AI demands more engineering discipline. Not less

# Summary AI has rapidly progressed from generating unreliable code in 2025 to producing work comparable to median software engineers by late 2025/early 2026, driven by advances in agentic systems and tool use rather than a single breakthrough. The author argues this exponential improvement demands stronger engineering discipline and validation practices—not abandonment of code review—as organizations must now manage AI-generated code at scale with rigor similar to the shift from manually-managed servers to immutable infrastructure.