The creator of one of the most widely used and admired AI-driven programming tools has expressed that the technique known as “vibe coding” possesses distinct advantages, yet its utility remains inherently limited. Boris Cherny, the accomplished engineer responsible for developing Anthropic’s Claude Code, discussed this nuanced perspective during an episode of *The Peterman Podcast* released on Monday. While he acknowledged that vibe coding—an approach characterized by intuitive, exploratory, or AI-assisted software creation—can be valuable in certain contexts, Cherny stressed that it cannot serve as a one-size-fits-all solution for every coding challenge.
Cherny elaborated that this freeform style of development works particularly well in low-stakes environments, such as when producing experimental snippets, temporary scripts, or prototypes that need not meet strict performance or reliability standards. In his words, vibe coding is ideal for “throwaway code and prototypes, code that’s not in the critical path.” He admitted that he frequently engages in this mode of programming himself, finding it useful for achieving rapid conceptual results. Nonetheless, he cautioned against overreliance on this method, remarking that while it can accelerate early ideation, it falls short when stability, scalability, and long-term maintainability become priorities. “You want maintainable code sometimes,” Cherny explained. “You want to be very thoughtful about every line sometimes,” underscoring the need for deliberate craftsmanship alongside creative experimentation.
Anthropic introduced Claude Code earlier this year as part of its broader mission to embed artificial intelligence more deeply into modern software development workflows. The tool represents an integral part of Anthropic’s ecosystem, with several major AI coding platforms—such as Cursor and Augment—built on the company’s underlying models. Even Meta has integrated these models into its own coding assistant, illustrating the reach of Anthropic’s technology. Beyond professional engineers, Claude Code has gained traction among non-technical creators who wish to construct functioning programs and applications simply by expressing their intentions in natural language. The company’s CEO, Dario Amodei, revealed in October that Claude is now responsible for generating an astonishing ninety percent of the code written internally at Anthropic.
When tackling critical or highly sensitive programming tasks, however, Cherny takes a more structured approach—one that balances AI collaboration with human oversight. He often begins by prompting the model to produce a detailed plan or outline, then proceeds through the implementation process incrementally, refining each stage with great care. “I might ask it to improve the code or clean it up or so on,” he noted. For subsystems where he holds strong technical opinions or where precision is paramount, he still prefers to write code entirely by hand. Despite the impressive progress of current AI models, Cherny remains candid about their limitations, asserting that they are “not great at coding” yet. Looking ahead optimistically, he emphasized that this moment represents “the worst it’s ever going to be,” implying that continued innovation will rapidly elevate model quality. Comparing today’s AI capabilities to those of merely a year ago, he characterized the transformation as “insane,” observing that coding assistants once limited to basic autocompletion have evolved into tools capable of generating sophisticated logic structures. “Now it’s a completely different world,” he said, expressing genuine excitement at the exponential pace of improvement.
The growing phenomenon of vibe coding has become central to the ongoing discussion about AI’s role in software creation. Across the technology sector, AI-assisted development tools are transforming how professionals and amateurs alike write code. Google’s CEO, Sundar Pichai, recently reflected on the trend, declaring that vibe coding has made the act of programming “so much more enjoyable.” He pointed out that individuals with little or no formal technical experience can now design simple applications or websites, dramatically lowering the barrier to entry in computer science. In a podcast conversation with Logan Kilpatrick, head of Google’s AI Studio, Pichai added that this accessibility has rekindled enthusiasm for programming: “Things are getting more approachable, it’s getting exciting again, and the amazing thing is, it’s only going to get better.”
The momentum is equally evident in quantifiable terms. During Alphabet’s April earnings call, Pichai stated that AI systems are now writing over thirty percent of all new code produced at Google—up from twenty-five percent just six months earlier. This rapid progression, he suggested, underscores a paradigm shift in how development teams collaborate with intelligent systems. Other industry leaders, including Andrew Ng, the founder of Google Brain, have echoed this optimism. In May, Ng described it as “fantastic” that developers can generate functional software at previously unimaginable speeds, sometimes “barely looking at the code.” At the same time, this convenience comes with a caveat. Business Insider recently reported that many non-technical professionals are using vibe coding to automate repetitive tasks, produce spontaneous prototypes, or realize creative side projects—activities that thrive on flexibility but do not demand industrial-strength robustness.
Despite the enthusiasm, experts and executives consistently warn that the technology’s current capabilities remain imperfect. AI-generated code, for all its fluidity, can contain subtle errors, inefficiencies, or inconsistencies in structure that complicate maintenance. The output might perform adequately in the short term but lack the rigor demanded of production-grade software, where reliability, clarity, and security are non-negotiable. As Pichai himself admitted in November, “I’m not working on large codebases where you really have to get it right, the security has to be there.” His comment underscores a shared industry understanding: while vibe coding fuels creativity and accelerates exploration, the discipline and precision of traditional software engineering remain essential for building durable systems.
In summary, Cherny’s remarks illuminate a crucial balance between innovation and responsibility in the evolving relationship between humans and AI within programming. Vibe coding embodies the thrill of creative spontaneity made possible by artificial intelligence, but its very freedom can become its weakness when applied indiscriminately. The future of software development will likely hinge on mastering the art of combining that excitement with the enduring values of clarity, maintainability, and thoughtful design—a synthesis that transforms fleeting inspiration into enduring technical achievement.
Sourse: https://www.businessinsider.com/claude-code-creator-vibe-coding-limits-boris-cherny-anthropic-2025-12
