Federal AI leaders say grounding, human oversight, cost visibility and workforce expertise are essential to scaling AI effectively.
The race to adopt AI is shifting from finding the most powerful tools to building the foundations that allow organizations to use them effectively and responsibly.
Federal and industry leaders said Thursday at the GovCIO Media & Research Federal AI Forum that agencies do not need to wait for perfect AI systems. Instead, they need to build the right foundations around the technology so they can use it with confidence and scale what works.
Brian Frutchey, CTO of Clear Fracture, said that starts with making sure that the decision-maker, whether human or AI, has the context it needs before taking confident action.
“AI agents quietly fail not because they lack reasoning power, they fail because they lack grounding,” he said. “An agent operating on vague instructions and unorganized situational awareness is just improvising. An improvisation leads to hallucinations, compliance failures and security risks. If your start and endpoints are fuzzy, agentic speed just gets you lost faster.”
NOAA is already applying that approach in the field. Stephen Bieda, acting deputy director of the NOAA Center for AI, described AI systems that help sift severe-weather reports, identify wildfire hotspots and accelerate weather prediction.
The agency’s AI-based wildfire detection has given wildland firefighters an average of 17 minutes of lead time before a 911 call The system is designed to alert forecasters, who then evaluate the information rather than allowing AI to make the final decision
NOAA is also using AI to reduce the computational resources required for global weather prediction, while continuing to incorporate large amounts of observational data. The agency is working to make its hundreds of petabytes of data more accessible for AI applications while developing policies and risk controls to support responsible adoption.
The same focus on building responsible practices is emerging around AI spending. Sidney Epner, a product manager at Apptio, said organizations are learning that managing AI requires more than tracking tokens and added he saw a 4.7-fold increase in model spending over six months across his customer base.
Epner said it’s important for organizations to have the right insight into their AI spending to connect their investments with measurable outcomes.
“Connecting that spend to the outcomes, whether it’s the cost per case processed, the incident resolved or mission outcome-not just what you spend, but whether it was worth it. Once you can prove it, the conversation changes,” he said. “It stops being about minimizing AI spend, and it starts being about optimizing it for cost, sure, but also for performance, risk and mission value, and for federal teams, that also includes data sovereignty, security review and procurement cycle.”
For the 4thMarine Logistics Group, the path to broader adoption is being built around people. Chief of Staff Col. Scott Wilson described how reservists with civilian expertise in AI and innovation helped him develop an AI-enabled command climate survey prototype in three hours.
As a result, surveys that previously took months due to low participation rates and other obstacles, now had a quicker turnaround, higher participation and a lower barrier to entry, Wilson said. Wilson and his innovation team are currently working to roll out the tool, PULSE Check, to other units to assist commanders in making real time command climate analysis.
“We’re hoping that in the next year, next time I come talk at this forum, we’re going to be able to show how we’ve rolled out the forward deployed engineer concept, how we have enabled AI innovation and modernization at the component MLGs, and we’ve been able to accelerate their decision-making and their resource optimization using AI tools that are available to us now. It’s incredible what’s out there. We’re making a lot of progress,” Wilson said.
This is a carousel with manually rotating slides.
Use Next and Previous buttons to navigate
or jump to a slide with the slide dots
