Dataiku Surveyed 685 CIOs. 81% Say They've Lost Oversight of Their Own AI Agents. 72% Can't Consistently Prove Those Agents Deliver Results.

November 10, 2026 · Josh Agusti, Founder, Aiven Consulting Group

For the past two years, most enterprise conversations about AI agents have centered on adoption speed: how many agents exist, how fast they are multiplying, how much work they are quietly absorbing. A new survey suggests the more urgent question is a different one. It isn't whether agents are spreading. It's whether anyone still knows what they're doing.

The Oversight Gap

The Harris Poll conducted the research on behalf of Dataiku between July 9 and July 29, 2026, surveying 685 CIOs across the United States, the United Kingdom, France, Germany, the UAE, Japan, South Korea, and Singapore for the company's Global AI Confessions Report: CIO Edition 2026. The headline finding is stark. 81% of those CIOs say they have lost oversight of their own organization's AI agents. Not agents built by an outside vendor or a rogue contractor. Their own agents, built inside their own company, now operating beyond the view of the people nominally responsible for them.

Shadow Agents, Real Consequences

The reason is structural, not accidental. 84% of the CIOs surveyed said employees are building new AI agents faster than IT can govern them. Business teams no longer wait for a formal rollout; a workflow automation here, a customer-service agent there, each one stood up by someone outside the team responsible for tracking it. The result is an inventory problem before it is a technology problem. Aiven's agent library catalogs how these use cases typically start small and multiply, which is exactly the pattern the survey describes playing out at enterprise scale.

Why Measurement Keeps Failing

Visibility problems compound into measurement problems. 72% of CIOs said they cannot consistently measure the business outcomes their AI agents are producing, and 83% said their organization lacks standardized lifecycle management for agents after they go live. Without a lifecycle process, an agent that works well in its first month can run unmonitored by its sixth, and nobody owns the question of whether it's still doing what it was built to do. Executives weighing new AI investment might start by asking a more basic question than ROI: can my organization currently name every agent in production and what each one is measured against?

Cleanup Has Already Started

Some organizations are responding by pruning rather than expanding. 47% of CIOs surveyed said they have already decommissioned more than 20 agents so far this year, retiring tools that were built quickly, used briefly, and never folded into any governance process. Aiven examined this same agent-sprawl dynamic in an earlier Insight on enterprise agent inventories, and the pattern Dataiku found looks like its natural next chapter.

What CIOs Should Do Instead

The common thread across the findings is sequencing. Agents got built before anyone decided how they'd be tracked, measured, or eventually retired, and 76% of CIOs surveyed said their own role is at risk if the company can't show measurable AI returns by the end of 2027. That is a career-level incentive to fix the sequence: inventory before expansion, a defined owner and success metric before deployment, and a governance process that already answers who decommissions an agent and when. None of that requires slowing down on AI. It requires deciding, in writing, how agents get approved, measured, and retired before the next one goes live, which is the same starting point Aiven uses when it helps a leadership team assess where its AI governance actually stands today versus where the organization assumes it stands.

Key Takeaways

In a Harris Poll survey of 685 CIOs fielded for Dataiku between July 9 and July 29, 2026, 81% said they have lost oversight of the AI agents running inside their own organization, and 84% said employees are building new agents faster than IT can govern them.
72% of CIOs said they cannot consistently measure the business outcomes their AI agents actually produce, and 83% said their organization lacks standardized lifecycle management for agents once they are built and deployed.
47% of CIOs surveyed have already decommissioned more than 20 agents so far this year, and 76% believe their own role is at risk if the company cannot show measurable AI returns by the end of 2027.