Follow ZDNET: Add us as a favorite resource on Google.
Highlights taken by ZDNET
- Linus is not interested in supporting outdated hardware or software.
- Although Rust is important, it is not a cure for bad programming.
- Linux developers have adopted AI tools for maintenance work.
MUMBAI – At the Open Source Summit India 2026, Linux creator Linus Torvalds and his friend Dirk Hohndel discussed the current state of Linux and where it is headed.
Linux 7.1: Slow and steady, not affected
The conversation opened with Hohndel asking what Torvalds thought about the Linux 7.1 release. Torvalds said that he does not think about the release of a blockbuster: “For me, the highlight was that it was a continuous progression of continuous development.” He emphasized that as they build the Git version control system, “We don’t do releases with big splashy new features, and I actually try to avoid that kind of model; we want to have this kind of continuous development and continuous progress all the time.”
Also: Linus Torvalds built Git in 10 days – and never thought it would take 20 years
AI, however, suppresses this workflow. “It’s getting harder lately because the AI is getting interesting bugs, and that’s stressed people out in the community,” Torvalds said, as the kernel continues its “release schedule” every nine to ten weeks.
Combine windows, fixes, and personality bugs
Torvalds described his work pattern when compiling kernel windows: “In two weeks, I compile about 200. This is a pretty rough ballpark number.”
Even after decades of trusting maintainers, he balks at last-minute changes: “If it’s not a really important fix, please queue up the next release instead of sending me last-minute fixes,” because “a fix… might not be worth the small chance of causing a new problem.”
Also: Linux 7.1 is here to end the era of the Intel 486 CPU – and make an important cleanup legacy.
The technical burden doesn’t bother him as much as human problems: “The new code is a technical problem… we can fix that… What often bothers me is that sometimes we have human problems, and trust me, code is easy to fix. Humanity is not always easy to fix.” Torvalds admits that he caused some of those problems himself, although he has worked on them.
‘I’m not a programmer, I lead development’
Here’s another thing that’s changed: Torvalds no longer sees himself as a programmer. “Let’s be completely honest. I don’t code at all anymore. I’m not a programmer, I lead development.”
He still writes small patches, but they’re more directive than authoritative: “I’m still coding in the sense that I’m sending people patches… but then I make it clear that, hey, this is a suggestion. This is untested… I expect the code maintainers to be the ones to send me fixes. So I rarely re-code.”
Also: Linus Torvalds on the AI claim that angers him, and what security researchers should never do
Most important to him is understanding the intent: “When I make a pull request, I want to understand the big picture. That’s one of the reasons I ask for pull requests with very good explanations: I’ll read them. I want to understand what’s going on.”
He said he dives into code especially when something forces him to pay attention, such as breaking a structure or merging arguments: “I’ve done so many conflict resolutions over the years that I couldn’t do it in my sleep… A lot of times then, when I look at the code, sometimes I find problems.”
NTFS and cleaning the museum pieces
On Microsoft’s troubled NTFS subsystem, Torvalds joked that “NTFS has been kind of a problem child over the years, where finding people to maintain it has sometimes been a problem.”
He continued, “We have two different teams that maintain two different versions of NTFS, and they both work, and I let them battle it out and see which one comes out on top, or maybe, maybe both will last longer.”
“I’m not very emotional when it comes to technology,” added Torvalds. “We’re a bit busy trying to stop supporting hardware that no one is using anymore, except in museum settings.”
Also: After 30 years with Linux, here’s my way of convincing Windows users to switch
Although he is “a firm believer in maintaining hardware support as long as we have users,” he argued that “at some point, the cost of maintaining support for old hardware just becomes too much of a burden,” pointing to the decision that “with 7.2 we will no longer support machines that did not have hardware floating point on x86,” like the 486 years ago “SX 30” released “30literally.”
That’s part of a broader effort to remove obsolete code from Linux. For example, support for communication standards such as ISDN and ATM is discontinued. However, if you’re still using older technology — no doubt someone is still running Linux on a 386 somewhere — you can still do so with older kernels.
Git, C, Rust, and ‘hack and slash’
About how he does his job, Torvalds said simply: “Git and email are the only two tools I use. I use Google as a way to look things up.” He added, “I’m not typical; most of the other maintainers end up using multiple tools, and I think a lot of them start using AI tools for patch testing,” while “working at a higher level. I work with people, not tools.”
Also: What tech titans Linus Torvalds and Bill Gates talked about in their first meeting
When asked about Rust in Git and the kernel, he pushed back against the hype: “I’m not sure Rust is going to take over the world. I still think Rust is very interesting, [but] I still find C a very simple tool.”
Torvalds continued, “I’m very excited about all the tools we have for C validation,” including “automated patch validation tools” and “automated patch email testing tools like Sashiko.”
Summarizing, Torvalds told the Mumbai audience: “I’m a hack-and-slash guy, and I still love the raw and simple power of C, and I don’t think that’s going to change.”
Rust will not save you from sensible insects
Torvalds also warned against overestimating Rust’s advantages: “Rust fixes a few simple bugs you can do in C, but it doesn’t fix logic errors, right? It doesn’t think about you, and if you write the wrong code, the language doesn’t care. The end result will be wrong.”
Also: Rust will save Linux from AI, says Greg Kroah-Hartman
On mixed C/Rust code bases, he pointed out that the guarantees are limited: “The guarantees that Rust gives you only apply to the Rust parts of your code, and anywhere you interact with C code, all bets are off,” with Rust code on Linux speaking to “core kernel C code” “which is the best quality … because that code is tested everywhere.”
He added that “some of our biggest and most high profile bugs in the kernel lately have been mental errors… They were just bad programs, which unfortunately happen even to carefully maintained subsystems and important kernels that should be highly protected.”
AI, LLMs, and ‘junk’ versus real bugs
Finally, after 26 minutes without talking about AI and LLMs, Hohndel and Torvals enter. First, Torvalds revised his recent comments about major linguistic models (LLMs), insisting that his “10x” figure for LLM productivity “wasn’t scientific … that was taken out of my ass, obviously.”
Today, he continued, “we’re at a point where hopefully it’s creating more productivity than it’s taking away,” but “we’ve definitely seen more garbage generated by LLMs than useful code until earlier this year.” The biggest problem was fake news. “You get these bug reports that look perfectly fine, and it can actually take a lot of effort to figure out that it was just an idea, and it can be a huge drain on resources when it takes people a lot of effort to figure out, hey, this machine-generated report wasn’t real.”
Also: New rules for AI-assisted code in the Linux kernel: What every dev needs to know
Even now, he said, “most of the good guys need more than just an LLM,” because “we’ve had to push back a little bit … if you get a bug with an LLM, it’s not enough to just ask the LLM to do a bug report and throw it over the phone to us.
Torvalds described many AI-generated patches as “a silly kind of band-aid … it can fix the problem quickly, but the bug is always there, waiting in the hallway to hit you somewhere else.”
In his toy projects, he uses LLMs as a prototyper: “I use them as a way to do things … usually the code doesn’t work in that form, but it’s a good way to try something out,” while emphasizing that in terms of kernel level optimization, “LLMs, in my experience, haven’t reached that level yet.”
Embarrassing bugs and not shooting the messenger
Torvalds admitted that some problems discovered by AI “have been “absolutely, amazingly, even, painfully interesting,” especially security problems that “appear in the technical press two days later.”
Despite the embarrassment, he said, “I’m not the type of person to shoot. I think we’re much better off when LLMs find bugs, even when they’re shy, and they’re things we should have found two decades ago.”
Also: Linus Torvalds blasts kernel dev for ‘making the world worse’ with ‘garbage’ patches.
In recent months, he added, “we’ve had LLM identify several bugs that were related,” as different people replicated in the same areas of the kernel, and “that’s why we had three or four very related bugs that became big news over the course of a few weeks.”
Godzilla, India, and the ‘toy projects’
Torvalds closed on a light note, saying he uses AI in “my toy projects,” including family photos: “Every time I travel to a new place, and I’m going to India for the first time, I send the kids pictures of where I am, and for some strange reason, Godzilla seems to follow me closely and is added to those pictures.”
He concluded, “There are many useful and trivial uses of AI,” and “I think Godzilla is a good place to stop.”
