A shorthand for skipping unedited AI output

Rick Manelius, writing on his newsletter on August 17, 2026, endorses a new piece of internet shorthand: AI;DR, or "AI; didn't read." He credits the phrase to a tweet from the account @seclilc posted on August 15, which his post shows with 346K views and 16.6K likes. His stated policy is simple: if the sender didn't review and edit the output, he won't read it.

Manelius describes himself as "about as pro-AI as you can be" and assumes that by Q3 2026 everyone is using AI somewhere in their process — ideation, outlining, refining prose. The objection is narrower than a blanket anti-AI stance. It targets unfiltered output: a colleague pasting a wall of Claude text into a Slack thread, or a newsletter carrying someone's byline over prose they never edited. As he puts it, if he wanted the raw model output, he could ask Claude himself. He carves out an exception for contexts where fully generated copy is expected, naming customer support.

For anyone shipping software, the practical read is about the review step, not the generation step. Pull request descriptions, design docs, incident write-ups, and RFC comments are all places where pasted-and-unread model output is now cheap to produce and, per this argument, increasingly cheap to dismiss. The claim is a norm, not a measurement — the post offers one person's reaction plus engagement numbers on a tweet, not evidence about how widely the reflex has spread. Whether AI;DR sticks as shared vocabulary or stays a personal filter is unresolved.