Managers are not becoming obsolete. Manual management is.

Middle managers are increasingly described as "redundant" in the age of AI. Headlines talk about the Great Flattening: layers removed, spans widened, autonomous teams that supposedly no longer need anyone in the middle.

This framing gets one thing right and one thing badly wrong — and eliminating managers because some of their tasks can be automated is a category error. The correct question is not whether we need managers. It is which parts of management should be done by humans, and which by systems.

The numbers say the confusion is understandable. McKinsey found that middle managers spend nearly half of their time on non-managerial work — about 31% on individual-contributor tasks and 18% on administration — leaving barely a quarter for the talent work that is supposedly the point of the role. Gartner reports that 75% of HR leaders say their managers are overwhelmed by the expansion of their responsibilities, and that the average manager now carries 51% more responsibilities than they can effectively handle.

A role where half the time goes to work that isn't the role, performed by people who are drowning — that is not a role that AI made obsolete. That is a role that was never redesigned for the world it operates in.

And though the debate names middle managers, the argument is about management work itself. In a 5,000-person company, the person drowning in status-chasing is a middle manager. In a 25-person company, it is the owner — spending the morning chasing job-site updates and the afternoon on the negotiation only they can do. Same two categories of work, same failure mode, no org-chart layer required. Everything that follows applies to anyone who manages — the middle layer is simply where the failure became most visible.

What the critics get right

Honesty first: the people arguing that management is bloated are not imagining things.

A large share of what managers do all day — collecting statuses, compiling reports, chasing updates, forwarding information up and down — is repetitive, rule-based, and consists of moving information rather than making decisions. Historically, middle layers existed partly to carry information through the organization, because information could not flow by itself. In a world of connected systems, information can flow by itself. A human whose job is to be the pipe is, indeed, replaceable by a pipe.

So yes: much of what today occupies a manager's calendar should be automated, and organizations that refuse to do it are paying director-level salaries for assistant-level work.

Where the critics go wrong is the conclusion. They observe that half the job is automatable and conclude the person is unnecessary. The inefficiency was never the person. It is the tasks the person is forced to perform.

What the manager's job really is

A manager sits between intent and execution — translating goals into operational outcomes while leading people and owning results. The role is defined by scope and accountability, not by title: a sales manager, an operations coordinator, an engineering lead with reports, a store or clinic manager, a project lead — or the owner of a small business who is all of these at once. What they share:

• Translate strategic goals into operational outcomes
• Align multiple people or functions
• Own cross-team execution and results
• Lead, coach, and develop people
• Keep day-to-day work connected to leadership intent

To see the two categories of management work in one place, watch a sales manager on an ordinary Tuesday. Deciding whether the 18% discount that closes the quarter is worth it; coaching the rep who froze in a negotiation; settling which segment to deprioritize — that is judgment work, and it is the job. Asking every rep "where does deal X stand?"; assembling Friday's forecast deck; chasing the proposal promised three days ago; checking who filled in the CRM — that is operations work, and none of it needs to be done by the most expensive judgment in the room.

Managers are the execution engine of the organization. Without them, strategy rarely survives first contact with reality — and when companies remove the layer without redesigning how work is coordinated, the coordination does not disappear. It gets pushed up to executives, down to overloaded individual contributors, or absorbed by nobody at all. Execution slows, burnout rises, accountability blurs.

The two kinds of management work

Management work falls into two distinct categories, and almost every bad decision about managers comes from confusing them.

Leadership and judgment work — high value, human. Coaching and feedback. Conflict resolution. Decision-making under ambiguity. Prioritization trade-offs. Motivation and morale. Sense-making during change. AI can support these activities; it cannot replace them. This is where the true value of managers lives.

Management operations — necessary, but automatable. The repetitive, rule-based work of moving information:

• Work tracking and status collection — chasing progress and compiling statuses, replaced by commitment-based tracking against an agreed plan, with real-time visibility.
• Reporting — weekly decks and narrative reports, replaced by automated summaries and exception-based reporting.
• Coordination and follow-ups — dependency tracking, reminders, and nudges, driven by the system rather than by a human's memory.
• Meeting-based alignment — most status meetings exist only to surface information; when information is continuously visible, meetings shrink to decisions.
• Process enforcement — checking compliance with templates and workflows, embedded in the system instead of policed by a person.

Automating management operations does not eliminate the need for them — it changes who performs them. That distinction is the entire argument.

What building this taught us: management operations are a job, not a feature

Months of building and operating this in production taught us that the obvious model — invisible software automating the mechanical work in the background, while the manager keeps managing — is one step short. Management operations are not a feature you sprinkle over a task tool. They are a job — and jobs are held by someone. The mature version of this idea is not automated management operations. It is an AI manager: an accountable role in the organization that runs the operations end to end — tracks commitments against a signed plan, follows up with the team, monitors deliverables at risk, intervenes first (reassigning, breaking work down, chasing the missing input), reports by exception, and escalates to a human only what actually needs human judgment.

That reframing changes the org design from two categories into three layers:

• Human leadership (the board layer). Objectives, budget, policy. Signing plans — the moment scope is reviewed by someone accountable. Deciding exceptions. Evaluating performance. Coaching people.
• The AI manager (the operations layer). Runs the day-to-day: assignment, tracking, follow-up, risk monitoring, reporting. Never approves its own actions — high-stakes moves route to the human by policy.
• The workforce (the execution layer). And here is the newest part of the model: this layer is no longer only human. Teams now mix human workers and AI workers — hired, given a work plan, tracked, and evaluated with the same yardstick.

In this model, the human manager is not "freed up by software". The human manager is promoted by structure — from operator to board. The work only humans can do is finally the whole job.

The gains

Hard gains. If you pay a manager $150k a year and half their time goes to work a system can run, you are spending roughly $75k a year on the wrong layer. Recovering it does not require firing anyone — it requires moving the operations to the layer that should own them. Span of control follows: a human manager typically saturates at 7–10 direct reports because tracking saturates first; with an AI manager running the operations, the same human effectively supervises 15–20 people. And decision latency collapses: the blocker that used to wait for Monday's status meeting is flagged the moment it appears, because the operations layer runs continuously.

Soft gains. Nobody likes being asked "is this done yet?" five times a week. When follow-up is done by the AI manager against a plan the person committed to, the chasing stops being personal — the human manager intervenes only when the system flags a risk, and the relationship is preserved for coaching instead of policing. Consistency improves for the same reason: the same rules apply to everyone, every cycle — the quiet employee gets the same follow-up as the loud one. And institutional knowledge stops living in the manager's head: when the way the team runs is recorded in the system — plans, decisions, evaluations, the reasons behind them — a manager's departure no longer takes the process with them. The organization accumulates precedent: a memory of how work is done that survives any individual, and, increasingly, any AI model or vendor.

Strategic gains. Forecasting stops depending on a manager's gut feeling — actual delivery velocity is visible, so slippage is flagged weeks before a human would notice. And the ultimate gain is what the freed-up human does with the recovered half of their working life: innovation, mentoring, strategy — the high-leverage work the role was always supposed to be about.

The inverted pyramid

There is one more consequence, and it is the one that settles the "are managers obsolete?" debate for good.

Every organization we know is shaped like a pyramid for a single reason: a human manager can only closely supervise a handful of people, so reaching a large base requires layers of managers supervising managers. AI work breaks that constraint from both ends. The AI manager multiplies how many workers one human can supervise; AI workers multiply how many workers there are to supervise. The pyramid inverts: not many humans under few managers, but many workers — human and AI — answering to few humans.

In that world, the manager's defining skill — translating strategy into execution, and holding accountability for outcomes — becomes more scarce and more valuable, not less. Someone has to sign the plan, set the policy, decide the exceptions, and evaluate the results. The organizations that thrive will not be the ones that deleted the management layer. They will be the ones that redesigned it — and gave it the supervision tools humans actually understand: plans, deadlines, cycles, evaluations. (How that supervision layer works — for AI agents from any vendor — is the subject of our companion article, Supervising AI agents the way humans understand.)

Conclusion

Managers are not becoming obsolete. Manual management is.

The future belongs to organizations that treat management operations as a system, leadership as a human responsibility, and AI as an accountable layer in the org chart — not a replacement for the people who lead it.

The question is no longer whether management survives the AI era. It is whether management is finally redesigned to work as it was always intended: humans leading people and owning judgment, with the mechanical half of the job done — at last — by something built for it.

References

• McKinsey & Company, Stop wasting your most precious resource: Middle managers (March 2023) — surveyed managers spend nearly half their time on non-managerial work (~31% individual-contributor tasks, ~18% administration), and almost three-quarters of their time on work not directly related to talent management.
• Gartner, Top 5 Priorities for HR Leaders 2025 (survey of 805 HR leaders, July 2024) — 75% of HR leaders say their managers are overwhelmed by the growth of their job responsibilities; the average manager has 51% more responsibilities than they can effectively handle.