AI was meant to free my managers. Why is it doubling their workload?

One of my best managers said it quietly last week. Every thirty minutes, she told me, someone on her team now produces something she has to read, check and approve. The drafts arrive polished. The decks look finished. The analysis reads clean. And she sits there at 7pm, queue still full, knowing that AI did not lighten her load. It flooded it.

If you lead managers, you have felt the shape of this even if you have not yet named it. Here is the name: AI moves the bottleneck to the manager. The technology made everyone downstream faster at producing, and production was never the constraint. Judgement was. Review was. Knowing what is true and what is plausible-but-wrong was. So the volume of work that must pass through one human's attention has climbed, while the human stayed the same size. This is a design problem, and it is solvable. It is not a story about a manager who cannot cope.

Why does AI make more work for the people meant to be freed by it?

Because the saved time does not stay saved. An eight-month embedded study of a 200-person technology company, run by researchers at UC Berkeley's Haas School and reported in Harvard Business Review, found that 83% of workers said AI had increased their workload. The mechanism they named is "workload creep": time saved on a faster task is reabsorbed into more tasks rather than into rest. The minute you draft faster, the expectation becomes more drafts. A ResumeTemplates survey puts the felt edge on it: 45% of workers say their managers explicitly reference AI when assigning extra work, and 59% report higher pressure to perform.

Now follow that current uphill. Every extra draft, every faster analysis, every "I had the AI take a first pass" lands in the same place. The manager's inbox. The team produces at machine tempo; one human still has to apply judgement at human tempo. That is the bottleneck, relocated.

What is the "megamanager" problem, and why is it a design failure?

The bottleneck has a face now, and the press has given it a name. Fortune calls it the "megamanager era," and frames AI as doubling bosses' workloads. Gallup's own span-of-control data puts numbers to it: the average American manager now oversees about 12 direct reports (12.1 in 2025, up from 10.9 a year earlier and near double the 2013 figure when Gallup began tracking), and increasingly without the administrative support that once made a span like that survivable. The same Gallup work found that 97% of managers still carry individual-contributor tasks of their own, spending a median 40% of their time as a doer rather than a leader. So the review queue lands on someone who was never given the room to clear it. Meta's applied AI engineering division, Fortune reports, runs at a 50-to-1 employee-to-manager ratio. Gartner found that 75% of HR leaders say their managers are overwhelmed by the expansion of their responsibilities, and 69% say their leaders are not equipped to lead the change being asked of them.

So we widened the span, stripped the support, and handed the manager an AI transformation she was not hired or trained to lead. Then we are surprised she is drowning. This is the part worth sitting with: it is a design failure, not a character flaw. We built a role that a coherent human cannot hold, and we are reading the symptoms of that design as if they were the manager's limitation.

There is a second, quieter tax. A BetterUp Labs and Stanford Social Media Lab survey of 1,150 employees found 40% had received "workslop" in the prior month: AI-generated work that looks finished but is not. Each instance cost the recipient nearly two hours to sort out. Guess who sits at the top of the funnel where that lands. The manager is now the quality-control layer for output produced faster than she can possibly inspect.

How do you redesign the management layer instead of adding more reports?

Start where the value actually lives. The 10-20-70 rule from BCG (Boston Consulting Group, the global strategy consultancy) holds that successful AI transformations spend roughly 10% of effort on the algorithm, 20% on technology and data, and 70% on people and processes: roles, workflows, change. The 70% is where most transformations break, and it is exactly the part we keep skipping. We bought the tools, we ran the training day, and we left the management layer untouched, then asked it to absorb the entire shock. The bottleneck is no longer the technology. It is the design of the human work around it.

This is the solution category: redesign the management role for a world where production is cheap and judgement is the scarce good. That looks like a smaller, deliberate span of judgement rather than a sprawling span of control. It looks like AI doing the first-pass triage so the manager applies attention where it counts. It looks like restoring the support layer we quietly deleted. Here is a sequence to begin.

1. Measure the review load, not the headcount. Count the items per day that truly require a manager's judgement. That number, not direct-report count, is the real span.
2. Move triage upstream. Put AI and clear standards between production and the manager, so only work that needs human judgement reaches her desk.
3. Set a workslop standard. Make "finished" mean checked by the producer, not by the manager. Quality belongs at the source, not the bottleneck.
4. Restore the support you stripped. Give the management layer back the coordination capacity that a 12-report span quietly assumes and rarely funds.
5. Protect the manager's capacity for judgement. Sustained judgement under load is a physiological capacity, not an infinite one. Neurovisceral integration research (work showing the heart and brain regulate each other) links heart rate variability (the natural beat-to-beat variation that signals a well-regulated nervous system) to executive function and self-regulation: a manager's ability to keep deciding well degrades as her state degrades. Design the role so her best judgement is available when it matters.

The promise of AI freeing your managers is still real. It simply arrives on the far side of a redesign, not before it. When you move triage upstream, set quality at the source, and size the role to the judgement it truly requires, the same technology that flooded your best manager becomes the thing that finally gives her room to lead. This is what Phase Three actually asks of us. Phase One was the Age of Effort: work hard, get a little more, linear growth. Phase Two was the Age of Scale: build once, sell to millions, exponential growth. Phase Three is the Age of Acceleration, where output is decoupled from human effort almost entirely, the phase AI unlocks. Production becomes near-free, and judgement becomes the constraint that decides who wins. The leaders who win the next decade are the ones who redesign the human work first, then let the machine serve it.

About the author

Thomas Green

British technology futurist, AI keynote speaker and advisor. Thirty years across enterprise technology and AI strategy, helping leaders navigate the future of work. The futurist who died.