AI is unbundling some parts of management while expanding others. The manager role is being reassembled around a different mix of responsibilities.

For decades, organizations have bundled a remarkable amount of work into one position called “manager”: gathering information, coordinating people, reviewing work, tracking progress, translating strategy, allocating resources, coaching employees, communicating upward, making decisions and carrying accountability. That bundle is starting to change.

At Edward Jones, CIO Kevin Adams has moved part of managerial memory and follow-through into an AI system that tracks context across enterprise programs and committees. At CVS, Alan Rosa described an operating process where agents now prepare the reports and dashboards that once consumed much of a two-hour daily management meeting. The meeting has fallen to roughly 15 minutes and is now largely about exceptions, problems and decisions.

In many companies, AI is helping reduce the need for layers whose primary role was to collect, repackage and transmit information. At the same time, the definition of management is expanding. Broadridge CTO Tyler Derr is already considering how to balance human teams with digital workers

Taken together, these examples point to two changes happening at once: Some traditional management work is moving out of the manager role, while entirely new management responsibilities are being added around AI and agents.The future manager may therefore be neither simply smaller nor simply more AI-enabled. The role itself is being reassembled.

Some management work is becoming infrastructure

The CVS example is useful because the change is not simply that a manager does the same work faster. The operating process itself changed. Agents run reports overnight, aggregate the results and prepare the dashboards before the team arrives. The daily stability meeting no longer has to serve as the mechanism for constructing a shared picture of what is happening. That picture increasingly already exists. The meeting can focus on the smaller set of issues that still need intervention.

Historically, a lot of management work existed because someone had to assemble and move information. These were activities somebody had to find time to perform. Increasingly, they can become persistent systems. At Edward Jones, CIO Kevin Adams has created such an AI system. Rather than using AI as a series of isolated prompts, he continuously builds context within each environment by adding relevant documents, decisions, history, open issues and commitments over time. The accumulated institutional context allows the system to surface connections, preserve continuity and prompt follow-through across work that would otherwise depend heavily on individual memory. A colleague described it as his “chief of staff.” Adams puts the boundary this way: “I don’t let the machine tell me what to do. I let the machine prompt me about what I should be doing.”

What interests us is less the metaphor than the organizational implication. Something that once lived almost entirely inside the manager including memory, continuity and follow-through can now partly live in the system around the manager. The same is increasingly true for reporting, preparation, monitoring and recurring operating rhythms. Those responsibilities do not necessarily disappear. They stop requiring the same amount of active managerial labor. Parts of management are becoming infrastructure.

That creates a second question: what happens to the capacity?

This is where the dialog elevates beyond simple productivity. If AI removes administrative work, what happens to the capacity it creates?

Pushpendu Pal, Chief Digital and Technology Officer at CVS Health, described spending less time on budget administration, project-status reviews and implementation reporting. But he has not simply ended up with a shorter day. He is spending more time coaching people, working directly with engineers and developers, developing organizational capabilities and generating ideas for service and business improvement. AI is assisting in creating more time for leaders to perform their primary task more effectively, like coaching and mentoring, being consultative with business stakeholders and working with engineers in sculpting the improved services. 

This demonstrates how removing management work and redesigning management are not the same thing. If AI saves a manager five hours and the organization simply fills those hours with more projects, more meetings or a larger workload, then AI improved capacity.  But the manager role itself did not necessarily change.

A real redesign requires asking what responsibilities should grow as others shrink. That question will differ by organization. But it has to be answered deliberately. Otherwise, the role gets compressed without being redesigned.

AI may also make certain management layers harder to justify

The changes at CVS and Edward Jones raise an organizational question as well: Which management layers still perform a distinct function when information, context and status updates are available through shared systems? That does not mean AI automatically creates flatter organizations. It means some layers need a clearer reason to exist. If a management layer primarily exists to collect information, summarize it, translate it, carry it upward, and distribute it downward, then AI makes that role easier to question.

The useful org design test is no longer simply “How many people report to this manager?” It becomes: What distinct management responsibility resides at this layer? If the answer is unclear, the layer may reflect an information architecture that AI is changing.

At the same time, management is expanding to include digital labor

While some traditional responsibilities are shrinking or moving into systems, another category of management work is growing. Someone has to direct the agents. That work can include assigning tasks, structuring workflows, inspecting outputs, coordinating multiple agents, resolving failures and deciding when a human needs to intervene. Individual contributors may take on some of it even when they have no human direct reports. At CVS, Alan Rosa described another part of that work: developing the agents themselves. Managers must give them relevant context, tune their performance and determine what they can safely handle. “I spend 70% of my time training it and 30% of my time using it,” he said. Directing digital labor takes managerial attention even when it reduces time spent on routine reporting. 

That is not really evidence that management is leaving the manager role. It is evidence that the definition of management is expanding. Today, organizations typically draw a bright line between people managers and individual contributors. AI makes that line less useful. An engineer may have no human direct reports but still direct five agents, and a finance professional may supervise an agentic workflow. These people may remain individual contributors in the traditional formal hierarchy while performing real management of digital labor. That means “management” may increasingly describe an activity, not merely a title.

Companies are rethinking the meaning of a subordinate

Tyler Derr, CTO at Broadridge, is already thinking about the AI impact on the manager’s role. He raised the possibility that some frontline managers may eventually manage agents almost exclusively. He also asked: “How do you manage the change when you’ve got a leader with three human employees and 19 agents?”

As this area develops, Broadridge is thinking about the ability to identify agents across the company, what work they would perform, what systems they would be able to access, who owns the agents, how the agents’ behavior is monitored so humans continuously remain in the loop, and what controls to implement.

The management requirements around three employees and 19 agents are fundamentally different from the traditional human-to-human management model. As a result, the traditional idea of span of control may need to be broken into multiple measures.

For example:

• There is a human span: the number of employees for whom the manager has direct people-management responsibility.
• There is an agent span: the number of digital workers or systems being directed.
• There is a workflow span: the bodies of work for which someone is accountable.
• And there is total productive leverage: the amount of work the combined system can produce.

Those numbers may diverge dramatically.

A manager with three employees and 19 agents illustrates why a single span-of-control number would obscure more than it reveals. The manager’s responsibilities for developing people, overseeing agents and owning the workflow are related, but they are not interchangeable. The question is not simply whether AI permits a wider span. It is which kinds of responsibility each person can effectively carry.

IBM points to another shift: management may attach more directly to outcomes

Matt Lyteson, IBM’s CIO for Technology Platform Transformation, added another dimension. He told us that organizations can easily mistake AI adoption for AI value. Usage is easy to measure but outcomes are harder. IBM is increasingly focused on whether an AI-enabled workflow actually changes something meaningful: revenue or margin, operating efficiency, risk, flow velocity or unit cost. Lyteson summarized the shift this way: “There is going to be a bigger focus on outcomes versus output.”

That matters for management because work itself is becoming more distributed. If an outcome is produced by two employees, six agents and an automated workflow, managing the people performing each individual step may no longer be enough. Someone still has to own the overall result. That creates another expansion of management. Managers may increasingly need to understand not only the people reporting to them but the economics, performance and health of the intelligent system producing the work. The object being managed expands from the team to the human-agent operating system.

Accountability gets more complicated as the work gets distributed

When most work flows through people inside a traditional hierarchy, accountability is relatively easy to map. When the work is split among people, agents and platforms, it becomes less obvious. If an individual contributor directs three agents, who owns their output? If one agent performs the task and another checks it, what exactly is the role of the human reviewer? If an enterprise platform determines what an agent can access, how much accountability belongs to the employee using the agent versus the team governing the platform?

Whirlpool drew a clear line. CIO Eduardo Salas described agents that create content and carry out legal, regulatory and brand review steps, while a human owner retains the decision to approve customer-facing material. The work can move through agents without leaving ownership of the consequential decision unclear. These questions require organizations to assign ownership at each point in the workflow: who directs the agent, who reviews consequential outputs, who governs its access and who owns the result. That gives us one anchor even as everything else changes: Management responsibility can move, but accountability still has to land somewhere. Where it lands may need to be redesigned explicitly.

The manager role is being reassembled, not simply reduced

Taken together, these conversations are pushing us away from a simple automation story. Some management work is moving into infrastructure. Some information-heavy management layers may become less necessary. Some existing management responsibilities may shrink.

At the same time, new responsibilities are appearing around agents, workflow orchestration, digital performance, access, oversight and the economics of intelligent work. And some of those responsibilities will be performed by people who are not formal managers today.

That means the future manager role is unlikely to be today’s job description with 30% of the tasks crossed out. It will be a different bundle.

CIOs and CHROs should redesign the bundle task by task

That is why we think organizations should stop beginning with the job title and instead start with the work of management. Take the role apart: Who gathers information? Who monitors progress? Who maintains context? Who follows up? Who allocates work? Who develops employees? Who owns the outcome?

Then decide where each responsibility should live. Some may stay with formal managers. Some may move to individual contributors. Some may become persistent AI infrastructure. Some may move to agents. Some may disappear. And entirely new responsibilities will be added.

Our conversations with CIO’s originally set out to classify management work as automated, augmented or preserved. The executive interviews are pushing us toward a broader question that is not just what AI can do, but how the total bundle of management responsibilities should be redistributed and rebuilt.

That is a much more consequential organizational-design problem than asking whether AI will replace middle managers. AI is not simply shrinking the management role. It is changing what belongs inside it. The real work now is figuring out what managers should still own, what should move elsewhere, and what new responsibilities AI is creating.

The opinions expressed in Fortune.com commentary pieces are solely the views of their authors and do not necessarily reflect the opinions and beliefs of Fortune.