Vibe Coding
also: agentic coding
Describing what you want and accepting AI-written code without reading it closely — prototyping’s new superpower and production’s new risk.
Vibe coding (Karpathy’s coinage, 2025) is building software by conversation: describe the goal, accept the diff, run it, describe the next change — engaging with behavior rather than code. Brilliant for prototypes and throwaway tools; dangerous when unreviewed code ships to production. It sits at one end of a spectrum with agentic coding — where engineers still own the architecture, review the diffs, and gate the agent with tests, budgets and a designed loop — as the production-grade discipline.
Worked example: building software mostly by describing what you want in natural language and letting an AI generate and iterate on the code, steering by the running result rather than reading every line — a fast path for prototypes and throwaway tools. Gotcha: it trades understanding for speed, fine for a demo but risky for code you must maintain, debug, or secure — the bugs you did not read are still yours; the durable skill is knowing WHEN vibe-coding fits (exploration) and when it does not (production, security-sensitive paths).