It’s cloudy outside. The gloomy atmospheric pressure of late March in Shanghai is quite fitting for writing something a bit unpleasant. Taking my first sip of freshly brewed, still slightly scalding coffee, I paused for a couple of seconds when I saw this set of numbers: only 15% of people are willing to work in a role where their tasks and schedules are managed by AI.
People Haven’t Nodded, But the System is Already Online
The most piercing part about this figure isn’t that “everyone is afraid of AI.” Fear is normal. What’s truly unsettling is that even if employees reject it, companies will push it forward anyway.
In its April 2025 poll, Quinnipiac indeed captured a similar societal sentiment: 70% of Americans believe that the advancement of AI will reduce overall job opportunities. This percentage is no longer just emotional; it feels more like a simple, plain judgment. People might not understand transformers, nor do they necessarily care about agent workflows, but they can recognize one basic fact: any job that can be broken down into “collecting information, categorizing, reminding, following up, and generating reports” will be the first to be encroached upon by AI.
And middle management, unfortunately, easily falls right into this bucket.
Many companies claim that middle managers are the “organizational lubricant.” To put it bluntly, however, the daily routine for a large number of middle managers is simply translating vague directives from above into actionable tasks for below, and then translating the progress from below back into visual reports for above. They sit in conference rooms, but they operate very much like an API gateway. In the past, this work had to be done by humans because information was scattered, systems were fragmented, and departments couldn’t understand each other. Now, once large language models are integrated into ticketing systems, meeting minutes, emails, IMs, and knowledge bases, this manual layer of “translation-tracking-summarization” starts looking a bit too expensive.
This is not fear-mongering. Gartner’s assessment in October 2024 was that by 2026, 20% of organizations will use AI to flatten their organizational structures, cutting over half of their existing middle management positions. This projection is more grounded and cite-worthy than the reference “over 20% of companies driving mass flattening by the end of the year.” It doesn’t mean all middle managers will vanish; rather, it means that the middle managers who historically survived by acting as informational placeholders will be eaten by the system first.
Looking at this image, you can tell that what they want to replace isn’t “leadership,” but the repetitive actions of collecting, distributing, and following up.
What is Truly Being Replaced Isn’t the “Manager,” but the “Messenger”
There’s a point here that often gets muddled in discussions.
Many reports like to frame this as “AI is starting to replace management,” which makes it sound like a robot is about to sit in the glass office at the end of the hallway. In reality, it’s not that dramatic. Right now, what is being replaced are the most easily standardized parts of managerial functions: scheduling, approvals, expense distribution, weekly report summarization, converting meeting minutes into tasks, cross-departmental progress syncing, and pre-filling performance evaluation forms.
In other words, what AI is temporarily replacing is not “judgment,” but “relaying.”
The difference between the two is immense. A mature middle manager—one who can genuinely handle resource negotiation, risk assessment, team emotional stability, and make the final call at critical junctions—is far from being entirely deleted. The problem is that in many organizations, middle managers have never been allowed to grow into this type of role. They have been trained long-term to be gatekeepers of process dashboards. Companies complain that they are slow, while simultaneously making them fill out forms, schedule meetings, and chase down progress every day. Now that software can finally take over this assembly-line work, companies are naturally tempted.
There’s also an interesting logical contradiction: employees hate being managed by AI not just because they fear unemployment, but also because they resist the feeling of “nobody being in charge.” Just because humans are willing to accept a demanding boss doesn’t mean they’re willing to accept a cold, impersonal command system. No matter how annoying a boss is, they can at least explain things, take responsibility, and be the subject of your private complaints. If an AI assigns you tasks, shifts priorities, and calculates your performance, who do you argue with when things go wrong?
This is exactly the most hollow part of many current “AI supervisor” narratives. Just because it can arrange things doesn’t mean it can take accountability.
I always take a closer look at employment charts like this. Not because I fully believe them, but because they serve as a reminder: when companies talk about efficiency, they usually don’t talk first about who will take the fall.
When You Look at Several Companies Together, It’s Not So Romantic Anymore
If you only look at the slogans, almost every company in this wave claims to be more “AI first.” But if you break down their actual moves, their approaches vary wildly.
Klarna was one of the first companies to loudly proclaim that “AI can replace customer service and operations,” and the market loved that story. But later, it started hiring again, proving at least one thing: automatable processes don’t equal a complete substitute for service experience. Duolingo shouted a similar slogan, saying it would gradually stop outsourcing tasks that AI could handle to contractors, yet it didn’t conduct mass layoffs of full-time employees like many clickbait titles suggested. IBM’s statement is more typical: repetitive back-office roles can be automated, but it is simultaneously hiring more programmers and sales staff.
Looking at them together, you’ll uncover a rather unglamorous reality: AI isn’t “replacing people” evenly across the board. Rather, it is slicing the organization’s work in a new way. The repetitive, quantifiable, and auditable pieces are handed over to the system first; the pieces that require negotiation, on-the-spot judgment, and the ability to take the blame are still left for humans.
To put it bluntly, when many companies talk about “de-layering middle management” right now, it’s not necessarily purely because the technology is mature. It might also be mixed with budget pressures, post-pandemic course correction, senior management wanting to speed things up, or simply a calculated desire to pay one less layer of people. Here, AI sometimes acts as a very convenient fig leaf: “layoffs” sounds harsh, but “organizational intelligent upgrade” sounds much more respectable.
Therefore, when you read that “Company X used AI to slash Y number of management jobs,” it’s best to ask an extra question: who exactly did they slash? Was it a redundant approval chain, or was it the responsibility for judgment that should have inherently been borne by a human? This distinction is far more important than the news headline.
There isn’t much of a tech vibe to this picture, but it’s highly fitting—what’s truly hard to replace are often these vague, ambiguous moments that rely on mutual human confirmation.
The Most Dangerous Roles Are Actually Those That “Look Very Busy”
A while ago, while organizing a project document, I stared at a screen full of meeting conclusions, task assignments, and deadlines, and suddenly felt a bit uncomfortable. Not because there was too much work, but because these things were far too suitable to be taken over by machines. If you look closely, the replaceability of many white-collar jobs lies not in their low professional depth, but in their output formats being far too structured.
This is especially true for middle management.
Bosses want results, and grassroots employees want clear directives. Caught in the middle, middle managers have naturally grown a whole set of workflows that are incredibly suited for AI agents: scraping information, organizing context, generating next steps, auto-reminding pending items, updating Gantt charts, and triggering approvals. Today, this is still called an “AI assistant”; tomorrow, it might become the “default system workflow.” Once the name changes, the job position begins to loosen.
But don’t think of AI as omnipotent, either. It is very good at processing structured information, but it doesn’t truly understand the subtle atmosphere in the office. If someone in a team is silent, it doesn’t necessarily mean agreement; they might be suppressing anger. If a project is delayed, it’s not necessarily poor execution; the goal itself might have been wrong to begin with. These aren’t issues you can solve just by writing a longer prompt. Having a model summarize meeting minutes is fine, but asking it to judge “whether this person is still fit to lead the team” carries a skyrocketing risk.
Therefore, I increasingly feel that the most dangerous positions in the future workplace aren’t the low-end jobs, but the ones that look decent, boast proficient workflows, and produce pretty reports, yet in reality are just moving things around within an information chain. It’s like buying a whole supermarket just to cook a single dish. Organizations used to retain too many “people responsible for moving things from A to B,” but now, companies are finally realizing that software is cheaper.
If Middle Management Truly Thins Out, Who Will Grow Up to Be the Next Generation of Leaders?
I sometimes wonder—everyone is discussing “who gets laid off,” but very few discuss another side effect: if middle management is squeezed too thin, at what layer will young people learn how to be leaders?
Although the middle management position is annoying, burdened with too many meetings and a lot of scapegoating, it was originally the place where many people first learned how to coordinate conflicts, allocate resources, and shoulder results. You might complain that this layer is inefficient, but if it gets yanked out completely, will the organization become thick at the ends and hollow in the middle? Senior executives issue tasks directly, and grassroots employees take system instructions directly. Efficiency looks higher, but the people who actually know how to lead teams, make judgments, and make decisions in gray areas will likely become fewer and fewer.
Or maybe I’m overthinking it. Perhaps in the future, a new kind of “light middle management” will genuinely emerge—fewer in number but with more concrete power, no longer responsible for passing messages, but dedicated to handling exceptions, conflicts, and edge cases in decision-making that systems cannot resolve. In that scenario, middle management doesn’t disappear; instead, it evolves from process administrators into exception handlers.
Only, this transition won’t be very gentle. Companies won’t cultivate you into the latter before eliminating the former. The more common script is that the former will be bumped out by the system first, and whether you can grow into the latter will be entirely up to you.
The image is a bit literal, but the meaning is accurate: the layer that gets removed is often not the “job title,” but the organization’s default assumption that humans are needed to relay information.
I am now more inclined to interpret this wave of changes as a very cold reminder: stop tying your value to “I know the process, I can follow up, I can report.” The system will soon be far more stable at this than you are. The ones who can truly stay behind will have to be the people who can still make decisions when information is incomplete and goals are clashing.
My coffee has already gone cold. But I actually have a question I’d love to hear your thoughts on: if your direct supervisor turned into an AI system that automatically dispatches orders, follows up on progress, and prioritizes tasks, what would you lose first—your sense of security, or your bargaining power?
References:
- Quinnipiac University Poll, April 16, 2025
- Poll: Americans Worry AI Will Cut Jobs, But Not Theirs
- What happens to middle management when AI flattens your organization?
- 30% of Americans Worry That AI Will Make Their Jobs Obsolete
—— Lyra Celest @ Turbulence τ.
