Artificial intelligence (AI) is transforming how enterprises modernize legacy systems, and they should start with augmentation rather than ripping and replacing foundational applications.
While augmentation may seem like a simpler route, organizations need to proceed with caution. Successfully enhancing legacy systems with AI depends on many factors, including data accessibility and in-house knowledge of existing systems. How AI will affect legacy applications isn’t entirely predictable; without the necessary rigor and risk management, systems can break and data can be compromised.
Research firm Gartner predicts that in the next couple of years, CIOs can expect generative AI to accelerate legacy modernization and reduce modernization costs as much as 70% by 2027. The caveat, however, is that organizations must balance modernization savings against rising compute, licensing and governance costs.
According to Grand View Research, the global application modernization services market size was estimated at $17.8 billion in 2023 and is projected to reach $52.5 billion by 2030, driven by the need to reduce IT complexity and operational costs.
Most significantly, AI adoption is driving application modernization. According to the 8th Annual Nutanix Enterprise Cloud Index, the rise of data-driven operations and AI workloads is moving enterprises to rethink their plans to modernize infrastructure.
The report notes that organizations are modernizing their infrastructure to ensure application and workload agility, portability, sovereignty and cost control, but cautions that implementing AI initiatives in isolation without consulting IT leads to inefficiencies and project delays.
Legacy Systems Still Have Some Life Left
Even with the ubiquity of cloud computing and Software-as-a-Service (SaaS) application delivery, legacy systems remain entrenched in many organizations.
Dave Trier, CEO of Chicago-based ModelOp, which offers an enterprise AI lifecycle management and governance platform, said just about every Fortune 500 company, especially those in the financial sector, has core ledgers running off a mainframe, and some still have Oracle databases that haven’t been migrated to the cloud.
He said those organizations with large on-premises data warehouses can’t wait to deploy AI because there’s so much demand for it from the business across different departments.
“They are looking to augment as opposed to waiting to replace that infrastructure,” Trier told The Forecast. “They've been talking about replacing the mainframe since the ’90s.”
Trier said that with any core legacy system, there is always a way to feed into a central data lake or data warehouse. “The AI is running against those data sets.”
Add AI Incrementally and Strategically
Data accessibility is essential for augmenting legacy systems with AI, advises Joel Haggar, executive VP of AI and technology at Radix IoT.
“AI systems are only as useful as the data they can reach,” Haggar said
He said if a legacy system has a well-documented API and it’s straightforward to build connectors, it’s almost certainly better to keep it and layer AI on top.
“If a system has no reasonable way to access its data programmatically, that’s a strong signal it should be replaced because a system you can’t get data out of is a liability regardless,” Haggar said in an interview with The Forecast.
“When you’re transforming a workflow that was a genuine bottleneck, the new complexity is a worthwhile trade.”
Dev Singh, founder and CEO of Chicago-based Airolabs.ai, a provider of AI and cloud solutions, said the biggest mistake companies make is treating AI as a reason to rip out systems that still work.
“The smarter approach is to layer intelligence on top of what’s already there,” he told The Forecast. “The goal isn’t disruption. It’s controlled augmentation.”
Singh recommends starting at the decision level — workflows, document processing and fraud detection, for example — before touching core systems of record.
“We’ve seen this work in a variety of verticals, including healthcare and banking, where organizations add AI-driven decision support on top.”
Balancing Risks and Trade-Offs
Singh said the real risk of adding AI isn’t only technical, but also underestimating how operational it is.
“It’s not a plug-in. It behaves more like a system you have to manage continuously,” he said.
He explained that legacy data is often messy, integrations are fragile and models don’t stay static.
“That creates ongoing overhead that teams don’t plan for,” he said.
Adding AI does run the risk of destabilizing legacy systems, Singh said, but most likely in subtle ways, manifesting as increased database load, timing issues and/or feedback loops no one anticipated.
“That’s where legacy systems start to strain,” he said.
Deploying AI with a clear architecture is essential if organizations want to truly modernize workflows and avoid adding layers of complexity, Singh said.
“Companies think they’re modernizing, but they’re actually creating a second layer of technical debt, just in a more sophisticated form,” he said. “You’re just stacking new complexity on top of old systems.”
Haggar echoes this sentiment. “If you’re hitting a legacy API with thousands of AI-generated requests without proper controls, you’re going to have a bad time.”
This isn’t an AI problem, he said, but an integration discipline problem.
Modernizing systems always has many moving pieces, notes Narayana Pappu, founder and CEO of Zendata, a San Francisco-based cybersecurity startup specializing in AI governance and data privacy solutions.
When you mix in AI, your data can end up going where you don’t want it to while failing to create the efficiencies or productivity gains you expected, he said.
“Adoption of AI is like fixing an airplane while it’s flying, and possibly getting fixed itself,” Pappu said
AI Compounds Compliance, Security Risks As It Augments
Layering AI onto legacy systems also poses security and compliance concerns.
Singh said healthcare data and logs fall under regulatory controls and AI models face the same scrutiny as traditional risk systems, as well as added pressure around explainability and bias.
“AI expands the compliance surface faster than most organizations expect,” he said.
The bigger issue, Singh said, is that AI moves faster than most governance frameworks were ever designed for.
“Organizations now have to rethink auditability, monitoring and human oversight in much more real-time terms.”
Pappu said it’s critical that organizations understand data flows and decision making boundaries, and that the inherent potential biases within some AI systems can increase risks for regulated industries.
Start Small, Access Risks and Respect Legacy
For IT decision-makers weighing how to embrace AI without compromising stability, security or long-term sustainability, Singh recommends starting small and focusing on use cases where failure won’t disrupt the business.
“Make sure your data is actually usable before investing in tools,” Singh said. “Architect everything at the start to fail safely, so you can roll it back or isolate it without impacting core systems.”
Trier said the journey of augmenting legacy systems must begin with understanding affected data from a business context, a clear business case and proper risk classification.
Identified risks should be monitored as part of a thorough testing process that accounts for the many nuanced, unplanned scenarios that might arise, he said. “At the end of the day, AI is risky. It’s inherently risky because it’s not deterministic.”
Haggar said it’s important to remember that established systems are the foundation of an organization.
“AI should observe, analyze and augment what’s already there. Treat AI as a sidecar, not a transplant,” he said.
Editor’s note: Learn how Nutanix Delivers Complete Platform for the Agentic AI Era with the Nutanix Cloud Platform (NCP) solution designed to help organizations operate reliably as AI workloads expand, cloud environments grow more complex, and hardware supply constraints drive the need for more flexible infrastructure platforms
Gary Hilson has more than 20 years of experience writing about B2B enterprise technology and the issues affecting IT decisions makers. His work has appeared in many industry publications, including EE Times, Embedded.com, Network Computing, EBN Online, Computing Canada, Channel Daily News, and Course Compare. Find him on Twitter.
© 2026 Nutanix, Inc. All rights reserved. For additional information and important legal disclaimers, please go here.
Related Articles
