Ben Horowitz and Martin Casado speak with TypeSafe AI founder Diogo Almeida about why extraordinary AI capabilities have produced less real-world automation than expected. They examine the limits of coding agents, the difficulty of reliable execution across long-tail exceptions, and the need for software guarantees. Diogo presents Jev as a new software primitive that embeds language understanding, probabilistic decisions, and state-machine behavior directly into applications. The conversation also considers why established SaaS companies may benefit most from this shift and how software could evolve toward systems that reliably do what people mean.