How leaders can make agentic AI ready to run

Agentic AI is moving quickly, but measurable value depends on more than launching pilots. Leaders need an operating foundation that connects investment to real workflows, human oversight, reusable architecture and business outcomes.

In this two-part discussion moderated by Arun Shastri, a global leader in ZS’s AI practice, leaders from Choice Hotels, Microsoft and Vanguard share how their organizations are moving agentic AI into enterprise execution. Together, the conversations show what leaders need to build, measure and govern as AI agents begin to support complex work across teams, systems and customer journeys.

What agentic AI needs before production

The first part of the discussion examines the practical work behind agentic AI adoption, from task forces and workflow design to governance, oversight and control frameworks. The conversation shows why production-ready AI agents need more than promising use cases. They need clear accountability, integration across complex stakeholder environments and ongoing monitoring as models change.

Their approaches focus on:

• Building the foundation for agentic AI, from task forces to production
• Designing workflows across complex stakeholder environments
• Moving from single-task agents to coordinated, multistep experiences
• Governing adoption through human oversight and control frameworks
• Running agents in production with monitoring, model updates and evaluation

How leaders prove agentic AI is creating measurable value

The second turns from deployment to enterprise value. As agentic AI scales, leaders need to measure what’s working, decide where to keep investing and build reusable architectures that support multiple use cases across the business.

Their approaches focus on:

• Building portfolios of AI investments with visible aggregate impact
• Measuring return through disciplined testing tied to business outcomes
• Creating reusable, modular agent architectures across use cases
• Turning customer conversations and behavior into next best action decisions at scale
• Shifting from build mode to run mode as operating models evolve