We Keep Renaming AI Coding. Here’s What I’d Call It.

Boris Cherny, who leads Claude Code at Anthropic, has grown frustrated with the term "vibe coding" and begun searching for better terminology to describe how developers actually work with AI coding assistants. The frustration reflects a broader problem in the industry: imprecise language obscures what these tools actually do.

"Vibe coding" suggests something loose or intuitive, yet the reality involves structured prompts, iterative refinement, and deliberate control over code generation. Developers aren't just vibing. They're exercising judgment at each step, directing AI models toward specific outputs, debugging generated code, and managing the system's errors.

The naming problem matters because language shapes how people understand technology. Current terms either romanticize the process or undersell it. "AI-assisted development" sounds passive. "Prompt engineering" overstates the precision involved. "Copilot" implies a backseat role that misses how central human direction actually is.

Cherny's complaint points to a deeper issue: the industry lacks vocabulary for a genuinely new development pattern. Traditional coding means writing instructions manually. AI coding means writing prompts that guide models, then validating and refining outputs. It's neither pure automation nor traditional programming. It sits in an uncomfortable middle ground that existing terminology fails to capture.

What matters here is that naming affects adoption and regulation. If policymakers hear "AI coding" and think fully automated code generation, they may craft rules that don't match reality. If companies market their tools using vague terms, developers form unrealistic expectations and blame the tool when results don't match the hype.

Cherny's search for better terminology reflects what the entire industry needs: clearer language that describes the actual workflow. Terms should acknowledge both the AI's role and the developer's required judgment. They should be honest about what automation exists and where human oversight remains essential.

Until