Gartner’s latest research on agentic AI for CHROs lands on a sobering point: managers are about to become accountable for the behaviour of autonomous agents, at a moment when 75% of CHROs already report that their managers are more overwhelmed than ever.
This is not a technology problem. It is an organisational design problem. Agentic AI changes what a manager oversees, how they intervene and what they are responsible for. Most organisations have not updated their management frameworks to match.
From managing people to managing people-plus-agents
For decades, span of control meant the number of direct reports a manager could effectively supervise. Agentic AI expands that definition. A manager may soon oversee a team of humans plus a set of agents that execute tasks, make recommendations and interact with customers or systems.
That changes the nature of oversight:
- Accountability without visibility. A manager may be held responsible for an agent’s decision even if they cannot easily inspect how it was made.
- Intervention at speed. Agents act quickly. By the time a manager notices a problem, the agent may have already produced customer-facing output or committed a transaction.
- Skill mismatch. Many managers are being asked to supervise systems they did not build, do not fully understand and were not trained to evaluate.
Gartner’s framing is important because it places the burden on the manager rather than treating the agent as a self-contained IT asset. That is the right place for accountability to sit, but only if the organisation gives managers the tools to discharge it.
Why this is happening now
Agentic AI is moving from experiment to operational deployment. Where earlier AI tools assisted individual contributors, agentic systems take action. They schedule meetings, process claims, respond to customers, triage support tickets and execute workflows.
Each of those actions used to sit within a human’s discretion. As agents take them on, the manager’s role shifts from direct supervision of tasks to supervision of boundaries: what the agent may do, when it must escalate and how its outputs are checked.
At the same time, manager workload is already high. Remote and hybrid work, rapid change and tighter margins have stretched many middle-management layers thin. Adding agent oversight without removing something else is a recipe for burnout and blame.
A practical playbook for managers
Clarify the accountability model. Be explicit about which agents a manager owns and what they are accountable for. Is it output quality? Customer satisfaction? Compliance? Cost? Without clarity, accountability becomes a trap.
Make escalation paths simple. Agents should know when to stop and ask. Managers should know what to do when asked. A clear escalation protocol protects both the manager and the customer.
Provide agent literacy, not just tool training. Managers do not need to become machine-learning engineers, but they do need to understand how agents make decisions, where they fail and how to challenge their outputs.
Review agent performance like human performance. Set metrics, review them regularly and hold retrospectives when things go wrong. An agent that is never reviewed will drift.
Protect manager capacity. If agent oversight is a new responsibility, something else has to come off the plate. Organisations that simply add agent ownership to existing roles will see quality suffer in both domains.
The broader implication
Agentic AI will not eliminate management. It will redefine it. The firms that succeed will be the ones that update their management model before the agents arrive in force. That means new role definitions, new training, new metrics and a shared understanding that a manager’s span of control now includes systems as well as people.
Gartner’s warning is timely. The window for getting this right is closing.