The newest shift in AI use is the growing use of AI agents, i.e. AI systems that can independently take multi-step actions and workflows towards achieving a set goal, rather than simply responding to individuals prompts. AI agents act with a degree of independence, which creates a new risk level.
McKinsey’s 2026 AI Trust Maturity Survey reports that 62 % of organizations are at least experimenting with AI agents, although deployment at scale remains limited in the current situation. The situation is likely going to develop fast and may provide yet another unprecedented change – and most definitely a new leadership challenge.
AI agents bring the governance challenge into focus.
AI agents provide an excellent example of how different elements of good governance are tied to the use of AI within an organization. Consider an organization that introduces an AI agent to handle customer complaints. The AI agent can access customer records, classify a complaint, determine an appropriate response, update the customer-management system and, within predefined parameters, send the response to the customer without human intervention.
From a technology point of view, attention may focus on whether the system performs these tasks effectively and without technical errors. From a legal perspective, the organization will need to consider whether its use complies with the applicable regulatory framework. A governance perspective places the use of the agent within a wider organizational context.
Introducing an AI agent means that the organization is delegating a degree of authority to a technological system
Introducing such an agent means that the organization is delegating a degree of authority to a technological system. This requires clarity about the scope and limits of that authority, including which actions the agent may take independently, what information it may access, when is human intervention necessary and how are its performance and consequences monitored.
But the governance analysis should neither start nor stop there. The organization should understand why this particular process is being automated in the first place. The use of the agent should connect with strategic and operational objectives, whether these relate to improving customer service, increasing efficiency, reducing costs or freeing employees for tasks requiring greater judgement. Those objectives in turn affect investment decisions, staffing, process design, risk management and the measures used to assess whether the initiative is successful.
The decision to introduce an AI agent is therefore not simply an AI decision. It is simultaneously a strategic, operational, financial, human and governance decision.
The decision to introduce an AI agent is therefore not simply an AI decision. It is simultaneously a strategic, operational, financial, human and governance decision. Good governance helps ensure that these dimensions are considered together rather than as separate conversations.
The AI-agent example demonstrates why the core elements of good governance remain highly relevant as technology evolves. Good governance requires clarity about decision-making authority, about roles and responsibilities. Organizations define which decision may be delegated, to whom, within which limits and subject to what type of oversight. The same logic needs to apply when tasks or decision-making capacity are delegated to AI.
The greater the autonomy of the system and the potential impacts of its actions, the more important it becomes to establish appropriate boundaries and oversight. Determining where human judgement is necessary and where automated action is appropriate is the key. “Human in the loop” is useful only when the organization has considered what the human is expected to oversee, when intervention should occur and whether that person has the required information and authority to intervene effectively.
Governance structures and processes, transparency and accountability remain at the core.
Transparency is equally important. Those responsible for an AI-enabled process need sufficient information to understand how it operates and to challenge its use where necessary. Meaningful transparency does not require every board member or senior executive to understand the mathematical workings of an AI model, nor does it mean providing as much information as possible. It means ensuring that relevant information is understandable and available to those who need it.
In the context of AI-agents, this includes clarity about its purpose, the information and systems it can access, the parameters governing its actions, the decisions it is permitted to make and the circumstances in which a human must intervene. Transparency may also be required towards customers, members, employees, business partners, regulators or other stakeholders affected but the use of AI.
Accountability completes the picture. Technology can perform tasks and increasingly exercise operational autonomy, but it cannot assume organizational accountability. Responsibility must remain identifiable for approving the use of the system, defining its mandate and limits, monitoring its performance, addressing problems and ultimately answering for its consequences and holding accountability.
If responsibility becomes blurred as soon as AI is introduced into a process, AI has not created the underlying governance problem; it has simply exposed the one the humans, and lack of proper governance, created.