Federal agencies are moving quickly from artificial intelligence (AI) experimentation to operational deployments, but the organizations seeing real returns are treating AI as a workforce and change management effort as much as a technology upgrade, according to Jim Kelly, vice president of federal sales at Google Public Sector.
In a recent interview with MeriTalk, Kelly said the current moment is “pivotal” because the question for government is no longer whether it will adopt AI, but “how fast” its new role as an AI accelerator will drive transformation.
“We commissioned a recent survey and found that nearly 90% of agency respondents are either planning to or are already using AI,” Kelly said. “The demand is real, but so is the need for trust, security, and measurable outcomes.”
Kelly positioned Google Public Sector’s role as helping agencies “bridge that gap from experimentation to production” with “secure, accredited AI capabilities, modern cloud infrastructure, and practical guidance.”
From experiments to operational use cases
Kelly said agencies are applying AI to “high-value use cases like document processing, data processing, workflow automation, decision support, and even predictive analytics,” with procurement vehicles helping speed adoption.
He pointed to the General Services Administration’s OneGov initiative as one lever that can help agencies procure technologies faster and at lower cost – and said the result is more AI moving into real-world environments.
On the deployment side, Kelly cited Gemini for Government deployments at the Department of Defense (DOD), rebranded as the Department of War by the Trump administration, and the Food and Drug Administration (FDA), as well as an agency-wide deployment of Google Workspace with Gemini at the Department of Transportation to more than 50,000 employees.
Kelly also highlighted work at the Department of Energy, where he said the department deployed AI co-scientist, a multi-agent AI system built with Gemini 2.0, in support of the Genesis Mission across all 17 national labs. “They’re using this multi-agent system as an autonomous research collaborator,” he said, describing early progress in hypothesis generation and automated literature synthesis to accelerate scientific discovery.
In another example, Kelly pointed to work with the U.S. Air Force Rapid Sustainment Office, digitizing paper-based maintenance workflows through a “digital maintenance binder.” The effort, he said, enables maintainers to save two to four hours per shift, speeding up aircraft release.
“What’s exciting is that many of these are enterprise-scale deployments, and they’re repeatable,” Kelly said.
Why some pilots stall and others scale
Kelly said the biggest separator between pilots that stall and programs that scale is whether agencies plan for operational use from the start.
“One of the biggest factors … is actually planning for real-world deployment from the start,” he said. That means thinking beyond proof of concept and investing early in governance, security, and workforce readiness – and selecting enterprise-grade tools that can move quickly from experimentation to production.
Kelly also emphasized metrics and mission outcomes as the litmus test for scale. “AI isn’t just a technology experiment; it’s actually a tool that can deliver changeable value,” he said. “When teams focus on [measurable outcomes], that’s when we’re seeing the real adoption.”
What’s overhyped – and what’s underestimated
Kelly argued that one of the most persistent misconceptions is treating AI as “just a more sophisticated chatbot.”
“If [agencies are] only using AI to summarize documentation or get basic queries, they’re really kind of missing the transformational portion of this technology,” he said. He also pushed back on fears of AI replacing employees, saying Google’s approach is to augment teams by offloading repetitive tasks so staff can focus on high-level mission-critical work.
What’s being underestimated, Kelly said, is the shift to agentic AI – systems that “coordinate complex workflows, support multistep decision-making, and operate across entire departments rather than in isolated pilots.”
He tied this directly to adoption. “The agencies that see impact are the ones that prioritize adoption and enablement,” Kelly said. “They don’t just deploy the technology, they empower the employees to actually manage and use it.”
Advice for CIOs: Invest in the workforce
For chief information officers (CIOs) and mission leaders trying to move quickly while staying responsible, Kelly’s guidance was direct: “Invest in your workforce as much as the technology.”
He pointed to the DOD and its recent GenAI.mil launch, which he says shows the appetite for AI paired with training. “When it launched, more than 5,000 personnel signed up for AI training within the first week,” he said, adding that earlier this month, the department announced GenAI.mil reached more than 1 million unique users in a little more than a month.
His recommended playbook: prioritize training, deploy “pre-packaged agents” that help employees learn quickly, and teach teams how to build their own agents to support focused workflows in their own environments.
Kelly said Google Public Sector differentiates itself through a “full stack approach” that brings together world-class infrastructure, “one of the largest networks in existence,” and AI tools on a unified platform. Unlike fragmented solutions, he said, the goal is to reduce integration hurdles and keep the focus on mission impacts.
“The faster we kind of weave AI into the entire fabric of government workflows, the faster we’ll start to see tangible results,” Kelly said.
What’s next
Google Public Sector’s Feb. 5 webinar on Gemini for Government – focused on how AI-powered agents can modernize public sector workflows and mission outcomes – is now available to watch on demand. View the recording here.