That's a very good point. I especially agree that those safeguards should be deterministic regardless of whether the change was written by a developer or an AI agent. It also makes me wonder where we draw the line between vibe coding and AI-assisted engineering. The original appeal of vibe coding is that intent can replace much of the implementation knowledge required to build something. Once we introduce structured tests, static checks, architectural controls and human review, AI is still making development faster—but perhaps we're no longer really “vibe coding.” We're engineering with AI as an accelerator. That distinction is something I've been thinking about more since writing this.
