"Modern software increasingly relies on large AI models whose behavior is defined by two core properties: an unbounded input space and a stochastic output. This breaks the deterministic assumptions of traditional engineering and product design. Inputs can no longer be exhaustively specified, and outputs can no longer be precisely predicted. As a result, systems must be built around probabilistic behavior, live feedback, and empirical validation. Development shifts from constructing fixed logic to managing uncertainty. Metrics shift from exact outcomes to distributions. The core constraint is no longer correctness, but the threshold of acceptable performance—what we define as minimum viable intelligence."

Directly from San Francisco, our guest Gian Segato, Core Contributor at Replit will share his perspective on this evolution.

At the end of the talk, there will be the usual network dinner.

See you there!