A county government plans a cross-agency generative AI assistant to help caseworkers screen residents for eligibility across health, social services, and housing programs. During planning, the county discovers that each department stores its client data in a siloed system with no agreed process for sharing records across agencies. What should the county establish before the assistant can move forward?

A cross-agency AI initiative that depends on combining records from multiple departments cannot function safely or effectively until those departments agree on how data moves between them. The scenario describes siloed systems with no agreed sharing process, which is a data governance gap sitting directly upstream of the AI assistant itself. A formal data-sharing framework addresses this by defining which fields can be exchanged, the legal basis for exchange, retention limits, and safeguards like access controls or audit logging, giving the assistant a lawful and consistent foundation to query. Assuming a larger model can reason across incomplete records misunderstands the problem: model capability does not create legal authority or technical access to data the departments have not agreed to share. Training caseworkers to manually cross-reference records substitutes a labor-intensive workaround for the actual fix and does not scale to the volume an automated assistant is meant to handle. A public awareness campaign, while eventually useful for transparency, addresses communication rather than the structural blocker preventing the assistant from functioning at all. A scope caveat: a data-sharing framework typically still requires legal review specific to each program's governing statutes. An operational check: verify whether a signed data-sharing agreement exists between all three departments before any integration work begins.