IBM Research Finds Tech Leaders Struggle With Agent Sprawl
In an era where artificial intelligence (AI) is rapidly transitioning from experimental pilot programs to full-scale production, a recent study by IBM highlights significant challenges faced by technology leaders. The survey reveals that approximately two-thirds of Chief Information Officers (CIOs) and Chief Technology Officers (CTOs) feel responsible for AI systems and their outcomes, even though they do not have comprehensive control over these systems. Conducted by IBM’s Institute for Business Value, this research surveyed 2,000 technology executives across 33 countries and 19 industries, underscoring a pervasive struggle to manage the burgeoning complexity of AI systems.
The study, which took place between January and April 2026 in collaboration with Oxford Economics, sheds light on the growing pressures these leaders face. With 80% of respondents indicating that their Chief Executive Officers (CEOs) are mandating faster AI deployment, there is an expectation that organizations will manage an average of 1,661 AI agents by 2027, marking a staggering 38% increase from the current figures. However, only 11% of these leaders claim to feel adequately prepared for such growth.
Matt Lyteson, CIO at IBM, articulated the core challenge confronting CIOs and CTOs today: the need to scale AI systems that function autonomously within governance frameworks that were designed for more stable environments. "It is no longer just about deploying AI faster; it’s about redesigning organizational control, governance, and investment strategies," he stated. This realization emphasizes the necessity of embedding control and visibility measures from the outset to facilitate robust scaling.
Compounding these challenges is the phenomenon referred to as "Shadow AI," wherein business units deploy AI solutions without full oversight from IT departments. A shocking 70% of executives reported that these units are advancing faster than they can monitor. This unregulated growth has led to organizations experiencing an average of 54 AI-related incidents over the past year, with 17% of those incidents being classified as high-severity—requiring over four hours to resolve. The fallout from these incidents is severe, with 37% of organizations experiencing data exposure, 33% facing cascading system failures, and 13% noting a decline in stakeholder trust.
Victoria Medina, Chief Technology and Data Officer at Allianz Spain, echoed these sentiments by pointing out the dual nature of AI. While many organizations are focused on harnessing AI’s potential, she cautioned against underestimating the vulnerabilities it introduces. “AI has both a light side and a dark side,” she explained, underscoring the need for a balanced perspective.
The survey further elucidates a phenomenon known as the "scaling trap," wherein the rapid growth of AI agents outpaces an organization’s governance capabilities. A staggering 77% of respondents acknowledge that AI adoption is advancing faster than their capacity to manage it, while 59% cite compliance concerns that hinder further scaling of AI agents. Moreover, more than two-thirds indicated that business units frequently bypass IT entirely in their quest to integrate AI tools.
This leaves CIOs and CTOs in a precarious position: prioritizing speed may yield short-term satisfaction for CEOs, but it heightens the risk of governance lapses due to uncontrolled growth. Afonso Eça, an executive board member at Banco BPI, likened the scenario to piloting an aircraft under impossible conditions—illustrating the immense pressure technology leaders face as they attempt to navigate these complex dynamics.
Amidst these challenges, the report suggests that organizations adopting "orchestrated control" systems—integrating essential guardrails, telemetry, rollback mechanisms, and kill switches into their architectures—are experiencing more favorable outcomes. IBM’s survey indicates that such organizations deploy 16 times more AI agents than those that rely on manual governance. Additionally, these companies incur four times lower expenditures in managing AI while achieving 18% higher operating margins and experiencing 25% fewer AI-related incidents.
As AI deployments accelerate, financial observations are also noteworthy. Projections indicate AI spending will rise from approximately 15% of IT budgets in 2025 to nearly 25% by 2027—a staggering 71% increase. Despite this growth potential, 84% of technology leaders confess they have not fully operationalized AI financial management, and 85% lack real-time visibility into AI expenditures.
To mitigate the challenges of agent sprawl, IBM advises technology leaders to focus on three key areas: infrastructure adaptability, governance by design, and stringent portfolio discipline. Organizations excelling in these areas report 38% higher expected revenue growth and 7% greater anticipated operating margins. These companies also deploy 2.6 times more AI agents compared to their peers.
The study reveals an encouraging trend, as 88% of organizations are either attempting or planning to shift workloads to alternative cloud providers. However, technology leaders report that only 25% of those workloads are easily portable. Furthermore, cloud costs have exceeded initial projections by an average of 48%, and 80% of tech leaders acknowledge experiencing higher-than-expected data transfer expenses.
IBM’s observations suggest a strategic approach to infrastructure management, urging organizations to prioritize high-value workloads instead of treating all workloads as equally crucial for migration. By carefully mapping exposure to vendor lock-in across infrastructure, data, and AI models, companies can identify strategic decisions that enhance flexibility without substantial disruption.
In the realm of governance, it is essential for CIOs to enforce production standards rigorously. If an AI agent or model lacks registration, ownership, observability, and stoppability, it should not be deployed. Furthermore, CIOs are encouraged to choose a critical functionality—such as customer service or claims processing—and to redesign governance processes comprehensively.
From a financial standpoint, technology teams should focus on initiatives with strategic value, discontinue funding pilots that lack explicit lines of ownership and success metrics, and prioritize workloads that deliver higher operational value.
Michael Voegele, Global CDIO at Philip Morris International, encapsulated the urgency of the situation succinctly: "AI isn’t just another technology disruption. Unlike past innovations, AI is a self-fueling organism that accelerates its own development." This underscores the critical need for businesses to adapt promptly and thoughtfully in order to harness the full potential of AI while managing its inherent risks effectively.