Managing AI agent sprawl across business units

This article presents a federated governance framework for managing AI agent sprawl across multiple business units, combining hub-and-spoke operations with risk-based classification and lifecycle management.

• Multi-BU organizations face agent sprawl: duplication, cross-BU data conflicts, hidden costs, shadow agents, and compliance fragmentation
• Hub-and-spoke model: central AI Governance Council sets standards; BU Agent Governance Leads ensure compliance within guardrails
• Self-service agents for single-BU scope; centrally governed agents for cross-BU data access or high blast radius
• Eight governance functions: registry, identity/access, guardrails, observability, cost governance, compliance, incident response, and AgentOps standards
• Four-tier agent classification: Tier 1 (safety-critical), Tier 2 (cross-BU), Tier 3 (BU-internal), Tier 4 (personal productivity)
• Agent lifecycle management with planned retirement to prevent sprawl accumulation
• Discover-before-you-build pattern with central marketplace to prevent duplication
• Platform maturity progression: Initial (centralized) → Repeatable → Reliable → Scalable (full self-service)

The framework enables fast agent development while maintaining enterprise visibility, cost control, and safety through progressive decentralization as business units demonstrate readiness.