The Hardest Part of Government Modernization Isn’t Technical—It’s Institutional

Why AI can accelerate legacy modernization but cannot replace institutional understanding

Over the past several years, I’ve worked on modernizing legacy analytical systems, translating SAS models into Python, migrating SQL workflows into PySpark, and rebuilding Stata analyses in modern environments. I’ve translated thousands of lines of legacy analytical code, and the difficulty was almost never the language change (although SAS occasionally tested that theory). Through that work, I realized that the hardest problems in modernization had very little to do with code at all.

One project in particular made this clear. My task was to modernize how data flowed into a loan reporting system by rethinking a set of long-standing input files that had accumulated over time. An issue was identified downstream, where several loans were not linked to their correct lender. As modernization work began, we traced the issue back to a legacy workflow that relied on an outdated identifier, which caused some lender relationships to be missed.

Resolving the issue required more than a simple correction. We already knew that the outdated identifier used historically could not be carried forward, but determining what should replace it required working closely with the client to understand the business logic behind how lender information should be sourced. The client explained when one data source should take precedence over another, reflecting business rules that were never fully documented in the code. Drawing on familiarity with the client’s data environment developed over years on the project, I was able to translate those rules into logic that correctly identified lenders and their active loans. The real work involved translating this institutional understanding into new logic that reflected how the process was intended to work. Without that shared understanding, the modernization effort would likely have reproduced the same errors in a newer language.

Experiences like this have changed the way I think about modernization. Legacy systems don’t only contain coding decisions; they encode years of institutional knowledge, policy choices, reporting requirements, and edge cases shaped by how government programs operate. Modernization requires understanding why those decisions exist, not just how the technical system works.

Faster technology has changed how we implement solutions but not the hardest part of modernization

At the same time, AI tools are rapidly transforming the most repetitive aspects of modernization work. Tasks that once required hours of manual translation and refactoring can now be completed in minutes. Today’s tools can translate code, suggest refactors, and even attempt full modernization efforts faster than ever before. Teams are already using assistants like GitHub and Copilot to accelerate development, while platforms such as Cognition’s Devin aim to function as autonomous software engineers. Government deployments like Microsoft’s Copilot for Government demonstrate how quickly these capabilities are advancing into regulated environments.

As a data scientist working on federal modernization projects, I’m excited about this wave of new tools. Rather than replacing the parts of the job I enjoy, they’ve opened the door to new ways of working and new problems to solve. What I’ve realized, though, is that faster technology has changed how we implement solutions but not the hardest part of modernization. The most important skills I’ve developed as a data scientist aren’t tied to any particular programming language. They involve framing ambiguous problems, interpreting the context behind the data, translating organizational needs into technical solutions, and working closely with subject matter experts whose knowledge lives outside the code.

AI is already changing how modernization work gets done, and that change is welcome. It reduces the repetitive effort required to translate and refactor legacy systems, allowing data scientists to focus more on the work that has always mattered most: understanding programs, interpreting context, and translating institutional knowledge into systems that function correctly. Technology can accelerate modernization, but understanding remains the foundation it depends on.

Photo by Conny Schneider on Unsplash