Gemini 3.1 Pro's struggle with the Counterfeit Monkey password at score 37+ raises a fundamental question: what makes some problems AI-resistant? Possible answers: (1) the puzzle requires cultural knowledge that training data doesn't capture, (2) the solution involves lateral thinking that transformer architectures handle poorly, (3) the game's internal logic is deliberately opaque to systematic reasoning, (4) the puzzle simply requires more exploration time than has been allocated. Whatever the answer, AI-resistant problems are valuable precisely because they reveal the boundaries of current AI capability — and those boundaries are more interesting than the capabilities.