I also imagine they will be left behind, for better or for worse as AI-authored software and hardware will satisfy people's needs faster than that developed by humans.
I don't see a way to merge these rapidly moving frontiers if software without radically different distributed software development and distribution networks where AIs collaborate on a global scale. Perhaps the only way to keep playing is to grow a "super cortex", a projection of ourselves into the digital. Like today's agents, but improved, grounded by our knowledge and understanding.
This smells a bit like Skynet. Also, Software Heaven.
I'm not sure how I feel.
And more importantly, documenting Mac hardware takes time. Components like the speaker on a Macbook can be permanently damaged if you don't write the drivers in a specific way. Other features like the Secure Enclave, power management, and TouchID might be unsupportable depending on how Apple designed it.
Asahi's stance isn't particularly crazy, and it probably aligns with how Linus feels about upstreaming code that nobody understands.
It is quite remarkable how far along support has come: https://aurorasilicon.org/devices/
> It is quite remarkable how far along support has come
Is it? All I'm seeing from that page is that they forked Asahi's M1-M4 work, and then failed to bring up M5 and M6. What's remarkable here?
> Please note that installing on an M-series Mac is not directly supported at this time.
And the way they have written that driver is absolute crap, not submimissible at all. They will need to write it from scratch, properly, using a TEE driver, a sensor driver, a userspace plumbing and an fprintd patch. What they did, instead, is they run pretty much everything in-kernel.
So, can't wait for ridiculous exploits that make your fingerprint effectively a backdoor to your OS.
Personally I don't think it's very likely the tables would turn and Asahi became dependent upon their contributions. I foresee a much bleaker future for Omarchy, crashing and burning from some reckless and unnecessarily abrasive whim of DHH. I'm not optimistic about Gravity Linux either, it's a single-dev project and that's prone to failure for many reasons.
Then those distributions will fail.
It's 2026. Any open source project with an anti-AI stance will get its milkshake drunk by a fork that embraces AI. How can people still not see this?
Perhaps the 30+ years of non-LLM generated Linux support for UEFI/ACPI and ARM device tree chips has something to do with it.
The state of the art has changed.
It may shock you to hear that coding was never the bottleneck for emulation or reverse-engineering.
Yes, and LLMs work well for the non-coding parts too. Pointing Claude at Gravity's repos has gotten Linux running well on my M4 Pro mini, with GPU acceleration as of a few days ago. So either I'm a genius or LLMs are pretty good at this, and I'm not a genius.
I have no particular love for Asahi, but they've made 10x more progress than the vibe coded alternatives.
I keep repeating myself, but make an effort to check the "state of the art" of the crap they added to their kernel tree.
The primary message we get out of the AI ecosystem is "MORE MORE MORE FASTER FASTER FASTER".
First off, that's not really what plenty of projects are about. You might like your favourite Linux distribution or desktop environment because it represents less, as a curated vision-- an AI "workalike" could offer ten thousand new features per week, but break the earned trust that a single clear vision provided.
Second, beyond a certain point, speed burns through the human components of a software project. There are only so many reviews that can be done, whether the source is AI or enthusiastic humans. Quality eventually suffers, contributors burn out, and the project dies.
If you point a well-funded slop cannon at an unfunded open-source project, they're not going to say "let's find 20 new reviewers to keep up with the firehose of commits!" They either have to circle their wagons and get finicky about what they allow, throw up their hands about quality, or rely on the machine itself, the dreamworld where they build enough test suites and harnesses to somehow keep Microsoft Paint from having zero-days.
Given those choices, I feel like I'd be happier to see "we circled the wagons". This doesn't necessarily mean "no AI" but it means keeping the project at a manageable scale, and right now, the way to handle that is to push away AI contributions without a bulletproof background behind them.
Going even further, it's high time we saw a "slow software" movement to counter the "move fast and break things" and its new cousin "move even faster and let the bot break things for you". I want smaller programs written by happier people under less psychic and economic assault.
> "MORE MORE MORE FASTER FASTER FASTER".
You people, if you had any clue about the code quality standards and cared about security, YOU would see this other way.
I have personally vetted their code for TouchID and some other aspects, including USB-C/Thunderbolt.
I absolutely cannot wait to hear your apologist brown-nosing when you start getting targeted with exploits.
b) being left behind is a good thing if your counterparts are headed in the wrong direction. I think projects which lean into LLMs are going to find they have a hot mess a year or two from now, whereas projects that eschewed them will be in a good place.
Because the "slop era" of AI-generated code is over. It ended in September. AI-generated code is now better than what the vast majority of practicing software engineers can produce. And AIs can more thoroughly test and verify the code too, both with programmatically automated unit and integration tests and with "manual" testing by the LLM itself.
If you are not letting LLMs write the code, then your code is not up to professional standard. This is the new reality now.
[1] https://asahilinux.org/docs/platform/feature-support/m3/
Also tried LordOS very quick, very easy to install, looks and feels like macOS, but Debian distro underneath.
I also tried GhostBSD. It's easy enough to work with, but it does not have any packages for "spice-vdagent," which I needed to resize the screen size and clipboard copy/paste support between host and guest devices when using Virt Manager :(