Enterprise adoption of autonomous AI agents is advancing, but moving those systems from pilots into production remains difficult. A new Harris Poll survey commissioned by Collibra found that nine in 10 surveyed organizations are actively deploying autonomous AI agents, while 76% reported critical roadblocks when attempting to move AI-agent initiatives into full production during the past year. Respondents increasingly point to data quality, context and governance as central barriers.
The enterprise AI market has entered a new phase. Companies are no longer simply experimenting with generative AI chatbots; many are testing autonomous agents that can make decisions, invoke tools and execute multi-step workflows.
But deploying an agent successfully in a pilot environment does not necessarily mean an organization is ready to let it operate at production scale.
That gap is the focus of Collibra’s 2026 Hallucination Tax Report, based on a Harris Poll survey of 306 U.S. adults who work full time in data management, privacy or AI decision-making roles at director level or above.
The survey found that 90% of respondents said their organizations are actively deploying autonomous AI agents. At the same time, 76% reported encountering critical obstacles while trying to move such initiatives from pilot to full production over the previous 12 months.
The findings suggest that the challenge is increasingly shifting away from simply building capable AI models and toward establishing the data and operational infrastructure required to make autonomous systems dependable.
Among respondents, 72% said that when their organization’s AI initiatives fall short, the root cause almost always comes back to an unaligned or poor data foundation.
That relationship is particularly important for AI agents. Unlike a conventional software application operating on predictable inputs, an autonomous agent may retrieve information, use enterprise data, call external tools and make decisions across multiple steps. If the context supplied to the agent is incomplete, outdated or poorly governed, errors can propagate through the workflow.
The survey found that 87% of decision-makers said their teams regularly re-verify whether the context available to their agents remains accurate and current.
That manual checking creates another problem: the more autonomous a system is supposed to become, the less sustainable human-by-human validation becomes.
More than half of respondents, 51%, said their organizations spend significant staff time manually reviewing and correcting autonomous AI-agent outputs before deployment. Among organizations generating at least $100 million in revenue, that figure rises to 64%.
The larger-enterprise findings also point to the importance of data infrastructure. In that group, 96% of respondents said AI project failures almost always trace back to poor data foundations.
Collibra argues that these challenges require enterprises to bring AI governance and data management closer together. The survey found that 53% of respondents said the reporting line for their AI function had moved closer to the primary data organization during the previous year. That figure reached 62% among organizations with revenue of $100 million or more.
Accountability is also becoming an organizational requirement. According to the survey, 84% of respondents said their organizations have clearly defined executive accountability when autonomous agents generate flawed or harmful outputs.
This matters as enterprises move toward agentic systems capable of taking actions rather than simply generating recommendations. Governance therefore needs to cover not only the model but also the data sources, context, permissions, actions and decisions surrounding the model.
Regulation is another driver. 90% of surveyed decision-makers said their organizations are actively preparing for changing AI regulations across federal, state and international jurisdictions.
The most commonly reported preparation measures included establishing clear internal accountability for AI outputs and decisions, cited by 58%, and investing in data lineage and documentation, cited by 51%.
Collibra’s own business focuses on data intelligence and governance, so the findings naturally support the company’s positioning around data context and AI controls. The survey should therefore be read as vendor-sponsored research rather than a neutral industry census.
Its methodology also matters. The study surveyed 306 U.S. decision-makers between August 5 and 11, 2026, and reports a Bayesian credible interval of approximately ±6.4 percentage points at a 95% confidence level. The population does not represent all employees using AI, and the results should not automatically be generalized to organizations outside the United States or to companies without dedicated data, privacy or AI leadership.
Nevertheless, the findings highlight a practical issue facing enterprise AI deployments: autonomous agents require more than increasingly capable foundation models.
They need reliable enterprise context, controlled access to data, traceable decisions, monitoring and clearly assigned responsibility. As organizations attempt to scale agentic AI beyond experiments, those operational controls are becoming part of the AI infrastructure itself.
Market Landscape
The enterprise AI stack is expanding beyond foundation models and inference infrastructure to include AI governance, data lineage, retrieval, observability, identity and runtime controls.
Agentic AI makes these layers more important because agents can retrieve information and execute actions across multiple systems. Enterprises therefore need to know what data an agent accessed, whether that information was current, which policies governed its behavior and who is accountable for its decisions.
Vendors across the AI ecosystem are responding differently. Data-governance companies are adding AI controls, cloud providers are developing agent-management frameworks, and model providers are introducing enterprise security and administration capabilities.
The competitive question is increasingly how organizations can provide agents with enough context to be useful while maintaining sufficient controls to make their actions auditable and predictable.
Top Insights
- Collibra’s Harris Poll survey found that 76% of surveyed AI decision-makers encountered critical barriers moving autonomous agents from pilots into production.
- 72% said AI initiative failures almost always trace back to an unaligned or poor data foundation.
- 87% reported regularly re-verifying whether agent context remains accurate and current, highlighting the operational cost of manual oversight.
- Data and AI functions are increasingly being aligned organizationally, with 53% reporting that AI reporting lines moved closer to data organizations.
- The findings underscore that production agentic AI requires data lineage, runtime governance, accountability and reliable context alongside capable models.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
