What Works, What Doesn’t - Lessons from CIOs

Leadership in the Agentic Era: When AI Becomes a Participant in the Enterprise

At CIONET’s discussions on ‘Leadership in the Agentic Era’, the conversation moved beyond the familiar question of how organisations should use AI. The more pressing question today is: what happens when AI systems no longer simply support human work, but instead operate alongside people – making decisions, initiating actions, and interacting with other systems on behalf of the enterprise. That shift changes the leadership agenda. In previous AI waves, the dominant concerns were productivity, experimentation and adoption. The current phase is different. Agentic AI introduces autonomy. It turns AI from a tool into an operational actor. For CIOs, that raises questions that are less about technology selection and more about control, accountability and trust.

TEXT: ROBBERT HOEFFNAGEL IMAGE: ENVATO

The tone of the discussion reflected this shift. There was curiosity about what agents can unlock, urgency about the speed at which capabilities are evolving, and a strong sense of responsibility about the risks that come with delegation. A leader from a creative business function described how experimentation with generative AI had started in the margins as Shadow AI, driven by curiosity and practical need. That early freedom helped teams discover useful applications quickly. But as the use of AI matured, the same organisation needed more structure, clearer rules and closer cooperation with legal and technology teams.

When AI stops assisting and starts acting

This tension ran through the discussion. Too much control can slow adoption and push innovation into the shadows. Too little control can expose sensitive data, create legal risk and allow unmonitored systems to make decisions no one fully understands. The leadership challenge is not to choose between speed and governance, but to design governance that enables responsible speed.

Several CIO-level perspectives pointed in the same direction: governance can no longer be a static policy document. In the agentic era, governance must become part of the operating environment. Employees should not simply be told what they cannot do. They need accessible, approved routes that make the right behaviour easier than the risky alternative. That requires AI literacy, clear boundaries, usable tooling and automated controls around data access, retention and monitoring.

Rethinking leadership in a world of autonomous systems

The discussion also exposed a deeper issue: the redistribution of control. When employees use AI assistants, control still largely sits with the human. When agents start completing workflows, connecting applications, accessing data and interacting with other agents, control becomes distributed across people, platforms and automated systems. A CIO at a legal services organisation noted that the enterprise is moving from prompt engineering towards “harnessing” applications and agents. With that approach, organisations may no longer only communicate with customers, suppliers or employees directly. Increasingly, their systems may communicate with the agents of those stakeholders.

“Move deliberately, build capability, clean up the foundations and prepare for a more autonomous operating model”

That prospect changes the meaning of trust. It is no longer enough to know whether a person has the right credentials. Organisations will need to know whether an agent is legitimate, what it is allowed to do, which data it can access and whether its actions can be traced. Agents become a new category of endpoints, with their own identity, permissions and risk profile.

Data emerged as the foundation for that trust. A technology leader warned that many organisations still underestimate the condition of their unstructured data: documents, emails, chats, shared folders and project sites. These are often overshared, outdated or poorly governed. In a traditional environment, that creates inefficiency. In an AI environment, it becomes a direct risk. An agent that can search, interpret and act on poorly managed data may surface information that should have been archived, expose confidential files or base decisions on obsolete policies.

When poor data becomes an enterprise risk

This is why the conversation repeatedly returned to data quality, data security and data governance. Without those foundations, agentic AI does not simply accelerate work; it accelerates confusion. And once employees lose confidence in AI outputs, rebuilding trust becomes difficult.

What made the event significant was not that CIOs discovered a new technology trend. It was that they collectively explored a shift in the nature of enterprise leadership. Leadership in the agentic era means preparing organisations for a world in which humans are no longer the only actors in business processes. It means deciding where autonomy is acceptable, where human oversight remains essential and how accountability is preserved when action is partly delegated to machines.

Who is accountable when machines act?

The mood was not defensive. There was broad recognition that inaction carries its own risk. A senior participant described the cost of not moving as one of the biggest failures organisations can make. At the same time, the room resisted easy optimism. The message was not “move fast and trust the technology.” It was: move deliberately, build capability, clean up the foundations and prepare for a more autonomous operating model.

That may be the defining leadership task for CIOs now. Not to block agentic AI. Not to unleash it blindly. But to create the conditions in which autonomous systems can operate within clear boundaries, with visible accountability and with enough trust for the business to use them confidently.

In the agentic era, leadership is no longer only about directing people and selecting platforms. It is about governing an ecosystem in which people, data, applications and autonomous agents increasingly act together. That is where the CIO’s role becomes more strategic than ever.