The rise of AI coding agents, like Claude Code, has led to "vibe coding," where developers rely heavily on AI, leading to a decline in personal code understanding and an increase in unreliable outputs. Andrej Karpathy coined the term, noting his own shift away from manual coding since December 2025. This seismic change in software engineering, while potentially accelerating development, brings anxieties about code quality and the "tidal wave of slop."

Experts like Grady Booch view this as the third golden age of software engineering, accelerated by AI. However, the current "vibe coding casino" approach, characterized by under-specified prompts and non-deterministic generation, erodes a developer's grasp of the codebase. This "black-box effect" can lead to blindly committing code and production outages. The core issues are a lack of trust in the agent's output and the loss of personal understanding.

Proposed solutions, such as markdown specs and skills, are largely immature. They often devolve into advanced prompt engineering rather than true intent distillation. To combat vibe coding, solutions must ensure agents reliably perform requested tasks, allow for auditable tracking without reading code, and maintain minimal friction. The disciplined path, currently the path of most resistance, needs to become the easy one.