Enterprise leaders are losing grip on artificial intelligence deployments happening beyond their view. According to research by Dataiku, the vast majority of chief information officers cannot see the full picture of AI agents their organizations are building through informal routes. The survey, which polled 685 CIOs across eight countries, found that 81% lack complete visibility into agents created outside established governance frameworks.
Harris Poll conducted the research online between 9 and 29 July on behalf of Dataiku, gathering responses from CIOs in the United States, United Kingdom, France, Germany, the United Arab Emirates, Japan, South Korea and Singapore. Dataiku, founded in Paris and now headquartered in New York, unveiled the Global AI Confessions Report: CIO Edition on Thursday during Succeed, the company's annual user conference.
Monitoring without management
A striking contradiction emerged from the data: while 90% of CIOs reported having complete tracking capabilities for their agents, 72% admitted they cannot reliably verify whether those same agents are actually delivering the business results they were designed to achieve.
The scale of deployment is substantial. Two-thirds of respondents estimate that 51 or more agents are currently running in production environments. Yet organizational readiness lags far behind: 83% lack standardized lifecycle management processes for agents across their companies, and 47% have already shut down more than 20 agents in the current year. Only 21% possess full, near-real-time visibility into the costs associated with AI spending broken down by team or specific use case.
The governance challenge is acute. 84% of CIOs acknowledge that their employees are building agents and applications at a pace that outstrips IT's ability to govern them. Meanwhile, 60% lack a centralized AI governance framework that spans their applications, tools and IT infrastructure. When agents malfunction, responsibility becomes murky: respondents distributed accountability among shared teams (23%), central IT (21%), data or AI teams (20%), and security, risk and compliance functions (18%).
Monitoring tells you an agent is running. Managing tells you whether it's earned the right to keep running, and right now, almost nobody can fire an agent.
Florian Douetteau, co-founder and CEO of Dataiku
Career stakes and budget pressure
The pressure on CIOs to deliver AI results is mounting. 88% said that their professional reputation or career trajectory hinges on AI success, and 87% reported that their chief executive has explicitly tied their job security to AI performance. Looking ahead, 76% expect their positions to be at risk if their organizations fail to demonstrate measurable AI gains by the end of 2027.
Financial consequences loom as well. 72% anticipate that missing performance targets by the end of 2026 will trigger cuts or freezes to their AI budgets. Across the board, 97% report experiencing at least some increase in board-level pressure to show returns on AI investments.
United States CIOs face particularly acute challenges. 94% of American respondents say their staff is building agents faster than IT can manage them. 87% expect budget cuts or freezes if they miss targets. 82% regret at least one significant AI vendor or platform decision made in the past 18 months.
On the technology front, 75% of CIOs plan to adopt more models or different ones. 74% are exploring open-source or open-weight models as insurance against the risk that a proprietary model becomes unavailable.
Dataiku's answer
Dataiku is addressing this governance gap with new software. The company launched Agent Management on the same day as the report release, a tool designed to consolidate agents into a single inventory system. The product integrates with AWS Bedrock, Databricks, Google Vertex, Microsoft Copilot Studio and Azure Foundry, Salesforce Agentforce and Snowflake Cortex. General availability begins in October, with pricing structured per instance annually and monitoring charges assessed per agent.
The challenges Dataiku is targeting align with broader industry concerns. Gartner projected in June 2025 that over 40% of agentic AI projects will be abandoned by the end of 2027, citing escalating costs, unclear business value and insufficient risk controls. At Amazon, AWS has observed that nearly 90% of early agent prototypes never advance to production.
Source: The Next Web
