How enterprise leaders are scaling AI agents across their organization

_Dee Fitzgerald (CDO, Danone), Prem Natarajan (EVP, Chief Scientist, Capital One), Ratheesh Kamoor (Group VP, Head of Data and Analytics, Warner Bros. Discovery), Razal Minhas (VP, Data, Engineering and ML Platforms, Ford Credit), Murali Vridhachalam (VP, IT Head of Cloud, Data and AI, Gilead Sciences), and Arsalan Tavakoli (Co-founder and SVP of Field Engineering, Databricks) share executive insights in_Leading the AI-Ready Enterprise.

What does it take to turn AI ambition into measurable business outcomes? We sat down with AI-driven executives from leading brands to understand how they’re thinking about ROI and tangible value within their AI initiatives– while keeping governance front and center.

What emerged from the discussion was a shared tension: executives feel pressure to deploy agents quickly without compromising trust, governance, or cost control.

I've now come to believe that deployment is the first step in the AI stairway to heaven… And everything after that, the monitoring, the observability, the performance assessment, the continuous learning, those are the value-adding steps. — Prem Natarajan, EVP, Chief Scientist at Capital One

Leaders described a "moment of the possible" where technological advances are unleashing creativity and mobilizing teams across the enterprise. With AI now a CEO-level priority, organizations are moving beyond simple experiments to green-light impactful use cases, while rapid improvements in model accuracy are expanding the scope of what is deployable almost monthly. As agents orchestrate complex, multi-step workflows, companies are finding that rigorous governance is a foundation for innovation.

Their discussion revealed five practices any organization can adopt to scale AI agents responsibly and effectively:

Embed Unified Governance into Your AI Agent Strategy

Leaders emphasized that data and AI governance must be part of the agent lifecycle, not a post-hoc checkpoint.

Murali Vridhachalam, VP, IT Head of Cloud, Data and AI at Gilead Sciences _,_ shared that every agent undergoes a formal risk review:

Even before an agent gets developed, it has to go through a risk assessment. And depending on the risk levels, the proper approvals are obtained. The very important thing for us is: how is the risk framework integrated along with the user experience?

As part of a comprehensive enterprise governance strategy, some organizations are establishing governance councils. These councils help set the strategic direction and policies for topics like data ownership and accountability, compliance, data quality, risk, and more.

Ratheesh Kamoor, Group VP, Head of Data and Analytics at Warner Bros. Discovery, shared how his organization utilizes a specialized council to prevent employees from inadvertently pasting sensitive PII into AI tools, requiring a cross-functional "green light" from C-level, legal, and technical leaders for every use case. Because AI is fundamentally probabilistic, Razal Minhas, VP, Data, Engineering and ML Platforms at Ford Credit, stressed that governance cannot be a "one-time approval" but must involve continuous re-evaluation to ensure a model’s risk profile hasn't shifted due to external environmental factors.

Ultimately, this centralized oversight prevents what Arsalan Tavakoli-Shiraji, Co-founder and SVP of Field Engineering of Databricks _,_ calls a "proliferation" of conflicting metrics, anchoring your agents in "certified definitions" and standardized data rather than allowing them to operate on "six different versions" of the truth.

Manage Complex Workflows with AI Agents

A recurring theme among the leaders was the strategic shift toward orchestrating complex tasks through specialized agents. Instead of merely deconstructing work into simple parts, organizations are now focusing on driving high-level outcomes through a multi-agent framework that autonomously manages sophisticated, multi-step workflows across the enterprise.

With AI agents, we're going away from a single task-based approach to more orchestrated, outcome-based. For example, employee onboarding - there are multiple tasks… issuing a laptop or registering the employee in Workday. Now it's outcome-based onboarding an employee that is autonomously trying to execute tasks independently across different systems. — Murali Vridhachalam

Natarajan noted that the real benefits come when you can automate these tasks: “If you can bring in an AI model that's actually capable of taking care of a particular specialized task on its own... the possibilities are kind of endless when you look around and say, how many complex tasks can I factor into smaller accomplishable tasks, in which I can take a specialized AI model... and actually automate complex workflows?”

Create Dedicated Spaces for AI Experimentation

As teams expand their curiosity and usage of AI tools, there’s a growing need for careful sandboxes and controlled environments. These environments will be sanctioned spaces for teams to audit the performance of agents against legacy systems without risking live operations.

Razal Minhas of Ford Credit described how his organization runs "shadow capabilities where something's running in production. But… it's running silently in the background as a challenger."

This approach allows organizations to validate accuracy before an agent ever touches a customer workflow. By carving out the space for experimentation, leaders can encourage their workforce to test bold hypotheses and discover new value while keeping the "blast radius" of experimentation firmly contained.

Showcase Early Wins to Build AI Momentum

All executives agreed that adoption accelerates when early wins are concrete and repeatable.

One concrete example of this approach is from Capital One, where the team prioritized "Chat Concierge," a customer-facing tool for auto dealers. This application represents a "low risk but useful way" to validate agentic software in the real world.

This measured approach allows organizations like Capital One to both establish early wins and build the institutional confidence necessary for more complex applications. As Natarajan put it, seeing these tools in action "has unleashed creativity at a place where everybody's now an empiricist."

Equip Your Workforce to Work with AI Agents

Responsible deployment requires preparing employees to collaborate effectively with agents. Dee Fitzgerald, Chief Data Officer at Danone, shared insights into how 90,000+ employees, many of whom sit in the factory or on the front line, are transforming their work with AI: “We spend a lot of time training and up-skilling how to prompt.”

Natural-language interfaces inside the platform are key to enabling non-technical users to work with data and AI safely, without requiring SQL or Python expertise.

One unifying message across the roundtable: agentic AI only works when data, governance, orchestration, and compute live within a single, secure architecture. Leaders repeatedly pointed to the need for certified data products, consistent guardrails, and a platform that can deploy and monitor agents across diverse workflows.

See the full discussion to learn how leaders are operationalizing agents across HR, finance, supply chain, and creative workflows—and what steps your organization can take in the next 90 days to deploy agents responsibly and accelerate business impact