Linus Torvalds now uses only two tools, bids farewell to programming

At Open Source Summit India 2026, Linux creator Linus Torvalds and his close colleague Dirk Hohndel offered a rare, candid look into the current state of the Linux kernel and its future direction. Torvalds made it clear he no longer considers himself a programmer, now identifying strictly as a development lead, and revealed that his workflow has been pared down to just two essential tools: Git and email.

When asked about the Linux 7.1 release, Torvalds downplayed the notion of splashy milestones. He emphasized that the project deliberately avoids blockbuster features, favoring instead a model of “incremental improvement and steady progress.” Since the creation of Git, he noted, the kernel team has moved away from dramatic releases. However, the rise of AI is beginning to stress this workflow. “It’s been getting a bit harder lately because of AI finding interesting bugs,” Torvalds said, noting that the community has felt the pressure even as the kernel maintains its reliable nine-to-ten-week release cadence.

During merge windows, Torvalds handles roughly 200 merges over two weeks. While he trusts his maintainers, he pushes back against last-minute fixes unless they are truly critical. The technical challenges of new code are manageable, he explained; the real stress comes from personality conflicts. “Code is easy to fix. Personality is not always as easy to fix,” he admitted, acknowledging his own role in some of those issues.

Torvalds now reads very little code himself. “I hardly read code at all anymore,” he said. “I’m not a programmer, I’m a development lead.” He still writes small patches, but they function more as suggestions than commands, and he rarely commits his own code. His focus has shifted to understanding the “bigger picture” behind pull requests, relying on thorough explanations from maintainers. He dives into the code itself only when forced by build breaks or merge conflicts, where his decades of experience often help him spot hidden issues.

On the subject of legacy support, Torvalds was blunt. He described NTFS as a “problem child” with two competing maintainer groups, and he is content to let them “fight it out.” More broadly, he is actively working to drop support for obsolete hardware. “I’m not very sentimental when it comes to technology,” he said, citing the upcoming removal of support for x86 machines without hardware floating point, such as the 486 SX from over 30 years ago. Similarly, support for outdated networking standards like ISDN and ATM is being phased out. Users of truly ancient hardware can still run older kernels.

Regarding programming languages, Torvalds pushed back on the hype around Rust. While he finds it interesting, he still prefers C for its simplicity and raw power. “I’m more of a hack-and-slash kind of person,” he said. He warned that Rust does not fix logic errors. “Rust fixes a few easy bugs that you can make in C, but it does not fix the logic errors, right? It does not think for you,” he explained. Furthermore, the guarantees Rust provides only apply within its own code boundaries; interactions with C code void those protections. He pointed out that many high-profile kernel bugs recently have been logic errors, not memory safety issues.

Turning to AI and large language models (LLMs), Torvalds revised his earlier comments about a “10x” productivity boost, calling that figure unscientific. He acknowledged that LLMs generate a lot of junk, including bogus bug reports that take significant human effort to debunk. “We certainly saw more junk being generated by LLMs than we saw useful code up until early this year,” he said. For kernel-level fixes, he insists that humans must act as intermediaries, providing suggested patches rather than simply throwing LLM-generated reports over the wall. Many AI-generated patches, he described as “mindless band-aid kind of patches” that fail to address underlying issues.

Despite these frustrations, Torvalds is not a “shoot-the-messenger” type. He appreciates that LLMs have uncovered embarrassing security bugs that should have been found decades ago. For his own personal projects, including adding Godzilla to family travel photos, he uses LLMs as prototyping tools, though he stresses they are not yet ready for kernel-level work.

In the end, Torvalds remains focused on the human side of development. “I work with people, not tools,” he said, a philosophy that has guided Linux through decades of steady evolution.