The development of the Partially-Averaged Navier-Stokes (PANS) method for turbulence modeling and simulations began more than two decades ago to address challenges in predicting complex turbulent flows. The objective was to achieve an optimal balance between computational cost and predictive accuracy, adapting to the flow complexity and specific applications. While sharing some features with hybrid RANS/LES methods, PANS has distinct original features that allow the adaptation of progress made in RANS methods for obtaining constitutive relations at different degrees of resolution. The presentation will commence with foundational advances addressing sub-grid turbulent transport, spatio-temporal resolution variation, near-wall physics, and low-Reynolds number effects. It will elucidate the physical and mathematical principles underlying the development, emphasizing internal consistency criteria that boost confidence in the model's validity for a given computation through the comparison of prescribed resolution degrees and result analysis. The main focus of the presentation is on future directions, addressing topics such as: 1. Automation of PANS computation by in situ specification of physical resolution for a given grid. 2. Ability to assess the optimal degree of resolution based on the relative importance of coherent structures and stochastic turbulence. 3. Extension of PANS scale-dependent constitutive relation to include compressibility, non-equilibrium, and body-force effects such as streamline curvature, rotation, and gravity. 4. Development of an error management/assessment strategy, including the ability to assess confidence in the calculation even without external data. 5. Judicious incorporation of Machine Learning techniques.