- The GenAI playbook lays out seven delivery phases
- StateChat, the State Department’s chatbot, anchors the guide
- The chatbot is backed partly by a GSA modernization fund
The Department of State has published a generative artificial intelligence playbook meant to guide federal agencies through building and scaling enterprise AI tools. The document, released last week, uses the department’s StateChat chatbot as its case study.
Kelly Fletcher, chief information officer at the State Department, and Amy Ritualo, acting chief data and AI officer, introduced the guide in a joint letter. Fletcher is a two-time Wash100 Award winner.
Scaling AI in government headlines the agenda at the Potomac Officers Club’s 2026 FedCiv Summit on Oct. 29. DOT Chief Digital and Information Officer and Wash100 awardee Pavan Pidugu will deliver the keynote. Register now!
What Does the State Department’s GenAI Playbook Cover?
The guide breaks enterprise AI delivery into seven stages and assigns a rough duration to each. Groundwork covering data and AI strategy runs for two to three years. Alpha and beta development take four to six months apiece. Launch occupies four to six weeks. Driving adoption runs six to eight months. Another four to six months covers deeper functionality, from complex workflow integration to AI agents. A closing impact assessment takes six to eight weeks.
Each chapter pairs the department’s own experience with a checklist and a set of lessons. An appendix catalogs other State AI tools, among them the Northstar media analytics platform, the SMART correspondence toolset and three developer environments named Proving Ground, Funhouse and Prompter API.
How Did State Build and Scale StateChat?
The Bureau of Diplomatic Technology’s Center for Analytics led the design and deployment of StateChat. Funding came in part from the General Services Administration’s Technology Modernization Fund, with that investment beginning in May 2024.
StateChat is approved for sensitive but unclassified data. It secured Federal Information Security Management Act High authority to operate following a yearlong process.
Alpha testing of the tool drew 150 participants. Beta expanded the base to 8,000 users over five months. Users reached 23,000 by the close of the first quarter of live operation, up from 12,000. State reported more than 62,000 users as of June 2026.
What Did the StateChat Impact Assessment Find?
State surveyed users between Aug. 26 and Sept. 26, 2025, after one year of enterprise deployment. Of the 3,700 employees who began the survey, 2,706 provided responses sufficient for analysis. The department said 90 percent of regular users reported time savings and 96 percent said they would recommend the tool. Average reported savings came to 1.6 hours per week per task.
The survey used an opt-in design, and the department said statistical inference is not appropriate for the results.
Regular users who reported no time savings pointed to response quality, time spent verifying outputs and weak integration with systems they already use.
The playbook identifies agent orchestration as the State’s next direction. FAMsearch, a tool that answers questions drawn from the Foreign Affairs Manual and Handbook, now runs in the background of StateChat rather than as a separate tab.
