A leading global automotive services and technology company faced a persistent paradox: whenever engineers needed to modify a feature or develop new software, they spent 40–60 per cent of their time simply searching for information across a massive codebase.
This was the challenge facing one of the world's largest automotive services and technology companies, headquartered in the United States, as it continued to operate software systems containing millions of lines of code accumulated over generations of technology. Built on legacy programming languages and repeatedly expanded over the years, the systems had become increasingly complex. The entire technology landscape resembled a vast underground city whose map was known by only a few people, where even a small change could trigger unpredictable ripple effects. Engineers therefore had to spend significant time tracing complex dependencies, assessing risks, and developing a deep understanding of the systems before making changes safely.
The pressure is growing as AI accelerates the pace of technological innovation. Smaller companies can experiment rapidly, while large enterprises must move more carefully because every change can directly affect business operations.
To support engineering teams, many enterprises have adopted AI chatbots and coding agents, which initially helped engineers find information and complete tasks more quickly. But as data grows larger and more fragmented, the limitations of these tools have become increasingly apparent. Without a complete understanding of the system context, AI can produce inaccurate analyses or even "hallucinations" - generating answers that sound plausible but are factually incorrect. At the same time, operating costs can rise rapidly as AI processes ever-larger volumes of data.
This is not a challenge unique to one company. After years of working with customers across the United States, Europe, and Japan, FPT has found this to be a common challenge among large enterprises, where operational knowledge, from business logic to troubleshooting experience, is buried beneath layers of source code, technical documentation, and the expertise of veteran engineers. In Japan alone, according to IDC, around 80 per cent of medium and large enterprises still rely on legacy systems.
Mr. Pham Minh Tuan, Executive Vice President of FPT Corporation and Chief Executive Officer of FPT Software, said these legacy systems represent decades of accumulated operational knowledge and business logic. They are assets that enterprises cannot afford to lose. FPT is working with large enterprises around the world to modernize these systems with AI.
One way FPT is helping enterprises address this challenge is through Flezi Metis, an AI platform that connects source code, technical documentation, and operational data into a unified knowledge map of an entire system. With a comprehensive view of the system, AI can reason based on actual system structures rather than guesswork, reducing the risk of inaccurate information caused by missing context and eliminating the need to repeatedly process the entire system for every new request.
When an engineer wants to modify a function, the platform can quickly identify related components, analyze potential impacts, and flag areas that may be affected. Data remains within the customer's internal infrastructure, meeting stringent security requirements.
In an initial pilot with the automotive customer, conducted on a project containing approximately one million lines of code, the time spent searching through source code was reduced by around 50 per cent, while the engineering team's task completion time decreased by approximately 30 per cent. Beyond helping engineers work more efficiently, Flezi Metis is also changing how enterprises manage their software systems. With institutional knowledge easier to access, technology leaders gain a more comprehensive view of their systems and a stronger foundation for decision-making.
Following its success with the US customer, Flezi Metis is expanding into industries with complex operating environments, including banking, insurance, and the public sector. It is part of FleziPT, FPT's AI ecosystem, which brings together technology platforms, experts, and partners to support enterprise-wide AI transformation.
To further accelerate this transformation, FPT has also introduced CASAN, a five-level AI maturity framework that helps enterprises structure their AI transformation journey, from isolated AI applications to operating models in which AI is deeply integrated across the organization.
Google translate