Skip to content

Knowledge base

Vibe coding

Where vibe coding is the right call, where it breaks down, and what a team switches to once software has to be run and maintained.

Written by

Updated:

Glossary 33

Vibe coding is development by conversation with a model: you describe what you want, take the code it returns, try it and describe the next change. There is no written plan, no agreed definition of done and no automated verification. The person at the keyboard tests and integrates everything by hand, and each session starts with little memory of the last.

Where does vibe coding work?

It works where speed matters more than correctness and no one has to keep the result running. Prototypes, experiments with a new API, one-off scripts and proofs of concept are a good fit.

Where does it break down?

It breaks down on software a team has to run, change and answer for. The output can look correct and still offer no evidence of how reliable or secure it is, or how it will change. Once a codebase outgrows what one person can hold in their head, or wherever compliance and audit apply, the missing structure shows. That is where agentic engineering replaces it: people state what done means, give the agent its context and check the result with tests and gates.

Working well with agents also takes skills vibe coding never builds: reading the shape of a diff, noticing when an agent writes too much code, knowing when to trust and when to verify. Those take months to learn, not hours. See The learning curve.