The Uncomfortable Convergence of Vibe Coding and Agentic Engineering

Two Concepts Collide

Simon Willison, author of Datasette and a leading voice on responsible AI coding practices, has shared an uncomfortable realization: the line he carefully drew between "vibe coding" and "agentic engineering" is starting to blur in his own work.

The Original Distinction

Willison had argued that vibe coding — where you don't look at the code, don't understand it, and just hope it works — is fundamentally irresponsible when building software for others. Agentic engineering, by contrast, means a seasoned developer leveraging AI tools while maintaining deep accountability for what gets shipped.

Where the Blur Starts

The problem: Willison now finds himself not reviewing every line of AI-generated code for standard tasks like JSON API endpoints or SQL queries. His reasoning is that Claude Code simply does these correctly, and he knows it will add tests and documentation. But if he hasn't reviewed the code, is it genuinely responsible to ship it?

The Question for the Industry

This piece cuts to the heart of a question the software industry hasn't fully answered: as AI coding agents become reliably correct for routine tasks, what does code review mean? Is not reviewing trustworthy AI output a failure of due diligence, or a sensible allocation of expert attention? Willison doesn't resolve the tension — but he's right to name it.