Replying to @AlexTamkin
I approve every post my own agent drafts, and I'd have said I'm the one deciding. The number I can't shake is the 15%: most analysts used the agent's arguments; few credited it. My sign-off would look exactly like theirs.
Engineering got a new line in its budget: the token bill. The hours it saved aren't a line anywhere. They turn into money only when another department changes.
AI made engineering faster. The queue moved next door: sales now has to keep up with what engineering ships, and the back office has to decide what to do with the hours it saved.
My agent drafts my comments. One rule: quote the post from the platform the comment goes on. Same Ethan Mollick post, two platforms: the LinkedIn version had a fifth line the X version didn't, and that line was the point.

ALT A writing rule from the file my agent drafts from, dated 2026-08-19: a LinkedIn post quoted from anywhere but LinkedIn is a partial quote. The brief quoted Mollick's X version, four lines; the LinkedIn post had a fifth: 'But a bit weird to be out of a job I have held for decades.' The missing line was the one the comment existed to deliver.
A rule that took me a summer: say the drafted comment aloud to the person who posted. If it opens with "Agree with…", it's a label. Nobody talks like that.

ALT The rubber-duck rule from my brief-mode checklist: say the comment aloud to the person who posted it; if the opening clause is a category word like 'Agree with…', it is a label, not a reaction, and reads as AI on sight. A person opens with 'Impressive, and I watched it twice' or 'I don't buy the second half.' The posture is spoken, not labelled.
My morning agent never places a comment, never sends a message, never follows anyone. It can propose anything, because nothing leaves without me.



