In February 2025, Andrej Karpathy — former director of AI at Tesla, co-founder of OpenAI — posted a casual observation that set the tech world on fire.
He described a new way he'd been writing software: you don't really think about the code at all. You just describe what you want. The AI writes it. You run it. If something breaks, you paste the error back in and say "fix it." You never actually read the code. You just vibe.
He called it vibe coding.
Over the next twelve months, vibe coding went from one man's tweet to a cultural movement. By early 2026, millions of people with zero programming background were shipping real products. Founders were building startups without a single technical hire. Students were creating tools that solved problems in their own lives. And software engineers were debating, loudly, whether their profession as they knew it was over.
The answer is more complicated — and more interesting — than either side admits.
What Vibe Coding Actually Is
Traditional software development requires you to understand the language you're writing in. Python, JavaScript, TypeScript — each has its own syntax, rules, and failure modes. You write code line by line. You debug it. You understand (or try to understand) what each part does.
Vibe coding flips that. You open a tool like Cursor, Windsurf, GitHub Copilot, or just ChatGPT, and you describe what you want in plain English:
"Build me a web app where users can upload a CSV and see it visualised as a bar chart."
The AI writes the entire codebase. You run it. It (probably) works. When it doesn't, you describe the problem:
"The chart isn't showing up. The page loads but it's blank."
The AI fixes it. You never needed to understand what a React component is. You never needed to know what a useEffect hook does. You just needed to know what you wanted.
That's vibe coding. The vibes are the spec.
Why This Is Bigger Than "AI Autocomplete"
AI coding tools have existed for years. GitHub Copilot launched in 2021. But those early tools were assistants for developers — they'd suggest the next line, complete a function, explain an error. You still needed to know what you were doing.
What changed in 2025–2026 was threefold:
1. Models got dramatically better at reasoning about full systems
GPT-4 could write a function. GPT-4o and Claude 3.5 Sonnet could write a whole module. GPT-5 and Claude 3.7 (released in late 2025) can hold an entire codebase in context, understand how pieces interact, and make changes across dozens of files simultaneously.
2. Agent-based IDEs emerged
Tools like Cursor and Windsurf don't just suggest code — they execute it. They run tests. They see the errors. They fix the errors. They can spin up a full-stack app in under ten minutes, including database schema, API routes, and frontend UI. You watch it happen like a contractor building your house.
3. Deployment became trivial
Vercel, Railway, Render, and similar platforms mean that once the code is written, putting it live on the internet takes thirty seconds. No DevOps knowledge required. The entire pipeline from idea to shipped product can happen in an afternoon.
The result: the barrier to building software has collapsed.
Who Is Vibe Coding Right Now?
Non-technical founders
The most significant early adopters. Founders who used to need a co-founder with technical skills — or a $100K+ annual expense in a developer hire — are now building MVPs themselves. A founder with a good idea, $20/month for an AI subscription, and a weekend can have something real.
Companies like Bolt.new, Lovable, and v0 by Vercel have made this even simpler — they're essentially vibe coding platforms designed for non-technical users. Bolt.new reportedly hit $20M ARR within months of launch.
Students and hobbyists
Teenagers and university students are building tools that solve problems in their own lives. A student builds a timetable optimizer for their university. Another builds a tool that automatically emails their landlord rent reminders. The barriers are gone. If you have an idea and can describe it clearly, you can build it.
Working professionals solving niche problems
A nurse builds a tool that calculates medication dosages based on patient weight. An accountant builds a spreadsheet automation that saves three hours a week. A teacher builds a quiz generator for their class. None of them are developers. All of them built something that works.
Developers themselves
Counterintuitively, professional developers are among the heaviest users of vibe coding tools — just for different reasons. Instead of replacing their work, it accelerates it. Tasks that used to take two days take two hours. Boilerplate that used to be tedious gets generated instantly. Senior developers report spending far more time on architecture and product decisions, and far less on typing.
What Vibe Coding Is Terrible At
Here's where honest analysis matters. Vibe coding is powerful. It is not magic. There are real, consistent failure modes.
Security is invisible
AI-generated code often has security vulnerabilities. SQL injection risks, exposed API keys, unvalidated inputs, broken authentication — the AI doesn't always flag these because it's optimised to make things work, not make things safe. If you're building anything that handles other people's data or money, vibe coding without a security review is a serious risk.
Debugging beyond the surface breaks down
When the AI's fix doesn't work, and the second fix doesn't work, and you're three layers deep in a problem you don't understand — you are stuck in a way that a real developer is not. Understanding what the code is doing requires knowing how to read code. Without that, you're just throwing prompts at the wall.
Scale introduces complexity vibe coding can't navigate
Building something for yourself or for a hundred users is different from building something for a million. Vibe coding produces working code for small-scale use. It is not good at producing code with the performance, fault tolerance, and operational robustness that real scale requires. Startups that vibe-code their way to product-market fit almost always need to bring in proper engineers to rebuild the foundation before they can grow.
Maintenance debt
Code you don't understand is code you can't maintain. If the AI that helped you build something changes, or if you hit a problem the AI keeps getting wrong, you may find yourself with a product you can neither fix nor improve.
What This Means for Developers
The honest answer: professional software engineers are not becoming obsolete. But the floor of what they're competing against has risen dramatically.
If your job was "write the code that senior developers spec out," that job is under real pressure. AI can now do that faster and cheaper.
If your job was "understand the system, make the architectural decisions, hold the complexity in your head, and ensure it works at scale" — that job is more valuable than ever. Because someone needs to do that work, and the AI genuinely can't (yet).
The developers who are thriving in 2026 are those who have moved up the stack: from writing code to designing systems, making product decisions, reviewing AI output for correctness and security, and doing the deep technical work that vibe coding genuinely cannot.
The developers who are struggling are those who treated their career as "write code" and nothing more.
What This Means for Non-Developers
This is the part that matters for the vast majority of people reading this.
For the first time in history, the ability to build software is not gated by whether you can memorise syntax, pass a bootcamp, or spend four years in a computer science degree. If you can think clearly about a problem, describe what you want, and iterate based on what you see — you can build things.
That is a profound shift. It means:
- Side hustles that would have required a developer can now be built alone
- Problems in your own industry that you understand better than any developer can now be solved by you
- The "I had this idea but couldn't build it" excuse has expired
What matters now is not knowing how to code. It's knowing what to build, who it's for, and why it matters. Those are product and business skills. They are, arguably, harder than coding — but they're the skills that most people already have or can develop without a computer science degree.
The Tools Worth Knowing in 2026
If you want to start vibe coding, here's the current landscape:
| Tool | Best For | Cost |
|---|---|---|
| Cursor | Full app development in your own editor | $20/mo |
| Bolt.new | Building full-stack apps from scratch fast | Free tier + paid |
| Lovable | Web apps, especially with Supabase backend | Free tier + paid |
| v0 by Vercel | Frontend UI components | Free tier |
| GitHub Copilot | Coding assistance inside VS Code | $10/mo |
| ChatGPT / Claude | Describing logic, explaining errors | $20/mo |
For beginners: start with Bolt.new or Lovable. Describe your app idea. See what it builds. Don't worry about the code. Focus on whether it does what you need.
The Bigger Picture
Vibe coding is not just a new developer tool. It is a redistribution of creative power.
For most of human history, if you wanted to build a physical thing, you needed raw materials and physical skill. The industrial revolution made manufacturing accessible but still required capital and machinery. Software was different — theoretically all you needed was a computer and knowledge. But the knowledge barrier was enormous.
AI has compressed that barrier to near zero for a wide range of applications.
What comes next is unpredictable. But if history is any guide, when the cost of creation collapses, the number of people who create explodes — and the world changes in ways nobody fully anticipated.
The question is no longer: can you build it?
The question is: what are you going to build?
If you found this useful, explore our related pieces on how AI agents actually work and prompt engineering as the skill everyone needs in 2026.