AI & Automation: In Depth

Original broadcast 6/3/25

Presented by Microsoft

AI & Automation: In Depth brings together federal and industry leaders to explore how artificial intelligence and automation are reshaping government operations, research, and service delivery. Host Francis Rose is joined by Nelli Babayan, FedCIV AI Director at Microsoft; Dr. Mike Horton, Acting Chief AI Officer at the Department of Transportation; and Matt Turek, Deputy Director of DARPA’s Information Innovation Office. Together, they examine how agencies are adopting AI to streamline workflows, enhance customer engagement, and advance national security. The discussion covers everything from proof-of-concept pitfalls and governance frameworks to infrastructure strategies and future planning—offering both strategic and operational insights into how AI is being implemented responsibly and at scale across the federal enterprise.

Driving Federal Efficiency with AI: Microsoft’s Vision for Secure, Responsible Innovation

The first segment of AI & Automation: In Depth, hosted by Francis Rose, features an insightful conversation with Nelli Babayan, FedCIV AI Director at Microsoft. Babayan discusses how artificial intelligence is actively transforming operations across federal agencies—streamlining repetitive tasks, enhancing employee experience, and improving citizen services. She emphasizes that AI’s impact goes beyond novelty; it's about aligning tools with real mission outcomes.

Babayan urges federal leaders to begin by deeply understanding their operational problems and goals before selecting or implementing AI tools. According to her, success starts with identifying clear use cases and defining measurable success criteria. Rather than adopting AI for its own sake, agencies should focus on whether the technology improves employee workflows, enhances decision-making, or elevates service delivery to the public.

One of the most significant shifts she notes is the accessibility of AI. With today’s low-code and no-code platforms, federal employees can integrate AI into their work with minimal technical expertise. This democratization of AI opens doors for innovation at all levels of government. However, Babayan cautions that AI also brings challenges—especially in data governance. Understanding what data exists, who has access to it, and how it’s being used is foundational before deploying AI solutions.

The conversation also delves into the integration of AI with automation. Babayan explains the rise of “agentic AI,” where AI isn’t just answering queries but performing structured tasks that are later refined or approved by humans. This human-in-the-loop model maintains accountability and ensures that AI augments rather than replaces human decision-making.

Ultimately, Babayan advocates for a thoughtful, mission-aligned approach to AI adoption. Agencies must weigh feasibility, mission impact, and risk—not just implementation cost, but also the risk of inaction in an increasingly AI-driven world. With proper evaluation, pilot planning, and clear goals, federal agencies can transition from experimentation to operational success.

Key Takeaways:

• Start with clearly defined problems and success metrics before selecting AI tools.

• Democratized access through low-code/no-code platforms is accelerating AI use in government.

• Data governance and human oversight are essential for responsible AI deployment.

From National Defense to DOT Operations: Scaling and Securing AI in Government

The second segment brings together a distinguished panel of leaders: Matt Turek, Deputy Director of the Information Innovation Office (I2O) at DARPA; Dr. Mike Horton, Chief AI Officer (Acting) at the Department of Transportation (DOT); and Nelli Babayan from Microsoft. The conversation, moderated by Francis Rose, explores two divergent but complementary perspectives on AI—DARPA’s role in national security innovation and DOT’s focus on scaling AI across agency operations.

Matt Turek outlines DARPA’s longstanding mission of creating and preventing strategic surprise—a mission that has defined its investments in artificial intelligence since the 1960s. He explains that DARPA’s AI research is designed to meet high-stakes national security demands, including military medicine and battlefield systems. Unlike traditional federal agencies that apply AI to optimize operations, DARPA’s programs are structured to push technological boundaries. Turek describes DARPA’s evolving focus through the “waves of AI,” including symbolic and statistical methods, now culminating in advancements like large language models. Central to DARPA’s approach is transitioning breakthrough capabilities from lab environments into the hands of military users—a challenge Turek refers to as crossing the “valley of death.” He notes that DARPA embeds liaison officers from military branches to ensure relevance and a clear transition path for successful AI applications.

Dr. Mike Horton offers a contrasting perspective, focusing on operational scalability within DOT. Horton emphasizes the distinction between “vertical” innovation—high-impact, narrow applications like those DARPA develops—and “horizontal” scalability, which aims to embed AI across large, complex agencies. Horton’s goal is to improve productivity and efficiency at scale, steering his agency away from flashy pilots and toward practical solutions that deliver real return on investment. He warns against “proof of concept purgatory,” where enthusiasm for innovation stalls before full deployment. Horton introduces the concept of personification—viewing each AI use case as an AI “employee.” This framing encourages federal leaders to think critically about governance, feedback loops, access controls, and the ongoing “care and feeding” of AI systems, much like they would with human staff.

Babayan reinforces the need to avoid endless piloting and emphasizes that success starts with identifying clear business goals. She lauds Horton’s “personification” framework as a smart way to bridge the technical and operational realities of AI integration. AI, she stresses, is not a magic fix—it must be one of many tools that align with mission needs.

Both Horton and Turek elaborate on how their agencies measure success. At DARPA, Turek describes customized technical benchmarks and operational transition metrics that vary by program. The goal is not just to prove the technology works, but to demonstrate its viability in real-world contexts. Horton adds that DOT evaluates success based on return on investment—not just in mission outcomes, but also in long-term sustainability and taxpayer value. He underscores that AI isn’t free; its ongoing use often increases cost due to continuous training, feedback, and maintenance. Horton uses this to advocate for smart investment strategies—only scaling tools that solve meaningful, high-impact problems.

Finally, the segment explores broader infrastructure and planning concerns. Babayan points out that cloud infrastructure, particularly high-performance computing from providers like Microsoft, is alleviating the burden on agencies to build or manage their own data centers. This platform-based approach allows agencies to focus on solving problems rather than managing infrastructure.

In closing, the segment paints a dynamic picture of how AI is evolving across government. From DARPA’s cutting-edge battlefield resilience programs to DOT’s operational scale-out of AI as workforce augmentation, the message is clear: AI must be planned, evaluated, secured, and maintained like any critical organizational asset. When done right, it becomes a strategic multiplier—not just of efficiency, but of mission impact.

Key Takeaways:

• DARPA’s mission is to create transformational AI in high-stakes environments and ensure successful transition to end-users.

• DOT emphasizes operational scalability and ROI, advocating for use cases that bring measurable mission and taxpayer value.

• Viewing AI systems as “employees” helps agencies plan for governance, feedback, and lifecycle costs.