Maor Shlomo had no formal coding background. In the spring of last year he sat down with a large language model, described what he wanted, and began tweaking. Six months later, his creation, Base44, a no-code development platform had attracted 250,000 users. Wix, the Israeli web-building giant, snapped it up for 80 million US dollars. Shlomo had, in the parlance of the moment, vibe coded a vibe coding platform.
The term was coined by Andrej Karpathy, a co-founder of OpenAI and former AI lead at Tesla, in a social media post in February 2025. He described an approach to software development in which the programmer “fully gives in to the vibes, embraces exponentials, and forgets that the code even exists.” Collins English Dictionary named vibe coding its word of the year for 2025. Even Linus Torvalds, who built Linux and has the temperament of a man who regards most software as insufficiently rigorous, was at it in January 2026, using Google’s Antigravity tool to vibe code a visualiser component for a personal project.
By the end of last year, 21 per cent of the winter cohort at Y Combinator, Silicon Valley’s most celebrated start-up factory, had codebases that were more than 91 per cent AI-generated. The line between having an idea and shipping a product has never been thinner.
The barrier that remains
None of which means the coders are redundant. A December 2025 analysis by CodeRabbit of 470 open-source GitHub pull requests found that AI co-authored code contained roughly 1.7 times more major issues than human-written code, with security vulnerabilities running at nearly three times the rate. A Veracode report put the share of AI-generated code samples that fail basic security tests at around 45 per cent. Someone has to catch all that before it ships, and that someone needs to know what they are looking for.
But the more interesting question is not who checks the code. It is who decides what to build, and why.
The promise of vibe coding is not that everyone will become a software engineer. It is that the software engineering part has become cheap enough to be almost free. What retains value is the problem (or rather, the ability to recognise one worth solving, understand it well enough to specify a solution, and know how to bring it to market).
Jensen Huang, Nvidia’s chief executive, put it bluntly: “General intelligence is becoming a commodity; the scarce skill will be applying it to specific domains.” The nurse who builds a patient-triage tool, the logistics manager who builds a route-optimisation app, the HR director who automates screening are not competing with software engineers. They are doing something software engineers typically cannot, namely combining intimate knowledge of a problem with the tools to solve it.
Replit’s chief executive, Amjad Masad, has observed that everyone from HR professionals to doctors can now develop apps based on their ideas. “People build so much domain knowledge about their field of work,” he told the Big Technology Podcast, “but they never could make it into software because they didn’t have the skill or capital.” That barrier has now largely fallen. Not for everyone, and not without caveats, but for a meaningful swathe of the workforce.
This is what might loosely be called entrepreneurial AI: the discipline of using these tools not merely to automate tasks but to build products, launch businesses, and test markets faster and more cheaply than was previously conceivable. It is less about prompt engineering and more about product thinking, understanding users, identifying pain points, knowing when to charge and how much.
Who teaches this?
The question of who equips people with these skills is largely unanswered. Computer science departments are, understandably, focused on producing people who can audit AI-generated code, not on training doctors to build apps. Business schools teach entrepreneurship in the traditional sense, which still tends to assume you either have a technical co-founder or a budget to hire one.
Coding bootcamps, which democratised software skills in the 2010s, are scrambling to adapt. Some are pivoting to vibe coding curricula, though many offerings amount to little more than prompt-engineering tutorials dressed up as business training. What is missing is something more substantive: a rigorous curriculum that combines product strategy, market validation, basic user-experience thinking, and just enough technical literacy to know when the LLM has gone badly wrong.
The gap is real and the demand pressing. In 2024, 67 per cent more entrepreneurs launched a venture after being laid off than in the year before, with AI displacement cited as a driver. Those people are not looking for Python tutorials. They want to know how to turn a redundancy cheque and a decade of industry knowledge into a functioning product.
The vibe and the hangover
Caution is warranted. Fast Company reported in September 2025 that the “vibe coding hangover” had arrived, with senior software engineers describing development hell when inheriting AI-generated codebases. One industry analyst has put accumulated technical debt from poorly structured AI code at 1.5 trillion US dollars by 2027. The ease of building something does not guarantee it is worth building, or that it will survive contact with real users.
The skills gap has not disappeared. It has shifted. Whether you can build is no longer the question given that most people, given the right tools and a clear problem, probably can. Whether you can build something that matters, that works at scale, and that people will actually pay for is rather harder. That, as it turns out, is the hard part. It always was.
Photo: Dreamstime.
