Linus Torvalds: AI Can’t Think Like a Programmer

Overview

Linus Torvalds discusses how AI is changing participation in the Linux kernel and other open-source projects. While he acknowledges that AI can help developers move faster—such as by reviewing patches, finding issues, and generating code—he argues that AI is also contributing to a “flood” of low-quality bug reports and drive-by contributions.

This creates additional work for maintainers, who may already be dealing with burnout. Instead of reducing workload, maintainers often must spend time clarifying questions and reviewing contributions that don’t follow through.

Impact on Maintainers and Smaller Projects

Torvalds explains that Linux kernel maintenance is relatively healthy today due to:

• A large contributor base
• Well-paid, solid maintainership

However, he emphasizes that many other open-source projects are much smaller—sometimes maintained by only a few people. In these cases, even a modest influx of automated, poorly supported AI-generated issues can overwhelm a small team and increase burnout.

Tool vs. Blind Reliance

On whether AI belongs in open source/kernel development, Torvalds distinguishes between:

• Using AI as a tool
• Relying on it blindly

He states that he doesn’t use AI for work that involves working with people, since his maintainer role is largely social and coordination-focused. He also cautions against becoming dependent on large corporate tooling.

He notes that the kernel community already uses automated and AI-adjacent tools, including patch-review assistance systems that:

• Scan mailing lists
• Ask questions

Additionally, individual developers increasingly use local AI tools.

“Most Code Is Written by AI” and the “Jobs/Doom” Debate

Torvalds pushes back on claims that “most code is written by AI.” He compares the AI era to earlier programming revolutions:

• Assemblers and compilers improved productivity
• Yet they did not change the underlying fundamentals

Similarly, AI can accelerate coding, but it does not remove the need for human understanding and judgment. He argues that developers who truly understand complexity will still produce good results (even with AI assistance), while those who don’t will generate code that fails.

Long-Term Maintenance Requires Understanding

Finally, Torvalds stresses that long-term maintenance depends on understanding what the end result actually is. Even when using AI, developers should inspect generated output—for example:

• Compiled output / assembly
• Whether it matches real intent

He highlights this as especially important for serious projects that must be maintained for years, not treated as one-off “vibe coding.”

Presenters / Contributors

• Linus Torvalds
• Durk (Durk leads Verizon’s open source program office)
• Jim (referenced via a keynote: “Jim’s keynote”)