“Vibe coding” is now a $4.7bn category — and a trap if you skip the rewrite
The fastest-growing idea in software right now is also the most misunderstood. Vibe coding — describing what you want in plain language and letting an AI generate the app — has gone from a phrase Andrej Karpathy coined in 2025 to an estimated $4.7 billion category in 2026, growing at 38% a year. Roughly 87% of Fortune 500 companies now use at least one of these tools, and the majority of people reaching for them aren’t developers at all.
It’s genuinely powerful for getting an idea on screen. You can have a working prototype of a booking tool, an internal dashboard, or a landing page in an afternoon. For validating whether something is worth building, that’s a real shift.
The catch shows up at scale. Industry reporting this year suggests that in roughly 70% of cases where an AI-built prototype needs to grow into a real product, a full rewrite is recommended — business logic gets tangled into infrastructure, database schemas aren’t built to grow, and APIs were never planned for load. A METR study even found experienced developers were 19% slower on certain tasks once the time spent reviewing and correcting AI output was counted.
Our read: treat vibe coding as a sketching tool, not a foundation. Use it to prove the idea cheaply, then build the version that has to survive real users on solid ground. The teams that get burned are the ones who mistake a convincing demo for a finished product.