As organizations accelerate their AI ambitions, many are discovering that the challenge is no longer simply building models or launching pilots. The harder problem is ensuring that AI systems can consistently operate on trusted, governed, and reliable data foundations.

That challenge is becoming even more urgent as enterprises move toward agentic AI, where autonomous systems increasingly access, interpret, and act on enterprise data without constant human oversight. In that environment, traditional approaches to data quality, governance, and trust are being pushed into unfamiliar territory.

In this first installment of a three-part interview series, Barr Moses, CEO of Monte Carlo, speaks with seasoned data leader Justin Heller about how the rise of AI agents is reshaping enterprise data strategy, why trusted data is becoming central to AI adoption, and where organizations are struggling to move from experimentation to measurable business value.

Moses also discusses the widening gap between AI “builders” and leadership teams, the growing governance complexity surrounding agentic systems, and why enterprises cannot afford to spend years perfecting data foundations before beginning AI experimentation.

The Shift From Human-First to Agent-First Data Consumption

Reflecting on how the market has evolved since Monte Carlo was founded roughly seven years ago, Moses says the core mission has remained remarkably consistent. “Data and AI teams are in the business of trust,” she says.

Monte Carlo was founded with the mission of helping enterprises adopt trusted AI, particularly by enabling organizations to deliver reliable insights, analytics, machine learning systems, and increasingly, AI-driven products and agents.

According to Moses, the scale and urgency of that challenge have changed dramatically as AI adoption accelerates across industries.

“A favorite part of my job is that I get to work with over 500 or so enterprises across different industries,” she says, pointing to organizations across retail, finance, healthcare, and transportation sectors, including companies such as Cisco, Rivian, and Roche.

The biggest transformation, however, is not simply the amount of data organizations manage. It is who, or what, is consuming it.

Moses explains that during the rise of cloud data platforms roughly a decade ago, enterprise data became accessible to far more people beyond traditional IT teams. Analysts, marketers, sales teams, governance professionals, and operational leaders increasingly relied on structured enterprise data for decision-making.

Today, AI agents are becoming a new category of data consumers.

“We have a new player, not just humans, but agents are going to access data now,” Moses says.

That fundamentally changes how enterprises must think about trust and governance.

“In the past, I could tap on your shoulder and say, ‘Hey, which table should I be using for this analysis?’” she explains. “When we live in a world where agents are accessing our data warehouse or lakehouse, they can’t be tapping on our shoulders.”

According to Moses, this transition toward agent-first data consumption will reshape enterprise governance models over the next several years.

“How we think about data quality and data trust has to change dramatically for that audience,” she adds.

Why unstructured data suddenly matters more

The rise of generative AI is also forcing organizations to rethink the value of their unstructured data estates.

For years, enterprises accumulated enormous volumes of PDFs, documents, images, emails, clinical records, claims files, and other forms of unstructured information. While much of that data held potential value, organizations lacked practical ways to operationalize it at scale.

That limitation is rapidly disappearing.

“The difference is that now, with AI, we can actually make better use of the unstructured data,” Moses explains.

AI systems can now process and synthesize large volumes of unstructured information at speeds and scales that were previously unrealistic for enterprises.

However, Moses emphasizes that the growing accessibility of unstructured data also increases governance complexity significantly.

Among the key concerns organizations must now manage are:

• Whether AI systems are pulling from approved and vetted data sources
• Whether the underlying information is current and accurate
• Whether sensitive or regulated data is properly protected
• Whether AI-generated outputs are based on reliable reasoning paths
• Whether both structured and unstructured datasets are properly governed for human and agent consumption

The growing gap between AI experimentation and business value

Further, Heller also raises a key question: Should organizations start AI initiatives by limiting agents to structured systems of record, or should they first clean up large volumes of unstructured data before deploying agentic AI?

Moses argues the reality is more nuanced. Monte Carlo recently conducted what she describes as a “builders and leaders” survey examining how enterprises are approaching generative AI adoption. One of the most significant findings was a widening perception gap between executives and practitioners.

“Leaders actually have a lot of confidence in how to go about this and think that organizations are doing well,” Moses says. “Builders are dramatically less so.”

According to Moses, many enterprises are still struggling to distinguish between successful experimentation and sustainable business impact.

Still, she believes organizations should not become paralyzed waiting for perfect governance or perfectly curated data environments before experimenting with AI.

“It does have merit to build things for the sake of experimenting and upskilling your team,” Moses says.

She encourages both leaders and technical teams to actively experiment with generative AI tools to better understand their capabilities and limitations.

“That’s going slow to go fast,” she adds.

Moses notes that even her own experimentation with AI workflows has changed how she understands what is possible operationally. “I’ve been able to build things that I didn’t even think were possible.”

Why organizations cannot spend years preparing for AI

One of the biggest risks Moses sees is organizations attempting to delay AI adoption until their data environments are fully modernized.

Instead, she recommends organizations identify trusted “gold layers” or curated domains of enterprise data that can safely support early AI initiatives while broader modernization efforts continue in parallel.

That balanced approach allows enterprises to continue learning, experimenting, and building internal capabilities without waiting for complete transformation programs to conclude.

Agent governance is becoming as important as data governance

As enterprises scale AI agents into operational workflows, Moses believes governance challenges are becoming substantially more complicated than traditional analytics governance.

She illustrates this with an example from a large airline organization that developed a customer-facing flight recommendation chatbot.

The AI system was designed to retrieve customer preference data before generating recommendations. However, the agent occasionally reversed the sequence of operations, generating recommendations before accessing the customer profile.

The result was technically functional AI behavior, but operationally flawed outcomes.

According to Moses, this highlights a broader governance challenge emerging in agentic AI systems:

• Agents must access trusted and authorized data
• The reasoning path itself must also be governed and observable
• Organizations need visibility into how agents select data and execute decisions
• Governance models must account for autonomous workflows, not just static datasets

For enterprise data leaders, the complexity increases because AI development is no longer isolated to centralized data teams.

“Every part of the business oftentimes is either building agents or will be building agents,” Moses says.

That shift is forcing organizations to rethink governance at both the data and AI layers simultaneously.

“Best-in-class organizations are building toward a world where we have data and agent governance in one,” she concludes.

CDO magazine appreciates Barr Moses for sharing her insights with our global community.