Agent-written PR descriptions are accurate but long enough that reviewers skip them. Keeping a short human summary visible and collapsing the rest preserves the detail without costing reviewer attentionThe mechanism that lets a model weigh which earlier words matter for the word it's currently processing — the core operation of a transformer.Full definition →.
Key takeaways · AI-distilled
Simon Willison says AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition →-written PR descriptions are accurate and complete but read like slop; his own reaction is "I'm not reading all of that", even when he drove the agent himself.
His fix: open the PR with the shortest human-approved summary of what it is for, sometimes with a screenshot, then put the full agent write-up inside an HTML details/summary block, which GitHub Markdown renders as one expandable line.
He labels the collapsed block with the model name and often the effort level, such as "Claude Opus 5.5 Medium", so readers can see that section was written by an agent.
The opener is not always his own prose. Sometimes he lifts it from the agent's text, but his promise to readers is that it reflects his judgment of what is worth their time.
Terms in this piece · Glossary
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
attention — The mechanism that lets a model weigh which earlier words matter for the word it's currently processing — the core operation of a transformer.
