Vibe coding and agentic engineering are getting closer than I’d like

Simon Willison reflects on a podcast conversation where he realized that vibe coding and agentic engineering are converging in his own practice. He describes how, as AI coding agents become more reliable, he finds himself not reviewing every line of generated code — even for production systems — and grapples with the ethical implications. He draws an analogy to trusting another team's internal service without reading its source code. He also discusses how AI has shifted software development bottlenecks upstream to design and requirements, why he still values experienced engineers, and why he believes real-world usage of a project matters more than the presence of tests or documentation when evaluating AI-generated code.