Vibe coding is building software by describing it to an AI and running what it writes. Where the term came from, what it costs, where it breaks, how to start.
Vibe coding means building software by describing what you want to an AI and running whatever it produces. When something breaks, you ask for a fix. You never read the code yourself. Andrej Karpathy, an OpenAI co-founder and Tesla's former head of AI, coined the term in a post on X in February 2025. He described giving in to the AI and forgetting the code even exists, and he pitched it for throwaway weekend projects. By November, Collins Dictionary had made "vibe coding" its Word of the Year.
In 2026 the phrase covers a lot of ground. It can mean someone who has never coded building a booking page in Lovable, or a senior engineer letting an agent draft a feature. What they share is the main skill involved: describing the outcome clearly and judging whether the result works.
The useful line is not which tool you use but whether you understand what was built. When you accept the AI's code because the app seems to run, that's vibe coding. If you read it, test it and could explain it to a colleague, that's AI-assisted programming, even when the AI wrote every line.
Both have their place. Vibe coding is fine for a prototype you'll throw away or a tool only you will use. It's also a cheap way to test whether an idea deserves a proper build. AI-assisted programming is what any software that holds other people's data, takes payments or needs to be maintained for years requires.
A typical vibe coding session in an app builder runs like this:
Developers vibe code in their own tools too. In an editor such as Cursor or a terminal agent such as Claude Code, the loop is the same, except the code lives in a repository they control.
| Tool | Free allowance | First paid plan |
|---|---|---|
| Lovable | 5 build credits a day, up to 30 a month | Pro, with 100 monthly credits that roll over for a month |
| Bolt | 1 million tokens a month, 300,000 a day | Pro, $25 a month, from 10 million tokens |
| v0 | $5 of credits a month | $30 per user with $30 of credits |
| Replit | No free plan listed on the pricing page | Core, $20 a month or $18 billed yearly |
| Figma Make | 150 AI credits a day, up to 500 a month | Included with paid Figma seats |
| Cursor or Claude Code | Cursor's Hobby plan; no free Claude Code | $20 a month for Cursor Pro or Claude Pro |
Prices from each vendor's pricing page, checked in late September and October 2026.
The table understates the real cost of a bigger project, because these tools charge for work rather than for time. Every change uses credits, and so does every attempt to fix a bug. Once an app grows past a few screens, a weekend of back-and-forth can burn a month's allowance. Top-ups or a larger plan are the usual next step.
Security. In 2025 a researcher scanned 1,645 apps built with Lovable. In 170 of them, anyone could read the database, because the generated projects had left Supabase's row-level security off or set it up wrong. The flaw was published as CVE-2025-48757 in June 2025, and it exposed names, payment details and API keys. The AI built what was asked for and nobody checked who could read the data.
Lost work. In July 2025, SaaStr founder Jason Lemkin was nine days into building an app with Replit's agent. During a declared code freeze, the agent deleted his live production database, then claimed a rollback was impossible. It wasn't. The data came back through Replit's rollback feature, Replit's CEO apologized, and the company now separates development and production databases automatically.
Code nobody understands. A vibe-coded app works until it needs a change the AI can't make cleanly. Without someone who can read the code, each fix risks breaking something else. The project slows down as it grows.
Fix loops. When the AI can't solve a bug, it may keep switching approach. Each attempt costs credits, and the problem moves around instead of going away.
Describing software precisely, breaking a project into steps and recognizing when output is wrong are real skills, and they transfer between tools. Product managers and designers use vibe coding to turn ideas into working prototypes instead of slides, and founders use it to test demand before hiring. What it doesn't replace is knowing how to debug and secure software. Teams that ship AI-built code to production pair it with people who can.
If you've never coded, start with an app builder such as Lovable, Bolt or v0 on a free plan, and build something small for yourself first. Designers can try Figma Make from an existing design. Developers who want the speed without giving up control should look at Cursor or Claude Code, compared in Claude Code vs Cursor.
Our vibe coding tools guide matches tools to each stage of a build, and best no-code AI app builders covers what breaks as an app grows. For the agents behind all of this, see what AI agents are.
Andrej Karpathy, an OpenAI co-founder and former head of AI at Tesla, in a post on X in February 2025. He described a way of programming where you give in to the AI, accept its changes without reading them closely and forget that the code exists, and said it suited throwaway weekend projects. Collins Dictionary named vibe coding its Word of the Year in November 2025.
You can start free. Lovable's free plan gives five build credits a day, up to 30 a month, Bolt's gives 1 million tokens a month, v0 includes $5 of monthly credits and Figma Make allows 150 AI credits a day on the free Starter plan. Serious projects usually need a paid plan, typically $20 to $30 a month, and every fix the AI makes uses credits, so bigger apps cost more.
Not for prototypes, internal tools and personal projects, where speed matters more than perfect code. It becomes risky when an app handles real users' data or money without anyone reviewing the code: AI-built apps have shipped with open databases, and in July 2025 an AI agent deleted a live production database during a code freeze.
It can help, mostly as a way to show what you can build. Product managers, designers and founders use it to make working prototypes instead of mockups. Engineering roles still expect you to read, debug and secure code, so vibe coding works best as a fast first step on top of real fundamentals rather than a replacement for them.
Whether anyone understands the code. In vibe coding you judge the result by running it and ask the AI to fix whatever breaks. In AI-assisted coding a developer reviews, tests and understands what the AI writes, using tools like Cursor or Claude Code to work faster. The same tools can be used either way.
No spam. Unsubscribe anytime.