Getting real about AI strategy and deployment – Federal News Network

Federal experts share real-world lessons on deploying AI across aging infrastructure, slow funding cycles, and workforce gaps.
Watch the entire discussion here.
If you’ve spent any time in federal tech circles lately, you’ve probably heard the phrase “AI is a journey, not a product.” It’s catchy but it’s also true, and it came up more than once during a recent panel I moderated with three seasoned enterprise architects: Kurt Steege from Thundercat, Landon Shaw from the U.S. Courts, and Erik Erichsen from Pacific Northwest National Laboratory.
We gathered to talk about what it really takes to deploy AI tools that can modernize how agencies operate. We wanted to get beyond the glossy sales pitch to the gritty reality where legacy systems, power-hungry data centers and workforce gaps collide with the promise of smarter, faster, more secure government operations.
Kurt Steege, Thundercat’s CTO, didn’t mince words when it came to the age and fragility of federal infrastructure. “Much of the critical infrastructure that’s out there is quite old,” he said. “You don’t have the ability to build something up completely in parallel, because many of these things are running important tasks… This sort of technical debt spiral, where you’ve got all of this stuff that’s been out there, in some cases still running mainframes, becomes quite difficult to fund and transfer over.”
That’s both a funding issue and a risk management problem. Agencies can’t afford additional downtime incurred with a system replacement, but they also can’t afford to keep patching systems that were never built for today’s threats or workloads. And as Kurt pointed out, agencies can’t skip over the physical layer of modernization that includes data centers, power grids and cooling systems.  Those are literally and figuratively foundational to any successful AI deployment.
Landon Shaw brought a unique perspective from the judicial branch. He’s worked in both defense contracting and government, and he’s seen firsthand how slow the funding cycle can be. “Even if I wanted to get a project funded now, it would be too late for me to do it the rest of this year and probably next year,” he said. “I would have to put it in next year and it would be the following year.”
That’s a problem when AI models are evolving weekly. Landon’s advice is to stop chasing the perfect model. Instead, focus on organizing your data, especially historical and archived datasets, into vector databases that can be queried by whatever model is best at the moment. “The industry is moving so quickly that that trained model you spent millions of dollars on last year has been obsolete for six months now,” he said.
It’s also a sobering reminder that those leading these projects have to keep an agile mindset and stay flexible as the technology races forward.
Erik Erichsen brought the conversation back to people. Technology is only as good as the teams deploying and maintaining it, and Erik’s an expert at leading transformation efforts across both public and private sectors.
“We’ve stood up what’s called the Center for AI within PNNL,” he said. “It brings together experts across the organization and gives people a sandbox to experiment with models, agents and tools. It’s about removing fear and building confidence.”
That kind of internal capacity-building is critical, especially when so much federal IT work is outsourced to contractors. As Landon pointed out, formal AI certifications or degrees don’t necessarily guarantee expertise in a field that is changing moment by moment. Agencies need to invest in upskilling their current workforce while also rethinking how they evaluate and onboard new talent.
Kurt added a generational knowledge transfer layer to that challenge. “Getting that information down from folks who are retiring and loading it into databases or LLMs so it can continually be worked on,” he said, is one way to preserve institutional memory while preparing for the future.
If there’s one thing all three panelists agreed on, it’s that modernization isn’t a one-size-fits-all process. Agencies are at different maturity levels.  They face different constraints and move at different speeds. But the core principles of planning deliberately, investing in infrastructure, organizing your data and supporting your people apply across the board.
We’re all on this journey together. Erik emphasized the value of sharing lessons learned, even when things don’t go as planned. And with the right mix of strategy and cross-functional teamwork, federal agencies can build AI infrastructure that’s secure and resilient so that they’re ready for what comes next.
Copyright © 2025 Federal News Network. All rights reserved. This website is not intended for users located within the European Economic Area.
Terry Gerton is host of the Federal Drive and has been working in or with the federal government for more than 40 years.

source
This is a newsfeed from leading technology publications. No additional editorial review has been performed before posting.