How AI is Delivering Real Mission Results Across Government

Federal technology leaders discussed how they are moving AI applications beyond experimentation and into production during GovCIO Media & Research’s Federal Tech Leaders Summit on Friday. While each speaker highlighted different use cases, a common theme emerged: AI succeeds when it empowers people, improves efficiency and supports rather than replaces human judgment. 

Building an AI Culture That Puts People First 

Brian Campo, chief data and AI officer and director of technology readiness at the Coast Guard, said the biggest challenge in AI adoption is cultural, not technical. 

“We’re really trying to get people to touch it, feel it, understand it and bring it into the work that they do,” Campo said. 

To accomplish that goal, Campo’s office launched a “digital roadshow” that brings AI training directly to Coast Guard personnel across the organization. During a recent visit to Sector Miami, Campo’s team worked with employees who had little or no experience using AI. After a week of training and hands-on workshops, participants created 83 AI agents and 25 automations. Campo estimated the effort generated approximately 45,000 hours of time savings. 

Yet, he stressed that those savings are not about workforce reduction. Instead, the Coast Guard measures success by “putting more minutes on mission.”  

“They spent time figuring out what was taking them away from the mission. Is it administrative paperwork? Is it drudgery?” he said. “By helping them understand that my goal with AI is not to replace them, it’s not to take away the fun part of the job, it’s not to move their job to a place that they don’t want to be or to make them do things that they’re not interested in. It’s to make them more effective at their job. It’s to help them get back to the parts of their job that they love.” 

Replacing Government Systems Through AI-Enabled Development 

The National Endowment for the Arts completed a project that would have taken more than a year in a week with the help of AI, according to CIO and CAIO Jim Tunnessen. 

When confronted with an outdated data center “knee deep in technical debt,” Tunnessen said one of the agency’s AI engineers was able to build a “modern tech stack replacement system with a user-friendly UI” in seven days. 

“Let me be clear about what that means,” he said. “This was not a vibe coding toy app. This was a production system, with documentation, with tests built by a human engineer using AI agents as force multipliers for every single phase of development.” 

For Tunnessen, the experience shifted the conversation from maintaining aging systems to creating AI-enabled DevSecOps pipelines. In this model, agents assist throughout the development lifecycle, generating code, conducting security scans, updating documentation and monitoring applications. 

Despite the advances, Tunnessen emphasized that AI is not a replacement, but rather a tool for employees who will remain accountable.  

“The developer who rebuilt the system in a week wasn’t superhuman. He had the right tools and the right model. AI agents as collaborative partners, not replacements, across the full development lifecycle,” he said.  

Preventing ‘Workslop’ Through Transparency and Oversight 

As agencies expand AI use cases, leaders also emphasized the need for governance and human oversight to ensure outputs remain accurate and useful. Harvard Business Review defines “workslop” as “AI-generated work content that masquerades as good work, but lacks the substance to meaningfully advance a given task.” In a September 2025 survey, Harvard Business Review found that 40% of 1,150 U.S.-based full-time employees reported receiving work-related stress in the previous month. 

At the Treasury Department, an emphasis on transparency is helping ensure quality output, according to Justin Cole, director of data transparency at the Bureau of the Fiscal Service. His team’s natural-language search model for USAspending.gov gives users the ability to search over 100 million records using the “type of terminology and language that we use every day.”  

In addition, the LLM explains its reasoning process for the answers it provides and gives users an opportunity to submit feedback so the model can improve over time.

Campo added that it’s important for people to use AI as a collaborator, not a decision-maker.

“Decision-makers still own the decision space. They still own the responsibility, the authority and the context that they are giving up the chain,” he said. “You do not get the opportunity to just put a query in, take the data and hand it off. I think you have to do it through responsibility, you have to do it through authority, and people have to understand that this is just a tool. You are still empowered to make decisions, you are still responsible for the decisions you make.” 

Tunnessen echoed the importance of accountability.  

“You own the work product that you do,” he said. “Just because you had AI assist you in this doesn’t mean that AI owns it.” 

He also challenged vendors to place greater emphasis on verification and validation. While many AI tools focus on generating outputs, he said agencies need evidence of accuracy, confidence levels and performance thresholds before relying on those systems.