When AI Agents Make the Call: Governing Autonomy at Scale

A conversation on The Agent Effect with the World Economic Forum on preparing enterprises for agentic AI.

Governing AI Agents in the Enterprise

AI agents are quickly moving from experimentation to real enterprise deployments. But many organizations are discovering that success in a pilot does not automatically translate into safe and reliable systems in production.

As agents begin making decisions, calling tools, and interacting with business systems, enterprises need new ways to evaluate performance, assess risk, and maintain oversight.

In the latest episode of The Agent Effect, Paul Jarratt speaks with Benjamin Larsen, Lead for AI Safety at the World Economic Forum, about what responsible deployment looks like as agentic systems scale across organizations.

Drawing on the Forum’s report AI Agents in Action, the conversation explores how enterprises can classify agentic systems, monitor their behavior, and design governance frameworks that enable innovation while managing risk.

From Experiments to Operational Systems

One of the key themes in the conversation is the shift from experimentation to operational use.

Many companies are currently piloting agents to automate tasks, support customer interactions, or coordinate workflows across internal tools. But deploying these systems in production introduces new questions. What level of autonomy should an agent have? What data and tools should it be authorized to access? And how can organizations trace the reasoning behind its decisions?

The discussion highlights why practices such as clear system classification, performance evaluation, and observability are becoming essential as agentic systems grow more capable.

Tune in: The Agent Effect

If you are a CIO, CTO, or AI leader exploring how AI agents could support enterprise workflows, this episode offers a practical perspective on moving from experimentation to production.

Listen to the full conversation with Benjamin Larsen on The Agent Effect.