Instances of agent AI, where groups of digital agents are given permission to automate multi-step tasks without human approval, remain rare in state governments, but a report released Tuesday by an industry group suggests adoption may be on the rise.
The report, published by the National Association of State Chief Information Officers, notes that state governments appear to be increasing the level of autonomy for artificial intelligence agents, although “this change will not happen overnight.” “I think this is still really new,” said Amy Glasscock, director of NASCIO’s Innovation and Emerging Issues Program and author of the report. “But there are some states that are starting to use it.”
“Agenttic appears to be the next step,” she continued. “[Generative] AI started out doing content creation, generation, summarization, things like that, but now states are figuring out how AI can actually do some things and do some of the work, but of course there’s still a lot of oversight. Glasscock said a recent survey of members of the group, which includes most state and territorial CIOs across the country, found eight states were using agent-based AI, but declined to name them.
However, the report cites several early state efforts to automate low-risk tasks using agent AI, based on public documents and news reports. That includes the Alaska Department of Government, which issued a request for information last November noting the state’s imminent plans to add agent functionality to its mobile app. The document cites interest in “advanced notifications, reminders, and personalized recommendations for relevant national services,” “context-aware assistance including dynamic form filling, document retrieval, and eligibility checks,” as well as proposed uses of AI in back offices, including “autonomous data exchange, workflow orchestration, and facilitation of service fulfillment across disparate platforms.”
Last summer, former Virginia Governor Glenn Youngkin issued an executive order launching a “first-of-its-kind pilot program” that uses agent AI to scan all federal regulations and guidance documents, find duplications and suggest “streamlined language.” His office dubbed this the “Virginia Model” of regulatory modernization, arguing that AI would “significantly enhance the regulatory reduction process.” But Youngkin’s order is no longer posted on the state’s website, and President Abigail Spanberger’s administration has not said whether it plans to continue using Agent AI.
In his report, Glasscock envisioned five stages of AI maturity that the state would move through, starting with “assisted” generative AI tools, where humans are the “primary performers and decision makers.” The majority of states report basic usage of such AI. Phase 2 is “more sophisticated” and “context-aware” generative AI. Phase 3 is “task-level automation.” This automation allows agents “limited actions within defined boundaries,” such as pre-filling resident forms using data already collected.
Phase 4 is “Stateful Multi-Step Workflow.” This workflow allows the agent to maintain context about the user, not just the most recent conversation, and progress is managed over time. This may be used to guide employee training and resident benefit applications. In Phase 5, “AI starts working before anyone asks it,” the report says.
Glasscock noted that governments tend to lag behind the private sector in adopting new technologies, and it could be years before fully autonomous AI agents can roam state agency networks. One barrier to adoption cited in Tuesday’s report is that state governments maintain many outdated computer systems that aren’t necessarily compatible with other systems, let alone the latest AI. Another barrier, of course, is that nations tend to be very risk-averse, and agent AI may be a tough sell, at least for some CIOs.
Glasscock said he expects states to start talking about the initial use of agent-based AI in the coming months. (The association plans to hold an interim conference in Philadelphia in late April.) She imagined that early uses of agent AI could include small, discrete tasks with minimal risk. In 2026, it will be even harder to imagine Phase 5 agent AI operating within state governments. “I think there’s a real possibility when it comes to automation, but it’s not just about doing all the work and making all the decisions right away,” she said.
