Why 'Vibe Coding' Won’t Replace Engineers: Insights from Infosys’s Nandan Nilekani

What Is “Vibe Coding”?

“Vibe coding,” also called “随性编码,” is a recently popular software development approach introduced by former OpenAI researcher Andrej Karpathy. Developers describe desired functionality to AI tools in natural language, and the AI generates complete programs, interfaces, front‑end pages, or even whole projects. Tools such as Cursor and Claude Code have lowered the entry barrier, allowing even non‑programmers to produce usable prototypes, which has sparked widespread anxiety about the potential replacement of traditional programmers.

Infosys founder‑chairman Nandan M. Nilekani told an industry conference that the threat of “vibe coding” is overstated because it misunderstands the full scope of software development.

Infosys Leadership’s Core Judgment

During the annual shareholders’ meeting, Nilekani warned that the industry is undergoing a major technological transformation, and AI’s disruptive scale intensifies doubts about the relevance of software engineers. He emphasized that AI can only generate code, not cover the complex, end‑to‑end software engineering workflow.

“The industry is experiencing a profound technical shift. When such shifts occur, people question our purpose, leadership, and ability to sustain growth and profit,” Nilekani said. “Given AI’s larger and more disruptive transformation, these doubts are louder and stronger. The fundamental question is: if programming becomes automated, what is our purpose?” “Enterprise environments are critical. Solutions must complement existing investments, undergo rigorous testing, have robust architecture, and ensure foundational cybersecurity.” He also believes AI will create more demand for services that only humans can provide.

Limitations of AI from a Project Delivery Perspective

Requirements and Business Understanding Cannot Be Handed to AI – Real business needs are often vague, fragmented, and embedded with industry‑specific rules, requiring engineers to iteratively communicate with stakeholders to produce clear, implementable specifications.

AI Generates Code Only from Text Prompts – It cannot proactively uncover hidden business pain points, identify contradictions or risks in requirements, and is suitable only for one‑off scripts or temporary prototypes, not for long‑term commercial systems.

System Architecture and Long‑Term Maintainability Are Human‑Centric – Enterprise systems that must operate for a decade need modular design, performance planning, scalability, and dependency management. AI produces isolated code snippets without a holistic view, leading to technical debt.

Security, Auditing, and Failure‑Handling Responsibility Remains with Humans – In case of outages, data leaks, or financial logic errors, legal and economic liability falls on the enterprise and its engineers, not on the AI tool.

AI‑Generated Code Lacks a Complete Audit Trail – Without clear provenance for each logic segment, diagnosing failures becomes difficult; professional engineers are needed to anticipate risks and perform rapid remediation.

Cross‑Team Coordination Requires Human Judgment – Large software deliveries involve product, testing, operations, customers, and third‑party vendors. Scheduling, conflict resolution, and delivery adjustments rely on human interpersonal and decision‑making skills that AI lacks.

Enterprise Adoption: AI as an Efficiency Tool, Not a Replacement

Infosys has fully introduced various AI coding tools internally, encouraging engineers to use “vibe coding” for repetitive boilerplate code and simple scripts, freeing them to focus on high‑value tasks such as architecture, compliance, and business design.

Internal data shows that after adopting AI code‑generation tools, basic coding effort decreased by roughly 40 %, while overall delivery quality and client satisfaction improved because engineers could concentrate on aspects AI cannot handle.

The company has not reduced headcount; instead, it continues to recruit senior talent with expertise in system design, security compliance, and industry knowledge.

Conclusion: Long‑Term Industry Trend

Nilekani summarizes that each technological iteration—assembly languages, compilers, development frameworks—has simplified coding but never eliminated programmers. AI‑driven “vibe coding” is merely the latest tool upgrade that removes mechanical, repetitive work, pushing engineers toward higher‑order engineering, business, and compliance capabilities.

The only developers at risk of obsolescence are those who merely copy‑paste, know only basic syntax, and lack systems thinking; professionals who can orchestrate business logic, control architecture, and uphold security and compliance will see their value continue to rise.

Relying solely on AI for “vibe coding” can serve as an auxiliary aid but can never independently deliver complete, reliable, and long‑lasting commercial software projects.