This was a very measured and reasonable post from the LibreOffice team explaining why they chose not to integrate any generative features into LibreOffice. I really appreciate that it neither hypes nor whips up the masses into a foaming frenzy, but instead just shows how the tech can’t yet meet their existing criteria for inclusion.
Please do not reflexively tag this as vibecoding. It isn’t about using LLMs, it’s about choosing whether to integrate generative models and features into software.
Please do not reflexively tag this as vibecoding. It isn’t about using LLMs, it’s about choosing whether to integrate generative models and features into software.
That is, quite literally, a discussion about the use of LLMs, in software, so the vibecoding tag seems appropriate. In fact, I'd say the tag is more appropriate on this article than on some other ones submitted recently.
But it's completely unrelated to the use of coding agents or developing software with LLMs. What does "vibecoding" mean to you that you consider it appropriate on this post?
The vibecoding tag has (culturally) come to encompass anything that uses/touches/mentions the zeitgeisty applications of generative AI. In this way, it is not about vibecoding at all.
A blog post that uses a ‘generated’ image but otherwise contains no other connection to generative AI would be sufficient, for example. By blocking the vibecoding tag, you could experience lobste.rs as though generative AI does not exist.
I’m not saying I agree with this, but it seems to be the case.
This was a very measured and reasonable post from the LibreOffice team explaining why they chose not to integrate any generative features into LibreOffice.
Yes, and their criteria are quite good. Basically, office suites should not send parts of your documents off to who-knows-where. So until there's a viable, 100% local solution, with no vendor dependencies, there's nothing to even discuss. Until then, users are invited to look at third party plugins.
This is an extremely sensible policy.
The reality is that local AI exists, but:
The minimum useful hardware for a local, agentic model is around 32GB of VRAM, or 96GB of slower unified RAM. This is unaffordable for most users, and it will likely remain so until the RAMpocalypse ends. You can absolutely get by with less memory, but the smaller you go, the more limited it gets.
Running local models and configuring other software to talk to them isn't super hard, but it still has terrible UX for end-users.
To get predictable results, you have to know what models your users are running. Nobody wants the support costs of dealing with users who configured a 2-bit quant of an 8B model, especially not if it's some weird fine tune they found on Hugging Face, with a name like l33thax0r/Qwen3.5-finetune-Fable5-ABLITERATED-TurboQuant-MTP-SUPERSMART-Updated3-Loops-Fixed-For-Real-This-Time.GGUF.
So I completely agree that "Keep all this AI stuff in a plug-in until things settle down" is a completely reasonable engineering approach. If you start losing enterprise customers who simply must have AI as a checklist feature, then maybe officially bless a plugin with a support agreement or something.
I only have two examples in recent memory (this and Waterfox), but maybe there are some others - why even do these FOSS projects feel the need to make hedging statements such as "well we don't hate AI, we have nothing against it! don't worry, it's useful!"
[OP] johnolinda | 12 hours ago
This was a very measured and reasonable post from the LibreOffice team explaining why they chose not to integrate any generative features into LibreOffice. I really appreciate that it neither hypes nor whips up the masses into a foaming frenzy, but instead just shows how the tech can’t yet meet their existing criteria for inclusion.
Please do not reflexively tag this as vibecoding. It isn’t about using LLMs, it’s about choosing whether to integrate generative models and features into software.
hackguy | 11 hours ago
That is, quite literally, a discussion about the use of LLMs, in software, so the
vibecodingtag seems appropriate. In fact, I'd say the tag is more appropriate on this article than on some other ones submitted recently.gulbanana | 10 hours ago
But it's completely unrelated to the use of coding agents or developing software with LLMs. What does "vibecoding" mean to you that you consider it appropriate on this post?
berserk | 10 hours ago
The vibecoding tag has (culturally) come to encompass anything that uses/touches/mentions the zeitgeisty applications of generative AI. In this way, it is not about vibecoding at all.
A blog post that uses a ‘generated’ image but otherwise contains no other connection to generative AI would be sufficient, for example. By blocking the vibecoding tag, you could experience lobste.rs as though generative AI does not exist.
I’m not saying I agree with this, but it seems to be the case.
pyfisch | 10 hours ago
The description on Lobsters is "
vibecoding: Using AI/LLM, coding tools. Don't also tag withai. "Its just the tag label chosen on this site, another site might have "Tag every story about sports as football".
Sanity | 16 minutes ago
Or tag anything about linux, unix, amiga or bsd as "linux". Not annoying at all.
emk | 8 hours ago
Yes, and their criteria are quite good. Basically, office suites should not send parts of your documents off to who-knows-where. So until there's a viable, 100% local solution, with no vendor dependencies, there's nothing to even discuss. Until then, users are invited to look at third party plugins.
This is an extremely sensible policy.
The reality is that local AI exists, but:
So I completely agree that "Keep all this AI stuff in a plug-in until things settle down" is a completely reasonable engineering approach. If you start losing enterprise customers who simply must have AI as a checklist feature, then maybe officially bless a plugin with a support agreement or something.
pyj | 7 hours ago
This seems like a good property for many programs in general!
Instead, we're just going to yank half of the package manager as dependencies, from random internet people. :P
Enterprise is going to want to use their own subscriptions for security purposes. I could see not having AI by default as a selling point.
prayerie | 2 hours ago
I only have two examples in recent memory (this and Waterfox), but maybe there are some others - why even do these FOSS projects feel the need to make hedging statements such as "well we don't hate AI, we have nothing against it! don't worry, it's useful!"