Government agencies are looking to use artificial intelligence (AI) and commercial tools to modernize their IT infrastructures while minimizing costs and risk. One powerful application is to use Generative AI (GenAI) to convert legacy systems and replace outdated languages and architectures. This approach, known as AI-enabled code conversion, delivers measurable value through increased speed, reduced waste, and greater operational resilience, all while accelerating the transition to modern programming languages.
Abt teams have the experience to know why the first output from a GenAI tool is often the best, especially when using deterministic settings: because GenAI is initially anchored on the most probable output. On follow-up, these tools may become less accurate as they navigate the embedding space, leading to less probable outputs. We know when to give feedback to GenAI in the same conversation, and when to initiate a new conversation to get the best possible results.
Why Legacy Systems Are Holding Agencies Back—And How AI Unlocks Progress
Advancements in AI, including GenAI, can provide significant advantages in modernizing federal IT systems, including improving code efficiency (i.e., minimizing compute resource utilization), managing technical debt more cost-effectively, and rapidly modernizing legacy systems without significant human intervention. According to a 2019 GAO report, the U.S. government spent $337 million maintaining just ten legacy systems, some of which were over 50 years old. Further, it was estimated that federal agencies spent 80% of their IT budgets, roughly $90B in 2019, on operations and maintenance, primarily focused on legacy systems. By replacing legacy codebases, agencies can shift away from aging infrastructure, such as on-premise, 3-tier architectures, and move toward modern, cloud-native environments that are more secure, scalable, and innovation-ready.
Outside of efficiency and cost savings, AI can help agencies with modernization efforts to support additional data integration efforts, handling of big data, and code parallelization efforts. Modern programming languages are also better suited for cloud computing, mobile applications, and the ability to support adoption of emerging technologies such as GenAI, edge computing, and quantum computing. AI-powered tools can also support compliance and security functions as they are able to identify security flaws in the original and converted code, ensuring safer transition, and can align the converted code to industry standards. Additional benefits include:
- Enforcement of consistent coding styles
- Detection of inefficient legacy code patterns
- Reductions in the need to maintain specialists for legacy languages
- Ability to suggest performance enhancements and ways to optimize code
- Generation of technical documentation
- Minimization of technical debt and future maintenance costs
Challenges of AI Code Conversion—and How to Overcome Them
Even with advancements in AI, automated code conversion models still struggle with complex and domain-specific tasks, such as interpreting the context of business logic, handling edge cases, and navigating differences in syntax, data structures, memory management, and functions. Successful conversion efforts require careful planning, the right tool selection, and extensive testing.
Abt teams have successfully transitioned legacy codebases, including updating legacy COBOL systems to use modern digital infrastructure for federal financial systems, and converting SAS to modern cloud-native environments such as Databricks (PySpark/SQL). These migrations involved complex data workflows, including the processing of sensitive and large-scale datasets where we apply GenAI tools to accelerate code conversion, validate outputs, and ensure semantic accuracy. In response to increasing security concerns, Abt has also supported agencies such as Centers for Medicare and Medicaid Services in replacing legacy automation tools, such as VBA macros, with modern programming languages such as Python, reducing risk and aligning with evolving cybersecurity requirements.
Securing Modern Systems with AI-Powered Compliance
Abt’s AI-driven approach accelerates code conversion and system modernization while maintaining the reliability, integrity, and security required in public sector systems. Our conversion process combines detailed planning, automated tests, manual reviews, and AI evaluation metrics to ensure that the translated code is reliable and production ready. We maintain a team of AI-specialized developers to select the best fit-for-purpose AI models and to review AI-generated code to catch subtle logic errors, refactoring where necessary to align with best practices in the target language.
Ready to modernize with confidence?
Abt helps federal agencies accelerate code conversion while maintaining security, compliance, and performance. Let’s build what’s next.
Greg Dupier,
Vice President, Technology and Innovation
Greg.Dupier@tspi.net