The authors argue that software is undergoing a third paradigm shift, from traditional instructions to data-driven machine learning, to context- and reasoning-driven Software 3.0. They propose that this new form converges to three elements: a generalized database, a large model for intelligence and generation, and an agent connecting the two. The authors formalize their convergence thesis, present a reference architecture, and analyze its limitations in terms of determinism, cost, security, and verifiability. --- Why it matters: This matters because it challenges traditional software development approaches and proposes a new architecture that could reshape the roles of developers and the database industry. Source: https://arxiv.org/abs/2608.20201