Andrej Karpathy: From Vibe Coding to Agentic Engineering

The transition to "Software 3.0" marks a fundamental shift where prompting replaces traditional coding, positioning LLMs as the primary interpreter for digital computation. While "vibe coding" democratizes software creation, professional development is evolving into "agentic engineering," which requires maintaining high quality standards while leveraging AI agents to accelerate workflows. AI performance remains "jagged," excelling in verifiable domains like math and code where reinforcement learning can be applied, yet struggling with tasks requiring common sense or aesthetic judgment. Consequently, the role of the programmer is shifting from writing syntax to designing specifications and overseeing AI agents. Despite the automation of technical details, human understanding, taste, and strategic oversight remain essential, as these qualities cannot be outsourced to models that lack intrinsic motivation, curiosity, or the ability to verify their own reasoning.