I expect they would find your response and conduct offensive, and your original post. Consdering the ammount of users who posted that they considered it so, or are they all unreasonable?.
In any case, you have been warned about your conduct, and I don’t care to respond to your offensive “trolling” behaviour any further.
Consider this an official moderation-hat warning and behave yourself in the future.
Hi,
I am the maintainer of EmailViews (the AI generated native email client for Haiku mentioned earlier in the thread), and I am most likely the reason why this thread was posted. I saw Andrea had Claude Code working in Haiku, so I asked him if he could share how he got it running in Haiku. Happy to see he did, but sorry for the extra heat this thread has picked up.
I can actually see and respect both sides here. My goal with EmailViews was to provide an email client that took advantage of Haiku’s unique database like file system in a way that actually benefits end users and as a way to showcase what makes Haiku special. Nothing more, nothing less. For me, the fact that it was created using AI is secondary. What’s more important is whether it fills a real gap for some users as a native solution. But of course, every user gets to decide if it’s for them or not — AI reservations included, totally valid.
I just hope AI-generated (or AI-assisted) apps don’t get systematically sidelined or dismissed. They can still be genuinely useful tools for end users, especially in a small ecosystem where volunteer time is limited like Haiku. That’s my only wish.
Anyway, just my humble (and perhaps biased) opinion, but with respect for everybody’s views.
Haiku can “claude”!
Thank you for the time you invested to make this Tutorial!
Great to know. Great to have another tool working on Haiku!
But how to use an API key and understand how it is working, would be another nice tutorial.
“export ANTHROPIC_API_KEY=“sk-ant-…”“
The strawman argument you made was you said there were no applications such as office suites and email applications on Haiku, which is clearly not true, and then used that as a reason to use LLM’s as it would speed up development and bring those applications sooner. While you mentioned the introduction to the tutorial was meant to add ‘spice’ to the discussion, it could come across as not caring about all the work of the Haiku/Haikuports contributors over the 20 years of ‘waiting’, which you later clarified as being respecting and thankful of.
It’s great you have been able to use an LLM to bring your own applications and functionality to Haiku, such as the Bluetooth stack improvements in the screenshots. One of the key reasons I’ve seen for such code not being allowed is the current legal grey area of such code generated by an LLM as they are trained on a vast array of codebases. A potential legal issue that could shut the project down simply because it added functionality quicker doesn’t sound like a gamble worth taking.
AI is one of those things where I would normally be positive, but the obnoxious non-stop shilling of AI from hardcore AI fans makes me want nothing to do with it.
When you run it for the first time, a link will appear for you to log into your Claude account. From there, you can get the API key and either paste it directly into the terminal where you launched Claude Code or add it to your ANTHROPIC_API_KEY environment variable.
@colmiga Thank you, this is exactly the kind of discussion I was hoping to start.
On the legal point you’re right that it’s a real grey area. But it’s worth noting that it affects all modern code: GitHub Copilot is used by millions of developers, including those who contribute to projects that end up in HaikuPorts. The line between “inspired by” and “derived from” has never been clear in software, with or without AI.
The legal risk exists, but it’s not new and it’s not exclusive to LLMs. Managing it with a clear policy makes sense. Managing it with a badge on HaikuDepot does not. 
@x68k I didn’t want to come across as an AI zealot, and I apologize if I did. My goal was exactly the opposite: to spark a discussion that a handful of people had already decided was closed. Apparently I succeeded, for better or worse. 
Would that I could be the peacemaker in your soul, that I might turn the discord and the rivalry of your elements into oneness and melody. But how shall I, unless you yourselves be also the peacemakers, nay, the lovers of all your elements?
-Kalil Gibran
Personally, I am not a fan of AI and I think it should be kept out of the core Haiku proper at all costs- just too dangerous there.. But for app development, if the source is open, it can be looked at as another tool. The hand saw verses power saw analogy fits.
The long term danger as a society is that we forget how to code by hand just like many people now do not know how to do math without a calculator.
This whole post makes me sad, I have a lot of love and respect for everyone here on all sides of this argument. Most of us have spent a good portion of our years here and it would be a shame to create any division in the Haiku community.
Many people were against ports like QT in favor of native software (me included) But I do see the value in the QT tools/apps available and I use them daily.
AI used as a toolkit is fine (IMHO). I use EmailViews now as my mail client, but I waited until the source was open as protection against any hidden malice in the code. And it does not matter who wrote the code (human or otherwise) if the source is closed it’s equally dangerous.
It’s an interesting topic. Publicly available LLMs were certainly not trained on proprietary or private codebases, companies backing them up emphasize that now and then. They might have been trained on GPL or other publicly available copyleft code. That makes it challenging to use with non-copyleft codebase like Haiku itself without switching the project to GPL license. However that issue doesn’t apply to third-party apps in HaikuPorts, as long as they use copyleft license. Apparently licensing issue also doesn’t really apply to Linux or other GPL licensed projects (or to commercial closed-source projects, because we all know why <trollface_emoji_needed>).
And then there are open source LLMs which you can train on publicly available dataset that excludes copyleft code. It’s not even that difficult to do.
For the record: I favor a blanket ban on slop in Haiku. Don’t ask me to explain, for two reasons:
Thank you to everyone who sides with common sense and human decency.
P.S. As for how we can enforce a ban like that? Simple: we catch anyone doing it, we show them the door. Much like in real life.
The issue is code that public LLM’s have trained on which is publicly available is copyrighted to some person or entity, even if it’s licensed under the GPL, which is different to Haiku’s licence as you mention. Some of those entities who own copyrights in open source projects like the Linux kernel for example are large companies, which is why I think it’s best to avoid such a potential liability in Haiku itself until it’s been litigated and resolved legally elsewhere, which may take some time.
Allowing a 3rd party LLM generated application into HaikuPorts is different to direct contributions of LLM generated code into Haiku. HaikuPorts hosts the compiled code etc. like it does with many other applications from various sources. It’s also mentioned by the OP LLM code could indeed be in those other applications by now, which is a valid point. That said, I can see why a policy as to LLM generated code needs to be introduced to HaikuPorts.
You definitely crossed a red line here. This is not funny. I don’t care about how strong you (or anybody else) feels towards AI (or any other topic for that matter), stuff like your original yellow star quote should not have any place on this forum. I’m not a moderator anymore but I definitely think you should receive an official warning about this.
I’m not even motivated to comment on everything else you said.
As with everything, we’ll do the best we can. Similarly to our “native” badge, which is also a grey area. Also, it’s rather the reverse of a “fair trade” badge. We don’t have to audit the whole tool chain of a software to make sure it’s “AI-free”. We’ll flag software that’s known to be largely AI-generated.
Call it “ideology” or “values”, people should be able to make an informed decision to use or don’t use some software that has profound consequences for our society and environment.
Maybe we should put a badge for applications that openly ban LLM contributions (hence the “fair-trade” certification). Because the “LLM-assisted” badge will get outdated really quick, since many people are already doing it.
I am not looking forward to using LLM-assisted software, however, there are shining examples of this type, like Ghostty. If the developers know what they are doing and able to audit code properly, I don’t see anything wrong with it.
I am not too bothered about copying other people’s work, everyone is already doing it anyway.
The legal issue is definitely problematic, although I don’t think there is a way back anymore. AI is used broadly in the corporate world, and I’m afraid that ship has sailed.
I think the ethical reasons behind not wanting to use AI are the strongest arguments. The resource waste, the stealing of code, the malpractice, the threat it poses to open source (like: hey Claude, I don’t like the license of this open source product. Please rewrite it for me), I probably don’t need to reiterate all the points.
That aside, though, I can understand that using AI to produce code can be appealing, if you consider it as a bit more intelligent IDE; if you let it produce code in very small steps, it can actually produce mostly useful code. I’m not sure it will be faster that way than doing it manually, but it certainly will be, if you’re not very familiar with the subject or language.
IMO, AI will only really become generally acceptable, when it can learn by itself, when it overcome LLMs. However, this is also where it’ll get really scary, and may make human output “superfluous”.
LLMs learn from all code, not just good, well designed, and correct code. It doesn’t really understand what it does once it gets a tad more complicated. It should not be used in any code where security matters.
Personally, if it weren’t for these ethical issues, I wouldn’t mind 3rd party software written mostly by AIs at all. I don’t think a badge makes any sense here, as AI really is everywhere nowadays. I’m a programmer, I like the challenge, I like finding nice solutions for problems, and I like to learn from what I do. But I totally get that some people just want to have the software they need without exercise, or months of learning stuff they don’t really care about.
I’m definitely in favor of keeping AI out of Haiku (for ethical reasons, as well as legal reasons, and code quality concerns), but I don’t really see the point for 3rd party applications in HaikuPorts.
I understand the tutorial about Claude Code but I’m not sure to understand the aim of the first part.
If it’s ironic maybe you can make another post about what you are trying to say because mixing it with a tutorial is a bad bad idea.
Personally I don’t find it very useful, and it’s counterproductive for the community
@axeld Thank you for the thoughtful response. I agree with you on HaikuPorts: I don’t see the point of restricting third-party software based on the tools used to create it. That was exactly the point I was trying to make, poorly, in the intro.
On the ethical and legal points you’re right that they’re real. I also agree that current LLMs shouldn’t touch code where security matters.
@DigitalBox You’re right. I came in “muso duro” (I’m not sure how to translate that into English, but I hope the meaning comes across), and that probably wasn’t the right approach. @axeld responded in a civil and rational way, and I think the discussion has already found its balance. I don’t think another post is needed.
@BlueSky You’re right about the yellow star, it was out of place and I apologize for it without reservation.
That said, I’d like to explain where that provocation came from. I don’t consider myself a developer, I’m a user. And as a user I genuinely find it absurd to profile a developer based on the tools they use. If tomorrow I use Vim to write an application that has nothing to do with AI, but the version of Vim I’m using has commits generated with an LLM, how is my software classified? Profiling a person who released a program based on the application they used to write it is dangerous, regardless of the topic.
cool ragebait