survive when the AI gold diggers finally settle.
If MS ever decided to discontinue VS Code or relicense it, there would be blood in the water. I guarantee you there would be multiple compelling competitors in under a year and probably a new open source winner with consolidation in 5.
So to answer your question: they would be forking Atom (which I think would’ve won otherwise).
Not saying there aren't people who care, there are, but they are a small minority.
https://cursor.com/cli
IMO sounds like natural foundation for Cursor
I haven't used it in a decade, Im sure it has has evolved
Nowadays I'm basically a Neovim purist, but I have positive memories of it. I'm kind of afraid to revisit it at this point, though, since everyone hates on it and I suspect I wouldn't like it as much.
Can’t say I miss eclipse, but a lot of the VSCode extensions seems to utilize old legacy eclipse stuff and has the bugs to match.
If not you really should. IntelliJ with Java is one of the best dev experiences I've ever had. I'm a VSCode fan for most other things but for Java I wouldn't even remotely consider using it over IntelliJ if I had the option :-)
This change is possibly too big and unless all my existing usage patterns are maintained or improved, I’ll likely give CC a try now. Not optimistic.
I should also add: Cline doesn't require any account at all. I just installed the extension and added my OpenRouter API key and that was it.
[0] https://shittycodingagent.ai/
The biggest concern is that if you want to use SOTA models I don’t see how they can match what you get with the subscription plans of Anthropic and Open AI, whether your spending $20 or $200 a month.
Even if they could match what you get in terms of token quantity, they are giving their tools away for free for the foreseeable future and Cursor is not.
I totally preferred the other way, but at some point , there is boiler plate or organizations you just want done and it does not make sense to put you waiting minutes a time to confirme few refactors. That literally killed the vibe for cursor to me
However, is this really a moat?
I think it's a really solid release, and while cursor seems to have fallen out of the "cool kids club" in the past three months it remains the most practical tool for me doing AI-first work in a large production code base. The new UI works better in a world where agents are doing most of the work and I can hop back into the IDE interface to make changes.
We've set up a linear integration where I can delegate simpler tasks to cloud agents, and the ability to pick that work up in cursor if I need to go back in forth is a real productivity boost. The tighter integration with cloud agents is something I've been hoping for recently.
I appreciate not being tied at the hip to one model provider, and have never loved doing most of my work from the command line. I was on vs code + meta's internal fork of it for years prior to the current AI wave, so that was a pretty natural transition. I'm pretty optimistic on cursor's ability to win in the enterprise space, and think we're going to see open source models + dev tools win with indie devs over things like claude code as costs start getting passed down more and the gap between frontier models and open source gets tighter.
https://github.com/rbren/personal-ai-devbox
Nerve wreaking race.
I think I'll switch over to cursor on trial basis.
What's the pitch for using Cursor now a days?
I don't know what you're talking about. My experience with Cursor (before this new v3) is that new Cursor agent tabs / cloud agents already intelligently manage worktrees to prevent conflicts.
[0] https://forum.cursor.com/t/working-with-worktrees-in-cursor/...
Cursor's inline autocomplete is very good though, much better than anything I could reproduce in Zed with various 3rd party "edit" LLMs (although checking google, they announced a new model since I tried it https://zed.dev/blog/zeta2)
But are they affordable already for developers who don't earn a Silicon Valley salary? Developers in 3rd world countries?
Your workflow is probably closer to what most SWEs are actually doing.
The only way you're going to let an agent go off on its own to one-shot a patch is if your quality bar is merely "the code works."
Agents today can generate solid code even for relatively complex requirements. However, they don't always make the right trade-offs.
Just because something works doesn't mean it scales. It doesn't mean it can handle unexpected user input. It doesn't mean it's easily extensible.
Today engineers really just need to define those high-level technical requirements.
At least within our company, this is quickly becoming what it means to be a software engineer.
Is "Cursor 3" == Glass? I get they feel like their identity means they need to constantly be pushing the envelope in terms of agent UX. But they could stand to have like an "experimental" track and a "This is VS Code but with better AI integration" track.
These AI companies are running out of ideas, and are desperate. I can't imagine investing in companies that are 3 month behind open source alternatives, and their target audience being the most experimental kind there is.
Looks pretty though.
Apparently if launched with --verbose it works, but that's the same crash I was seeing without the verbose flag
At least before they were tangentially still an actual developer tool, standard vsc windows, the code was the point etc.
Now they offer really nothing interesting for professionals.
That's a curious statement given that what they're doing is just becoming more like Claude Code, which seems extremely popular on this forum.