Supervision of AI Agents: New Roles and Governance Challenges

The role of those who oversee agents

AI agents can access internal systems, consult data, execute process steps, and make decisions without human approval for each action. While this autonomy enhances technology’s utility, it raises a critical question: who monitors what the agent is doing? The answer is emerging through dedicated oversight roles. Responsibility may fall to AI governance, security, operations, product professionals, or managers who now include agent supervision as part of their duties.

This role goes beyond verifying the correctness of the final answer. Those monitoring an agent must define what it can do, what information it can access, and when it should stop and call a human.

Consider an agent responsible for handling customer requests. It can review an account’s history and offer a solution within predefined limits. If a situation arises outside these rules, the system must know not to continue acting independently. It is the responsible team’s job to set these limits and ensure they are being followed.

Analyzing the logs generated by the agent is also part of the routine. When a task starts producing repeated errors, someone must identify the pattern, determine the cause, and adjust the rules, permissions, or instructions used.

Gartner recommends different control levels based on the degree of autonomy. Agents that only consult information may require simpler controls, while those capable of executing actions independently need continuous monitoring, mechanisms to halt operations, and clearly defined responsibilities.

Why responsibility is changing

The growth of AI agents is making this role increasingly important. A KPMG survey released in April 2026 found that 54% of organizations already implemented AI agents, up from 12% in 2024.

The survey also revealed that 63% required human validation of agents’ responses, compared to 22% in the first quarter of 2025.

Deloitte identified a similar gap in control structures: only 21% of technology and business leaders surveyed across 24 countries reported having mature governance for AI agents.

The identified gaps were precisely elements such as clear limits of action, real-time monitoring, and records of actions performed.

This helps explain why there is growing demand for professionals dedicated to this area. As more systems operate without constant human intervention, it becomes increasingly important to have someone responsible for verifying whether they are acting within the limits defined by the company.

A role that blends technology and business

The professional does not necessarily need to be a programmer. Depending on the organization, the role may require expertise in technology, security, processes, risk management, and the business itself.

The World Economic Forum, in a 2026 guide, advocates that agents should have defined authorization, traceable actions, and accountability mechanisms throughout their usage cycle.

For those working with AI, a new field is emerging that combines technical knowledge with the ability to evaluate processes and make decisions about when autonomy should end.