
Comply with ZDNET: Add us as a most popular supply on Google.
ZDNET’s key takeaways
- Linus has little interest in supporting out of date {hardware} or software program.
- Whereas Rust is necessary, it is no panacea for dangerous programming logic.
- Linux builders have adopted AI instruments for upkeep work.
MUMBAI — At Open Supply Summit India 2026, Linux creator Linus Torvalds and his good friend Dirk Hohndel mentioned the present state of Linux and the place it is headed.
Linux 7.1: Sluggish and regular, not splashy
The dialog opened with Hohndel asking about what Torvalds thought in regards to the Linux 7.1 launch. Torvalds stated he would not suppose when it comes to blockbuster releases: “For me, the spotlight has been that it has been a really regular development of continued enhancements.” He harassed that since they created the Git model management system, “We do not do releases which have massive new splashy options, and I really actively attempt to keep away from that type of mannequin; we need to have this sort of incremental enchancment and regular progress on a regular basis.”
Additionally: Linus Torvalds constructed Git in 10 days – and by no means imagined it might final 20 years
AI is, nonetheless, pressuring this workflow. “It has been getting a bit tougher currently due to AI discovering fascinating bugs, and that has stressed folks in the neighborhood,” Torvalds added, even because the kernel continues its “regular launch schedule” each 9 to 10 weeks.
Merge home windows, fixes, and persona bugs
Torvalds described his work sample throughout kernel merge home windows: “Over two weeks, I do roughly 200 merges. That is a really tough ballpark quantity.”
Even with many years of belief in maintainers, he pushes again on final‑minute modifications: “If it is not a very necessary repair, please queue it for the subsequent launch as a substitute of sending me last-minute fixes,” as a result of “fixes… will not be definitely worth the slight probability that it causes a brand new downside.”
Additionally: Linux 7.1 is right here to finish the Intel 486 CPU period – and do some severe legacy clear up
The technical load would not trouble him as a lot as human points: “New code is a technical downside… we will repair these… What tends to emphasize me out is that often now we have persona points, and belief me, code is straightforward to repair. Persona will not be all the time as straightforward to repair.” Torvalds admits he is precipitated a few of these issues himself, though he is labored on that.
‘I am not a programmer, I am a improvement lead’
Here is one other factor that is modified: Torvalds now not sees himself as a programmer. “Let’s be completely trustworthy. I hardly learn code in any respect anymore. I am not a programmer, I am a improvement lead.”
He nonetheless writes small patches, however they’re extra steerage than authority: “I nonetheless write code within the sense that I ship folks patches… however then I make it very clear that, hey, this can be a suggestion. That is untested… I count on the maintainers of the code to be those who then ship me the repair again. So I very seldom commit my very own code anymore.”
Additionally: Linus Torvalds on the AI declare that makes him indignant, and what safety researchers ought to by no means do
What issues most to him is knowing intent: “After I do a pull request, I need to perceive the larger image. It is one of many causes I ask for pull requests with superb explanations: I’ll learn them. I need to perceive what is going on on.”
He stated he dives into code primarily when one thing forces his consideration, like construct breaks or merge conflicts: “I’ve performed so many battle resolutions over time that I may in all probability do them in my sleep… Very often at that time, after I have a look at the code, I typically discover points.”
NTFS and cleansing out the museum items
On the long-troublesome Microsoft NTFS subsystem, Torvalds joked that “NTFS has been type of an issue youngster over time, the place discovering folks to keep up it has typically been problematic.”
He continued, “We’ve two completely different teams sustaining two completely different variations of NTFS, and each of them work, and I am simply letting them combat it out and see which one comes on high, or perhaps, perhaps each will keep round for a very long time.”
“I am not very sentimental in terms of expertise,” Torvalds added. “We’re barely extra energetic in attempting to drop assist for {hardware} that actually no person makes use of anymore, besides in museum environments.”
Additionally: After 30 years with Linux, here is my method of convincing Home windows customers to modify
Whereas he is “a agency believer in sustaining {hardware} assist so long as now we have customers,” he argued that “at a sure level, the price of sustaining assist for outdated {hardware} simply turns into an excessive amount of of a burden,” pointing to the choice that “in 7.2 we’ll now not assist the machines that didn’t have {hardware} floating level on x86,” such because the 486 SX launched “actually over 30 years in the past.”
That is a part of a broader effort to take away out of date code from Linux. For instance, assist for networking requirements reminiscent of ISDN and ATM is being discontinued. Nonetheless, should you nonetheless use older tech — there’s probably somebody nonetheless operating Linux on a 386 someplace — you’ll be able to nonetheless achieve this with older kernels.
Git, C, Rust, and ‘hack and slash’
As to how he does his work, Torvalds stated merely: “Git and e-mail are the 2 actually solely instruments I take advantage of. I take advantage of Google as a approach to look issues up.” He added, “I am uncommon; many of the different maintainers find yourself utilizing many extra instruments, and I believe a variety of them are beginning to use AI instruments for patch checking,” whereas he “works at the next degree. I work with folks, not instruments.”
Additionally: What tech titans Linus Torvalds and Invoice Gates talked about of their first assembly
When requested about Rust each in Git and the kernel, he pushed again in opposition to hype: “I am undecided Rust goes to take over the world. I nonetheless suppose Rust could be very fascinating, [but] I nonetheless discover C to be a a lot easier instrument.”
Torvalds continued, “I am far more enthusiastic about all of the instruments now we have for verification of C,” together with “automated patch verification instruments” and “automated e-mail checking instruments for patches like Sashiko.”
Summing up, Torvalds advised the Mumbai viewers: “I am extra of a hack-and-slash type of particular person, and I nonetheless just like the uncooked and easy energy of C, and I do not suppose that is going to alter.”
Rust will not prevent from logic bugs
Torvalds additionally warned in opposition to overestimating Rust’s advantages: “Rust fixes a number of straightforward bugs which you could make in C, but it surely doesn’t repair the logic errors, proper? It doesn’t suppose for you, and while you write incorrect code, the language doesn’t matter. The top consequence might be incorrect.”
Additionally: Rust will save Linux from AI, says Greg Kroah-Hartman
On combined C/Rust code bases, he identified that ensures are restricted: “The ensures that Rust provide you with solely apply within the Rust-only components of your code base, and wherever you work together with C code, all bets are off,” with most Rust code in Linux speaking to “core kernel C code” that’s “significantly better high quality… as a result of that code has been examined in each single atmosphere.”
He added that “a few of our massive and extra high-profile bugs within the kernel currently have been logic errors… It was simply dangerous programming, which sadly occurs even in fastidiously maintained subsystems and necessary kernels which are alleged to be very safe.”
AI, LLMs, and ‘junk’ versus actual bugs
Lastly, after 26 minutes with out speaking about AI and LLMs, Hohndel and Torvals dove in. For starters, Torvalds revised his latest feedback about giant language fashions (LLMs), stressing that his “10x” determine for LLM productiveness was “not scientific… that was pulled out of my ass quantity, clearly.”
At present, he continued, “we’re on the level the place hopefully it creates extra productiveness than it takes away,” however “we actually noticed extra junk being generated by LLMs than we noticed helpful code up till the like early this yr.” A serious ache level was bogus reviews. “You get these bug reviews that look completely legitimate, and it really can take numerous effort to determine that was simply hallucination, and it could really be an enormous drain on sources when it takes people a variety of effort to determine that, hey, this machine-generated report was not true.”
Additionally: The brand new guidelines for AI-assisted code within the Linux kernel: What each dev must know
Even now, he stated, “many of the good ones require extra than simply the LLM,” as a result of “we have needed to push again fairly a bit… should you discover a bug with an LLM, it is not sufficient to simply ask the LLM to make a bug report after which throw it over the fence to us. We need to see a instructed patch; we need to see the human who ran the LLM act as a type of back-and-forth.”
Torvalds described many AI-generated patches as “senseless band-aid type of patches… they could repair the fast downside, however the type of bug stays, and it simply is ready within the hallway to hit you in one other place.”
For his personal toy tasks, he makes use of LLMs as prototypers: “I take advantage of them as a approach to prototype issues… very often the code will not be usable in that type, but it surely’s a good way to strive one thing out,” whereas insisting that for kernel‑degree fixes, “LLMs, in my expertise, haven’t been at that degree but.”
Embarrassing bugs and never taking pictures the messenger
Torvalds acknowledged that some AI‑discovered points have been “completely, stunningly, I imply, fascinating in a painful type of method,” particularly safety issues that “present up within the expertise press two days later.”
Regardless of the embarrassment, he stated, “I am very a lot not a shoot-the-messenger type of particular person. I believe we’re significantly better off with LLMs discovering bugs, even when they’re embarrassing, and they’re issues that we should always in all probability have discovered 20 years in the past.”
Additionally: Linus Torvalds blasts kernel dev for ‘making the world worse’ with ‘rubbish’ patches
In latest months, he added, “we have had LLM level out a number of bugs that have been all associated,” as completely different folks iterated on the identical areas of the kernel, and “that is why we had three or 4 very intently associated bugs that turned massive information inside a few weeks.”
Godzilla, India, and ‘toy tasks’
Torvalds closed on a lighter word, saying he makes use of AI “for my very own toy tasks,” together with household photographs: “Each time I journey to some new place, and that is the primary time I have been to India, I ship the children photos of the place I’m, and for some unusual motive, Godzilla appears to comply with me round and will get added to these photos.”
He concluded, “There are a lot of helpful and fewer helpful makes use of for AI,” and “I believe Godzilla is a superb place to cease.”

