AI Legacy Systems to Outnumber Traditional Ones by 2031, Infosys Chair Warns

AI Legacy Systems to Outnumber Traditional Ones by 2031, Infosys Chair Warns

As enterprises globally accelerate their adoption of artificial intelligence, Infosys co-founder and chairman Nandan Nilekani has issued a stark warning: the very tools meant to modernize legacy infrastructure are rapidly generating a new class of obsolete systems — AI legacy systems — that will soon outnumber traditional ones.

"Five years from now, there will be more AI legacy systems to deal with than any other kind," Nilekani stated in a recent analysis published by Diginomica. His assertion underscores a paradox at the heart of the AI revolution: while organizations are leveraging generative AI to automate, optimize, and replace outdated codebases, they are simultaneously creating new technical debt that may prove even more difficult to maintain.

Infosys, a global leader in technology services and consulting, has spent decades helping clients navigate complex digital transformations. According to the company’s public-facing mission, it specializes in delivering "end-to-end digital transformation" across industries — a mandate that now includes managing the unintended consequences of AI adoption. Nilekani’s comments reflect an insider’s pragmatism: enterprises are deploying AI tools faster than they can establish governance, documentation, or long-term maintenance strategies.

Many organizations are using AI to rewrite COBOL systems, migrate mainframes, or automate legacy ERP modules. But in their haste, they often rely on proprietary AI models, opaque training data, and vendor-specific APIs that lack interoperability. These systems become "legacy" not because they’re old, but because they’re opaque, unstandardized, and tightly coupled to ephemeral technologies.

"We’re seeing clients deploy AI-powered chatbots for HR onboarding, then realize those bots were trained on outdated employee handbooks from 2019," said an Infosys senior consultant familiar with the trend, speaking anonymously. "When the vendor updates the model, the output changes. There’s no audit trail. No version control. It’s legacy by design."

The problem is exacerbated by the lack of regulatory frameworks and industry standards for AI system lifecycle management. Unlike traditional software, where source code can be reviewed and refactored, many AI systems are treated as black boxes. Even when they’re built on open-source foundations, the fine-tuning, embeddings, and prompt chains used in production are rarely documented.

Infosys’s own Knowledge Institute has begun tracking this phenomenon as part of its annual technology trends report, noting that 68% of enterprises deploying AI in core operations lack a formal deprecation strategy for AI models. This creates a cascading risk: when an AI system fails or becomes obsolete, organizations may have no way to replicate its functionality — even if they retain the original data.

Meanwhile, the market is racing ahead. From China’s sword-wielding humanoid robots to Vietnam’s 25-kilometer overwater drone delivery networks, AI is being deployed in increasingly visible and critical applications. In Australia, courts are overwhelmed by AI-generated legal filings, many of which contain hallucinated case law. These are not edge cases — they are early symptoms of a systemic issue.

Nilekani’s solution is not to slow adoption, but to institutionalize discipline. "Organizations no longer have any excuse to retain legacy systems," he said, "but they must not replace them with unmanaged AI systems. We need AI governance frameworks as rigorous as those for financial auditing."

Infosys is already developing internal tools to map AI system lineage, track training data drift, and auto-generate documentation for AI workflows. The company’s global workforce — including thousands in its Americas careers division — is being trained in "AI archaeology," the practice of reverse-engineering black-box systems to understand their logic and limitations.

The message is clear: AI won’t eliminate legacy systems — it will multiply them. The organizations that survive the next decade won’t be the ones that adopted AI fastest, but those that managed it most responsibly.