Survey: 57% of Data Teams Run AI Agents, but Only 8% Trust Their Data

More than half of enterprise data teams are now piloting or running AI agents, but fewer than one in ten say the data behind those agents is trustworthy enough for production. That is the headline from interim results of The Modern Data Company's annual survey of data leaders, reported by IT Brief this week. For CIOs and chief data officers, the numbers put a figure on a problem many already feel: agent adoption is moving faster than the data foundations those agents depend on.

What did the Modern Data survey find?

The survey drew more than 540 responses from data leaders and practitioners in 66 countries, and the findings released this week are interim rather than final, according to IT Brief. Among the main results:

• 57.3% of organizations are piloting or running AI agents in data and analytics workflows, with 23.5% in production and 33.8% in pilots.
• Only 8.4% say the data feeding their AI systems is trustworthy enough for production use.
• Even among companies already running agents in production, just 21.7% are very confident in that data.
• 75.9% name data quality and trust among their top three barriers to production, ahead of missing context and lineage (63.5%) and security concerns (61.7%).

Skills and tooling ranked well below those, cited by 25.9% and 19.5% of respondents. In other words, most teams do not think they lack the tools or talent to build agents. They lack data they are willing to let agents act on.

Saurabh Gupta, president and CEO of The Modern Data Company, summed up the first half of that picture: "Enterprises have proven they can put AI agents to work."

Why does a context layer matter for AI agents?

The survey's most striking finding concerns what it calls the context layer, the metadata, definitions, lineage and business rules that tell an AI system what the data means. About 61% of respondents said a reliable context layer is necessary for agents, yet only 16% said their organization deliberately designs and engineers that layer as a product. One in four has no formal context layer at all. Asked to choose, respondents favored a better context layer over better tools by roughly six to one.

Organizations running agents in production were nearly four times as likely to have built their context layer intentionally, and almost three times as likely to trust the data underneath their AI, IT Brief reported. The survey notes these are correlations, not proof of cause.

The finding echoes a point made repeatedly at AI Connect events. Human analysts catch numbers that look wrong. Software agents generally don't. "Agents don't do that. They take the data at the face value and act," Sam Mittal of Snowflake's partner solutions engineering team said during a AI Connect webinar on why agents need metadata, not just clean data. A separate expert panel examined what belongs in the enterprise context layer, from semantics to memory.

Who is accountable when an AI agent gets it wrong?

Governance results were weaker still. Nearly two-thirds of respondents (65.1%) said AI-enabled decisions must be explainable, traceable and defensible. But only 39% keep either an audit trail of AI inputs and outputs or a link from decisions back to source data, and just 10% keep both.

Accountability is often unclear. Only 17.7% have a clear, documented AI accountability framework. A quarter describe accountability as shared but unclear, 19.1% call it poorly defined and 11% say no one is formally accountable.

The survey also found 47% of organizations actively consolidating toward fewer data platforms, with another 17.2% evaluating it. Nearly 90% of those consolidating still rely on a best-of-breed point tool somewhere in their stack.

What should enterprise data leaders do now?

The data points toward a practical sequence. Before adding agents, teams can check whether the data those agents will touch has documented definitions, lineage and an owner. Monitoring also needs to cover the data, not only the model. "If you don't have good data, you're never going to have good agents," Ethan Post, principal FDE for agent observability at Monte Carlo, said in a AI Connect webinar on early warning signals for production agents.

Accountability is the cheapest gap to close. Naming an owner for each production agent, and keeping a record that links its outputs to source data, costs far less than a platform migration and answers the question regulators and auditors are likely to ask first.

The survey is a vendor study and still interim, so the exact percentages may shift in the final report. But the direction matches what many enterprises report: as past COLLIDE sessions on the AI production gap argued, data rather than models remains the main thing holding agents back from production.

Frequently asked questions

What share of companies trust their data for AI agents?

Only 8.4% of respondents to The Modern Data Company's interim survey said the data feeding their AI systems is trustworthy enough for production.

What is a context layer for AI agents?

It is the metadata, definitions, lineage and business rules that tell an AI system what enterprise data means and how it can be used.

What is the biggest barrier to putting AI agents into production?

Data quality and trust, named among the top three barriers by 75.9% of respondents, followed by missing context and lineage and security concerns.

Filed under: Agentic AI, Observability & Data Quality, Context, Semantics & RAG