A synthesis of your own material flatters you; assign a second AI to attack it.
A synthesis of your own material flatters you; assign a second AI to attack it.
When AI synthesizes your own notes, data, or writing into a plan or summary, the result sounds wise — because it is you, distilled and rhythm-corrected. That fluency is the danger: reading it feels like insight when it may just be a mirror. Borrow the security/military move — the red team, a reviewer assigned to attack rather than affirm. After the AI builds something, spin up a separate pass whose only job is to refute it, told to assume the output is self-flattering and to hunt the blind spots. The critic must be separate from the builder (the writer is invested in its work), instructed to be adversarial not balanced, and given the source material so it can catch where synthesis smoothed a contradiction into a decision. Pair it with an external-research pass and you catch two kinds of blindness: the internal skeptic finds self-serving logic, the fact-check finds where you're simply wrong. The honest twist: the red team is a sparring partner, not an oracle — it overreaches, flags non-problems, invents risks. You referee. A red team you obey blindly is just a differently-flattering mirror. This is structure beats magic applied to judgment: reliability comes not from a better model or prompt but from building the adversary into the loop.