IDC's research reveals a critical governance gap as 69% of organizations plan agentic AI implementation without adequate controls in place. Dave Schubmehl, research vice president at IDC, identifies "decision debt"—the accumulation of ungoverned autonomous decisions that compounds risk as agent populations scale—as the primary concern CX teams face. The core problem lies not in the technology itself but in how organizations manage non-human identities (NHIs) that dynamically acquire permissions and spawn subagents. For teams already deploying agentic AI through platforms like Genesys or emerging solutions, this means the gap between perceived policy enforcement and actual agent behaviour widens precisely as you scale operations, creating a compounding liability that mirrors technical debt but operates across decision-making authority and accountability.
The practical implications demand immediate operational changes. Organizations must implement pre-deployment testing that includes red teaming for prompt injection, stress testing across connected environments, and validation of reversible controls—kill switches, clamps, and rollback mechanisms—before agents touch production systems. Equally critical is establishing real-time monitoring for specific signals: anomalous agent behaviour, unexplained permission escalations, cost or token consumption spikes, performance drift, and breakdowns in human-in-the-loop escalation. For CX leaders, this raises a pointed question: if your current governance frameworks struggle to audit human agent decisions, how will you maintain visibility and control over autonomous agents operating at scale across customer interactions?
The timing matters. As vendors like Genesys and others roll out agentic orchestration stacks, the window for establishing governance baselines is closing. Teams that treat agentic AI deployment as a straightforward technology upgrade—rather than a fundamental shift in decision-making authority—will face compounding risk exposure. Audit trail completeness and real-time instrumentation are not optional enhancements; they are prerequisites for production deployment. Without them, your organization accumulates decision debt that becomes exponentially harder to remediate as agent populations grow.
Analyst Dave Schubmehl speaks with No Jitter about agentic AI and what organizations should know when implementing this tool.