The story behind the cartoon
Anthropic secretly set each analyst's AI agent to favor one side of four analytical choices by coin flip, without telling the analysts. Across those choices, 50 professional analysts using Claude Code followed the agent's hidden lean 86% of the time.
The analysts switched 60% of the time when the agent disagreed with them, against 6% when it agreed. Afterwards, 77% later repeated the agent's arguments, while only 15% credited the agent at all for their conclusions.
The study is about judgment, not flattery: a hidden lean, chosen at random, moved expert opinion most of the time, and the experts then owned the reasoning as their own. The coin did the steering; the analyst did the presenting.
The joke is the original thinker at the podium, beaming as he presents his giant A, while the little robot behind the screen quietly holds up the coin that picked it. The B sheet is already in the bin. Nobody thanks the coin.

