“Vibe coding” and “AI-assisted coding” overlap, but they emphasise different working relationships. AI-assisted coding usually leaves the human in direct control of implementation. Vibe coding delegates more of that implementation and steers through intent, behavior, and feedback.

The practical difference

In AI-assisted coding, a developer might request a function, accept autocomplete, ask for an explanation, or generate tests while continuing to design and edit the surrounding code directly.

In vibe coding, the human more often asks for an outcome—“add account export,” “make this mobile layout work,” or “trace and fix this failure”—then evaluates a multi-file result. The conversation becomes the primary programming interface.

Neither label proves quality. An experienced engineer can vibe code with strong tests and careful review. A manually written system can be unsafe and unmaintainable.

A spectrum, not two boxes

It is more useful to describe increasing delegation:

  1. Explanation: AI answers questions; the human writes the change.
  2. Completion: AI suggests local code while the human controls structure.
  3. Task generation: AI implements a bounded function, test, or component.
  4. Repository editing: an agent inspects and changes several files, then runs checks.
  5. Outcome steering: the human primarily specifies behavior and evaluates the running system.
  6. Agent-led maintenance: agents plan and execute broad work with humans reviewing checkpoints and releases.

A single project can move across this spectrum. A prototype may begin with outcome steering, then receive conventional review for authentication and deployment.

Better disclosure questions

Percentages such as “90% AI-generated” sound precise but are difficult to define. Lines changed do not measure design influence, debugging, review, or generated code later replaced.

Instead, disclose:

  • which tools materially contributed;
  • whether AI produced isolated suggestions or whole features;
  • what humans reviewed;
  • what automated checks run;
  • whether sensitive or high-impact areas received specialist review;
  • what remains experimental.

Choosing a badge

Use a general Proudly Vibe Coded badge when conversational or agent-led generation was central to the project. Add tool-specific badges when naming the model or coding agent provides useful context. The badge should link to a README note or project documentation when the development method itself matters.

The badge is a disclosure signal, not a certification. Its value comes from the transparent context around it.

Further reading