This is moving in the direction towards a final interface, very good.
There aren't enough digital french cleat systems.
The true benefit of a French cleat isn't storage or organizing the way a fiddly pegboard and hanger approach uses, it's the ability to move all tools for a specific project onto a single rail without having to fiddle around, reducing friction for engaging and re-engaging with a project.
That's what it is, it's basically two boards with angled cuts to match. One mounted to the wall, the other to the thing you want to hang from the wall. No fasteners needed other than for those two attachments (or more if you use more boards). Gives you some flexibility in arranging things in a shop (or, that's my reason for using it with our 2-car garage).
Yeah, now that agents are doing most of the work, and we want to give them as much freedom as possible to run code, take screenshots, click on things, etc, it really doesn't make much sense to have a setup where all of this happens on your local computer. IMO it makes more sense to have this hosted somewhere, and then have UIs that make this visible to you on your devices.
Exactly. They sit in their little world in my homelab and I control everything through networking and virtualization restrictions. They run on a beefy VM, can spawn their own VMs (Ubuntu, Debian, Fedora, Arch, Windows, even macOS on an old mac mini), run their own browsers, access their own secrets, etc. There's no way I would go back to agentic development on my local machine anymore.
Any good writeup on how to set this up? Not that it's a rocket science, but would be good to have a bootstrap how to with good practices and a general showcase.
Not that I can find. I'm a Senior SRE so it's more or less an extension of my domain. The hardest part is really just knowing the "best" way to do something which generally just comes from experience. But most frontier agents can probably get you close enough without too much experience.
This sounds like a short-term strategy at best
You only need an agent to click things long enough to build an API on top of the thing it's clicking on. And running a full windowed virtual machine is expensive!
I'd also caution against saying things like llms are doing most of the work I don't think that's true. And where it is true it shouldn't be. These things are still new tech and they need a lot of hand holding and oversight still.
I am curious about Pi, but my experience with a seemingly more fully-featured Hermes has been endless tinkering and fixing minor issues. Have your experience been at all similar?
I've been using it without thinking much about it for 3 months now. Have yet to run into any bugs or strange behaviors. But bear in mind I'm using oh-my-pi, not Pi. I found the latter to be mostly useless in terms of what I needed for my day to day work. I'm sure I could tinker with it to get it right, but that's literally what oh-my-pi is doing for you. And like I said, zero problems thus far.
It's probably sorted, but for a while there was an escaping issue. Then our company forced us to use Claude via CoPilot in PyCharm and that may have been the other issue. I just stick to a command prompt in a shell now unless there's a pressing advantage to doing it in any IDE.
Looks interesting, but they need to clean up the claudese in those bullet points holy crap. It's like they purposely want the "what you get" section to be as obtuse as possible.
I do not get the appeal of web-based GUIs now that we have LLMs.
Migrating my screenwriting software[0] from OpenTUI to Roc thought me what were once rough edges that take a lot of time are now just a prompt away from a solution.
It now uses 10x less RAM, bundle is half the size, and my initial implementation is 700-800 fps for scrolling and 200 fps for writing.
Granted, a TUI is less complex than a full-blown GUI, but I built a Roc Iced platform just as easily that abstracted Iced flux away
JetBrains UIs aren't web-based, unless this is an exception.
JetBrains tools are heavier and can take a few seconds to launch, but they also have some of the most responsive typing out there. They did some good work on zero-latency typing modes when that was a big topic and they were very good at typing latency relative to most editors.
They will never be the right tool for the personality type that gets angry when their IDE takes a few seconds to launch at the beginning of the day or when an application takes more than 1GB of RAM, but I am not that type of person so I enjoy them.
I think the TUI people are well served by vim or emacs and should just stay there and be happy.
The rest of your post feels like an injected ad for your project underneath a topic where it's not relevant
This post is about a TypeScript repo, which is what I commented on. It may as well be a necessity since Pi is using a JS runtime as well. It has nothing to do with JetBrains and everything to do with the codebase in the link.
It also has nothing to do with TUI vs. GUI, since my comment addressed both (anecdotally).
If anything, your mentioning of their performance work is exactly my point.
Why does AI-generated imply "not bothered" to you? If I take all of my design docs/specs, all of the existing repo, and I tell an agent that I'd like to make a readme that contains things X, Y, and Z, what's the problem?
Rejecting slop doesn't imply rejecting all AI-generated text wholesale.
> Why does AI-generated imply "not bothered" to you?
1. AI generated text is just hard to read and verbose, if you worked so hard on your docs/specs, why would you want your README to be unreadable?
2. There's so much garbage written with AI that I'm just going to assume yours also is. Would you sign up to a gym with broken windows? It could be the best gym ever, but I wouldn't count on it.
What JetBrains needs is nothing short of a revolutionary product, which they would need to sell with an excellent introduction.
Instead they phoned it in and pasted that slop:
> Once a GitHub PR is open it shows the CI checks and review decision, and lets you mark it ready, hand failing checks or a diverged branch to the agent, merge it with the repository's last-used method, reopen it, or delete the remote branch.
But given you dont know me, im going to guess you wont be foolish enough to try and guess especially considering there are at least 100 different flavours.
Software is even more complicated. How is any person even the smartest software type in the world or AI going to know all the exact details of your wish if nobody tells them exactly every painful detail ?
Not sure why it matters? I can't tell the difference, personally, but even if I could the AI is much more likely to have given complete and accurate information than any human..
Do we have to
explain in every bloody thread why human-generated Readmes, with their faults and omissions, are preferable to overly verbose and flowery AI slop prose?
Conciseness and highlighting relevant points is much more important than word count. Machine generated readmes are anything but. Sorry you can’t tell the difference between that and human prose, you should try reading more human content.
Left to themselves, LLMs prefer writing that's really atrocious. e.g. The writing will try and make everything sound profound.
The other problem I've noticed is e.g. LLMs will write in stuff that's inappropriate for the content of the text. e.g. Writing out stuff that looks like a chat reply, in commentary or descriptions or whatever.
I'd distinguish, though, between "sloppy writing" & "LLM generated writing". I'd agree if you didn't put in the effort to check the writing is of high quality, then perhaps the other parts are also not high quality.
I don't think just because "AI" wrote it it's bad. I wouldn't prefer e.g. sloppy human writing over well-crafted LLM-written text.
This readme the main thing which sticks out is the "what you get" bullet points are rather dense/verbose for bullet points. But aside from that it seems fine.
I'm a technical writer and I use agents a lot for my work and personal projects. I also created a basic open-source preset for agents to create end-user documentation.
During the preset testing, I created dozens of READMEs using agents. Although I can't say that any of them were perfect for me, most of them were good enough.
README is a very special kind of document. For small projects, it's often the only one that potential users will read. The README should explain what the product is, what problems it solves, how to get started, provide some examples, and explain the restrictions. It should cover all these topics in a concise way so that even readers who aren't used to read long texts can easily read through it. It's definitely up to human to make it clear and easy to understand.
> I refuse to take a repo with an AI-generated readme seriously. If you haven't bothered to write a readme, what else have you not bothered to do?
I care whether the readme is any good, not who wrote it. If you haven't bothered to critique the readme and instead critique its suspected creator — something entirely in your head — what else have you not bothered to do?
> (100% AI according to Pangram btw)
So what? Who cares about Pangram? I certainly don't! You cannot detect AI — it's impossible. Perhaps it's not AI, just Kenyans?[0]
majority of people here, have to read ai for work. that's why you're getting paid nowadays, read, understand, steer and fix errors produced by ai. this is an ai product. It's a bit weird why readme of ai product has to be written by human?
If anything spend your time on the code and let AI make the readme. If you’re basing the code quality on the readme maybe look at the architecture is a better way to judge a project. Seems a bit superficial like judging a book by its cover…
Such an idiotic take. You can't possibly have ever built anything yourself that you actually cared about. If you had, you would know what you're saying makes no sense at all.
"A pi session and the change it made, scoped to an isolated git-worktree workspace."
Yep. The "A $thing, $elaboration." has joined "$title. $thing, $past_verb." in my "this team didn't care enough to do a basic pass over the main page introducing their product" tells.
> the app owns the workspace, the editor, and the wire — it never assembles prompts behind your back.
Thanks JetBrains, I assumed the app would assemble prompts behind my back, so I'm glad they clarified that.
If they bothered reading this that would have been removed. I understand smaller players... Vibe Coders, Solo projects, something on its initial commit (to be edited / cleaned up / improved).
But, from JetBrains? Bad sign for the industry, IMO. Is it that fucking hard to have a decent technical writer collect some some human-readable documentation? This just exudes lack of effort.
It probably came from some out of scope call out in the original prompt, or a redirect. I have found that AI really loves patting itself on the back like this.
It really really loves mentioning what something "is not" right after it just said what it is. I don't like this pattern and see it more and more in READMEs, probably part of recent Claude RL. I don't think its as prevalent in the zeitgeist as the emdash, "Its not X its Y" and the emoji bullet list.
Some repo owners just review code that others submit, and don't write anything themselves. If the review process stays the same, does it matter? The fault is not generating the code with AI, rather, with faulty review process or not caring about the code, or not reading it. A different problem.
README seems to be where $AGENT dumps some kind of context for itself. I've come to peace with that and instead create a Github page for user level shit.
I’ve been using JetBrains IDEs for over a decade but rarely open them anymore because working on several worktrees concurrently is extremely frustrating and wasteful. Their AI strategy is all over the place and they always wanted to sell their opaque and overpriced tokens, which made no sense. I hope they get their act together.
I thought so too until I learned that there is a dedicated worktree switcher hidden away as a tab in the GitHub pain. Now I can open each worktree in a separate window as if it were its own project. Maybe that's not everyone's ideal work though, but it's exactly what I want 99% of the time.
But I don't use it for AI assistant development, it's definitely meant for humans and not AI agents.
Yeah, their worktree implementation was very recently hacked in, and it really doesn't work well. The IDE project list duplicate entries for every worktree, IDE project settings don't transfer across worktrees, and every time I switch to another worktree it asks me again what project I want to open, instead of just opening the same solution I have opened.
It's wild that a set of IDEs that are generally fairly good just completely fail in this basic workflow. I really wonder what went through the minds of the product owners and developers that implemented this.
I canceled my subscription after like 12 years because I realized all the things that made it great aren’t really needed anymore. The “it just works” tooling, refactoring, jumping around code, etc.
Now my agents mainly write code under my direction and I review it either in the harness or on GitHub.
I hadn’t opened it in ages and when I did need an IDE something like Zed was more than sufficient.
Code review is the problem I'm currently trying to figure out for myself. I've never liked GitHub PRs but I'm not 100% sure what I want instead (which is possibly just: it feels like it should be a VS Code plugin... But maybe I don't like VS Codes interface that much either?)
Part of what I've loved about JetBrians tools has been the navigation and deep semantics for large code bases. Zed is pretty basic in that regard. My agents usually get what I need, but when they go off the rails, cleanup gets interesting.
But I'll admit to keep my skills from atrophying, there are some projects of mine I don't touch with AI.
Same. I like their products but they have a problem with feature creep and design debt, and their AI tools are extremely in your face while also being underwhelming. I don't know if I'm going to renew my subscriptions.
I'm inclined to agree. Jetbrains' own agent host 'Junie' was such a paim to set up and get working with other providers that I switched to Cline. I had moderately good results with using that as a coding agent, but eventually I switched to Deepseek Harness for generative tasks and am now using my IDEs more for review.
Never used their agent. They added a lot of completions and smart features into the editor. I never got along well with that. I've reverted to static completions now, like before AI. It made the editor less distracting, more performant and much more predictable to use.
If I edit code I know what I want to edit. If I want AI to do it, I prompt an agent.
Like pretty much every traditional tech company, it looks like JetBrains is looking at the AI bubble and panics. They feel like they have to do something, anything, and quickly, when, at least in my opinion, they should take it slowly, capitalize on what they do well and progressively integrate AI where it really makes sense, rather than jumping on the (already crowded) vibe coding bandwagon.
Hopefully, all these companies will go back to reason in a few years, when the bubble will burst. By bubble bursting, I don't mean that AI will disappear, but the nonsense and crazy money around it will.
Didn't JetBrains release JetBrains Air recently? This looks like it's basically the same basic premise, except implemented differently. What's their strategy here, are they just launching competing products?
It seems that Air is a strategic product, and Thinkrail is more of an experiment.
While they are both "agentic development tools", it seems that Air is mostly concerned with managing multiple agents, whereas Thinkrail is about spec-first development.
They also started developing Fleet several years back and made a few alpha releases, only to shutter it before it ever was ever fully released for reasons that essentially seemed to boil down to "we want to focus on our core IDEs instead, which didn't really explain what changed. I was interested in Fleet, so at this point I can't help but be skeptical of anything new they've done since then because it's not clear to me why the same logic for discontinuing Fleet couldn't apply to pretty much anything else.
> Many of the improvements and refreshed design were "backported" into the traditional IDEs.
Okay, but what stops any of that for happening for any of the new things since then? To me, a large part of the appeal of Fleet was that it wasn't the entire IDE; adding a new skin to something that feels much too heavyweight for me does not resolve the fundamental problems I had with it.
Oh, I agree it's not ideal. But the kind of people who are willing to move to a new traditional IDE are now also the people who would ditch IDEs completely. The people who stuck to their traditional IDEs are also the ones likely continuing their subscriptions now.
I actually think they could do something more radical than Fleet today, after seeing the (alpha, but still) ArtCraft apps, and translations like Bun. They could embrace still being the best traditional IDE, with a lot of the heaviness removed.
I even forgot that it exist. I tried it many times but it was not good enough. Eventually I switched to Zed since that was better. Probably Zed killed it.
Huh, I went back to look at the announcement of halting development on Fleet, and they went back and updated it six months after the fact to announce the rebrand. It never occurred to me to go back and look.
I think they said they will discontinue Air, it currently supports codex and claude as harness as opposed to pi, and own UI lib (this one seems web/electron based), shame.
NGL Jetbrains has massively helped my developer journey and be a great tool for many many years. I don’t open it that often anymore, but I see that they’re trying hard to stay relevant, which is probably not easy as probably the whole market moved against you.
The problem with all the AI IDE's is you can't use them with most enterprise/teams accounts, you have to use API which can cost upto to 10x more over an enterprise plan. I really like JetBrains, used them for a LONG time, but just find myself doing 99% of my day to day with iterm and sublime now.
Ah this reminds me I need to cancel my jetbrains subscription. I started with CLion and used RustRover, PyCharm, WebStorm, DataGrip over the years but now they all feel outdated and unnecessary.
Edit: oh shit I am on the cancel screen and reminded I also used ReSharper. Best of luck jetbrains I have given you a lot of money figure your shit out and I would happily return.
154 comments
[ 1.5 ms ] story [ 25.1 ms ] threadThere aren't enough digital french cleat systems.
The true benefit of a French cleat isn't storage or organizing the way a fiddly pegboard and hanger approach uses, it's the ability to move all tools for a specific project onto a single rail without having to fiddle around, reducing friction for engaging and re-engaging with a project.
This is exactly what I was missing
I'd also caution against saying things like llms are doing most of the work I don't think that's true. And where it is true it shouldn't be. These things are still new tech and they need a lot of hand holding and oversight still.
To be clear, it clicks things to do things like test software. If you build an API, you're not testing if the clicking actually works.
I don't think harness like Claude Code or Codex support jj very well for working in parallel
Also, reading through their "SPECS".... https://github.com/JetBrains/thinkrail/blob/d299f20d2ba014f3... or https://github.com/JetBrains/thinkrail/blob/d299f20d2ba014f3...
How is this llm-vomit "specs that are ground truth" etc.?
Migrating my screenwriting software[0] from OpenTUI to Roc thought me what were once rough edges that take a lot of time are now just a prompt away from a solution.
It now uses 10x less RAM, bundle is half the size, and my initial implementation is 700-800 fps for scrolling and 200 fps for writing.
Granted, a TUI is less complex than a full-blown GUI, but I built a Roc Iced platform just as easily that abstracted Iced flux away
[0]: https://blisswriter.app/
JetBrains tools are heavier and can take a few seconds to launch, but they also have some of the most responsive typing out there. They did some good work on zero-latency typing modes when that was a big topic and they were very good at typing latency relative to most editors.
They will never be the right tool for the personality type that gets angry when their IDE takes a few seconds to launch at the beginning of the day or when an application takes more than 1GB of RAM, but I am not that type of person so I enjoy them.
I think the TUI people are well served by vim or emacs and should just stay there and be happy.
The rest of your post feels like an injected ad for your project underneath a topic where it's not relevant
It also has nothing to do with TUI vs. GUI, since my comment addressed both (anecdotally).
If anything, your mentioning of their performance work is exactly my point.
JetBrains, explain concisely why this isn't written in C/C++/Rust?
(100% AI according to Pangram btw)
Rejecting slop doesn't imply rejecting all AI-generated text wholesale.
I can do it. Probably better for my needs than someone else’s agent can.
The README is meant to communicate substance and intent. Without the latter it’s just a landing page.
1. AI generated text is just hard to read and verbose, if you worked so hard on your docs/specs, why would you want your README to be unreadable?
2. There's so much garbage written with AI that I'm just going to assume yours also is. Would you sign up to a gym with broken windows? It could be the best gym ever, but I wouldn't count on it.
Skill issue, it’s easy for a guy like me
If it is blatant unfixed llmese, then it means that no effort whatsoever was put into making this text.
What JetBrains needs is nothing short of a revolutionary product, which they would need to sell with an excellent introduction.
Instead they phoned it in and pasted that slop:
> Once a GitHub PR is open it shows the CI checks and review decision, and lets you mark it ready, hand failing checks or a diverged branch to the agent, merge it with the repository's last-used method, reopen it, or delete the remote branch.
The answer is very short.
But given you dont know me, im going to guess you wont be foolish enough to try and guess especially considering there are at least 100 different flavours.
Software is even more complicated. How is any person even the smartest software type in the world or AI going to know all the exact details of your wish if nobody tells them exactly every painful detail ?
Conciseness and highlighting relevant points is much more important than word count. Machine generated readmes are anything but. Sorry you can’t tell the difference between that and human prose, you should try reading more human content.
This is a useful criticism.
Left to themselves, LLMs prefer writing that's really atrocious. e.g. The writing will try and make everything sound profound.
The other problem I've noticed is e.g. LLMs will write in stuff that's inappropriate for the content of the text. e.g. Writing out stuff that looks like a chat reply, in commentary or descriptions or whatever.
I'd distinguish, though, between "sloppy writing" & "LLM generated writing". I'd agree if you didn't put in the effort to check the writing is of high quality, then perhaps the other parts are also not high quality.
I don't think just because "AI" wrote it it's bad. I wouldn't prefer e.g. sloppy human writing over well-crafted LLM-written text.
This readme the main thing which sticks out is the "what you get" bullet points are rather dense/verbose for bullet points. But aside from that it seems fine.
During the preset testing, I created dozens of READMEs using agents. Although I can't say that any of them were perfect for me, most of them were good enough.
README is a very special kind of document. For small projects, it's often the only one that potential users will read. The README should explain what the product is, what problems it solves, how to get started, provide some examples, and explain the restrictions. It should cover all these topics in a concise way so that even readers who aren't used to read long texts can easily read through it. It's definitely up to human to make it clear and easy to understand.
I care whether the readme is any good, not who wrote it. If you haven't bothered to critique the readme and instead critique its suspected creator — something entirely in your head — what else have you not bothered to do?
> (100% AI according to Pangram btw)
So what? Who cares about Pangram? I certainly don't! You cannot detect AI — it's impossible. Perhaps it's not AI, just Kenyans?[0]
[0]: https://marcusolang.substack.com/p/im-kenyan-i-dont-write-li...
Well, good luck in professional development then. Love it or hate it, this is the way everything is moving.
If you care about these things you are either not the intended target or you are a hypocrite.
Yep. The "A $thing, $elaboration." has joined "$title. $thing, $past_verb." in my "this team didn't care enough to do a basic pass over the main page introducing their product" tells.
Thanks JetBrains, I assumed the app would assemble prompts behind my back, so I'm glad they clarified that.
If they bothered reading this that would have been removed. I understand smaller players... Vibe Coders, Solo projects, something on its initial commit (to be edited / cleaned up / improved).
But, from JetBrains? Bad sign for the industry, IMO. Is it that fucking hard to have a decent technical writer collect some some human-readable documentation? This just exudes lack of effort.
But I don't use it for AI assistant development, it's definitely meant for humans and not AI agents.
It's wild that a set of IDEs that are generally fairly good just completely fail in this basic workflow. I really wonder what went through the minds of the product owners and developers that implemented this.
Now my agents mainly write code under my direction and I review it either in the harness or on GitHub.
I hadn’t opened it in ages and when I did need an IDE something like Zed was more than sufficient.
I use it when I need to deeply understand a PR.
But I'll admit to keep my skills from atrophying, there are some projects of mine I don't touch with AI.
An IDE still has it's uses, and it's perfectly fine without AI. Agents and IDEs have never felt like something that should be combined in my opinion.
If I edit code I know what I want to edit. If I want AI to do it, I prompt an agent.
I just could not understand their AI strategy.
I know because I've tried practically all of their AI product offerings and not a single one feels coherent.
Hopefully, all these companies will go back to reason in a few years, when the bubble will burst. By bubble bursting, I don't mean that AI will disappear, but the nonsense and crazy money around it will.
https://www.jetbrains.com/air/
While they are both "agentic development tools", it seems that Air is mostly concerned with managing multiple agents, whereas Thinkrail is about spec-first development.
And fleet was made to attract the VSCode generation, who they were starting to lose new user momentum to. But VSCode is no longer the biggest threat.
Okay, but what stops any of that for happening for any of the new things since then? To me, a large part of the appeal of Fleet was that it wasn't the entire IDE; adding a new skin to something that feels much too heavyweight for me does not resolve the fundamental problems I had with it.
I actually think they could do something more radical than Fleet today, after seeing the (alpha, but still) ArtCraft apps, and translations like Bun. They could embrace still being the best traditional IDE, with a lot of the heaviness removed.
I suppose they were in a tough spot regardless though. They started developing it before the AI age, so they would have to pivot either way.
That doesn't exactly inspire confidence that this product will stick around.
Also... why bun?! I just finished purging it from all my systems, ha! It'd be cool to have plain pnpm support.
So I’m willing to give it a try.
I tried Air before and found it underwhelming. I tried using it around Pi through ADB, but it didn't any benefit around using Pi purely.
Edit: oh shit I am on the cancel screen and reminded I also used ReSharper. Best of luck jetbrains I have given you a lot of money figure your shit out and I would happily return.