In February 2025, Andrej Karpathy — former AI lead at Tesla and co-founder of OpenAI — fired off what he later called a "shower thoughts throwaway tweet." It described a new way of building software where you "fully give in to the vibes, embrace exponentials, and forget that the code even exists." The post collected over 4.5 million views and a new vocabulary entered the tech lexicon: vibe coding.
Fourteen months later, Collins Dictionary has named it Word of the Year, MIT Technology Review listed it among its 10 Breakthrough Technologies of 2026, and 92% of US developers now use AI coding tools daily. What started as a casual observation about prompting Cursor Composer with voice commands has become the defining paradigm shift in how software gets built.
The adoption curve has been steep. GitHub reports that 46% of all new code is now AI-generated. Among Y Combinator's Winter 2025 cohort, one in four startups reported codebases that are over 91% written by AI. And 87% of Fortune 500 companies now use at least one vibe coding platform.
The most visible beneficiary of this wave is Lovable, a Stockholm-based platform that lets users describe what they want and get a working application in return. Lovable hit $200 million in annual recurring revenue within its first twelve months — the fastest in software history. By March 2026, that number had doubled to $400 million, with 15 million daily active users creating 200,000 new projects every day. The company raised $330 million at a $6.6 billion valuation and is on pace to cross $1 billion ARR before year's end.
No story captures the promise — and controversy — of vibe coding better than Medvi. Matthew Gallagher spent $20,000 and two months building a telehealth company from his house in Los Angeles. AI wrote the code, designed to reach the website, generated the ad creatives, and handled customer service. The only other employee: his brother. Medvi pulled in $401 million in its first year and is projected to reach $1.8 billion in 2026.
The story went viral for obvious reasons, though it comes with caveats. The FDA has already issued warnings to Medvi, and The New York Times profiled the company with a healthy dose of scepticism. But the underlying mechanics are instructive: by using AI for development, marketing, analytics, and support — while outsourcing regulated functions to licensed professionals — Gallagher compressed what would traditionally require a team of hundreds into a two-person operation.
Sam Altman predicted in 2024 that AI would make the solo billion-dollar company possible. Gallagher may have delivered the proof of concept.
The more relatable side of vibe coding lives in smaller stories. Priscilla Tina, a 28-year-old product manager in San Francisco with a mechanical engineering degree and zero coding experience, sat down one evening, opened Claude, and described an app that turns personal photos into mailed postcards. Four hours later, Postcard Press was a working prototype. She charges $2 per postcard, about 100 people have used it, and her Instagram reel about the build has over 80,000 views.
Then there's Jacob Klug, who runs a vibe-coding agency building and selling apps on Lovable. His best month: over $170,000 in revenue.
These aren't outliers anymore. They're becoming the norm.
The picture isn't entirely rosy. Developer favorability toward AI tools has actually dropped from 77% in 2023 to 60% in 2026, and only a third trust AI-generated code for accuracy. Research shows AI co-authored code contains 1.7x more major issues than human-written code, and 63% of developers report spending more time debugging AI output than it would have taken to write the code themselves.
Fortune recently framed the core tension well: in the age of vibe coding, trust is the real bottleneck. The tools are fast, often astonishingly so. But speed without reliability creates a different kind of technical debt — one that accumulates quietly until it doesn't.
Vibe coding isn't replacing traditional software development. It's creating a parallel track where the barrier to entry has collapsed entirely. A product manager can prototype in an evening. A solo founder can scale to nine figures. An agency of one can out-ship a team of twenty.
The question is no longer whether AI can write code. It's whether we can build the review, testing, and trust infrastructure fast enough to keep up with what it produces.
One year in, Karpathy's throwaway tweet reads less like a prediction and more like a timestamp — the exact moment the rules of building software changed for good.
Sources:
