The AI-powered, intuition-first approach to coding is gaining momentum. But is it ready for prime time?
In February, OpenAI cofounder Andrej Karpathy introduced the internet to a new kind of software development: vibe coding. In a viral post on X, he described it as a workflow where developers “give in to the vibes,” relying almost entirely on AI assistants to build software, skipping traditional syntax and debugging in favor of intuition and prompts.
“It’s not really coding—I just see stuff, say stuff, run stuff, and copy paste stuff, and it mostly works,” Karpathy wrote.
That offhand comment has since taken off, capturing the attention of developers, hobbyists, and tech thinkers alike. As reported by MIT Technology Review, the term has come to define a growing trend in how people use powerful AI tools like ChatGPT, Claude, and GitHub Copilot to create apps, websites, and digital experiences—often with little to no formal programming experience.
Coding by Instinct, Not Instruction
While AI-assisted development has been around for several years, new tools like Cursor (formerly known as Composer) and advanced versions of Copilot now offer real-time suggestions and full code generation capabilities. This shift has enabled a new class of software creators to emerge—ones who build through prompting rather than programming.
“Autocomplete tools have gone from finishing lines of code to rewriting entire files,” said Barron Webster, a designer at interface company Sandbar, in an interview with MIT Technology Review.
At the core of vibe coding is trust: trusting the AI to build, fix, and adapt software without micromanagement. Karpathy himself described simply pasting error messages back into the tool and moving on, rarely stopping to understand what went wrong.
A Divisive Approach
But not all AI-assisted coding qualifies as vibe coding. To truly embrace the philosophy, developers must surrender control—resisting the urge to analyze or manually edit the code AI outputs. Critics argue this hands-off approach can quickly become risky, particularly in complex systems.
“Vibe coding can make a lot of errors,” said Canva engineer Sergey Tselovalnikov. “In a small game or a lightweight app, those don’t matter much. But in larger projects with sensitive data or user interactions, it’s a different story.”
This distinction is especially important for non-technical users. One such enthusiast, known as Leo on X, built a SaaS application entirely through Cursor. But within days of sharing his work publicly, users identified glaring security vulnerabilities. He later admitted the project had overwhelmed him: “I’m not technical, so this is taking me longer than usual to figure out.”
Who Should Be Vibe Coding?
According to MIT Media Lab researcher Tobin South, the approach is most useful for two groups: skilled developers who can fix AI-generated mistakes if needed, and complete beginners with ideas but no way to execute them.
“I’d define vibe coding as having a vision that you can’t execute, but AI can,” South said.
Used correctly, vibe coding can accelerate prototyping and experimentation. However, it is not a substitute for secure, production-ready development practices. Experts caution that while AI lowers the barrier to entry, it doesn’t eliminate the need for software engineering fundamentals, especially when building for real-world users.
The Future of AI-Led Development
Despite the current limitations, the rise of vibe coding signals a fundamental shift in how software is created. As AI tools continue to improve, and as platforms integrate them more deeply into development environments, the traditional boundaries between coder and non-coder may continue to blur.
“The cost of producing software is dropping exponentially,” South told MIT Technology Review. “The world will have to adapt to this new reality. It isn’t going anywhere.”
For now, vibe coding remains a useful tool—particularly for creators experimenting with simple digital projects. But as it grows in popularity, it also raises deeper questions about trust, accountability, and the role of human intuition in an increasingly automated development process.
