A global IBM survey of 2,000 CIOs and CTOs finds that two-thirds are held responsible for AI systems they can’t fully supervise, and only 11% feel ready for large-scale AI agent deployment. Respondents expect a 38% rise in AI agents by 2027, yet 77% say their current governance lags behind adoption. Manual governance correlates with more incidents, while those embedding controls into AI see 25% fewer problems. Last year organizations logged an average of 54 AI agent incidents needing human intervention; 17% were high-severity, with 37% causing data exposures or breaches, 33% triggering cascading failures, and another 17% sparking compliance breaches. Nearly 60% of tech leaders cite security and compliance as their top barriers to scaling AI. Financial management of AI remains weak: 84% haven’t operationalized AI budgeting, and 85% lack real-time visibility into AI spend. Yet firms that build controls into their AI deploy 16 times more agents, achieve 18% higher operating margins, and use only 25% of their AI budget. Teams with strong financial discipline deploy 2.4 times as many agents without hiking their AI/IT budget and are three times likelier to feel fully prepared for scale. Early focus on adaptability also pays off. Organizations keeping workloads portable and models replaceable report a 10% higher return on AI investment in 2025. According to Matt Lyteson, IBM’s CIO, architecture, governance, and investment choices now determine pace, risk tolerance, and which AI opportunities succeed.