Artificial Intelligence & Machine Learning
,
Government
,
Industry Specific
State CIOs Are Exploring How AI Agents Can Boost Productivity, Efficiency
Reputation aside, most pen pushers in state governments don’t actually like pushing paper. They also don’t care to force citizens to fill out forms in triplicate. Two decades of promises to minimize those chores may be on the cusp of gloriously coming true with the advent of agentic artificial intelligence.
See Also: New Trend in Federal Cybersecurity: Streamlining Efficiency with a Holistic IT Approach
A report from the National Association of State CIOs outlines how states are beginning to explore ways to use agentic AI across state agencies in ways that could transform how government work gets done and how citizens interact with government.
Agentic AI systems, capable of not just generating content but of orchestrating tasks and autonomously completing workflows, have the power to reshape everything from citizens services to internal government IT.
“The easiest way to describe the biggest difference is that generative AI was generating content, writing things, summarizing things,” said Amy Glasscock, program director for innovation and emerging issues at NASCIO. “Agentic AI is AI that can actually do things. It can take steps to move processes along.”
GenAI is now used throughout state governments. Nearly all are running a GenAI pilot project, several of which are now being scaled across agencies, according to NASCIO. But taking the leap to agentic AI is one that requires even more sophisticated systems and more trust on the part of both employees and citizens.
Adoption, to date, is limited, Glasscock said. “About eight states said they had already started using agentic AI, so that’s not a lot. It’s still very early,” Glasscock said. “States aren’t going to be where the private sector might be on that yet.”
States are also moving cautiously, especially when AI systems could interact with citizen data or government decision-making. “Usually states are going to be trying things out internally before using them externally for citizen services,” she said, and “need to be really careful, because we’re talking about public trust and citizen data.”
But as states develop their systems, citizens could benefit greatly from automating high-volume administrative processes with paperwork-heavy workflows, she said. Agents could help with tasks related to applying for state services like filling out forms and uploading documents, and help the state be faster in responding to citizen inquiries. Manual review can be replaced by AI systems that check applications for completeness and identify missing documentation.
“If you can automate some steps in that process to help citizens hear back from the state sooner, they’re going to have what they need in hand sooner,” she said.
States that have projects underway include Alaska, Tennessee and Virginia. In Alaska, the state is evaluating using agentic AI in the myAlaska portal, to add features such as “autonomous completion of multi-step government transactions on behalf of users.” Tennessee is considering next-generation ERP systems with agentic AI features, and Virginia is piloting a tool that it hopes will trim the time it takes to review regulations.
Government agencies handling large volumes of citizen interactions are likely to be early adopters, Glasscock said. Human services agencies currently top the list, followed by transportation and health services.
States are also looking at agentic AI as a way to address workforce shortages and backlogs across agencies. “There’s a lot of work still left to be done in state government that governments just don’t have the workforce for,” Glasscock said. “Productivity is definitely one of the biggest plus sides.”
But although agentic AI has the power to deliver productivity and efficiency boosts across state agencies, it creates a challenging new security and governance landscape for IT teams. Agentic systems could touch multiple state systems and databases, meaning errors or a breach could have widespread consequences.
“There’s an additional layer of concern if hackers are able to control your government agents and make them do things,” Glasscock said.
She recommends that states start with NIST’s AI framework. “States are already looking at that with generative AI, hopefully they have already been implementing governance,” she said. “AI is here, it’s advancing very quickly, and if they don’t have governance in place already, it’s time to get it in place.”
For now, agentic AI remains largely experimental across state government, but NASCIO expects the conversation to accelerate quickly.
“It will be really interesting to see where we are a year from now,” Glasscock said. “We’re going to be hearing a lot more about agentic AI in 2026.”