The pyramid wavefront sensor (PWFS) can provide the sensitivity needed for demanding adaptive optics applications, such as imaging exoplanets using the future extremely large telescopes of over 30 m of diameter (D). However, its exquisite sensitivity has a limited linear range of operation, or dynamic range, although it can be extended through the use of beam modulation—despite sacrificing sensitivity and requiring additional optical hardware. Inspired by artificial intelligence techniques, this work proposes to train an optical layer—comprising a passive diffractive element placed at a conjugated Fourier plane of the pyramid prism—to boost the linear response of the pyramid sensor without the need for cumbersome modulation. We develop an end-2-end simulation to train the diffractive element, which acts as an optical preconditioner to the traditional least-square modal phase estimation process. Simulation results with a large range of turbulence conditions show a noticeable improvement in the aberration estimation performance equivalent to over 3λ/D of modulation when using the optically preconditioned deep PWFS (DPWFS). Experimental results validate the advantages of using the designed optical layer, where the DPWFS can pair the performance of a traditional PWFS with 2λ/D of modulation. Designing and adding an optical preconditioner to the PWFS is just the tip of the iceberg, since the proposed deep optics methodology can be used for the design of a completely new generation of wavefront sensors that can better fit the demands of sophisticated adaptive optics applications such as ground-to-space and underwater optical communications and imaging through scattering media.