Simba
Collins Dictionary has crowned “vibe coding” its 2025 Word of the Year, officially enshrining the practice of writing software you don’t understand into the English lexicon. The term, coined by OpenAI co-founder Andrej Karpathy in February, describes the revolutionary process of instructing an AI to generate code while you “forget that the code even exists”. Karpathy urged developers to “give in to the vibes,” a phrase that would make any computer science professor weep into their copy of “Introduction to Algorithms.” Within nine months, this celebration of willful technical incompetence has been validated by one of the English language’s most venerable institutions and backed by nearly $2 billion in venture capital.
This isn’t democratization. This is the AI bubble’s most honest admission yet: in 2025, understanding how your software product works is considered an optional feature, not a core requirement. The vibes have never been more expensive, or more profitable, or more destined to catastrophically fail.
Following the Money: The $2 Billion Vibe Economy
The evidence of collective delusion is hiding in plain sight, conveniently compiled in Series A through F term sheets. Swedish startup Lovable achieved unicorn status eight months after launch, raising $200 million at a $1.8 billion valuation from Accel. The company now claims 8 million (initially had 2.3 million) active users who can build apps by simply describing them to an AI, no coding knowledge required. Eight months. From zero to nearly two billion dollars. For a platform that helps people write code they won’t and will never understand.
The feeding frenzy extends well beyond Lovable. Cursor, developed by Anysphere, raised $900 million at a $9.9 billion valuation in June 2025, reportedly crossing $500 million in annual recurring revenue while doubling every two months. Vercel secured $300 million at a $9.3 billion valuation in September, boasting a 45x revenue multiple on $200 million ARR—stratospheric even by inflated AI standards! Replit announced a $250 million round at $3 billion valuation and projects hitting $1 billion in revenue by the end of 2026. Magic AI raised $320 million in August 2024.
The arithmetic is straightforward: nearly $2 billion has been invested into platforms whose core value proposition is “typing is hard.” Y Combinator, Silicon Valley’s most prestigious accelerator, reports that a quarter of its Winter 2025 batch consists of startups with 95% of their codebases generated by AI. YC CEO Garry Tan has declared that 10 vibe coders can now match the output of 100 traditional engineers. Even Google CEO Sundar Pichai is personally experimenting with Cursor and Replit to build his own news aggregator, describing it as “delightful”. When a company’s chief executive finds joy in tools that eliminate the need for his own employees’ expertise, the disruption narrative has achieved terminal velocity.
The Code Quality No One Wants to Discuss
Beneath the soaring valuations and breathless press releases lies a technical reality venture capitalists would prefer remain unexamined. Research reveals that 40% of AI-generated database queries are vulnerable to SQL injection attacks, one of the most basic and preventable security flaws. Code churn—the rate at which code is rewritten or discarded—has doubled since the rise of AI coding tools, indicating that the initial output is frequently unusable in production. Developers report AI-generated code suffers from “opaque logic,” “inconsistency in style,” “hidden bugs,” and “documentation gaps” that make maintenance a nightmare.
The defining characteristic of vibe coding, according to programmer Simon Willison, is accepting AI-generated code without fully understanding it. This is not a bug in the system; it is the system’s core feature. Replit CEO Amjad Masad has openly stated, “We don’t care about professional coders anymore”—the company is targeting “the billion people who want to build software but can’t code”. The implication is clear: technical competence is now an impediment to market expansion, not a prerequisite for building reliable software.
When products break—and they will—developers are left with codebases they cannot explain, adapt, or audit effectively. Debugging degenerates into a “trial-and-error loop” where inexperienced developers prompt the AI repeatedly with slight variations, accepting unverified fixes without tracing root causes. This isn’t engineering; it’s digital divination. The software works until it doesn’t, and when it fails, no one in the room understands why.
Collins Dictionary describes vibe coding as “a major shift in software development, where AI is making coding more accessible”. Accessible to whom? Certainly not to the developers who will inherit these opaque, undocumented, security-riddled codebases. The word “accessible” is performing heroic rhetorical labor here, reframing technical debt as democratic progress.
The Bubble’s Signature: Legitimizing Ignorance
The selection of “vibe coding” as Word of the Year is not a linguistic observation; it is a cultural capitulation. Collins Dictionary’s managing director, Alex Beecroft, praised the term as evidence of “how language is evolving alongside technology,” celebrating “the seamless integration of human creativity and machine intelligence”. This is the vocabulary of inevitability, the surrender disguised as insight. When a phrase that literally means “I don’t know how this works” achieves official lexicographical status within nine months of its coinage, the bubble has metastasized beyond financial speculation into institutional validation.
Karpathy introduced the term in February 2025, and by November it had been canonized by Collins Dictionary. No other tech buzzword in recent memory has traveled so rapidly from Twitter thread to dictionary definition. “Cloud computing” took years to achieve mainstream recognition. “Blockchain” required countless explainers and debates. “Vibe coding” required only the spectacle of founders raising billions while openly admitting they don’t understand their own products.
The VCs have spoken with their checkbooks, and the lexicographers have followed with their imprimatur. In the AI bubble, the fastest way to achieve legitimacy is to brand your ignorance with a catchy name and raise a Series A. The fact that Lovable, a company founded in 2023 and launched commercially in 2024, could raise $200 million at a $1.8 billion valuation just eight months later suggests that due diligence has been replaced entirely by vibes. When revenue multiples reach 45x and founders declare professional coders irrelevant, the market is not pricing in future cash flows—it is pricing in the collective delusion that this time, the exponential growth curve won’t revert to the mean.
History doesn’t repeat itself, but it does rhyme. The Dutch bought tulip bulbs they couldn’t plant. The dot-com era funded businesses with no path to profitability. Today’s believers are funding platforms that generate code no one can maintain. The asset class changes; the madness of crowds endures.
The Verdict
“Vibe coding” is not Word of the Year because it represents linguistic innovation. It is Word of the Year because it perfectly captures the moment when the tech industry stopped pretending that competence matters. In an ecosystem where eight months from launch to $1.8 billion is considered a reasonable trajectory, where a quarter of Y Combinator startups consist almost entirely of code their founders don’t understand, and where dictionary publishers rush to legitimize a term that celebrates ignorance as disruption, the bubble has achieved something remarkable: it has made its own absurdity official.
The billions pouring into vibe coding platforms are not investments in sustainable businesses. They are bets that the music will keep playing long enough to exit at a higher valuation. When Garry Tan declares that 10 vibe coders equal 100 traditional engineers, he is not describing productivity gains—he is describing cost-cutting through mass deskilling. When Sundar Pichai finds it “delightful” to use tools that could render his own employees obsolete, he is not celebrating technological progress—he is celebrating margin expansion. When Replit’s CEO announces his company “doesn’t care about professional coders anymore,” he is not targeting an underserved market—he is admitting that quality and maintainability are business impediments to growth at any cost.
The vibes are immaculate. The code is not. The difference between the two will become apparent approximately 18 months after the first major production failure, when a hospital system or financial platform built entirely by AI collapses under its own technical debt and no one in the building understands how to fix it. At that moment, “vibe coding” will earn a different kind of dictionary definition: a cautionary tale, filed under “hubris.”
The Aftermath: Questions for the Digitally Disillusioned
Have you inherited a vibe-coded codebase at work yet? What was the first production bug that made you question whether anyone actually understood the system they built?
When did you realize your company was prioritizing “move fast and break things” over “build things that don’t break people”?
What’s your over/under on how many months until the first major vibe-coded disaster makes international headlines?
GIPHY App Key not set. Please check settings