The Business and Enterprise Systems (BES) Department of the Air Force Bot Operations Team (DAFBOT) is employing Generative Artificial Intelligence (AI) and Large Language Models (LLMs) to modernize select legacy code bases.
This initiative seeks to transform Independently-Developable End-User Assessable Logical (IDEAL) and Common Business-Oriented Language (COBOL) code into a functional Java application, representing a significant advancement in legacy software system management.
In collaboration with industry partners, including a SBIR III contractor, the effort utilized an AI platform that integrates multiple LLMs.
"The use of reasoning models like o3 Mini, combined with the expertise of a joint contractor and civilian team, facilitated this breakthrough,” said DAFBOT Chief Technology Officer Jude Stanley. “Refactoring code with an LLM is a complex endeavor. While the LLM may initially translate from COBOL to Java, the resulting Java code can often be incoherent. This process is akin to separating a combined Mandarin and Sanskrit text into its original languages, then converting it into Latin, and finally drafting it into English. The challenge lies in ensuring that the English version conveys the same narrative as the original combined text, capturing all its nuances. Legacy code bases are highly customized and unique to their capabilities, making the task of producing the same functionality in a modern code base with LLMs demanding but ultimately rewarding. When successful, a coherent narrative emerges that translates the COBOL to Java, including all dependencies. This is something to watch and appears magical, yet the LLM is ultimately a creation of math and code, built upon hundreds of billions of binary transistors."
DAFBOT Program Manager, Matthew Roberts highlighted the partnership as an example of effective public-private collaboration in addressing complex government challenges.
The DAFBOT team within BES was tasked with using AI to reconstruct the lost technical and functional baseline of the legacy code. During this process, they identified a model that enabled the joint team to develop LLM coding agents to rewrite the code into a more modern and sustainable language. This achievement enhances the Air Force's operational efficiency and sets a precedent for future advancements in defense technology, potentially saving billions when applied to other legacy code bases.
The government faces numerous complex challenges, and this partnership underscores the importance of engaging contractor communities to address these issues. Historically, many technologies integral to daily life, such as those in cell phones, originated from military designs. Once unclassified, the private sector leveraged them to create innovations like computers, the internet, and cell phones. This public-private partnership is crucial for driving Gross Domestic Product and fostering technological advancements.
By harnessing AI, the Air Force has demonstrated a successful approach to modernizing legacy systems, paving the way for improved supply chain and maintenance processes, and informed decision-making.