AI PR descriptions work better when a human edits the entrance

Simon Willison uses the pull request as the basic unit of work with Claude Code and Codex. Instead of deleting the long machine written description, he places it inside a collapsible element and puts a few human approved sentences in front.

GitHub can hide machine detail without discarding it

GitHub Markdown supports the HTML details and summary elements in pull request descriptions and comments. Willison places the agent generated explanation inside them. A short introduction remains visible, sometimes with an image and a label identifying the model or effort level.

He shows five concrete pull requests. One opener explains a layout problem involving a table with 46 columns. Another calls out the unnecessary loading of a file of roughly 220 KB. The detailed account of implementation and verification stays available after expansion.

Reading has become the bottleneck after writing became cheap

A coding agent can produce an accurate, comprehensive description at almost no cost to the author. The reviewer still pays with time. When every pull request arrives with several equally thorough paragraphs, completeness starts to obstruct orientation.

Willison separates two jobs. The short opener answers whether the change deserves attention. The collapsed section preserves decisions, implementation details and tests for the moment somebody needs them. Human work shifts from producing prose to curating it.

A collapsible element cannot replace judgment or review

The pattern does not guarantee correct code or an accurate description. A poorly chosen opener can hide risk as easily as it hides filler. Long text inside the element also remains long, even if it no longer blocks the first glance.

The convention works only when the team shares a clear promise: the visible paragraph reflects human approved priorities rather than another automatic summary. Without that responsibility, the element becomes a digital cupboard that everyone fills and nobody opens.

Faster decisions will matter more than longer descriptions

The useful signal is a shorter time to first review and fewer missed risks, not the number of words behind an arrow. Teams should watch whether the opener consistently states the purpose, scope and most important risk of a change.

If the pattern spreads, the change will look modest. The agent supplies the full case file, a person selects the cover page and the reviewer decides whether to open the rest.

Lilith's verdict

The agent brings a full box of files to review. A person still has to write on the lid why a colleague should open it.