The Urban Institute released a new playbook to guide state and local governments in deploying agentic AI. Unlike standard generative models, these systems possess autonomy to plan and execute tasks independently. The institute emphasizes the need for structured oversight as these tools are increasingly adopted by non-IT staff, which creates risks for workflow disruptions and errors.
Practical applications are already underway. Virginia utilized agentic AI to review regulatory documents, identifying $1.4 billion in annual savings and cutting permit processing times by nearly 80%. In Utah, a pilot program with Doctronic tested AI for prescription renewals, where the system recommended renewals in 72% of cases, with a 91% physician agreement rate.
The Urban Institute’s framework suggests that agencies document the purpose, target population, and estimated costs of AI agents before deployment. It also recommends defining specific metrics for accuracy and fairness tailored to the service provided. Oversight levels should vary based on whether the system is for internal use or direct constituent interaction.
To ensure accountability, the institute proposes a governance structure involving seven distinct roles, including an accountable owner, security lead, and independent reviewer. These roles are responsible for maintaining audit trails and monitoring potential harms. The goal is to move AI management beyond the IT department and integrate it into broader organizational processes.
