The Mississippi Artificial Intelligence Network (MAIN) is announcing its AI Agents Playbook, a free resource for Mississippi organizations deciding what an AI agent may do, who can approve its actions and how its work will be checked.
Available online since October 16, 2026, the 22-page guide is written for process owners, IT and security leads, data owners, purchasers and people who review agent work. Readers can use the web version or download the PDF.
An AI agent can select steps and use connected tools to carry out a task. The playbook helps teams decide whether that flexibility is needed, compare it with an assistant or fixed workflow, and define the limits before connecting systems or information.
Plan one task before connecting tools
The guide starts with the existing human workflow, the intended work product and the people responsible for accepting it. Seven working tools cover the agent charter, permissions, action review, evaluation, supplier and configuration review, pilot acceptance, and incident response and restart.
A worked example follows a hypothetical community college workforce office researching public funding opportunities. The proposed agent gathers evidence from approved public agency sources and prepares a lead table for staff review. When an authoritative eligibility notice cannot be retrieved, the example calls for an unresolved finding instead of an invented answer. The agent cannot submit an application or contact a funder.
Other scenarios address service-request routing, maintenance-document review and software-change preparation. Each identifies permitted work and the decisions that remain with qualified people. The scenarios are examples for local adaptation and do not report tested deployments or measured performance.
Keep approvals tied to institutional authority
The playbook distinguishes approval to use a product from approval for a particular action. Its planning records ask teams to identify the adopted policy and approving authority. A routine operational problem can be escalated to a person; a proposed exception to policy goes through the organization’s existing exception process.
Its security sections reference sources including the joint agentic AI guidance coauthored by NSA and CISA. MAIN’s worksheets help organizations apply the material to a specific task. They do not grant deployment authority or certify that a system is secure.
Teams can pair the playbook with the relevant MAIN sector policy guide and their own approved policies. Pilot decisions account for useful results, failures, review effort, costs and unresolved issues before the authorized owner decides whether to continue, restrict or stop.
Read the AI Agents Playbook or download the 23-page PDF. The resource is also listed in MAIN’s AI playbook collection.