Enterprise AI is moving from experimentation toward autonomous workflows, but a new survey suggests the data infrastructure supporting that shift is struggling to keep pace. Early findings from The Modern Data Company’s third annual Modern Data Survey show that organizations are deploying AI agents faster than they are building trusted data, business context and governance around them.

The enterprise AI conversation is changing. Companies are no longer asking only whether generative AI can summarize documents, write code or assist analysts. Increasingly, they are giving AI agents access to enterprise data and systems and asking those agents to make decisions and execute work.

The infrastructure underneath those decisions may not be ready.

That is the central tension emerging from early findings in The Modern Data Company’s third annual Modern Data Survey. More than 540 qualified respondents across 66 countries have participated so far, and the company says the interim results reveal a widening gap between the speed of AI-agent adoption and the maturity of the data foundations supporting it.

According to the survey, 57.3% of respondents are piloting or running AI agents in data and analytics workflows. Of that group, 23.5% say their agents are already in production and 33.8% are running pilots. Another 15.6% expect to begin within six months.

Only 4% say they have no plans to use AI agents.

The numbers suggest agentic AI is moving quickly into enterprise operations. But the survey’s more consequential finding concerns trust: only 8.4% of respondents say their data is trustworthy enough for production AI.

That creates a potentially uncomfortable equation. An AI agent can act faster than a human analyst, but if the information guiding that action is incomplete, poorly defined or difficult to trace, automation can also accelerate the wrong decision.

The problem becomes more serious as agents move beyond recommendation systems.

Traditional analytics generally puts a human between the output and the action. An analyst reviews a dashboard, questions the underlying data and decides what to do. An AI agent can collapse those steps. It can interpret data, select an action and trigger a workflow.

That makes data quality, lineage, context and governance part of the AI system itself, rather than back-office concerns.

The survey supports that interpretation. Data quality and trust were among the top three barriers to putting AI agents into production for 75.9% of respondents. Missing context and lineage followed at 63.5%, while security concerns ranked at 61.7%.

Interestingly, skills shortages and immature tooling ranked substantially lower, at 25.9% and 19.5%, respectively.

That challenges a common explanation for slow enterprise AI deployment. The obstacle may not primarily be a shortage of AI engineers or better models. In many organizations, the harder problem is making existing information sufficiently reliable and understandable for machines to act on.

The context layer becomes an enterprise AI priority

One of the survey’s strongest findings concerns business context: the definitions, relationships, lineage and policies that allow an AI system to understand what enterprise data actually means.

Some 60.9% of respondents consider a reliable context layer essential for AI agents. Yet only 16% say they deliberately design and engineer that layer as a product, while roughly one-quarter have no formal context layer.

That gap matters because foundation models do not automatically understand an organization’s internal definitions.

A model can recognize the term “customer,” for example, but an enterprise still has to determine whether that means an account holder, a billing entity, an active user or a specific customer segment. The same issue applies to revenue, churn, inventory, risk and thousands of other business concepts.

This is where technologies such as data catalogs, semantic layers, knowledge graphs, metadata management and data lineage increasingly intersect with enterprise AI architecture.

The survey found that respondents would prioritize a better context layer over better tools by roughly six to one if given a single investment to improve their data and AI programs.

Organizations already running agents in production also appear to be further along in building these foundations. The Modern Data Company says they are nearly four times as likely to have intentionally developed a context layer and almost three times as likely to express confidence in the data underpinning their AI.

The company correctly cautions that these relationships show correlation rather than causation. The survey cannot establish whether agent adoption causes companies to build stronger data foundations or whether mature data architectures make agent deployment easier.

AI adoption is broad, but enterprise scaling remains difficult

The findings fit into a larger enterprise AI pattern.

McKinsey’s 2025 global survey found that 88% of respondents said their organizations regularly use AI in at least one business function, up from 78% a year earlier. Yet most organizations remained in experimentation or pilot phases, with only about one-third reporting that AI programs had begun scaling.

AI-agent adoption is similarly uneven. McKinsey found that 62% of respondents said their organizations were at least experimenting with AI agents, but only 23% reported scaling an agentic system somewhere in the enterprise. In individual business functions, no more than 10% reported scaling agents.

That distinction is important for technology leaders. Buying an AI platform is increasingly straightforward. Making the underlying enterprise information reliable enough to support autonomous decisions is considerably harder.

The Modern Data Company’s research also points toward another structural shift: data-platform consolidation.

Nearly 47% of respondents say their organizations are actively consolidating toward fewer platforms, while another 17.2% are evaluating consolidation. Yet almost 90% of organizations pursuing consolidation still retain at least one best-of-breed point solution.

That suggests the enterprise data stack is unlikely to become completely standardized. Companies may consolidate core infrastructure while retaining specialized tools for particular workloads.

Governance is becoming an AI infrastructure problem

Governance may ultimately be the most consequential gap.

The survey found that 65.1% of respondents believe AI-enabled decisions must be explainable, traceable and defensible. Yet only 39% maintain either an audit trail connecting AI inputs and outputs or a link from decisions back to their source data. Just 10% maintain both.

Accountability is similarly fragmented. Only 17.7% report having a clear, documented AI accountability framework.

For enterprise teams deploying agents, this changes the architecture conversation.

An AI system that merely generates a draft can often operate under relatively loose controls. An agent that can query financial records, modify customer information, approve transactions or trigger operational workflows requires much stronger boundaries.

That is why major technology ecosystems—including Microsoft, Google Cloud, Amazon Web Services, Salesforce and NVIDIA—are increasingly competing not only on models and compute, but on the infrastructure surrounding enterprise AI.

The market is moving toward systems in which models, data, identity, permissions, observability and governance work together.

The Modern Data Company’s interim findings reinforce a simple conclusion: enterprise AI cannot scale independently of enterprise data maturity.

For CIOs and chief data officers, the priority may therefore be shifting from adding another AI tool to building a trusted environment in which agents can safely operate.

The companies that solve that problem will have an advantage as AI moves from answering questions to taking action.

Market Landscape

The enterprise AI stack is evolving from a model-centric architecture toward a data-and-agent architecture.

Foundation models from providers such as OpenAI, Google, Anthropic and Amazon provide increasingly capable reasoning engines. Cloud platforms from Microsoft, AWS and Google Cloud provide compute, identity and orchestration. Data platforms and governance vendors provide the context agents need to interpret enterprise information.

The bottleneck is increasingly at the intersection.

McKinsey’s latest research shows that AI adoption is widespread, but enterprise-scale deployment remains limited. Only 7% of respondents in its 2025 research said AI had been fully scaled across their organizations.

That helps explain why data quality, lineage and governance are becoming strategic AI concerns. As agents move from copilots to autonomous workflow participants, enterprises need systems that can answer four questions reliably:

1. What does this data mean?
2. Where did it come from?
3. Can the agent use it?
4. Who is accountable for the resulting decision?

Organizations that cannot answer those questions may still deploy AI, but scaling autonomous systems safely will be much harder.

Top Insights

• More than half of survey respondents are piloting or running AI agents, showing enterprise adoption accelerating even as trusted production data remains surprisingly scarce.
• Data quality and trust outrank skills shortages as barriers to agent deployment, shifting enterprise AI priorities toward governance, lineage, context and reliable data infrastructure.
• A reliable business context layer is emerging as critical AI infrastructure, helping agents understand enterprise definitions, relationships, policies and the meaning behind raw data.
• Organizations running agents in production report stronger data foundations, although the survey shows correlation rather than proving that mature context directly causes successful agent deployment.
• Enterprise consolidation is advancing while best-of-breed tools remain common, suggesting future AI stacks will combine centralized data platforms with specialized capabilities and governance layers.

Power Tomorrow’s Intelligence — Build It with TechEdgeAI