Hacker News Reader

Comment by sippeangelo | Hacker News Reader

That's a hefty payday for a model that barely functions! Every time I run out of API credits and get kicked back to Composer 2 I feel like I'm better off just packing up for the rest of the month.

I feel like we're finally at a point where you don't have to constantly argue with and constantly babysit coding models, which makes it even more frustrating when you're suddenly forced to deal with one that ignores your instructions and gets stuck in thinking loops again.

I suspect it's the vast troves of training data rather than any tech that Cursor possesses that SpaceX is after...

View on HN

Cursor is still the best coding environment and hardness. It's actually not really close. They are so good that they actually made Gemini usable.

The problem is they can't compete with Anthropic and OpenAI because they can't sell Opus and GPT at a discount to subscribers like OpenAI and Anthropic do with their subscriptions.

So they either need to build a competing model or slowly die.

Bet they will become tied to grok pretty soon.

They seemed to be doing fine with Kimi distillation. Not speaking from experience though, I prefer to use my editor.

I personally disagree on the first point. Claude code in a terminal with vim is much nicer. I just don’t see the need for the bloat of an IDE when the CLI versions work so damn well now.

And Claude can use CLI too. It's the perfect environment for coding agents.

They have Cursor CLI.

Cursor is essentially all the Claude Code products but without the horrible bugs of Claude Code products.

You can transfer from CLI to web and it actually works.

> They are so good that they actually made Gemini usable

I think Gemini is best model out there, and it's not Cursor who you should praise. I use it with jetbrains junie. Vastly cheaper than claude, faster, produces better quality code, actually listens to your instructions, more accurate. I'm sure claude code cli has some cli magic that I'm missing out on, but having everything just work in a nice IDE (and llm to actually understand your symbol table) is like magic.

Are you using Gemini 3.1 Pro? Subscription or paying for the tokens?

Tried 3.1 pro preview today a little bit, definitely blowing thru credits quicker, not sure about being better quality, but achieved all tasks perfectly.

IDK how Junie does it, but I spend less than $50 USD per month and I'm on it 30 hours per week.

That's why I'm so puzzled to why Composer doesn't work better when they have the ability to train it from scratch for their agent harness! Yet it still fails to apply edits, gets confused why it can't call some commands in its sandbox, the list goes on...

I doubt they're buying it for Composer, I imagine they're buying it for the agent harness. It's arguably the best non-Anthropic agentic coding harness, and you get _all the models_ for one subscription price.

JetBrains is crying in the corner...

Jetbrains has gone so far downhill

I honestly can’t believe how poorly JetBrains has done. I used to love PyCharm but now it’s so far behind. I still use DataGrip but it is absolute dogshit when it comes to agentic coding.

I was a massive jetbrains fan - still believe it's the best IDE even with it's massive performance issues.

But I just... barely use an IDE anymore. I think I have the lowest possible subscription price for "all products" you can have (at least as an outsider) and I think I'm going to cancel this year. I've been paying for a decade+

I am subscribed to their all you can eat plan and use their Junie coding agent which is included with subscription with some free tokens. I then pay for extra tokens on on-need basis and all works like a charm. So far I pay (well my clients do as I bill separately for that) about $100 a month to cover my current coding needs. All works as a charm. I mostly use their CLion, Webstorm and PyCharm IDE's for development, sometimes other as well. All in all dev experience is excellent and far exceeds that of Cursor I was trying to use for a while.

Not sure what problems people here have with JetBrains offerings

Once you work somewhere that gives you unlimited opus 4.6 and learn how to use it properly, your perspective of what you should be doing day to day shifts.

Honestly unlimited codex with 5.4 high has a similar effect.

SOTA models + harnesses used together is very different than it was 6 months ago. People that have significant software engineering experience can get so much done it's scary.

I have what you call "significant engineering experience", decades of it to be precise and have designed and developed many complex products successfully used in various industries.

I do not need to "shift my perspective" since I do use agents to the degree that I need and it helps help me very much. I am way more productive with them.

Generated code is still not perfect disregarding of any particular model (I have access to all). I have to watch and fix, sometimes by supplying more precise specs, sometimes asking to rewrite piece of code in such and such manner using this and that structures.

Wasn't meant to be personal- I was using the proverbial "you".

I keep seeing what I'm referring to happen - folks are using / opening their editor less and less.

What's crazy is a developer can go on a walk and use tmux/tailscale and keep working as if they were sitting at their desk.

I keep hearing this, but I have yet to see “so much getting done” anywhere. I’d sure like to but things seem to be pretty much be business as normal.

This was absolutely the case - not actually that much more productive - until only a few months ago.

We hit some sort of tipping point between models and harnesses and people learning how to use the tools idk.

And directs / engineers / friends seem happier.

Simon Wilson recently did a podcast where he discussed his experience and it felt very familiar.

Worth listening (ignore the click bait title) https://www.youtube.com/watch?v=wc8FBhQtdsA

Unless you do Jakarta EE development, where Cursor with their simple LSP support is far, far behind. Cool for generating a bean, but when you got to debugging deployment descriptors you wish you were in IntelliJ.

I use JetBrain's all you can eat subscription that comes with their Junie coding agent which includes some free tokens to cover my coding needs. I then top up tokens on on-need basis. Costs me about $100 / month in AI tokens (well I bill my clients for that separately so do not really care bout the price). All works as a charm. I mostly use their CLion, Webstorm and PyCharm IDE's for development, sometimes other as well. All in all dev experience is excellent and far exceeds that of Cursor I was trying to use for a while.

Not sure what problems people here have with JetBrains offerings

IntelliJ is a bit dated, and its plugins are too. I use IntelliJ all the time, in its various incarnations, but vscode is really up there now.

I use both (not IntelliJ but other IDEs) and quite frankly I fund VS Code and derivatives very much inferior. For C++ development for example CLion vs VS Code (needed plugins installed) is night and day and not to the benefit of VS Code.

I know JetBrain product could be sluggish on "normal" computers however all 4 of my development machines run on 16 cores AMD with 128GB RAM. It flies in environments like that

Yeah, and it seems to be completely self-inflicted. I created a small personal skillset that explains to the agent how to use the JetBrains MCP tools for refactorings/find-usage/navigation, and it improved its performance by a lot.

Yet JetBrains tried to do everything themselves and failed :(

I've subscribed to Jetbrains all product for years. If the agent coding is going to be the next wave. Jetbrains is really behind. Even Microsoft offer better agent coding with VScode and Github copilot cli.

They’re definitely playing catch up, but the IDE integration makes interactive development really nice. Claude is good for one-shotting things, but I find JetBrains AI integration really useful for working with large codebases where I may be unfamiliar with things.

I think they’ve been caught in a bad spot. They’re a profitable company, but nowhere on the scale of Microsoft. And they don’t have billions of VC to effectively price dump. Other tools that can focus on one thing and burn cash are advancing quickly and some of them don’t really need an IDE at all.

The semi-recent introduction of ACP integration in the JetBrains IDEs has been a nice bridge. But now it’s confusing how everything comes together. I really hope they can survive.

Maybe vertical integration is the main business case.

A controlled environment to determine effort and token usage, and to get plenty of exclusive training on code.

It could end up making sense. Idk if they needed to offer 60B though.

I'm not willing to give them the benefit of the doubt. I think this is purely Elon trying to take a pot shot at Anthropic.

Cursor is great. I was using it up until recently. Then I switched oh my pi, and honestly I haven't looked back. I've also heard great things about open code.

I actually really like Composer 2. For my use case, between the planning tool, and getting it to ask a lot of clarifying questions, I regularly get very good results. I'm not doing anything complex though; mostly staying in the lane of very common web app type code.

Composer 2 is really good for me too.

It definitely feels sufficient for questions and planning, but it is surprisingly lacking in the actual coding department once you go for edits that need changes in multiple files. Which is surprising considering they should have been able to train it on their own harness!

They still just bought access to all the code you've ever fed into the model...

Cursor very reasonably had a “no retention” checkbox available to everyone, including those on free plans.

I'm sure those work as well as the "don't collect my data" checkboxes too.

I don’t think this is the case. With “accidentally” collecting an individual’s data, the company’s risk is that somebody cares enough to sue them based on vague and poorly defined damages. With “accidentally” collecting source code, you’ve not only violated your contract with 98% of your enterprise customers (many of which have dedicated legal departments) providing a very real and obvious path to lawsuits, you’ve also gained a strong reputation as a vendor never to be trusted. My employer uses cursor, and I strongly suspect we would cut ties and blacklist them at the first sign of them inappropriately retaining data.

Is Composer 2 a bad model because Cursor are bad at training models, or because they are compute constrained? This deal will provide the answer to that question.

Reply on news.ycombinator.com