Vibe coding
Vibe coding is building software by telling an AI what you want in plain language and letting it write the code, often without reading that code closely. You describe the app, try what comes back, say what’s wrong, and repeat. The AI researcher Andrej Karpathy gave it the name in early 2025, and it stuck because it described something a lot of people were already doing.
Why it matters
It lets people who don’t write code build working software: a marketer makes a calculator for the website, an ops lead builds an internal tracker, a founder mocks up a product over a weekend. For engineers, it’s a fast way to throw together prototypes and one-off scripts that were never worth an afternoon of careful work.
The trade is that nobody fully understands what got built. That’s fine for a demo. It’s a problem once real users, real data or real money depend on the result.
Where it goes wrong
Vibe-coded apps tend to work on the happy path and fail at the edges. Common trouble spots are security (API keys left in the code, missing access checks, unvalidated input), data handling, and code that grows into a tangle the AI itself starts to struggle with. Because the person driving never read the code, these problems surface late, often in production.
A good rule: vibe code freely for things you’ll throw away or that only you use. Before anything ships to customers or touches company data, have someone (or an agent with clear instructions) review it the way they’d review any other code.
Vibe coding is not the same as agentic coding
Professional developers increasingly hand whole tasks to coding agents too, but they set the goal, keep tests and review in place, and read the diff before it merges. The tools can be identical. The difference is whether anyone is checking the work.