The latest release of the most widely used Linux init system is here, and between dropping init script support and AI-assisted coding, we feel sure that this release will win it yet more admirers.

Systemd 260 delivers one of the changes that the developers have been promising for at least a few years – we reported that init script support was going back in 2023.

According to the release notes:

It also needs a newer Linux kernel (minimum 5.10, 5.14 recommended, and 6.6 required for full functionality), plus a selection of other libraries and supporting files. Notably, it drops support for version 1 of libidn – now you'll need libidn2. None of this is likely to be hugely disruptive for new versions of distros using systemd 260.

There's also a new Markdown file in the systemd GitHub repository, which may set alarm bells ringing for some developers. The new file is called AGENTS.md, and replaces one that was called CLAUDE.md. The new AGENTS.md file provides instructions to help guide AI agents. (We are restraining ourselves from scattering quotation marks around this paragraph like confetti.)

Given the long history of AI agents ignoring instructions, which The Register was reporting in 2024 and still is this year, this strikes us as rather like King Canute ordering the tide not to rise (for non-Brits, this is a famous legend). It is pretty much doomed to fail.

The existence of a file with instructions for agents doesn't necessarily prove that systemd 260 itself was built using LLM coding assistants. For now, their use appears to be limited to reviewing changes, as the code suggests. Its description says:

However, systemd does now feature on at least one edition of the OpenSlopware list of slop-contaminated FOSS. We covered the rise and fall of the original controversial list back in January.

According to the OpenSlopware list, for now, bot-generated code is confined to one specific sub-element of systemd, sd-bus, which is a "lightweight D-Bus IPC client library." The list highlights Commit 744d589 from late January, summarized as "add test cases for truncated fds," in which Red Hat developer Allison Karlitskaya notes that she did this "with Claude's help."

This is a requirement – the AGENTS.md file specifically stipulates:

The extra 244 lines in sd-bus are spread across two relatively small changes, which add more tests to a small submodule. Appearing on OpenSlopware probably won't faze the systemd developers much. They are by necessity a thick-skinned crew, as the facetious comment on the release of systemd 256 – "now with 42 percent less Unix philosophy" – emphasized. That said, we suspect more AI code changes will follow.

The other changes in this version are mostly relatively minor, including tweaks to handling of intermittent network connections, aid the display of friendly distribution names, add the ability to handle OCI images via systemd-mstack, and other low-level adjustments that most users will never notice.

Just like the removed support for traditional init scripts, this consent for LLM-generated changes will likely further deepen and widen the divide between the many folks who use systemd and say that it makes life easier, and the determined holdouts who want nothing to do with it. In practice, we suspect it will make no visible difference. ®