AI sounds confident even when it's wrong. Learn to spot the three failure modes, run a 30-second verification, and build the daily habit that keeps AI useful instead of dangerous.