46% AI-Generated Code — Vibe Coding Explained

GitHub says Copilot writes 46% of code, 84% of developers now use or plan to use AI tools, and the biggest AI coding startup sold for $60 billion, all within about 18 months of Andrej Karpathy coining the term "vibe coding" in February 2025. That's the short answer. The longer answer is that the term itself is already being retired by the person who invented it, even as the workflow it describes has become the default way most software gets written.

It's not a fad — Cursor alone went from $100 million to roughly $4 billion in annualized revenue between January 2025 and June 2026. But the data also shows a real quality gap that founders raising on "AI-native" engineering claims need to reckon with before an investor asks about it in diligence.

What Is Vibe Coding in 2026?

Vibe coding is a software development approach where a developer describes what they want in plain English and an AI model generates the code, often accepted with minimal or no line-by-line review. Coined by Andrej Karpathy, OpenAI co-founder and former Tesla AI lead, in a February 2025 post, the term captures a shift from writing syntax to orchestrating outcomes — the developer states intent, the model produces the implementation, and the human's job becomes describing, testing, and steering rather than typing.

Karpathy himself said in December 2025 that 80% of his own code is now written by agentic AI systems, up from writing 80% manually just a month earlier — a personal data point that mirrors the industry-wide inflection. But by early 2026 he had already moved past his own term, preferring "agentic engineering" to emphasize that overseeing AI agents is a discipline requiring real expertise, not a hands-off vibe. You can see the same tools and adoption curve reflected in our ranking of the top AI coding tools in 2026.

Vibe Coding Adoption Statistics for 2026

Adoption has moved from novelty to default in roughly two years. Stack Overflow's 2025 Developer Survey found 84% of developers use or plan to use AI tools in their development process, up from 76% a year earlier, and 51% of professional developers use them daily. Vibe coding in the strict sense is rarer at work: 72% said it is not part of their professional work. Among Y Combinator's Winter 2025 cohort, 21% of companies reported codebases that are more than 91% AI-generated — a figure only possible for startups with no legacy code to migrate around, which is why adoption skews so much younger than at incumbent tech companies. For each company's latest valuation, revenue run-rate and user count, with a source for every figure, see our vibe coding market data.

The trust gap is the underreported half of the story: more developers distrust the accuracy of AI tools (46%) than trust it (33%), even though most of them use the tools. Reviews have found AI co-authored code contains roughly 1.7x more major issues than human-written code, which is exactly why Karpathy's rebrand to "agentic engineering" — emphasizing oversight — landed with practitioners faster than most terminology shifts do.

How Much Code Is Actually AI-Generated? The Numbers Disagree

Ask five sources what share of code is AI-written in 2026 and you'll get five different numbers, because "AI-assisted" and "AI-generated" get conflated constantly. GitHub said in March 2023 that Copilot was "already writing 46% of code", a vendor figure rather than a platform-wide measurement. Sonar's January 2026 State of Code Developer Survey put committed AI code at 42%. An empirical study covering 4.2 million developers from November 2025 through February 2026 found a lower, stricter figure of 26.9% for AI-authored production code — closer to Microsoft's internal estimate of 20-30% of code inside its own repositories.

Google is the outlier at the high end, reporting 75% of new code was AI-generated and engineer-approved as of April 2026 — nearly triple the more conservative empirical estimates, likely reflecting both Google's internal tooling maturity and a looser definition of "AI-generated." The spread matters for anyone diligencing an engineering team's claims: a founder saying "90% of our code is AI-written" could mean anything from light autocomplete to full agentic generation with no human review, and those are very different risk profiles for a VC underwriting technical debt.

Vibe Coding Market Size and the Companies Behind It

There is no reliable published estimate of the vibe-coding tools market's total size, but the companies show its scale, and a few have captured most of the attention and capital. Cursor-maker Anysphere raised a $2.3 billion Series D in November 2025 at a $29.3 billion valuation, then grew annualized recurring revenue from roughly $1 billion in November 2025 to about $2 billion by February 2026 and toward $4 billion by June 2026 — a pace of revenue growth almost unheard of even by AI-era standards. In August 2026, SpaceX closed its $60 billion acquisition of Cursor. We've covered how AI companies are getting priced at these multiples on our AI valuations dashboard.

Lovable raised $330 million at a $6.6 billion valuation from CapitalG and Menlo Ventures, then hit $500 million in annualized run-rate revenue in June 2026 and raised $400 million at a $13.3 billion valuation in August. Replit raised a $400 million Series D at a $9 billion valuation in March 2026 and hopes to reach $1 billion in ARR by the end of the year; Sacra estimates it reached $525 million in annualized revenue in April 2026 (Replit has not disclosed a current figure). Cognition, which owns Devin and Windsurf, raised $2 billion at a $48 billion valuation in September 2026 with a $900 million run-rate. All three are growing faster than almost any SaaS category comp from the prior decade, and newer entrants are keeping pace — Emergent went from $300M to $1.5B in four months.

Why Karpathy Renamed Vibe Coding to "Agentic Engineering"

The core critique that killed the "vibe" framing is discipline, not capability. Karpathy's new framing — agentic engineering — describes a world where "you are not writing the code directly 99% of the time, you are orchestrating agents who do, and acting as oversight." The distinction matters because "vibe coding" implied you could skip review entirely, and the data shows that's exactly where quality breaks down: 1.7x more major issues in AI co-authored code, and only 33% of developers trusting the accuracy of AI output (Stack Overflow 2025) despite widespread use.

For founders, this is a fundraising and hiring signal, not just a terminology footnote. VCs increasingly ask not "do you use AI to code" — the answer is almost always yes at 84%+ adoption — but "what's your review and testing discipline around AI-generated code." Teams that can answer with specifics on test coverage, agent oversight workflows, and code review gates are underwriting a fundamentally lower risk than teams that just cite a high percentage of AI-generated code as a selling point.

What Vibe Coding Means for VCs Underwriting Engineering Teams

VCs increasingly ask every technical founder the same follow-up question after they mention AI-generated code: what does your test suite look like, and who reviews agent output before it ships to production? The honest answer separates two very different companies that might otherwise look identical on a pitch deck. One has 60% AI-generated code and a rigorous CI pipeline with mandatory human sign-off on anything touching payments, auth, or data models. The other has 90% AI-generated code and effectively no review gate, which is a ticking technical-debt liability dressed up as a velocity story.

This matters more at seed and Series A, where a two- or three-person engineering team can plausibly ship an entire product using Cursor, Replit, or Lovable in weeks instead of months — which is exactly why time-to-MVP benchmarks have compressed industry-wide. But it also means the traditional signal of "how big is the engineering team" is a weaker proxy for defensibility than it used to be, since a five-person team using agentic tooling can now out-ship a twenty-person team from 2021. Track how this shows up in valuation multiples on our SaaS valuations dashboard, where AI-native companies with small headcounts are increasingly commanding revenue multiples that would have required much larger teams five years ago.