5 Takeaways from AI in Action

Retrieval-augmented generation (RAG) combines large language models with external knowledge bases, giving federal employees the ability to tap into their own expertise while pooling insights with colleagues.

Alexis Bonnell, partnership manager at OpenAI, described RAG as a way for individuals to “curate their own little knowledge universes” that can be shared across teams, breaking down silos and making scattered data actionable.

Bonnell highlighted her experience consolidating more than 183 IT tools into fewer than 47 during her time as Air Force Research Lab CIO and underscored how RAG can streamline systems while fostering collaboration. She emphasized that RAG makes AI a more personal and intimate technology, shaped by each user’s unique knowledge and experiences, adding that its effectiveness depends on how intentionally humans engage with it.

Glenn Parham, CEO of GovBench and former Defense Department technical lead for generative AI, stressed that scaling AI is more about people than tools. Agencies must identify “champions” within their workforce who are eager to adopt AI and can serve as early adopters to carry momentum forward.

By cultivating these leaders, CIOs and program offices can bridge the gap between technical solutions and operational needs, ensuring AI is integrated in ways that support real mission requirements rather than creating unnecessary complexity.

Parham emphasized that this approach requires CIO shops and leaders to take stock of their teams’ readiness and capabilities. Scaling AI depends on finding where enthusiasm and expertise already exist and pairing that energy with practical training and support.

By empowering employees who understand operational pain points but may not yet be technologically fluent, agencies can create “problem owners” who collaborate directly with AI practitioners. This combination helps ensure AI deployment stays grounded in real-world use cases while accelerating adoption across the enterprise.

Customs and Border Protection (CBP), which manages over half of the Department of Homeland Security’s AI inventory, is seeing significant gains in both efficiency and workforce morale through AI adoption.

Chief AI Officer Joshua Powell highlighted that document processing times in HR have been cut by 50%, reducing the effort from as long as 24 hours to just two or three. This time savings allows employees to redirect their focus to higher-value work, reinforcing AI’s role as a force multiplier rather than a job replacement.

Central to this success is CBP’s investment in workforce training, which builds trust in AI systems and equips employees with the confidence to integrate them into daily operations. By framing AI as a tool for efficiency and empowerment, rather than a threat, CBP is cultivating a culture of adoption.

While enthusiasm for AI is strong, experts warned that agencies must get the basics right to achieve meaningful and sustainable results. Johns Hopkins APL’s Julie Obernauer-Motley pointed out that poor-quality or inaccessible data remains the number one reason AI projects fail. She urged agencies to treat data as a living resource that must be continually cleaned, maintained and governed.

Ali Naqvi, CEO of the American Institute of Artificial Intelligence, added that research and education are critical if the U.S. is to compete with adversaries like China and Russia in AI innovation. He argued that true progress hinges on capabilities in mathematics, quantum computing and technical expertise.

“You cannot avoid the use of artificial intelligence. It’s in health care, education, employment, finance, law enforcement, government, the military; this technology has exploded, and it will continue to explode. The uses, when properly used, are remarkable,” said Hon. Paul Grimm, Director at the Bolch Judicial Institute at Duke Law School.

Agencies like the Internal Revenue Service (IRS) and the Defense Information Systems Agency (DISA) are demonstrating how AI can reduce burden on the workforce and citizens while strengthening mission outcomes.

IRS Executive Director of the Office of Online Services Karen Howard explained that the agency is using AI at critical taxpayer touchpoints. With over 100 active AI use cases already in play, Howard emphasized the importance of governance, transparency and human oversight to ensure reliable outcomes.

DISA is applying AI to maximize “decision space” for warfighters by rapidly delivering accurate and trustworthy information to commanders. CTO Steve Wallace noted that predictive and inferential models are already embedded across DOD networks, enhancing cyber defenses and enabling dynamic access control.

“The faster we get that information to [the commander] and the more trustworthy that information is, the better decision that they can make,” Wallace said. “The best case is when the user doesn’t even know that [AI is] there.”