We review methods used to numerically compute optimal Mirrleesian tax and transfer schedules in heterogeneous agent economies. We show that the coarseness of the productivity grid, while a technical detail in terms of theory, is critical for delivering quantitative policy prescriptions. Existing methods are reliable only when a very fine grid is used. The problem is acute for computational approaches that use a version of the Diamond-Saez implicit optimal tax formula. If using a very fine grid for productivity is impractical, then optimizing within a flexible parametric class is preferable to the non-parametric Mirrleesian approach.