'Vibe coding' is here. It's an early look into how AI will disrupt knowledge work
What the rise the of AI-assisted software development tells us the urgent need to adapt education
Over the last year, there’s been a growing chorus of discussion in tech circles about ‘vibe coding,’ or the use of language models to aid in software development. This trendy Gen Z-approved buzzword might conjure images of engineers casually chatting with their AI co-pilots, but it’s probably the single most significant development in the AI space. It is a glimpse into the near future of knowledge work.
If you’re an educator, a parent, or anyone who cares about ensuring students have the skills and knowledge to thrive in the age of AI, pay close attention.
What is ‘vibe coding’?
Earlier this year, Andrej Karpathy—a key early researcher at OpenAI and and director of AI and Autopilot Vision at Tesla—coined vibe coding (I’m going to drop the quotes in part to acknowledge the assent of this term to common vernacular in the AI field) to describe the act of programming with the help of large language models (LLMs). Instead of methodically typing every single line of code, vibe coding entails the use of LLMs to speed up the process. Initially, vibe coding looked a lot like the autocomplete feature in iMessage, where you start typing some code and “tab” forward to accept the suggested line of code. With the growing power of LLMs, software developers are able to communicate a general concept in plain english (e.g. “help me add a dynamic call to action button based on the user’s selection of the drop-down box at the top of this page”).
Vibe coding has rapidly evolved from an experimental curiosity into a core part of software development workflows. Tools like Cursor, above, can instantly suggest whole blocks of code and significantly reduce the tedious aspects of coding. With the help of AI, developers are able to quickly generate content from abstract ideas, shifting fluidly between writing their own code and collaborating with one or more AI co-pilots to create, enhance, or (occasionally) de-bug.
Tools like Cursor streamline tasks that developers have dreaded: documenting codebases, refactoring messy legacy code, and debugging complex errors. By describing a desired outcome in plain English—such as “refactor this file to be more memory-efficient” or “find where this function is causing an error”—developers can offload cumbersome tasks to AI, with the net effect of increasing their productivity.
“This isn’t a fad… it’s actually the dominant way to code…”
There have been smoke signals about the prevalence of vibe coding. Google CEO Sundar Pichai recently reported on an earnings call that AI systems now generate a quarter of new code for the company’s products. The Federal Reserve of St. Louis published data from a Real-Time Population Survey (RPS) that revealed 40% of software engineers are using generative AI at work.
But arguably the most prominent data point came from a survey conducted by Y Combinator of their most recent cohort of technical founders, which they review in detail in an entire episode of their Lightcone Podcast (which should be required listening for all of my readers). YC found that a quarter of founders reported that over 95% of their companies’ codebase was entirely generated by AI.
“This isn’t a fad. This isn’t going away. It’s is actually the dominant way to code and if you’re not doing it you might just be left behind,” said Gary Tan, President & CEO of Y Combinator
Founders described becoming far less attached to their code, finding it easer to “scrap and rewrite” because of how quickly the code could be generated. One respondent reported having multiple AI coding windows open simultaneously, prompting different features in parallel. Another founder put it it simply, “I don’t write code much, I just think and review.”
Another founder highlighted a fascinating generational shift: one of YC’s top performing startups from this batch was founded by a pair of highly technical individuals who never received “classical computer science” training (a term now being bandied about that I personally find both hilarious and eye-opening) because their entire coding careers occurred within the last two years. They have never known a world before tools like Cursor. Vibe coding is creating an entirely new category of “AI-native” developers who approach building software in fundamentally new ways.
What does this mean for computer science education?
The field of computer science education is understandably reeling from the rash of eye-popping statements from CEOs at the frontiers of the AI space.
At the World Government Summit in Dubai, Nvidia CEO Jensen Huang put it plainly: “Over the last 10-15 years, almost everybody who sits on a stage like this would tell you that it is vital that your children learn computer science, everybody should learn how to program. In fact, it is almost exactly the opposite.” He suggested that AI will handle programming tasks with human language becoming the primary way that people develop code.
Sam Altman, in an interview with Stratechery, opined about the trajectory of the field, stating, “my basic assumption is that each software engineer will just do much, much more for a while. And then at some point, yeah, maybe we do need less [sic] software engineers.”
But the YC survey, as well as my own conversations with developers, suggest it’s not so simple.
While vibe coding is extremely powerful in getting from zero to one—in other words, rapidly iterating toward an MVP—once the codebase starts to scale, the cracks show. Today’s models largely struggle with debugging (in fact, many founders report it is faster to ‘take another spin at the slot machine’ and re-try a prompt rather than try to de-bug). They also struggle as a codebase becomes large, interconnected, and mission-critical. When things break (and they will), you need humans who understand how systems work. You need people trained to peel apart complex architectures, evaluate subtle flaws, and design robust solutions from the ground up—even if those people are still employing AI tools to increase their productivity.
This is a broader pattern we’re seeing across other fields where LLMs are being deployed. Whether it’s coding, writing, design, law, or medicine, the most effective AI users are people who already have deep domain expertise. Expertise isn’t obsolete; it’s more important than ever—because the value isn't just in producing outputs quickly. It's in being able to vet, steer, and improve those outputs.
The future of computer science education isn’t about teaching less. It’s about teaching differently. We still need students who can understand how software works at a fundamental level. But we also need to train them to collaborate with AI—to become fluent in prompting, reviewing, debugging, and refining AI-generated outputs. Mastering this hybrid skillset will be critical not just for engineers, but for anyone hoping to thrive in a world where knowledge work is increasingly AI-augmented.
Practically speaking, AI could dramatically lower the barrier to entry for students. When I was in high school, it would take months (if not years) of training in CS before you could create a game or app that was genuinely cool to people that aren’t inherently curious and nerdy. Today, the first day of Introduction to Computer Science could consist of having students use AI to build a playable video game or fashion app. Vibe coding can help teachers create that spark of excitement to draw students in and capture their interest so that they commit to the arduous process of building their computational thinking skills and learning to code.
This playable flight simulator game was created with a single (“one shot”) prompt with Google’s latest reasoning model: Gemini 2.5 Pro
That is, what we teach might not change dramatically. But how we teach CS (and other subjects, frankly) needs to evolve. As AI saves students hours of time writing code and doing rote tasks, we can use that time for engaging project-based learning activities that simultaneously build durable skills like creativity, problem solving, teamwork, and critical thinking. CS education should ensure that students are best equipped to use their coding and AI capabilities to solve real-world problems that they find meaningful.
The overall vibe I’m feeling: urgency
I hear a lot of complacency about the AI revolution. After all, past technology revolutions have happened gradually. Yes, humanity has many times navigated new forms of automation and changes to the way we work, but as David Autor put it in a recent lecture titled “The Work of the Future: Where Will it Come From?”: there’s no rule in labor economics that automation has to happen at a gradual enough pace to allow society to adapt. I worry that we are sleepwalking into massive near-term disruption to traditional career pathways. And the rapid ascendance of vibe coding in the field of computer science portends a trend that is certain to affect other knowledge work fields. How far away are we really from ‘vibe accounting’ and ‘vibe marketing’?
Our educational leaders and policymakers need to take swift action to ensure the education system adapts. We cannot afford to underestimate the speed or extent of this technological shift. Vibe coding isn’t just changing how software is built, it’s redefining what it means to add value in the modern workforce.
The stakes for students, schools, and society couldn’t be higher. We need immediate and wide-ranging investments in teacher training and capacity building across school systems to ensure students are building the durable skills along with the subject matter expertise to effectively augment AI systems. At aiEDU, we believe this means doubling down on long-standing efforts to center durable skills in the educational experience.
Schools that fail to act will risk their students being left behind by the most powerful economic force of our generation. The communities that will be hit the hardest are already the ones who were left behind 25 years ago during the shifts ushered in by computers and the internet. As someone born and raised in Akron, OH (which was recently named among the top 20 cities most at risk from job automation), I’ve witnessed this divide first-hand and don’t want to repeat the same mistakes with a new generation.
The good news is that, in the hands of effective teachers, AI can be a powerful tool to help us in that endeavor.