Aleksandra Konoplia/Moment via Getty Images
Follow ZDNET: Add us as a preferred source on Google.
ZDNET’s key takeaways
- Vibe coding enables one to program in plain English.
- However, it means speed over code review and rigor.
- Code quality may be inconsistent.
Vibe coding has become the new must-do in technology shops. The term and methodology were first coined by Andrej Karpathy, co-founder of OpenAI, in February 2025. “I just see stuff, say stuff, run stuff, and copy-paste stuff, and it mostly works,” he said at the time.
Also: I’ve tested free vs. paid AI coding tools – here’s which one I’d actually use
“Unlike traditional AI-assisted coding or pair programming, vibe coding means you describe what you want in plain English,” according to a report from v0. It also enables both professional and citizen developers to “let AI suggestions guide projects with little or no code review.” One can even “trust the AI to handle heavy lifts like syntax, structure, and implementation. This shift transforms coding from a technical skill requiring years of training into a conversational interface that anyone can master in hours.”
(Disclosure: Ziff Davis, ZDNET’s parent company, filed an April 2025 lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.)
‘Businesses don’t run on vibes’
Early 2025 data show that 25% of Y Combinator startups had over 95% AI-generated code, commented Varun Badhwar, founder and CEO at Endor Labs. Vibe coding has emerged as the go-to approach to quickly building and developing new products.
Also: I did 24 days of coding in 12 hours with a $20 AI tool – but there’s one big pitfall
Is vibe coding too good to be true?
Veteran industry watcher David Linthicum thinks it is, especially at the enterprise level. “Vibe coding may work for personal projects or hackathons, but the real world demands something more substantial,” he said in a recent LinkedIn post. “Businesses don’t run on vibes — they run on reliability, scalability, and maintainability. The longer an enterprise indulges in vibe coding, the harder and more expensive it becomes to standardize, refactor, and secure their systems.”
The lack of standards that come with vibe coding “means that code quality is wildly inconsistent,” Linthicum also pointed out. “Features implemented one week are duplicative or incompatible with code written the next.” And since vibe coding is a shortcut, it racks up technical debt. The cost of cleaning up after vibe coding is huge. What may start as ‘move fast and break things’ too often becomes move fast and break everything, then spend a fortune rebuilding it.”
Vibe coding apps “keep hitting vulnerabilities: exposing secrets, access misconfigurations, hardcoded credentials,” according to v0.
Also: Why AI coding tools like Cursor and Replit are doomed – and what comes next
For startups and smaller businesses, the risk grows. “As bad actors grow more sophisticated and find new ways to achieve remote code execution, the stakes are going to grow for amateur vibe coders,” Badhwar cautioned. The smaller the organization, the more difficult and costly it is to recover from a breach. “Basing products entirely off AI-generated code is risky.”
How vibe coding can amplify human coding issues
In some ways, vibe coding may amplify and scale issues with human coding. “People don’t do code reviews enough,” Brandon Evans, senior Instructor at SANS Institute, told ZDNET. “This is accelerated by vibe coding. We have been neglecting the reviewing of human-written code as well.”
Vibe coding, powered by GitHub Copilot or any AI tool, “helps in generating responses quickly but often skips best practices, documentation, and structured design,” agreed Naga Santhosh Reddy Vootukuri, principal software engineering manager at Microsoft and author of Vibe Coding with GitHub Copilot. “You need to iterate multiple times or pass in right context a right prompts to achieve your desired results.”
Also: Bad vibes: How an AI agent coded its way to disaster
Vootukuri identifies the primary dangers of vibe coding as “security vulnerabilities, rapid technical debt, fragmented architectures, and code nobody understands or can maintain.” Security is a worrisome issue, as “AI-generated suggestions may look functional but hide subtle bugs or even create new attack vectors,” he added. “Security oversight and deep expertise are critical in avoiding “time and effort in rewriting, support headaches, and unreliable software,”
Code generation tools aren’t new
At the same time, the issues with vibe coding aren’t necessarily new. “Code generation tools aren’t new,” said Louis Landry, chief technology officer at Teradata. “We’ve had scaffolding, templates, and code generators for decades. What’s different now is the scope of what’s possible. It feels magical because it is. You can rapidly prototype and explore experiences and integrations that would’ve taken too much time to justify before. But the fundamentals haven’t changed: Dev teams are responsible for the code they ship, whether a human or machine wrote it. Code review is critical regardless of the source.”
However, Landry continued, “the problem is when teams skip that review process because the output looks polished. We’re early in the maturity curve here. The technology is powerful, but the discipline around it hasn’t caught up yet.” Landry also cautioned against technical debt that may build up with vibe-coding shortcuts.
Also: How to vibe code your first iPhone app with AI – no experience necessary
“Bridging speed with discipline is key — success comes when teams merge AI-driven creativity with robust workflows and clear standards,” said Vootukuri. “Treat AI outputs as rough drafts, never production-ready code. Maintain rigorous reviews, run static analysis, and follow strict coding standards. Document each use of AI and always cross-check security-especially on anything customer-facing or sensitive. With experience and time, you can improve on the prompts and ask the right questions to LLMs, enforce mandatory peer reviews within your team, automate CI tests, and clear alignment with business goals, all help ensure that AI remains an accelerator, not a liability.”
