Original broadcast 2/3/26

Presented by Microsoft

Dave Hinchman, Director of IT and Cybersecurity at the Government Accountability Office, and Wole Moses, Chief AI Officer, US Federal Civilian at Microsoft, focus on the infrastructure decisions agencies need to make to support AI adoption, modernization, and long-term scalability. Their conversation highlights that while AI interest is expanding rapidly, agencies are still early in their journey, and the foundational infrastructure choices made now will determine whether AI becomes a scalable capability or a collection of isolated tools.

Hinchman describes federal AI adoption as being in “baby steps,” noting that while many tools and applications exist, government has not yet reached wholesale implementation. In his view, agencies are still working to understand how to integrate AI across organizations and how to apply this transformational technology in ways that deliver measurable mission benefits. From GAO’s experience auditing major system implementations, Hinchman stresses that planning should happen early and deliberately. He encourages agencies to begin planning even before engaging vendors so they understand what they are pursuing, what outcomes they expect, and how they will evaluate success.

A key concern Hinchman raises is flexibility. AI is changing quickly, and agencies must build infrastructures that can adapt as new tools, models, and use cases emerge. He emphasizes that agencies need to ensure what they invest in today will remain useful further downstream, especially given that technology evolution will continue during the life of these systems. For government, where large investments can take years to deliver, adaptability becomes a core requirement.

Moses explains that major cloud providers have simplified many of the infrastructure requirements for AI adoption. Managed services offer access to AI models, GPUs for training and inferencing, and platforms and tooling for building AI applications. This reduces the burden of agencies having to design everything from scratch and can accelerate deployment timelines. However, Moses emphasizes that agencies still need to make several important architectural and governance decisions early, even in a managed environment.

Moses highlights decisions around model selection and benchmarking, as well as AI observability and governance. Agencies must understand how to select the right model for a specific use case, how to validate performance, and how to monitor outcomes after deployment. Observability is critical because AI systems can behave unpredictably if inputs change or if the model begins producing outputs that are inconsistent with mission needs. Governance ensures that AI use remains controlled, auditable, and secure.

Integration with existing systems is another major theme. Moses notes that many of the most valuable AI applications involve connecting to case management platforms, records management systems, and other mission-critical applications that agencies already rely on. These systems often contain the data and workflow logic necessary for AI to deliver operational value. Without integration, AI tools may remain disconnected from mission execution, limiting their impact and increasing friction for employees.

A challenge Moses acknowledges is that many agency systems are legacy environments that were never designed for modern AI workloads. To address this, he describes building abstraction layers that enable connectivity. One approach is using REST APIs to expose functionality and data from legacy systems. Another is using AI-native connection methods such as model context protocols, which are designed to integrate AI applications into broader enterprise environments. Moses also notes that agencies can use other integration technologies that connect even to systems that do not support APIs, allowing agencies to modernize incrementally rather than waiting for full replacement.

Hinchman reinforces that AI adoption appears more grounded in use cases than many previous technology shifts. He points to rapid growth in documented government AI use cases, including repositories that track this expansion, as evidence that agencies are tying AI adoption to real operational needs. He describes this as an exciting development because it signals that agencies are not simply experimenting with AI for novelty, but increasingly applying it to solve specific mission problems.

However, Hinchman identifies cultural barriers as one of the most persistent roadblocks to scaling AI. He explains that agencies need executive sponsorship and vocal leadership to help federal employees understand why transformation is necessary. AI will reshape business processes, and without strong communication around the value proposition, employees may resist change or struggle to understand how AI is intended to help them. Culture and change management, in his view, are as critical as technology itself.

Moses expands this perspective using the framework of people, process, and technology. He describes challenges in the people category as uneven understanding of AI capabilities and uneven trust across organizations. In some environments, AI may be over-trusted, while in others it may be rejected entirely, leading to uneven adoption. He notes that trust and understanding determine whether AI tools are embraced and integrated into everyday work or remain underutilized.

On the process side, Moses describes a common problem: organizations have a reflexive tendency to bolt AI onto existing workflows instead of reimagining those workflows around new capabilities. This mirrors broader modernization challenges where agencies attempt to preserve legacy processes while applying modern tools. Moses suggests agencies should use AI as an opportunity to rethink how work is done, modernize processes, and design workflows that maximize productivity and mission performance.

Technology presents a newer tension. Moses notes that in the past, technology was often stable enough that people and processes were the main barriers. In today’s AI environment, technology itself is changing faster than organizations can absorb. This rapid change creates challenges in training, planning, and keeping architectures adaptable. Moses suggests one key pathway to success is agencies learning from each other, especially because many AI use cases share common patterns and challenges even if mission specifics differ. Sharing best practices and lessons learned can help agencies accelerate adoption without repeating the same mistakes.

Hinchman offers that measuring culture and trust can be difficult, describing it as more art than science. However, he suggests indicators such as workforce training can provide meaningful signals. He notes that federal technology workforce challenges—particularly in areas like cybersecurity, IT, and AI—make training and skills development essential. Agencies that build trained teams who understand AI and can evangelize its value internally will be better positioned to scale adoption.

The conversation also touches on cloud computing benefits and the need to manage security responsibilities carefully. Hinchman emphasizes that as agencies move into cloud environments, attack surfaces expand. He advises agencies to ensure cloud providers meet security commitments outlined in contracts, while also maintaining strong internal responsibilities. Cloud adoption does not eliminate agency accountability; it shifts responsibilities and requires updated security practices and change management.

Looking ahead, Moses encourages agencies to begin preparing for the next wave of AI evolution: agentic AI. He describes progression from AI assistants that support human work today to human-plus-agent teams, and eventually toward more autonomous agent scenarios. In his view, agencies that become fluent in this progression and begin preparing for the underlying technologies will be better positioned to adopt emerging capabilities responsibly and strategically.

Across the discussion, the core message is that AI success depends on infrastructure readiness, integration strategy, governance, workforce adoption, and leadership-driven culture change. Agencies that plan early, design for flexibility, modernize connectivity, and invest in people and processes will be better equipped to scale AI in ways that improve mission performance and deliver long-term value.

This segment was part of the program Mission Innovation: Becoming a Frontier Agency.