Vibe coding is a style of machine intelligence-assisted programming where the developer describes what they want in natural language, accepts the generated code, and moves on without deeply reviewing it. Coined by Andrej Karpathy in February 2025, describing his own workflow: see a problem, ask the model, run the result, copy-paste what works.

It has a legitimate place. Prototypes, internal scripts, throwaway tools, and personal projects where the cost of a subtle bug is low and speed is what matters — fine. The problem starts when vibe-coded output reaches production without anyone understanding what it does. You accumulate technical debt you can't describe, bugs become opaque because no human wrote the logic, and security vulnerabilities hide in code nobody has actually read.

Simon Willison drew a useful line with "vibe engineering": machine intelligence-assisted development where you still review, understand, and take ownership of the output. Same tools, different discipline, very different risk profile. The leadership question is straightforward: which one is your team actually doing? If you don't know, assume the riskier one.

Responsible AI-assisted coding is a genuine competitive advantage. Rubber-stamping output is a liability that only surfaces when production breaks — or when something worse happens first.