Linus Torvalds has addressed the growing use of AI assisted coding tools and offered a balanced view on their role in software development. Speaking about recent trends, he said he recognises that some developers may already experiment with AI tools while working on kernel related code, although he is not testing these systems himself. He noted that the rise of automated crawlers scraping kernel.org has created inaccurate reports of bugs and vulnerabilities, adding noise and unnecessary work for maintainers.
Torvalds outlined a clear distinction between using AI tools for exploration and relying on them for production work. He said he supports what he called “vibe coding” as a way for users to learn or experiment. He compared this approach to his own early computing experiences when he typed programs from printed magazines. He argued that modern systems are far more complex, and tools that help beginners understand concepts or complete small tasks can be positive.
His support, however, stops at the boundary of mission critical software. Torvalds said AI generated code should not be applied to core systems such as the Linux kernel because it risks creating long term maintenance problems. He said that while AI might seem useful for rapid development, the kernel requires careful control and predictable behaviour that automated tools cannot currently guarantee. He described using AI in that environment as an unwise choice.
He also discussed the growing frustration maintainers face due to low quality automated vulnerability reports created by crawlers collecting code at scale. Many of these reports link to irrelevant or misleading sources and require review even when they lack substance. Torvalds said he rarely replies to emails but reads messages, implying that the noise created by automated systems adds friction to routine review processes.
The discussion touched on wider topics such as the role of Rust in the kernel and his reflections on two decades of guiding kernel development. When asked about workforce reductions blamed on AI systems, Torvalds said the situation is complicated and long term effects will take time to understand. He compared AI to earlier technological advances such as compilers, which increased productivity without removing the need for programmers.
Torvalds suggested that the industry may eventually view AI as standard infrastructure, used alongside cloud hosting and storage systems rather than treated as a novelty. His comments imply a future where development splits into two paths. Experimental workflows may rely heavily on AI tools, while critical and long lived code will remain under strict manual control. He said the enduring question for AI driven systems is whether they can achieve consistent accuracy rather than surface level output.
