Posted by FinnLobsien 5 hours ago
I want them to succeed, but can't see how they can win in this agentic world, where people are barely opening their IDEs, and living in agents terminal whole day and from time to time opening a PR to get a review
The volume of code is only increasing. Problems that used to be solved with a few lines (or even removing lines) are now solved with changes across 10 different files.
My IDE is invaluable for me when trying to navigate this stuff. I am completely convinced that people reviewing LLM code only via PR diffs are consistently missing massive systemic issues. You *need* to review changes in context of the surrounding code.
And doing this without an IDE is just making your life harder for no reason.
We have put a lot of thought into how we can add value for the workflow you just described. Here's the way I think about it: FSD for a car. Can you imagine trying to sell a car without a steering wheel? I still want to code, I still want the deterministic side, even in cases where I'm letting the machine do the work.
But the choice of AI providers was very limited, and I found no ways to add the ones I use (the only options for providers were Anthropic, OpenAI and Google).
Any plans of opening up for more providers?
That's a small subset of the functionality you would need if you're actually authoring code. So the IDE maybe be more valuable, but the feature set you need is actually much smaller and can probably be redesigned given your narrow workflow.
Is this accurate?
Just some important things off the top of my head that I use daily:
* Instantly jumping between definitions and usages
* Folding and unfolding code blocks
* Quick overview of structures in files
* Database access, so I can browse the data and the code at the same time
Also, sometimes I just want to quickly refactor/rename some variable across all its usages or whatever. It's just quicker to do it in the IDE with a simple keyboard shortcut, rather than prompting an agent "rename oldName to newName".
Unless you are blindly vibe coding everything without an IDE to help navigate what is being generated to review the output, working would be hellish.
And if you are blindly vibe coding everything, I am just glad I don't have to work with you. While LLMs mostly speed me up, they make serious mistakes often enough that a review is necessary.
I have opened an IDE for very very few things in the last 9 months or so. Im still renewing a license at this point mostly because I like JetBrains the company and it's not a financial burden but they are in a bad place.
After using Air in PyCharm, it's some good stuff. It's re-imagined for this hybrid world. But my question for you, honest question: do you think you still want to look at code? How about write code? If so, what are some things you'd do, if you were in charge of our IDEs?
Note https://youtrack.jetbrains.com/projects/LLM/issues/LLM-17450... is still open although https://youtrack.jetbrains.com/issue/LLM-16479/Be-able-to-ac... appears to be closed as completed now
Yes, we're writing less code, but the job of "a programmer" has always been about so much more than _writing_ code. It's about code reviews, QA, debugging, discovery, and maintenance; much more than the "writing" part. Anyone with production code will agree, I think: the writing part is the easy part. That's where I still very much prefer a proper IDE over a text editor or terminal. I want easy code navigation, I want proper static analysis without relying or waiting on third party tools, I want a visual overview of changes that happened. I want to stay in charge of whatever is writing my code. On top of that, I want my agents to have access to the tools that have made me the most productive developer for years.
Curious about your thoughts?
AI can do all of that quite well. If you take your hands off of the wheel, lean into it, start building systems for it, start figuring out where it goes wrong on your codebase and building systems to prevent that... you might get 10x more done. I do, and the results are generally better than anything I'd have coded by hand, because there was never enough time.
I don't look at or try to grok all changes, though. I couldn't get near as much done, otherwise, and I'm not convinced that 1 hr of me looking over it is better than 4 hrs of AI looking over it, and the AI is far cheaper. I look over the code, occasionally, to look for code that's objectively terrible, but I don't fix it directly - I use it as a test-case for improving my AI stack, to make it able to find and fix or prevent that class of terrible.
I probably spend 9 out of 10 credits on review and refinement, not writing fresh code. In spite of that, I'm still producing things way faster than I ever did by hand, with only a single $200/mo Codex plan.
I still keep my JetBrains' subscription, so that agents can use the MCP. I probably spend <5h per month in an IDE myself, though.
Of course, but agents are also opening up new possibilities to do these jobs.
For example, you can do quick first code review with agent, or you can ask agent to add couple of print statements and debug the code until it figures out the issue, also for discovery, you can do discovery with agents and generate nice output with the details you want (although not accurate sometimes)
As a JetBrains user myself, whose organization cancelled the IntelliJ license recently, I can attest things I am doing inside IntelliJ is getting less and less. I am primarily using IntelliJ for code navigation at the moment, which is still the best one
I suspect we won't for several years yet - and it's likely going to be way more integrated/specialized for specific technology than our previous ides were.
But you're unlikely to go back once you've the full stack integrated with full observability, release pipeline, system integration tests etc pp
Still, it needs way more integration to work well and not be DOA, hence this is still far out I think.
But there's still a group that needs to own the result, who still read code and write code. Folks who still believe in developers and development. This group might not be loud in SV, but they are big out in the world.
I'm a Python advocate at JetBrains and our team runs into folks like this all the time.
I've been enjoying using Air inside the IDE as a way to get the acceleration without giving up coding. I get the speed side from the agent and the trust side from the deterministic IDE.
Not to mention all the lies about the new UI adoption and when faced with backlash, they put the old UI into a plugin that "might" receive updates and support in the future.
I actually did try their latest IDE a few months ago to see if their AI integration was worthwhile. There appeared to be no functional way to use a local LLM. You can set up OpenCode as a provider, but any time you click anything it resets your model to one of the free API services, with no discernable way to change the default.
I don't have high confidence in Air. So far JetBrains has not shown they know what a good AI integration is like. Frankly I don't think they know who their customers are or what they want.
For my part, I'm still using 2024.1 versions on my perpetual fallback license. I had Claude slap together an MCP plugin so now I have my own AI integrated IDE that actually works and functions as a usable tool. All my buttons have text labels instead of hieroglyphs and padding.
I’ll try out Air but I don’t have high hopes, my current workflows work great and I’m still not sold on what Air offers even after reading this post and the page about Air.
But I have been using their jetbrains remote system, and you can clearly see how incredibly high the RAM usage is on both server and client.
Personally I am also looking at the door, but not because of the resource usage… but because they keep putting resources into things which aren’t their core offering and it makes me wonder if they have run out of innovation in that area.
… then again, they did recently figure out how to make development on a Windows remote work… so I can keep using my linux machine while developing Windows software, so I think that earned a lot of goodwill from me.
I do feel however they have been neglecting anything non-AI for a while now. I in particular loved their auto complete, and with the new guesswork AI auto complete it's just not the same. I want deterministic behaviour from my IDE.
On the LLM side, I couldn't use the Anthropic subscription with their offerings, so I just use it from a harness outside the IDE, and switch to the IDE for review. This may not be JB's fault, but it still makes their offering useless to me.
I may be one of the few remaining but I also still program myself sometimes (I weigh it based on the task). The IDE has been pushing so much AI it has been getting in my way sometimes, and it is definitely using more resources for stuff I don't use.
I too am on the fence on whether I'll resume my subscription. Luckily for JB I dearly dislike VSCode's code navigation and interface, and would severely miss their debugging tools integration.
I'm happy about our Air launch because we're being AI "realists"...neither too much sunshine pumping nor too much skeptic. We're doubling down on developers and development, both when you type and when you don't.
On to the point..I can promise the PyCharm is doing some pretty massive non-AI stuff on the deterministic side. I think determinism is the best path to trusting the code generated by the box.
We tried to make this work, but unfortunately hit a hard stop with Anthropic: their current policies do not allow us to support this integration :c
We still hope to resolve this in the future
About the terminal, honest question...is that because you are looking at code less and less? Writing code less and less?
I know harnesses (and harness harnesses) have a long way to go and lots of improvements even without models getting “smarter”, I’m rooting for JetBrains but so far IDEA has been more of hindrance in the LLM-era. The CLI TUI’s have become the most important tool I have and the GUI versions haven’t kept pace.
I’d love to live in IDEA instead of the CLI but so far I’ve found that to be too limiting and not as adaptable to our rapidly changing field. Maybe Air will help with that, we will see.
I want something like herdr in IDEA, a way to monitor agents, smart worktree integration, easily visualize changes and use the power of the InteliJ brains to assist in coding (lint/checks/etc).
Right now my IDE gets more in my way than assists me in shipping code.
I don't think that there's a better alternative at the moment, and I get the need to pump out features to keep the revenue stream going, but I miss the slimmer versions
There's so much going on right now, and the push to PaaS is not performing as well in my opinion.
Take a look at how Herdr got viral, and how it exposes protocols that let others come and build with it. Multiplexers are a good bet, though there are plenty of other verticals out there too.
I would say try to chase another ReSharper moment for you guys.
Their IDEs have been nosediving for years now in a way usually reserved to "has been acquired by private equity". But they have not been acquired by private equity as far as I can tell.
I'm writing this from my laptop, where each keypress in a .md file in WebStorm makes the fan spin up (not hyperbole).
What happened, JetBrains?
Most development was done in Russia (disregard tales sold to the West about them supposedly being a Czech company, they were Czech mostly in the name, real work happened elsewhere). Exited Russia in the beginning of 2022, leaving many talented people behind: not everyone wanted or could leave because of families, obligations, personal beliefs.
Remembering where most of JB's money was coming from, can't really say they had a choice.
They will bring a 9950x with nothing else running to a crawl. Then complain they are out of memory. (When it is obviously the IDEs themselves running the system out of memory)
I <3 JB IDEs for their functionality, but resource pigs they are.
The answer is more like the obvious one...our category is getting redefined, at hyper-speed, and all of the players are all over the map trying to sort this out.
I like what we're doing with JetBrains Air. We know what we bring to the table, who it's for, and we're doubling down on developers and development.
That said, CPU fan on Markdown is not cool. Want me to look through it with you?
Honestly not really.
I mean I am happy that there is some kind of response, but this is just damage control.
And, frankly, you do not need me to fix that. Of that I am certain.
I'm not running any weird setups or something out of the ordinary. If you actually test your software on a normal computer (think: random thinkpad with 8th gen intel cpu), running some random boring linux distribution (think: debian) you will run into these problems.
The bad experiences all so far have been smoke-test-level issues.
Or.. well. Product strategy. Not even bugs or performance issues at all.
__
But, yeah, you had to say something. I understand that, given your role as "developer advocate" (as per the HN account about section). This is your job after all.
But, honestly, sticking ones head in the sand and hoping that this passes might've been the strategically more wise move.
The problems your employer has cannot be fixed by one person taking their time and looking at an individual issue some other vocal guy on the web has. I think we both know that.
I'm sorry, genuinely.
- This is a real problem with JB IDEs
- It is not specific to an individual's setup
- It has existed for many years
- Anyone at JB who actually is interested in diagnosing or fixing it does not need to extract details, support tickets, or logs from users, as it's trivial to reproduce on any system.The OP post feels like completely marketing department, no skin in the game. I'm afraid they are becoming Borland.
What else can one use?
Excluding VSCode and all those AI-first IDEs I mean. JetBrains' git features have been excellent. I have not yet seen them be reproduced anywhere, but, eventually, they will kill that classic UI plugin. :(
I find this setup works great for actually participating in the development. To understand what is happening and keep the mental model of what the project is about. Review in progress work in IntelliJ, use the Git, Database and REST(.http files) in the IDE as the agents generate the code.
I don't see a problem with this setup is it really that uncommon? Have people stopped using TOOLS completely? That said I don't have much interest in these JetBrains AI products, at least not currently.
What is the problem with using Codex in Rider, can you please share? It's generally available, with your own Codex subscription or via JetBrains AI.
DataGrip is seriously under-rated, IMO.
For a while I used Junie for small refactorings, and it generally does a good job, quite quickly. A couple of times I tried asking it for more complex things, and it didn't do well. I do find the UI to be a bit tricky, so it's quite possible I did something wrong or didn't add something to the context I was supposed to. Hopefully they sort out the "AI Chat" panel soon.
But now I use Codex/ChatGPT... and what it can do is frankly incredible. And it might just be me, but I do like having my AI chat UI completely separate from my IDE.
edit: Sorry; misread. No, and I am very confused by their diaspora of AI offerings!
- Natural treatment of multi-file projects
- Introspection
- Refactoring
- Keyboard shortcuts, specific types of search etc
- The build/run windows
Falls at the first hurdle...