Linux creator Linus Torvalds has offered a qualified defense of artificial intelligence in programming, describing it as a useful tool for enjoyment and bug finding while drawing a firm line around its use in high-stakes software development.
Speaking at Open Source Summit Europe in Prague, Torvalds said he now likes using AI after previously viewing it as insufficiently capable. But his enthusiasm came with a warning that is especially relevant to the Linux kernel: generated material can create more work for maintainers when it is used carelessly.
A tool, not an automatic substitute
Torvalds framed AI as a tool whose value depends on context. For hobby programming and learning, he said, it can make coding more accessible and restore some of the pleasure of working on a small project. He described AI-assisted “vibe coding” as a way people can find joy in programming.
That is different from relying on generated output in software that is central to systems used by many people. Torvalds said developers need to be careful when AI is applied to work that is real and important.
The distinction reflects the practical demands of an open-source project such as Linux. The kernel is maintained through a large contributor community, and maintainers must assess patches, reports and proposed changes. A flood of low-quality AI-generated submissions can consume review time even if some generated material appears plausible.
Bug finding is a more promising use
According to the report, Torvalds sees AI as becoming a vital component in finding and fixing bugs in the Linux kernel. That use case is narrower than handing responsibility for design or code review to a model: the technology can help identify problems, while people remain accountable for determining whether a proposed correction is appropriate.
It is a useful distinction for teams adopting coding assistants. Generating code is only one part of software work. Maintaining a major project also requires understanding system behavior, reviewing changes, testing them and accepting responsibility for their consequences.
Torvalds noted that he is principally a maintainer rather than someone doing large amounts of kernel programming himself. That role gives his comments particular relevance to the practical issue of how AI output enters a mature, collaborative codebase.
Lowering the barrier for new programmers
Torvalds also connected AI to changes in the experience of learning to program. He began programming in the early 1980s, when computers and the programs written for them were much simpler. Today, he said, the bar for software engineering is far higher, making it harder for newcomers to see small efforts as meaningful.
AI can help reduce that barrier by giving people a way to experiment, ask questions and build something without first mastering every layer of a modern computing stack. In that setting, the value may be motivational as much as technical.
That does not remove the need to learn fundamentals. It suggests, instead, that an assistant can make the early stages of exploration less intimidating. For people working on personal projects, a tool that shortens the path from an idea to a working experiment can make programming feel more immediate.
Maintainers still need signal, not volume
The same accessibility that helps learners can create difficulties in established projects. If AI makes it easy to produce patches or bug reports at scale, maintainers may receive more submissions without receiving more useful contributions. The cost is paid during evaluation, when experienced developers must identify mistakes, duplicates and changes that do not fit the project.
Torvalds’s position is therefore neither a blanket endorsement nor a rejection. AI can be useful when it supports a developer’s work, particularly in experimentation and bug discovery. It is risky when it encourages people to submit work they have not understood or validated.
For Linux and other critical open-source projects, that leaves human review at the center of the process. An AI system may assist with locating an issue or drafting an idea, but maintainers must still judge technical correctness and project suitability.
A practical view of AI coding
The broader significance of Torvalds’s comments is their emphasis on use rather than ideology. Arguments about AI in programming often collapse into a choice between total adoption and total rejection. His remarks describe a more operational standard: use the technology where it helps, and apply greater caution as the consequences of failure rise.
That approach is likely to resonate across software development. A personal experiment, an educational project and a production kernel release carry very different risks. The appropriate role for an AI assistant should change accordingly.
Torvalds’s message is straightforward. AI can make programming more enjoyable and can assist in finding bugs. But in software that matters, generated output cannot replace the care, expertise and accountability of the people responsible for it.
This article is based on reporting by ZDNET. Read the original article.
Originally published on zdnet.com
