Improving the precision of atmospheric neutrino flux predictions has been always challenging. Due to more sophisticated data analysis techniques and growing exposure of large volume Cherenkov telescopes (such as IceCube), progress will eventually hit a wall since flux uncertainties may become the dominant systematic uncertainty. The tool MCEq has resolved several problems for the high-energy flux calculations. MCEq almost entirely eliminated the uncertainty related to the method of solving the one-dimensional cascade equations. But the flux prediction will only remain as certain as the combination of cosmic ray flux and hadronic interaction model used in the calculation. In this seminar, I will present recent advances in modeling hadronic interactions for atmospheric lepton flux calculations, and discuss how the particle production phase space seen by accelerators is probed by the different lepton flux observations. I will lay out a path toward achieving a sub-10% flux uncertainty over almost the entire energy range probed by large volume neutrino telescopes. This level of precision may trigger new approaches to analyses of already available data and shed light on some systematic uncertainties, which are degenerate with the atmospheric flux model.