What is vibe coding?

Vibe coding is building software by describing what you want in plain language and letting an AI tool generate the code. The human picks the destination and the feel — the vibe — and the AI handles the syntax, the boilerplate, and most of the typing.

You're not reading every line. You're reading diffs, testing screens, and prompting again when something is off. The unit of work shifts from 'function' to 'intent'.

Andrej Karpathy posted in February 2025 that he was 'fully giving in to the vibes' and forgetting the code even existed. Within a week the phrase was everywhere — partly because it was funny, mostly because it named a thing thousands of people were quietly already doing.

Before 'vibe coding' the same activity went under names like 'prompt-driven development' or just 'using Copilot'. The new name made it a category. Categories get tools, agencies, conferences, and arguments.

Lovable is the full-stack option — describe an app, get a working app with a database, auth, and a public URL. Cursor is the IDE most serious vibe coders open every morning; it pairs with you on a real codebase. Claude Code lives in the terminal and runs as an agent — give it a task and walk away. Codex is OpenAI's answer to Cursor and Claude Code, fast and increasingly capable.

Most people who vibe code professionally use at least two: Lovable for the first 80% of a new app, Cursor or Claude Code for the long tail of fixes and refactors.

Speed from zero. An MVP that used to take a team three months now takes one person five days. That changes which ideas are worth trying — the cost of being wrong collapses.

Internal tools. Throwaway dashboards, scripts, one-off forms. Vibe coding is perfect for software that only needs to work once or twice.

Founders prototyping in customer conversations. You can change a screen while the customer is still on the call.

Security. AI tools default to insecure patterns — exposed keys, missing row-level security, no input validation. A vibe-coded app put into production without a review is a data leak waiting to happen.

Codebase rot. After a few weeks of prompts the project has dead files, three competing patterns, and the AI starts contradicting itself. The vibe-coded speed only lasts as long as the codebase stays readable.

Hard engineering problems. Anything involving payments reconciliation, real concurrency, complex permissions, or third-party integrations needs a human who understands what the AI is doing.

Pair the vibe with a review. Either learn enough to read your own diffs, or hire someone who does. Even one hour a week of senior eyes will catch the worst category of mistakes.

Keep the project boring. Resist the urge to add a third framework, a fourth state library, or a custom build step. Boring vibe-coded apps stay vibe-codeable.

Audit before launch. Run a security pass before you put a payment form online. Most launch-day disasters were preventable in an afternoon.