The origin of life and the emergence of biological complexity pose a persistent combinatorial challenge, exploring a flat 4ⁿ sequence space within known physical limits is insufficient to account for the appearance and scaling of functional polymers. Modern evolutionary theory including the Extended Evolutionary Synthesis addresses aspects of this problem through mutational biases, developmental constraints and canalisation, yet these refinements do not provide a structural mechanism capable of navigating astronomically large configuration spaces. This preprint proposes that far-from-equilibrium chemical systems naturally give rise to a universal gradient-based optimisation dynamic. This dynamic constrains exploration to a narrow, low-dimensional manifold of viable configurations and, once a persistent molecular memory substrate exists, enables cumulative refinement across evolutionary time. Formally, the process is mathematically analogous to stochastic gradient descent where randomness acts only as local noise, while the direction of change is imposed by the structure of the underlying physical landscape. This framework termed biological anthropicity does not replace empirical evolutionary theory but supplies a physically explicit mechanism underlying its observed regularities. It motivates testable predictions about manifold structure, evolutionary trajectories, and the geometry of functional sequence space, suggesting a potentially unifying, computationally informed framework for origin-of-life chemistry and macroevolution without presuming that the analogy is yet complete. Within this framework, life with its functional complexity becomes the cumulative output of a deep-learning-like optimisation algorithm running on chemical hardware for four billion years.