As enterprises across Asia accelerate their modernisation efforts, artificial intelligence is increasingly being seen as a shortcut to faster code conversion, lower technical debt, and more agile IT environments. But while AI is proving valuable in boosting the modernisation journey, it is often misunderstood, particularly by organisations hoping it can compensate for legacy complexity created over past years.
In conversation with iTNews Asia, Dr Vishnu Nanduri, AI Innovation Leader, ASEAN & Korea, Kyndryl shares his perspective on where AI is genuinely accelerating application modernisation, where expectations are running ahead of reality, and why successful AI-led transformation depends as much on organisational readiness and human oversight as it does on the technology itself.
According to Nanduri, AI is most effective in modernisation tasks that are well-defined, repetitive, and measurable, such as dependency mapping, code analysis, test generation, and documentation. In these areas, it can reduce manual effort and significantly shorten timelines.
He explained that these use cases work because the outputs can be validated against something concrete. “If a test fails, the error is visible. AI can extract structure from messy systems and surface relationships that would take humans much longer to identify,” he said.
That makes testing and dependency discovery some of the strongest near-term use cases for AI in modernisation, especially in enterprises dealing with sprawling legacy estates and fragmented application portfolios.
The limits appear when AI is asked to replace understanding
Nanduri cautioned that expectations begin to break down when organisations assume AI can compensate for a weak understanding of their own legacy environments.
“If the architecture is unclear or the business logic is buried in decades of workarounds, AI produces output that looks plausible but is often incorrect in subtle ways. That introduces risk rather than real progress,” he said.
He argued that one of the biggest misconceptions among CIOs is the belief that AI-led code conversion alone amounts to modernisation. Simply translating code into a new language, he said, does not resolve deeper issues around architecture, resilience, scalability, or operational complexity.
Instead, he positioned AI transformation as both a technology and organisational challenge. “It’s more than just deploying new applications; it’s preparing people and creating the right culture and support to help employees embrace AI as an enabler rather than a disruptor,” he explained.
Code conversion remains one of the riskiest areas for AI-led modernisation
While syntax translation is becoming increasingly common, Nanduri said preserving the business behavior embedded inside legacy applications remains far more difficult where systems are poorly documented and full of implicit logic.
This is also where many modernisation programs run into trouble. He pointed to common pitfalls such as incomplete system knowledge, weak validation, and attempts to convert entire environments in one sweep instead of breaking work into smaller, controlled phases.
He added that organisations often place too much trust in code that compiles or passes basic tests, without examining whether it behaves correctly across integrations, workflows, and edge cases.
“Teams assume AI-generated code is correct because it compiles or passes basic tests. That ignores deeper issues in logic and integration,” he explained.
Foundation gaps are stalling projects
Nanduri said a large share of modernisation efforts stall because organisations have not invested enough in understanding the current state of their environments before introducing AI. In many cases, the issue is less about legacy-system knowledge in isolation and more about foundational weaknesses across architecture, data and infrastructure.
He added that many enterprises are trying to run modern AI capabilities on infrastructure that was never designed to support them, while fragmented data environments further undermine AI reliability. For organisations looking to avoid those traps, he recommended a more structured readiness approach that evaluates operating environments, workflows, controls and production readiness before AI is scaled across modernisation programs.
Nanduri also argued that failures typically begin with organisational issues rather than technology choices.
A common issue is misalignment between leadership ambition and execution capability. Teams are expected to move quickly but often lack the skills, training, or clarity to deliver.
- Dr Vishnu Nanduri, AI Innovation Leader, ASEAN & Korea, Kyndryl
He also pointed to underinvestment in change management as a recurring weakness. As AI changes how software is built, tested, and operated, enterprises need to prepare teams for new workflows rather than simply deploying tools and expecting adoption to follow.
Governance, testing and human oversight remain non-negotiable
For organisations moving beyond pilots, Nanduri said the central challenge is proving that AI-generated outputs will work reliably in production. He stressed that AI-generated code should go through the same production controls as any other software change, with additional scrutiny around how the code was produced and validated.
“At a minimum, this includes code verification tools, followed by review from experienced engineers, and testing in a controlled environment before deployment,” Nanduri said.
He added that human oversight remains essential throughout the lifecycle, not just at the final release stage. Teams need to review logic, edge cases and integration points continuously if they want to move quickly without compromising control.
That becomes even more important after AI-led code conversion, where the testing burden extends beyond the code itself into live data flows, workflows and system-wide dependencies.
Warning signs usually appear early
In Nanduri’s view, the clearest warning signs that an AI-led modernisation project is heading off course tend to appear early and are rarely limited to technical execution. Weak leadership alignment, unclear ownership, poor change management and a narrow focus on isolated AI use cases are among the strongest indicators of trouble ahead.“If there is no clear ownership or buy-in at the top, priorities shift and momentum slows,” he said.
Projects also run into difficulty when success is defined only at the pilot stage rather than around production outcomes, governance and operational scale. In those cases, organisations may prove that AI can generate outputs, but fail to build the controls and delivery model required to sustain it in live environments.
Rather than applying AI indiscriminately across entire modernisation programs, Nanduri advocated for a more selective approach, particularly in complex environments such as mainframes or tightly coupled legacy systems.
“Full-scale automation often looks efficient upfront, but it breaks down when systems are tightly coupled or poorly documented,” he said.
Instead, organisations should start by identifying where AI can create measurable business value, whether that is surfacing hidden dependencies, prioritising high-impact applications, or flagging operational risk.
Success will be measured by business outcomes
For CIOs deciding whether to modernise, replace, or retain a system, Nanduri said the starting point should be business impact. Systems that are stable and effective may not need intervention, while those that slow down change, create compliance exposure, or sit in critical revenue paths require closer attention.
When evaluating AI-led modernisation, he said the real metrics are operational rather than experimental: release speed, incident reduction, efficiency gains, and the ability to scale into production.
“AI can handle the heavy lifting and scale, while people focus on direction, constraints, and accountability. Organisations that get that balance right will tend to move faster without taking on unnecessary risk,” he added.
