Elastic metasurfaces enable precise manipulation of elastic waves, promising transformative advances in vibration isolation, sensing, and imaging. However, conventional designs—often based on the Generalized Snell’s Law (GSL)—suffer from significant efficiency losses and parasitic scattering, limiting their practical utility. While newer approaches such as impedance-matching and diffraction-grating metagratings have improved efficiency, they remain constrained in generating complex wavefields like focused beams. In this talk, we introduce a direct wave-shaping (DWS) topology optimization framework that bypasses traditional design paradigms to automatically generate high-performance, monolithic elastic metasurfaces. To overcome the prohibitive computational cost of full-scale optimization, we employ a parametric geometry model based on movable, deformable, and interactable elliptical voids. This approach drastically reduces design variables while holistically accounting for nonlocal inter-cell coupling—a critical factor often neglected in unit-cell-based methods. We demonstrate the power of this framework through several challenging design tasks: • High-efficiency longitudinal-to-transverse wave conversion with large-angle beam steering, • Wavelength-multiplexed beam steering, • Reflective and transmissive metalenses with numerical apertures exceeding 0.99. Compared to state-of-the-art gradient-index, impedance-based, and hybrid designs, our metasurfaces consistently achieve superior efficiency, significantly reduced spurious scattering, and enhanced focusing performance. We also present experimental validation of the most demanding design—a high-NA transmissive metalens—confirming its practical feasibility. This work establishes a scalable and computationally efficient pathway to realizing practical, high-performance elastic metasurfaces, opening new possibilities for advanced elastic wave control in real-world applications.