Kentucky inventory connects school AI to existing student and workforce systems

Separate the district's student-support problem from its educator-service navigation problem. Ask counselors what happens after an alert and ask workforce staff which information requests generate repeat contacts. A bounded engagement could map one existing workflow and establish its baseline before adding an assistant. The value hypothesis is better routing or more timely human follow-up, subject to testing. Include the student-services director, IT and the staff who actually handle exceptions in discovery. KDE's inventory supplies examples, not evidence of savings or improved retention. Do not turn an implementation list into a claim that the same intervention will work in another state or district.

For early warning, validate source data, refresh timing and role permissions before exposing a score to staff. For educator navigation, use approved public content with visible source links and a maintained human contact route. These are separate engineering proofs with different sensitivity and failure consequences. Prerequisites include a documented current process and a representative test set. Test stale information, missing records, ambiguous questions and unavailable downstream services. Measure false alerts and missed cases where suitable labels exist; measure answer accuracy and handoff completion for navigation. Neither the inventory nor its use of the word agent demonstrates a validated autonomous architecture, model choice or hosting requirement.

Appoint a student-services owner for the alert workflow and a separate educator-services owner for navigation. Inventory dependencies, document the current handoff and train staff on exceptions before a small pilot. Establish privacy approval and a review calendar with IT.

Proposed acceptance criteria

sampled alerts reach an authorized person with a recorded disposition, navigation answers cite approved current material, and users can reach staffed support when the tool fails. Agree targets after collecting a baseline. Measure staff effort alongside completion, and evaluate any claimed learning benefit separately with independent assessment. Risks include stale data, extra alerts without response capacity and historical descriptions being mistaken for current operational assurance.