Raymond Chen separates pull request descriptions from durable code comments

Raymond Chen, identified as a veteran Microsoft engineer writing on his The Old New Thing blog, drew a clear line between two forms of written communication in software development. A pull request description is a point-in-time statement whose job is to persuade the approver that the change should be accepted. Code comments, on the other hand, describe the code itself: how to call a function, what prerequisites it has, and other details that remain useful after the PR is merged.

The Register frames this distinction as timely because AI coding tools are now generating large volumes of pull requests alongside annotations of varying quality. The article does not cite specific tools or studies, but the editorial context links the guidance to a broader concern about AI-generated code and its accompanying comments. The piece also references past examples of poor human-authored comments, including pages of apologies and self-described 'self-commenting' code, suggesting the problem predates AI.

Chen also weighed in on code formatting, stating he does not care how source code is formatted but recommending that wholesale layout or formatting changes be made in a separate check-in. The rationale is to avoid forcing maintainers to review a diff dominated by style changes rather than substantive logic. The article connects this to a prior tabs-versus-spaces debate involving another Microsoft veteran, Larry Osterman, who argued spaces now make more sense because they are always consistent.

This is a single-source opinion column, not a product announcement or policy shift. No benchmark, version number, or measurable outcome is provided. The value for an engineering team is prescriptive guidance: treat PR descriptions as ephemeral persuasion and code comments as durable documentation, and isolate formatting changes from functional ones. The article does not address enforcement mechanisms, tooling support, or how teams should operationalize the distinction in automated review pipelines.

The broader context links in The Register's sidebar, AI slop polluting CVE pipelines, Git being unprepared for AI coding volume, and GitHub uptime complaints, suggest the editorial concern is about AI-generated content overwhelming existing review and documentation workflows. Chen's advice is positioned as a human-judgment counterweight to that trend, but the article itself does not propose tooling or process changes beyond the conceptual separation.