Nandan Nilekani AI Legacy Code Warning: The New Debt [2026]

Will AI-generated code create a new crisis of technical debt?

This is where the conversation takes a sharp turn. When humans write code, they write it at human speed. When AI writes code, it generates it at machine speed. Nilekani predicts that this explosion of generated content is going to result in a pile-up of what we might call "AI legacy."

"Five years from now, there’ll be more AI legacy systems than any other legacy system – all the kind of stuff that will have been generated – and we’ll have to clean that up as well," Nilekani said.

Think about it. If an AI agent generates millions of lines of code to build an application today, who maintains that code tomorrow? If the AI generated "slop"—functional but poorly optimized or messy code—companies might find themselves drowning in a sea of synthetic software that no human engineer understands. The market is effectively pivoting from a cleanup of human-made mess to a future cleanup of machine-made mess.

What is the ‘fake productivity’ trap in AI adoption?

Beyond the code itself, Nilekani highlighted a behavioral loop that suggests we might be fooling ourselves regarding efficiency. He cited a phenomenon of "fake productivity" that is becoming increasingly common in the corporate world.

The example he gave is relatable to anyone working in a modern office: one employee uses an AI tool to expand a simple one-paragraph point into a ten-paragraph email to sound more professional. The recipient, seeing a wall of text, uses their own AI tool to summarize those ten paragraphs back down to one.

Lots of computing power is used, AI models are queried, and energy is consumed, but the net result is zero actual value added to the communication. It is activity masquerading as productivity.

How will AI change the role of the software engineer?

With AI taking over the heavy lifting of code generation, the role of the human technologist is shifting dramatically. According to Nilekani, the era where "writing code" was the primary goal is ending. The new objective is "making AI work."

This is not just semantics. It represents a move toward "Agentic AI"—systems that don’t just chat with you but execute complex workflows autonomously. Infosys recently announced a collaboration with Anthropic to integrate Claude models into their services, specifically to power these agentic capabilities.

This shift has massive implications for the job market. Nilekani cited data suggesting that while AI could put 90 million traditional tech jobs at risk, it has the potential to create 170 million new roles. However, these new roles won’t look like the old ones. Infosys is already ramping up hiring for "AI Engineers" and "Forensic Analysts"—people who can investigate what the AI did, orchestrate different models, and ensure the output is safe and accurate.

The Bigger Picture

Nilekani’s comments reveal a lucrative cycle for IT service giants like Infosys. First, they charge enterprises billions to modernize human-written legacy code using AI. Then, five years later, they will likely charge billions more to optimize, secure, and "clean up" the sprawling, unmanageable codebases that the AI generated. For the enterprise client, the promise of a "clean slate" remains a mirage; technical debt isn’t disappearing, it’s just changing authors from humans to algorithms. The winners in this scenario are the firms selling the orchestration and cleanup services, while companies that rely blindly on generative code without strict governance may find themselves bankrupt by the maintenance costs of their own "efficiency."