As enterprises continue advancing beyond early AI experimentation, the challenge is increasingly shifting from whether AI can create value to how organizations can deploy it responsibly at scale. Questions around autonomy, human oversight, and risk are becoming more immediate as AI systems begin moving deeper into enterprise workflows.

In this second installment of a three-part interview series, Barr Moses, CEO of Monte Carlo, continues her conversation with Justin Heller, CDO Magazine Editorial Board Member, discussing how organizations are approaching the balance between human intervention and autonomous AI systems. She also shares the three recurring priorities she is seeing among Chief Data Officers (CDOs) as enterprises move from experimentation toward operational AI deployment.

Part 1 focused on 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.

Finding the right balance between autonomy and oversight

As Heller reflects on organizations increasingly beginning with a “human-in-the-loop” approach for AI deployments, he raises a broader question: How should enterprises determine where human oversight remains essential and where autonomous systems can operate with more freedom?

Drawing on conversations with data leaders across industries, Moses explains that organizations are currently operating at different stages of maturity, but common patterns are beginning to emerge.

Spending time regularly with CDOs has allowed her to identify what she describes as a continuum of priorities across organizations. While approaches vary, she says most enterprises are largely concentrating on three major areas:

1. Preparing AI-ready data foundations
2. Improving internal productivity through AI-driven automation
3. Building AI agents that directly support business functions

Collectively, these priorities are helping organizations determine where they can move faster and where stronger controls may still be necessary.

AI-ready data starts with strong foundations

The first challenge Moses identifies centers on preparing enterprise data environments for AI adoption.

For many organizations, that work starts with fundamentals rather than sophisticated AI use cases. Enterprises are focusing on understanding both structured and unstructured data environments, moving workloads to cloud platforms, establishing semantic layers, and ensuring data remains accurate, current, and reliable.

According to Moses, many teams are also increasingly automating foundational capabilities such as:

• Security
• Governance
• Observability
• Data quality
• Monitoring and controls

She describes this stage as getting the organization’s “house in order” before moving aggressively into AI deployment.

Notably, Moses says this pattern extends across industries, including organizations that might otherwise be expected to move quickly because of lower regulatory constraints.

Rather than being driven solely by compliance concerns, many organizations are simply recognizing that trusted AI begins with trusted data.

Why data teams are becoming their own first AI users

The second pain point, according to Moses, involves CDOs being under pressure to identify their own teams. Elaborating, she says, data teams often spend significant time on manual activities, including identifying important datasets, building pipelines, configuring monitoring systems, and troubleshooting incidents.

Moses notes that many of these activities can now be accelerated through purpose-built AI agents.

“All those things can actually be done much faster and better with purpose-built agents for those teams,” she says, and then illustrates this through the example of a data incident.

Traditionally, when data problems emerge, teams often spend days developing and testing multiple hypotheses. Teams may investigate whether a schema changed, whether an ETL process failed, or whether a third-party data source introduced errors.

The process can involve significant manual investigation before teams identify the underlying issue. “The process can be condensed to just a few minutes when you have agents and sub-agents coming up with those hypotheses, recursively looking at the issue,” she explains.

For Moses, this represents one of the strongest near-term opportunities for enterprise AI. Beyond efficiency gains, she says this also creates an opportunity for teams to gain experience working with AI systems in production environments before expanding those capabilities elsewhere across the business.

Business-facing AI agents create greater opportunity and greater risk

The third area Moses highlights involves CDOs building AI agents themselves that directly support broader business functions.

These may include agents designed to answer business questions for financial teams, support product organizations with analytics insights, or serve other operational functions across the enterprise.

While these use cases often offer stronger business value, they also introduce more governance complexity because they expand access to enterprise information across larger groups of users.

“I think the ROI to a certain degree is higher because it’s tied to the revenue and accelerating priorities of other functions across the business,” Moses says.

However, because these deployments involve broader access to enterprise data across more teams and stakeholders, they introduce greater risk, placing added responsibility on CDOs to ensure AI agents remain secure, reliable, and trustworthy.

According to Moses, two concerns repeatedly emerge as organizations move into these deployments:

• Are the agents secure?
• Are they producing reliable and trusted responses?

Failures in either area can quickly stall progress.

“POCs or pilots often get killed because of security or wrong answers,” she concludes.

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