What is vibe coding? Definition, examples, and where it stops

Vibe coding is building software by describing what you want in plain language and judging the running result, not the code. You prompt, the AI writes, you click around, you prompt again. The code itself becomes something you glance at, or never read at all.

The term is not yet two years old and already has a dictionary entry. This guide covers where it came from, how the loop works, what it is reliably good at, and the line where it stops being enough.

Where the term came from

Andrej Karpathy, the AI researcher and former Tesla AI director, coined it in a February 2025 post: a style of programming where you "fully give in to the vibes, embrace exponentials, and forget that the code even exists." He described talking to his editor, accepting every suggestion, and pasting error messages back in until things worked. He was half-joking about weekend projects.

The internet was not half-joking. Within a year "vibe coding" went from a wry post to a category with dedicated tools and job listings, and in November 2025 Collins Dictionary named it Word of the Year. The joke became an industry because the underlying capability was real: models crossed the line where a described app usually runs on the first or second try.

The loop, concretely

1. describe   "an invoice tracker for freelancers,
               with clients, due dates, and a paid toggle"
2. generate   the AI writes the app
3. run        click around the live preview
4. react      "overdue invoices should sort to the top
               and turn red"
5. repeat     until it feels right

The defining move is step 3. You evaluate behavior, not implementation. Traditional programming reviews the code and trusts the behavior follows; vibe coding reviews the behavior and takes the code on faith. Everything good and everything dangerous about the practice flows from that swap.

Vibe coding vs no-code vs AI-assisted engineering

• Not no-code. No-code tools give you sliders and blocks inside a walled platform; the ceiling is whatever the vendor built, and the "app" often can't leave. Vibe coding produces actual source code in ordinary frameworks. The ceiling is what code can do, and (with the good tools) you can export the repo and go.
• Not AI-assisted engineering. A developer using Cursor or Claude Code to move faster, reviewing diffs as they go, is doing something adjacent but different: the human still holds the code in their head. Vibe coding, in Karpathy's original sense, is defined by not holding it. In practice most builders slide along a spectrum between the two, prompting freely early and reading carefully once something matters.

What it's genuinely good at

• Zero to working prototype. Idea to clickable product in an afternoon. This is the headline capability and it's not oversold; it's why the practice exploded.
• Personal software. Tools with one user: the spreadsheet replacement, the habit tracker shaped exactly like your habit. Software too small to have ever justified a developer now gets built.
• Validation. Founders put real products in front of real users in a weekend and learn whether anyone cares before investing a quarter. The cost of testing an idea fell to nearly zero, which quietly rewrote startup math.
• Learning by doing. Beginners report that watching generated code evolve under their prompts teaches faster than tutorials ever did. The code is right there when curiosity strikes.

Where it stops

The failure line is consistent across every tool: vibe coding is excellent at producing software that works and unreliable at producing software that's safe to run for strangers. Generated apps recur to the same holes: permission checks in the browser, secrets in the repo, databases with security rules off because that made the demo work. We catalogue them (with fixes) in the security guide.

The second stop is subtler: a working app is a small part of a working business. Identity and tenancy, billing that survives declined cards, deploys with rollback, monitoring, marketing that keeps happening. None of that appears in the canvas, because none of it is code you can vibe. We map that gap in vibe coding to production and, for any SaaS, in what a business needs beyond the app. Closing it is the difference between a demo with a URL and a company.

Is it a skill? Increasingly, yes

Watching a good vibe coder and a frustrated one work the same tool is instructive. The good one decomposes ("build the list view first, then filters"), gives concrete feedback with reproduction steps instead of "it's broken," resets to a checkpoint when the model spirals, and knows when to stop prompting and read the code. The frustrated one writes longer and angrier prompts. The tools reward product taste and clear thinking; they only amplify whatever direction you supply.

Common questions

Is vibe coding production-ready?

The code can be, after a hardening pass that generated apps don't ship with (see above). The honest answer for "can I paste a prompt and take strangers' money" is no; for "can a careful person get from prompt to production in weeks instead of months," emphatically yes.

Will it replace programmers?

It's replacing a certain kind of programming: the routine assembly of standard apps. Demand moved up the stack, toward people who can specify systems, judge quality, and own what the machine can't: architecture, security, and the business around the code.

Which tool should I start with?

Depends on who you are; we keep an honestly-ranked list current. Short version: Lovable if you've never coded, Replit if you want one tool for everything, Cursor or Claude Code if you read code.

Where this is heading

The interesting frontier is no longer generating the app; that's table stakes. It's what happens after: the deploys, the operations, the marketing, the running business. The vibe moved up a level, from "describe an app, get an app" toward "describe a business, get a business." That second sentence is the one we're building.

It is also how we built ListMyCar.ai. Claude Code wrote the product features inside a repository Solo had scaffolded, and Solo ran the deploys, monitoring, and search content around it. From April through August 2026, more than 50 people used it to create a listing. The case study shows the operating loops and the numbers, including the ones that are still small.

Vibe-code the app. Let Solo run the business around it. Solo assembles your product on a standard stack of Next.js, FastAPI, Postgres, Stripe, and AWS, then keeps the deploys, monitoring, and marketing moving, with human approval at the consequential steps. Your tools, your repo. Join the waitlist →