Linux Kernel Developers Consider Adding AGENTS.md To Help Guide AI/LLM Agents

While the Linux kernel continues to be bombarded with patches from AI/LLM agents, to date the kernel hasn't carried an AGENTS.md Markdown file with instructions catering to AI/LLM agents. But a patch set out today would finally introduce one.

Longtime Linux developer Sasha Levin who has worked on other AI initiatives for the kernel like helping to determine patches for back-porting to the stable kernel tree and AI-powered merge conflict resolution is now looking to add an AGENTS.md to the kernel source tree.

The proposed AGENTS.md is just linking to the Linux kernel's README file. From there it points to documentation on coding assistants and other information relevant to kernel developers.

The rationale for adding AGENTS.md to the Linux kernel tree is that in testing on one AI agent without the file, the AI agent added a "Signed-off-by" tag to the proposed kernel patch even though it shouldn't have without the user explicitly signing off on their own. It also resorted to its own attribution tag rather than the kernel standardized Assisted-by tag.

Testing on a second unnamed AI agent without AGENTS.md added no attributions. But after adding the AGENTS.md, both agents behaved to kernel standards and the second one better conformed to the kernel development best practices.

This kernel mailing list thread laid out the patch proposing the AGENTS.md introduction. While it's just linking to the README flle, not all are in favor at this point. There have been some objections on the LKML to the patch given that making AI/LLMs consume the entire Linux kernel README and associated documentation would increase token consumption. More catered documentation most relevant to AI/LLMs would be one way to help reduce token consumption, but we'll see what comes of this new Linux kernel proposal around enhancing AI workflows.