---
title: "Managers Own the Workflow"
date: 2026-10-04
description: Most AI business cases count the time agents save; the saving becomes real or not in the manager's workflow. Five questions Boards should put to the executive.
author: Mario Thomas
canonical: https://mariothomas.com/blog/managers-own-the-workflow/
---

In [Governing the Redeployment Dividend](/blog/governing-the-redeployment-dividend/) I set out who owns what when AI enters an organisation: AI owns the execution of the work, managers own the workflow it sits within, and the Board owns whether the freed capacity creates value.

For as long as the role has existed, managers have been responsible for the people in their teams and how well they perform. In organisations deploying AI, those same managers now find themselves responsible for fleets of software agents as well. The agents do some of the work those people used to do, and new work the team was never able to do. So the workflow the manager owns now means deciding which work goes to which team member, person or agent, checking what comes back, and deciding what to do with the time this frees.

The role has changed in kind, and its span of control has grown with it, because every agent the team runs is one more thing the manager supervises, whether or not it appears on an organisation chart. In most organisations the manager's job description has not changed, and the business case that put the agents there says nothing about how the change is to be managed.

Most AI business cases count the time the agents will save and ignore the time it takes to check what they produce, which is what [The Balancing Item](/blog/unpriced-cost-ai-oversight/) was about. For managers that checking compounds, because a manager is responsible for the output of every agent the team runs, not one person's, and carries that on top of a job that was already stretched before the agents arrived ([Gartner](https://www.gartner.com/en/newsroom/press-releases/2024-10-15-gartner-survey-finds-leader-and-manager-development-tops-hrleaders-list) 2024). And the routine work through which the next generation of managers would have earned their judgement is the work AI takes first. Little of this is in the business case.

## The saving depends on the manager

A [Gartner analysis published in December 2025](https://www.gartner.com/en/articles/ai-is-coming-for-inefficiency) found that AI saves about **five hours per person per week**, and that most of that time is then spent on work that adds no value. A business case that assumes that saving expects it to be available for more work. The manager may see nothing like five hours, because the team spends some of it checking what the AI produces, some of it coordinating work between people and agents, and some of it dealing with the exceptions the agents hand back. The team can end up busier than before, and the manager is usually the person best placed to see what saving has been realised and what additional work the team has absorbed.

In June 2026 [Harvard Business Review published interview research](https://hbr.org/2026/06/ai-adoption-is-overloading-your-middle-managers) from two large consulting firms. The middle managers were the ones validating the AI's output, catching its mistakes, and teaching their teams how to use it. They did this under the same delivery pressure as before, or more, and with no formal support. The people above them and below them saw the benefit, while the managers did the work that made the benefit real.

In its [State of the Global Workplace 2026](https://www.gallup.com/workplace/708071/global-employee-engagement-continues-decline.aspx), Gallup reported that manager engagement fell from **31%** in 2022 to **22%** in 2025, a nine-point drop in three years, while the engagement of the people managers lead held roughly flat. Two of Gallup's observations belong side by side. The first is that manager engagement can decline with larger spans of control. The second is that the strongest predictor of whether employees use AI, aside from technical integration, is whether their direct manager actively champions it. Organisations that widen managers' spans while relying on those same managers to champion AI are putting both pressures on one job.

There is an older reason to expect engagement to fall as spans grow. A [1983 paper on the ironies of automation](https://www.sciencedirect.com/science/article/pii/0005109883900468) showed that automating the routine work turns the person who did it into a monitor, and nobody can keep watch over something that rarely goes wrong for long. A manager whose team's routine work has gone to agents, and whose span has grown, is a monitor by design. The saving the business case counted on now depends on a manager who has been turned into a monitor.

## The job changed by accretion, not by design

McKinsey's July 2026 piece on [the rise of the 'agent manager'](https://www.mckinsey.com/capabilities/tech-and-ai/our-insights/rewired-takes-how-ai-is-unlocking-creativity-and-heralding-the-rise-of-the-agent-manager) asks three questions that a Board will recognise as its own: whether agents belong on the organisation chart, how to think about agent capacity alongside human capacity, and what happens to spans and layers. Its description of the task is blunt. Managing agents is "something no manager in history has ever had to do before." McKinsey is equally candid about the people being asked to do it. Earlier changes to the operating model, agile and DevOps among them, had already strained managers who were often never trained for them, and whose spans grew at the same time. [The Conference Board's 2026 survey](https://www.conference-board.org/press/ai-skilling) of nearly 1,300 workers found the same gap from below: only a third had received any employer-provided AI training in the six months before the survey, and where training existed it mostly covered AI literacy and prompting, with far fewer organisations teaching anyone to manage agents.

McKinsey's [November 2025 piece on management for agentic AI](https://www.mckinsey.com/capabilities/people-and-organizational-performance/our-insights/the-organization-blog/rethink-management-and-talent-for-agentic-ai) lists the moves it says organisations should make now, and the first of them is one a Board can check for itself: redefine the role, and change the job descriptions and performance reviews to match. A Board does not need to see the redesign to know whether it has happened. If the job descriptions have not changed, the Board has good reason to conclude that the job changed without anyone deciding to change it.

That is how the agent manager role is arriving in most of the organisations I see: by accretion. Each agent deployed adds a supervisory duty to whoever owns the workflow the agent sits in. Each duty is small enough to pass unremarked, and the total is rarely approved because it is rarely presented. A cost left out of a budget does not disappear; something absorbs it. Here it is the manager's job that absorbs it, and the job description rarely says so. **Whether an AI business case pays off is decided in the manager's job, not in the approval.**

The Board's question is not what the redesigned role should look like. That is management's to answer. The Board's question is whether the role has been designed at all.

## The next managers are not being grown

The work AI takes first is the work that used to train people for responsibility. The [Stanford Digital Economy Lab's August 2026 update](https://digitaleconomy.stanford.edu/news/canariesaug26/) on young workers found that employment of 22 to 25 year-olds in jobs where AI can already do many of the core tasks is now about **19%** below where it would be had it kept pace with workers in less exposed jobs. Stanford calls that a pattern, not proof that AI caused it. [McKinsey's July 2026 research on building expertise](https://www.mckinsey.com/capabilities/people-and-organizational-performance/our-insights/building-expertise-in-the-age-of-ai-who-trains-the-next-generation) describes which work that is: research, documentation, data clean-up, basic coding, and preliminary analysis. That work was never only output. It was how junior people built the instincts that earned them more responsibility, and eventually a team. McKinsey reports what happens without it. When workers used generative AI to perform technical tasks they could not do themselves, "the capability vanished the moment AI access was removed. No durable skill had formed." The lesson it draws: "Passive reliance builds output; structured comparison builds experts." The design McKinsey recommends, in which a person attempts the work first, the AI grades the attempt, and a manager talks through the differences, is a decision someone has to take.

Managers learned their judgement the same way. A new team leader used to learn by doing the routine work themselves: reconciling the forecast by hand, writing the monthly report, working out what had gone wrong before anyone told them. That is the work the agents now do. At the same time the agents produce more output that someone has to interpret, so that judgement is needed more than before, while the work that used to build it is being done by the agents instead. The next managers are not being grown.

Succession is a Board responsibility, and the nomination committee's horizon is not this year's executive bench. It is the layer that bench will be drawn from in the years that follow. If the route by which people earn the judgement to run a team is being automated away now, the pipeline is thinning today and the shortage arrives later. That decision is being taken by default, inside the same tool rollouts the Board approved as productivity investments. Not every skill needs preserving, and some should fade as AI absorbs them. Complex judgement, including the judgement to tell when an AI output is wrong, is not one of them, and the difference between the two is a choice the organisation either makes or has made for it.

## What the Board owns

The Board owns whether the freed capacity creates value, and in the manager's job that ownership comes down to five questions the Board should put to the executive. The answers say more than any adoption dashboard.

1. Does the AI plan include a redesign of the manager role, or only a rollout of tools to the people who hold it?
2. Where do agents sit on the organisation chart, and who is responsible for what they produce?
3. Has anyone recounted the span of control since the agents arrived, with the agents included, or is it absorbing the change without anyone having looked?
4. Is supervising the agents someone's job, with time allowed for it, or does it land on whoever is nearest?
5. How will the next managers be grown when the work they would have learned on has gone to the agents?

The answer to the first question is already in the job descriptions and performance reviews, whatever the transformation deck says. The rest matter because of Gallup's finding: aside from technical integration, the direct manager is the strongest predictor of whether employees use AI at all. A business case for AI that nobody uses does not happen, so the job the business case ignores is the job it depends on.

## What the Board actually approved

AI owns the execution of the work, and the Board owns whether the freed capacity creates value. Between them, managers own the workflow, and the workflow is where the saving becomes real or does not. Nothing in the business case says so, and the Board is rarely asked to decide it. The choice in front of the Board is whether to go on approving AI cases on that basis, or to ask management to show how the manager's job changes when the agents arrive. **An AI business case that leaves the manager's job unchanged has priced the technology and assumed the operating model.**

## The Questions Considered {#questions-considered}

### Why does the saving in our AI business case depend on managers?

Gartner's December 2025 analysis found AI saves about five hours per person per week. The manager may see far less, because the team spends some of it checking AI output, coordinating between people and agents, and handling exceptions the agents hand back. The manager is usually best placed to see what saving has been realised and what additional work the team has absorbed. Whether an AI business case pays off is decided in the manager's job, not in the approval.

### What questions should our board put to the executive about managers and agents?

Five questions: does the AI plan redesign the manager role or only roll out tools; where do agents sit on the organisation chart and who is responsible for their output; has span of control been recounted with agents included; is supervising agents someone's job with time allowed; and how will the next managers be grown? The first answer is already in the job descriptions and performance reviews.

### How do we tell whether the agent manager role has been designed?

McKinsey's November 2025 piece says the first move is to redefine the role and change job descriptions and performance reviews to match. If our job descriptions have not changed, the Board has good reason to conclude the job changed without anyone deciding to change it: the role arrived by accretion, each agent adding a supervisory duty small enough to pass unremarked, with the total never approved because it was never presented.

### Why is our managers' engagement falling as agents arrive?

Gallup reports manager engagement fell from 31% in 2022 to 22% in 2025, while the engagement of the people managers lead held roughly flat. Gallup observes manager engagement can decline with larger spans of control, and that the strongest predictor of whether employees use AI, aside from technical integration, is whether their direct manager actively champions it. A 1983 paper on the ironies of automation adds that automating routine work turns the person who did it into a monitor.

### How do we grow the next managers when agents do the routine work?

With difficulty unless someone decides to. McKinsey reports that when workers used generative AI for technical tasks they could not do themselves, no durable skill formed; passive reliance builds output and structured comparison builds experts, so a person attempts the work first, the AI grades the attempt, and a manager talks through the differences. Succession is a Board responsibility, and if the work that builds managers' judgement is being automated away, the pipeline is thinning and the shortage arrives later.