AI Agents: Costs Soar, Security Risks Emerge

As enterprises move agentic AI beyond pilot programs, significant challenges in cost discipline, security, and organizational adoption are becoming apparent. Brian Gracely, senior director of portfolio strategy at Red Hat, noted that many companies overestimate their lag in AI agent deployment, as learning curves are often steeper than anticipated. This rapid progress, however, directly correlates with rapidly increasing AI costs, shifting cost management from an engineering issue to a boardroom concern.

Controlling AI expenditures is becoming critical, especially as organizations recognize their dependence on a few major model providers. Gracely suggests that right-sizing AI models for specific tasks, through mechanisms like semantic routing and caching, is the most effective way to cut costs. This approach avoids defaulting to the most powerful, expensive models when simpler ones suffice. The financial discipline required for managing token spend will mirror established FinOps practices for cloud compute.

Furthermore, the speed at which AI tools discover vulnerabilities necessitates a reevaluation of traditional patch management cycles. Enterprises may face short windows, potentially seven to 14 days, to identify, validate, and deploy patches as AI accelerates both vulnerability discovery and exploitation. This rapid pace transforms software update management into a strategic capability.

Ultimately, the successful scaling of agentic AI hinges on organizational factors. Gaining the trust and sustained involvement of subject matter experts is essential. Companies must address incentives, ensuring that experts feel their roles are valued and not threatened by AI, thereby fostering cooperation and enabling widespread innovation.