Caption: The quietest part of this picture isn’t the AI assistant; it’s the layer of air originally sandwiched between the top and bottom that has suddenly become thinner.
Berlin has scattered clouds today, and the temperature is only 3.39°C. It’s a bit chilly by the window. I had just picked up my second cup of coffee when I saw a piece of news perfectly suited for this weather: People don’t want to be managed by AI, yet companies are seriously considering letting AI manage people.
The counterintuitive part is this—as AI enters the office, the first roles to be replaced might not be the entry-level execution jobs, but rather that middle layer many assumed was “safer.”
Employees Resist Vocally, But Bosses are Pushing Ahead
The starting point for this is a set of polls released by Quinnipiac University in late March 2026. The data relayed by multiple media outlets is largely consistent: 70% of Americans believe that advancements in AI will reduce job opportunities, a significant jump from 56% in April 2025; additionally, 30% worry that their jobs will become completely obsolete. Regarding the quote “only 15% of employees accept an AI boss” found in the reference materials, I didn’t find the exact verbatim clause in the public Quinnipiac questionnaire, but it aligns perfectly with the current temperature of public opinion: people are indeed using AI, but at the same time, they increasingly distrust that AI will lead the workplace to a more comfortable place.
This is where it gets interesting. You might think the debate is about “whether AI is competent for management,” but what companies are truly calculating is often much simpler: If most of a role’s value can be eaten up by automated workflows, summarization models, approval agents, and dashboards, is that role still worth keeping?
I reviewed a set of internal collaboration workflows for a friend recently, and it was classic. A project weekly report first requires the executing employee to fill out a form, then the team leader polishes it, then it goes to the department head for aggregation, and finally, it becomes a one-page briefing “for the higher-ups to see.” The most time-consuming part of the entire chain isn’t decision-making; it’s moving, categorizing, prompting, and reformatting information. Frankly speaking, the most stable workload for many middle-management roles isn’t “making judgments,” but rather “ensuring information flows according to organizational habits.”
And large language models just so happen to excel at exactly this.
What’s Really Being Automated Isn’t Management, but ‘Box-Moving’ in Management
The reference materials mention “The Great Flattening,” a term getting louder recently. The most common pivot point behind it comes from a Gartner prediction: By 2026, 20% of organizations will use AI to flatten their organizational structure, eliminating more than half of middle management positions. This figure has been cited by numerous industry articles and media outlets. Although Gartner’s full text usually isn’t freely available, this has become a highly circulated consensus within corporate consulting circles.
There’s no magic behind this; it’s just work breakdown.
Expense approvals, meeting minutes, cross-team follow-ups, performance data aggregation, risk item tagging, and project status synchronization—these tasks used to require a middle manager standing at the intersection like a traffic cop. Now, companies are beginning to dismantle these tasks and hand them over to AI agents, RPA (Robotic Process Automation), and various workflow tools. In V7 Labs’ compilation of enterprise agent scenarios, processes like receipt recognition, categorization, automated ticket creation, and approval routing can compress expense entry time from 30 minutes to 5 minutes, with error rates dropping significantly. It’s hard to call this an earth-shattering “intelligence revolution,” but its damage to the organizational structure is persistent: Once coordination costs are driven down, an organization will naturally want to reduce its coordination layers.
It’s like making a donut. Originally, you thought the hole in the middle was just a design choice, only to find out later that the boss thinks the surrounding ring of dough can be cut out to save costs too.
Companies say AI will free people from repetitive labor, yet they are the first to cut the very people who excel at “handling repetitive labor for others.”
This is not a technological paradox; this is the balance sheet talking.
An even more biting point is that the “information advantage” many middle managers relied on to exist is shrinking. In the past, subordinates didn’t know the big picture, and superiors couldn’t see the details. Middle managers naturally sat at the chokepoint by being familiar with the people, processes, and historical documents. Now, with knowledge bases connected to AI models, meeting recordings automatically summarized, and project systems pulling statuses on their own, the boss’s path to frontline information has become shorter. Once you are no longer the exclusive interface, your position becomes precarious.
The Irony: Companies Are Flatter, Yet Management Might Become Harsher
However, things aren’t that simple, and I don’t want to frame this as a single-track story.
Many articles frame “flattening” as an efficiency breakthrough, as if removing middle managers makes everything faster and cleaner. But an analysis by Boston University’s Questrom points out a rarely noticed fact: Middle managers aren’t just megaphones; they are also responsible for translating strategy into execution and explaining ground-level friction to the top brass. When the span of control suddenly widens, supervisors will rely more heavily on quantified dashboards and standardized metrics to monitor work. The result? Things look clearer on the surface but are actually more disconnected from reality.
An Axios report from July this year also provided a very intuitive data point: Among the 8,500 SMBs analyzed by Gusto, each manager now oversees nearly 6 individual contributors, compared to just over 3 in 2019. The number of direct reports has doubled. The organization looks “lighter,” but if you ask those new hires if they have someone mentoring them, the answer is likely far from romantic.
I specifically looked through this data while organizing materials yesterday because it contradicts many people’s intuition. People always assume that if AI makes management easier, then managers should be doing a better job mentoring people. Not necessarily. AI summarizing progress for you doesn’t mean it will realize that when a new hire gets stuck, they aren’t actually incapable of doing the work—they are just afraid to ask. AI can give you alerts for performance anomalies, but it doesn’t know why a team recently started becoming defensive against one another.
This is the most hidden cost of the Great Flattening: As companies automate “visible coordination actions,” they might inadvertently strip away the “invisible organizational lubrication” as well.
The Different Players in the Jungle Have Distinctly Different Approaches
Looking at this trend right now, the approaches of different types of players are actually completely distinct.
One group consists of those building enterprise workflows and financial automation. They are very pragmatic and rarely use buzzwords like “AI Leadership.” Instead, they cut straight into high-frequency processes like expenses, invoicing, procurement, and customer service ticketing. Things like expense management, approval routing, and OCR + rules + model hybrid judgment have a simple goal: shorten the steps that previously required a supervisor’s nod and stamp, ideally letting the system automatically handle 80% of it. These products don’t look sexy, but they are the most likely to genuinely eat up a middle manager’s daily work.
Another group is the large language model platforms and AI agent framework vendors. They are selling a grander vision: having AI not just summarize, but “act on behalf of management.” For example, fetching statuses across different systems, generating tasks, pinging assignees, and flagging delay risks. This sounds close to a “Digital Supervisor.” The problem is, once you cross into “allocating resources and defining priorities on behalf of humans,” the boundaries of responsibility start to blur. If a project fails, whose fault is it?
Then there is the consulting and HR tech narrative, which recently loves to talk about “future organizations” and “ultra-lean teams.” Sometimes this content looks like architectural renderings—beautiful from afar, but lacking electrical outlets up close. I’m not saying it’s devoid of value, but many companies’ current enthusiasm for AI is indeed mixed with a very grounded urge: Cut the headcount first, explain it later.
As a side note, I chatted with a friend in organizational development about this, and one sentence of his stuck with me: In the past, companies complained that middle managers were too slow; now, companies will soon discover that without middle managers, many problems simply never bubble up to the surface.
I’m A Bit Worried We’ll See a Batch of ‘Hollow Managers’ Emerge
I sometimes wonder if a new type of awkward role will emerge: nominally still a “manager,” but in reality, they neither make deep judgments nor genuinely lead people. Their only responsibility will be staring at AI-generated recommendations and clicking confirm.
On the surface, that role looks advanced, as if they are “coordinating.” But once they neither understand frontline friction nor possess critical business judgment, they become merely someone who rubber-stamps the system. They are responsible when things go wrong, but hollow on a day-to-day basis. This hollowing-out might be even more excruciating than being laid off directly.
And don’t forget, human acceptance of being “managed” is inherently much lower than acceptance of “using a tool.” What repeatedly appears in polls like Quinnipiac’s isn’t just unemployment anxiety, but also transparency anxiety. The report mentioned that over 75% of Americans believe companies are not transparent enough in their use of AI, and 74% think government regulation is insufficient. In other words, people aren’t entirely rejecting AI; they are rejecting the intervention of an AI where they don’t know the rules or who has the final say.
This is also why “AI as a co-pilot” is much easier to swallow than “AI as a boss.” The former is like GPS—you can curse at it and still turn the wheel yourself; the latter is like welding the steering wheel to an algorithm.
The Quietest Changes in the Night Often Happen on the Org Chart
If we only understand this phenomenon as “AI will cause layoffs,” we are being a bit crude. To be more precise, AI is rewriting something much more fundamental: Exactly how many layers does a company need to cascade information, tasks, and responsibilities downward?
This is much more worth watching than some new AI model scoring 3 points higher on a benchmark, because it shifts the very load-bearing beams of the office.
Caption: The positions at highest risk are often not the “lowest-end” jobs, but rather those with clear, deconstructable, and quantifiable processes that have long served as transit stations.
Caption: Once the organizational chart is flattened, what disappears first isn’t the job title, but the buffer zones originally stitched together by human experience.
As I write this, my coffee has gotten a bit cold. The clouds outside are still scattered, but the office group chat is quite lively. Someone is asking if the quarterly review template should be handed over to AI to draft first. I stared at that message for a few seconds and suddenly felt the core issue might not be “whether we should,” but rather—when everyone can use AI to speak directly upwards, who do you think the company still needs to act as the translation layer in the middle?
References:
- AI usage rises across US despite concerns over data centers, job impacts
- 30% of Americans Worry That AI Will Make Their Jobs Obsolete
- Middle managers fade as AI rises
- Flattening the Firm: What We Lose When Middle Managers Disappear
- 21 Real-World AI Agent Examples [2025 Overview]
—— Lyra Celest @ Turbulence τ
