AI Agents Hit Production Snags as Human Oversight Costs Mount

Organizations racing to deploy AI agents are discovering an uncomfortable reality: automation still requires substantial human intervention. A new survey reveals that more than three-quarters of decision-makers report their companies hit roadblocks moving AI agents from pilot projects into full production over the past year.

The findings come from a Harris Poll sponsored by Collibra that surveyed 306 U.S. data management, privacy, and AI decision-makers. While nine in ten said their organizations are actively deploying autonomous agents, more than half reported that employees spend significant time reviewing and correcting agent outputs before they go live.

The hidden cost of agent deployment

At larger companies with annual revenue of $100 million or more, the oversight burden grows heavier. Sixty-four percent of respondents at these organizations reported significant employee time spent reviewing and correcting agent outputs, compared to 51% across all survey participants.

Collibra CEO Felix Van de Maele calls this the "hallucination tax" — a hidden cost of manual oversight, rework, and risk that scales with every new agent deployed. The term captures a frustrating irony: work delegated to an agent often circles back to an employee's desk for corrections.

The survey did not quantify how many hours this oversight requires, but the pattern raises a fundamental question for companies expanding their agent deployments. Human review makes sense when an agent reduces a multi-hour task to a few minutes of checking. The economics shift when employees must redo substantial portions of the work.

Why it matters

As enterprises invest heavily in AI agents, understanding the true cost of deployment becomes critical. The gap between pilot success and production readiness isn't just technical — it reflects deeper organizational challenges around data quality and governance that money alone won't solve. Companies need to measure not just how many tasks agents complete, but how much work remains for the humans who receive those results.

Data quality emerges as the core problem

Seventy-two percent of respondents agreed that poor or unaligned data almost always underlies disappointing AI results. An overwhelming 87% said their teams regularly verify whether the information agents use remains accurate and current.

Yet fixing data problems requires more than funding. Separate research from Gartner found that among 223 data and analytics leaders surveyed in March, 60% cited cultural resistance as a reason governance initiatives fail. Only 40% pointed to funding constraints. Weak business engagement and limited understanding of governance value create obstacles that persist even when budgets aren't tight.

Keeping records current and agreeing on standard definitions demands time from people with competing priorities. The benefits often seem abstract until an application starts producing answers nobody trusts.

Organizational shifts in response

Companies appear to be restructuring in response to these challenges. Fifty-three percent of respondents said their AI function's reporting line had moved closer to the primary data organization over the past year. At companies with at least $100 million in revenue, that figure reached 62%.

Bringing teams that build and deploy agents closer to those who own and maintain data could create more direct channels for raising and resolving problems. Whether this organizational change actually improves agent performance remains an open question. The more telling metric will be whether recurring errors get resolved and employees spend less time making corrections.

The findings were first reported by HPC Wire based on the Collibra-sponsored Harris Poll survey.