Workflow & practice

Vibe coding

Vibe coding is building software by prompting an agent and steering on feel, accepting the code it writes without reading every line. It is fast and freeing for prototypes and personal tools, and genuinely risky the moment the code has to run in production.

James Phoenix
Understanding Data Updated July 2, 2026

Vibe coding is building by prompting an agent and steering on feel. You describe what you want, run whatever it produces, glance at the result, and prompt again. You accept the code without reading every line. The term is only half a joke: for a lot of work this is genuinely how people build now.

Where it shines

For throwaway work it is excellent. A quick script, a UI you are exploring, a prototype to test an idea: speed matters far more than craftsmanship, and vibe coding lets you cover ground in minutes that used to take an afternoon. If the code is going to be deleted next week, reading every line is wasted effort.

Where it bites

The risk is shipping code nobody understands. When you have not read it, you cannot reason about its edge cases, its security, or how it fails under load. Bugs hide in the parts you skimmed. And because a model sounds confident whether it is right or wrong, "it looked fine" is not evidence that it is fine.

  • Great for prototypes, demos, and personal tools.
  • Dangerous for anything with users, money, or data behind it.
  • The line is production. Once code ships, someone has to actually understand it.
Watch out
Vibe coding is not a substitute for human review. It is a way to move fast up to the point where correctness starts to matter. Past that point, slow down and read the code, or have something read it for you.

Related terms

Building with AI agents?

This dictionary is part of how I think about agentic engineering. If you want the same thinking applied to your codebase, that is what I do.

See how I can help