The Vibe Coding Debate: Evolution or Crisis in Software Development?

AI generated Image

A new term emerged in early 2025 that would spark one of the most contentious debates in modern software development: "vibe coding." Coined by Andrej Karpathy, co-founder of OpenAI and former AI leader at Tesla, the concept describes a fundamental shift in how software gets created—from writing code to orchestrating AI agents through natural language.

The term was introduced in February 2025, and by year's end, the industry was grappling with both unprecedented productivity gains and alarming warnings about code quality, security, and the future of engineering itself.

What Is Vibe Coding?

Karpathy described vibe coding as "fully giving in to the vibes, embracing exponentials, and forgetting that the code even exists". Rather than manually writing code line by line, developers provide natural language descriptions of their goals and let AI agents handle implementation details.

While Karpathy originally described this as "not too bad for throwaway weekend projects," the concept quickly spread beyond experimental prototypes into production environments and startup codebases.

The Rapid Adoption

The speed of adoption has been striking. Searches for "vibe coding" reportedly jumped 6,700% in spring 2025, reflecting widespread experimentation across the industry.

The numbers from major tech companies tell a compelling story. Microsoft CEO Satya Nadella estimated in April that between 20% to 30% of his company's code had been generated by AI, similar to an estimate by Google CEO Sundar Pichai from October 2024. Anthropic CEO Dario Amodei has said that 90% of Anthropic's code is now being written by AI.

The startup ecosystem has embraced vibe coding even more aggressively. Y Combinator reported in March 2025 that 25% of startup companies in its Winter 2025 batch had codebases that were 95% AI-generated.

The Productivity Promise

For experienced engineers, vibe coding can dramatically accelerate development. The most striking example came in early January 2026, when Jaana Dogan, a Principal Engineer at Google working on the Gemini API, shared her experience with Claude Code.

Dogan reported that she gave Claude Code a problem description and got back a result in one hour that matches what her team had been building for the past year. The task involved distributed agent orchestrators—systems that coordinate multiple AI agents—a problem Google had been working on without reaching consensus.

The prompt wasn't particularly detailed, just three paragraphs, and she built a simplified version based on existing ideas since she couldn't use internal company details. While the output wasn't perfect and needed refinement, the ability to rapidly prototype complex architectures demonstrated the power of these tools in expert hands.

Seventy-four percent of developers report increased productivity with vibe coding tools, and the paradigm is clearly shifting from writing to orchestrating code.

The Dark Side: Security and Quality Concerns

However, enthusiasm has been tempered by concerning findings about code quality and security. A landmark Veracode study in 2025 analyzed over 100 large language models across 80 coding tasks and found that 45% of AI-generated code introduces security vulnerabilities. These aren't minor issues—many are critical flaws from the OWASP Top 10 list of dangerous web application security risks.

AI models "hallucinate" non-existent software packages and libraries—commercial models do this 5.2% of the time, while open-source models hit 21.7%. Malicious actors have begun exploiting this through attacks on AI-generated code.

In May 2025, Lovable, a Swedish vibe coding app, was reported to have security vulnerabilities in the code it generated, with 170 out of 1,645 Lovable-created web applications having an issue that would allow personal information to be accessed by anyone.

The Technical Debt Problem

Beyond security, developers are experiencing unexpected maintenance challenges. Over 40% of junior developers admit to deploying AI-generated code they don't fully understand. Sixty-three percent of developers report spending more time debugging AI-generated code than they would have spent writing it themselves—at least once.

AI-generated code tends to include 2.4 times more abstraction layers than human developers would implement for equivalent tasks, creating unnecessary complexity that becomes technical debt over time.

In September 2025, Fast Company reported the 'vibe coding hangover' is upon us, with senior software engineers citing 'development hell' when working with AI-generated code. In an August 2025 survey by Final Round AI, 18 CTOs were asked about vibe coding and 16 reported experiencing production disasters directly caused by AI-generated code.

Professional Developers Don't Actually "Vibe"

Despite the hype around vibe coding, research reveals that experienced developers use AI tools very differently than the term implies.

A study found that professional developers do not vibe code. Instead, they carefully control the agents through planning and supervision. They are looking for a productivity boost while still valuing software quality attributes, they plan before implementing and validate all agentic outputs, they find agents suitable for well-described, straightforward tasks but not complex tasks, and yet they generally enjoy using agents as long as they are in control.

The reality is nuanced: AI tools are copilots, not autopilots. They can help accelerate work, but require human judgment and oversight, especially for production code.

The Shift to Context Engineering

By late 2025, the industry began moving from vibe coding to what's being termed context engineering. A loose, vibes-based approach has given way to a systematic approach to managing how AI systems process context.

This evolution recognizes that effective AI-assisted development requires structure, not just natural language prompts. Antipatterns have been proliferating, including complacency with AI-generated code, and early ventures into vibe coding exposed a degree of complacency about what AI models can actually handle—users demanded more and prompts grew larger, but model reliability started to falter.

What Vibe Coding Really Reveals

The vibe coding phenomenon has exposed several important realities about AI-assisted development:

Not a Replacement, But a Role Transformation: The most successful use cases involve experienced engineers using AI to accelerate implementation of solutions they already understand conceptually. Dogan recommended that skeptics of coding agents try them in areas where they have deep expertise.

The Experience Gap Widens: Over time, junior developers will likely become less skilled at writing their own code and more dependent on vibe coding tools, widening the skills gap between junior and senior engineers.

Understanding Remains Critical: Computer scientist Gary Marcus noted that vibe coding enthusiasm often stems from reproduction, not originality. Developers may use AI-generated code without fully comprehending its functionality, leading to undetected bugs, errors, or security vulnerabilities.

Appropriate Use Cases Matter: While vibe coding may be suitable for prototyping or "throwaway weekend projects," it is considered by some experts to pose risks in professional settings, where a deep understanding of the code is crucial for debugging, maintenance, and security.

Industry Expert Perspectives

Simon Willison stated: "Vibe coding your way to a production codebase is clearly risky. Most of the work we do as software engineers involves evolving existing systems, where the quality and understandability of the underlying code is crucial".

Andrew Ng has taken issue with the term, saying that it misleads people into assuming that software engineers just "go with the vibes" when using AI tools to create applications.

The Real Story: A Non-Technical Writer's Experience

Perhaps the most revealing perspective comes from actual vibe coding attempts by non-developers. Kevin Roose, a New York Times journalist who is not a professional coder, experimented with vibe coding to create several small-scale applications in February 2025. He described these as "software for one" due to the ability to personalize the software. However, Roose also stated that the results are often limited and prone to errors. In one case, the AI-generated code fabricated fake reviews for an e-commerce site.

A Stack Overflow writer who tested vibe coding found that while it initially "felt like hitting one of those 'That was easy!' buttons from Staples," upon handing the output to someone with technical expertise, "the holes began to show".

The Path Forward

The vibe coding debate isn't about whether to use AI tools—that question has been answered. By late 2025, surveys indicated 84% of developers were already using or planning to use AI coding assistants. The real questions are how to use them responsibly and what skills matter in an AI-assisted future.

The emerging consensus suggests that successful developers will need to:

Master Orchestration Over Implementation: The skill shifts from typing syntax to architecting systems and guiding AI agents toward correct solutions.
Develop Critical Review Capabilities: Reading and evaluating AI-generated code becomes more important than writing it from scratch.
Maintain Deep Understanding: Surface-level prompting without comprehension leads to security vulnerabilities and unmaintainable systems.
Apply Appropriate Context: Tools like CLAUDE.md and structured prompting frameworks help AI produce better results aligned with project standards.

Conclusion

Vibe coding represents a genuine inflection point in software development, but not the one its name implies. Rather than eliminating the need for engineering expertise, it's redistributing where that expertise gets applied—from implementation details to architecture, validation, and system design.

As Dogan observed about the rapid evolution of AI-assisted programming: in 2022, systems could complete individual lines. In 2023, they handled entire sections. By 2024, they could work across multiple files and build simple apps. In 2025, they can create and restructure entire codebases.

The technology has advanced faster than most experts predicted. Dogan noted that in 2022, she didn't believe the 2024 milestone could be practically feasible to scale as a global developer product. In 2023, today's level seemed five years away.

The question isn't whether AI will transform software development—it already has. The question is whether the industry can develop the practices, standards, and skills needed to harness these tools without compromising the reliability and security that production systems demand. Early evidence suggests this will require more discipline and expertise than ever, not less.

cursor

Related Topics

The Vibe Coding Debate: Evolution or Crisis in Software Development?