Vibecoding is attracting real money. Is it producing real software?

The debate started, as many do, with a question that sounds stupid until you sit with it. A Hacker News post from earlier this week didn't bother with nuance: has vibecoding actually produced anything real, or is this just a new way for developers to feel productive while generating garbage?

Andrej Karpathy coined the term in early 2025 with characteristic directness — stop reading the code the LLM writes, stop second-guessing it, just accept the output and ship. It was partly a provocation, partly a genuine workflow philosophy. Fourteen months later, nobody seems entirely sure which.

The investment case for the tooling around vibecoding is, at least, not ambiguous. Cursor's parent company Anysphere hit a reported $2.5 billion valuation after its Series B. Lovable — previously GPT Engineer, rebranded with better marketing — claims $17M ARR. Bolt.new says it had millions of users churning out hundreds of thousands of apps at its peak. Replit crossed a billion-dollar valuation by repositioning as AI-native. The VCs clearly think something is happening here. What they're actually betting on, though, is mostly the shovels: the editors and platforms, not whatever gets built with them.

That gap is where the real argument lives. The HN thread is full of developers who say they've shipped genuine things with these tools — side projects, internal dashboards, small SaaS products — mostly pairing Claude or GPT-4o with Cursor or Bolt. The failure reports are just as specific: APIs that don't exist, security holes that pass a visual scan, codebases that work on day one and become a nightmare the moment someone tries to change anything. Nobody is seriously disputing that vibecoding is fast. The fight is over whether fast is the same as good, and at what point the speed advantage gets eaten by the maintenance bill.

The harder problem for anyone thinking about this as an investment thesis rather than a workflow: if two founders with identical LLM access can build the same product in the same weekend, what exactly is the moat? The no-code wave from 2019 to 2022 hit this wall — the platforms attracted capital but the things built on them rarely did. Vibecoding's defenders would say the capability ceiling is meaningfully higher this time, which is probably true. Whether it's high enough to produce something defensible is still unanswered, and right now investors are placing bets on the outcome before the results are in.