Procurement should govern agentic AI through shared ownership with IT, clear decision rights, documented data controls, human approval for consequential actions, and auditable records of how recommendations and actions were produced. The goal is to automate routine work without transferring commercial accountability to a system.

At ProcureCon Indirect West, Marla Bradstock, Director – Procurement, IT Business Partner at Gilead Sciences, described why governance must extend beyond data access and technical performance: "The results of that analysis need to go to a human who questions it." Her point applies especially to supplier decisions, category strategies, and contractual choices, where internal context, changing business requirements, and risk appetite cannot be fully captured in a model.

The 2026 ProcureCon Insights report An Achievable Future for AI in Procurement found that 44% of procurement leaders see better compliance and audit trails as the strongest business case for agentic AI, ahead of measurable cost savings at 34%. This indicates that leaders may be more likely to support agentic tools when they can review activity, understand outcomes, and preserve control over important decisions.

CPOs, CIOs, procurement risk leaders, and legal teams should create a joint governance group before deploying agents beyond low-risk tasks. Define which activities an agent may complete, which require human review, what data it can access, how exceptions are escalated, how outputs are tested, and how audit records are retained. Start with bounded tasks such as classification, intake routing, document extraction, or routine follow-up before extending agent authority into sourcing, contracting, or supplier management.

Join us at ProcureCon West to examine how procurement and technology leaders are building trustworthy agentic AI programs with practical controls, human accountability, and measurable value.