The physical world is vast, noisy and dynamic. Nonetheless, sensory perception of it is reliable, allowing accurate judgments and actions. This conjunction of the variable world and trustworthy perception of it, suggests that sensory neurons are under selective pressure to exploit environmental regularities, such as contrast or colour in natural scenes, generating structural and functional adaptations, which improve sensory information estimates [1]. However, little is understood how the sensory structures and their neural information processing evolved together. Here we show that by sequestering “unreliable” biochemical reactions into semi-autonomous transduction units, insect phototransduction evolved to generate reliable neural representations of natural light changes, from single photon regimes to full sunlight [2]. To better understand the mechanistic basis of this performance, we synthesized a virtual Drosophila photoreceptor from its biophysical parts. By combining stochastic simulations with intracellular experiments in Drosophila photoreceptors, we show how discrete sampling by phototransduction units (~30,000 microvilli), each producing variable quantum bumps to captured photons, governs macroscopic responses to light changes. Feedbacks and stochasticity in these reactions oppose saturation, dynamically adjusting availability of microvilli, whilst intracellular calcium and voltage adapt their sample size (bump waveform), generating reproducible responses. We further demonstrate how these rules for adaptive sampling predict information processing across a range of species with different visual ecologies. Our results clarify why fly photoreceptors are structured the way they are and function as they do, linking sensory information to sensory evolution and highlighting the benefits of stochasticity for neural information processing [3].