Governments test autonomous AI systems

State and local governments are experimenting with agentic AI that can independently plan and take action toward goals, unlike conventional generative AI that only produces content.

Virginia finds $1.4 billion in savings

Virginia's pilot used agentic AI to review regulatory documents for redundant language, identifying over $1.4 billion in annual savings and cutting permit processing times by nearly 80%.

Utah and Boston deploy their own pilots

Utah tested autonomous AI for prescription renewals with 91% physician agreement on recommendations. Boston integrated agentic AI into its open-data portal to analyze municipal data and inform policy.

Non-IT staff can now build AI agents

Unlike traditional AI, agentic systems can be created by employees outside IT departments, accelerating innovation but also creating risks for errors and workflow disruptions in sensitive functions.

“What's different about the agentic generative AI is that the IT person is not the only person who is empowered to build it.”

Document purpose before deploying agents

The Urban Institute recommends agencies document an agent's intended purpose, target population, measurable outcomes and estimated costs before deployment, with accuracy and fairness defined separately.

Oversight should match the stakes

Internal systems can operate at smaller scales with less stringent controls, but constituent-facing systems like disability benefits require greater scrutiny and fairness metrics tailored to specific populations.