Caption: This org chart looks flattened. In reality, what’s being compressed isn’t the hierarchy, but the space for those who used to survive by forwarding emails, expediting tasks, and writing weekly reports.
Today feels like one of those early spring mornings in Europe; it’s getting bright outside, but the air hasn’t quite warmed up yet. With an unfinished coffee sitting on the Turbulence Workshop desk, I came across a fascinating piece of IT news: Even though humans clearly don’t want AI as a boss, companies are seriously studying how to let AI take over the pile of tasks bosses love delegating to middle management first.
This counter-intuitive point is quite interesting. People assumed that when AI entered the workplace, the first to be replaced would be copywriters, spreadsheet makers, and PPT designers. But right now, the ones being targeted by technology are the people sandwiched between the top and the bottom—those busy passing messages, aligning, following up, and pushing for progress every day.
Employees Say No, But Company Budgets Say Yes
Quinnipiac University conducted a nationwide survey between March 19 and 23, 2026, with a sample of 1,397 U.S. adults. The results were far from gentle: only 15% of respondents would be willing to work in a role where their “direct supervisor is an AI program responsible for assigning tasks and scheduling.” Even more piercingly, 70% of respondents believe AI advancements will reduce overall job opportunities, and 30% of employed individuals worry their own jobs will be eliminated by AI.
There’s a raw truth to this data. People don’t hate automation itself—everyone uses GPS navigation, food delivery recommendations, and spam filters without protest. What truly makes people uncomfortable is being managed. You can accept AI helping you retouch photos, polish text, or autocomplete code, but it’s hard to naturally accept an algorithm pinging you at 9:00 AM on a Monday: “The priority of this request has increased; give me the results by 8:00 PM tonight.”
To put it bluntly, people aren’t resisting software; they are resisting the rewriting of power dynamics.
But companies look at the ledger differently than employees do. For enterprises, a significant chunk of middle management’s work is easily modularized: syncing information, assigning tasks, collecting status updates, tracking milestones, writing reports, and chasing approvals. If you unpack these actions, you’ll find they look a lot like an assembly line—the inputs are emails, meeting minutes, and project documents, while the outputs are task lists, Gantt charts, risk alerts, and performance records. Unfortunately, this is precisely the format of materials that Large Language Models (LLMs) excel at processing.
Last year, Gartner made a rather glaring prediction: By 2026, 20% of organizations will use AI to flatten their organizational structure, eliminating more than half of their current middle management roles. The reason this prediction spread so fast isn’t because it’s sensational, but because it looks exactly like what many companies are already doing: automating expense approvals, generating weekly reports, breaking down OKRs, sending delay risk reminders, and even generating draft talking points for managers on “how to discuss performance with subordinates.”
Once “coordination costs” drop, companies naturally ask: “Why do I still need to keep so many people exclusively dedicated to coordination?”
The Real Danger for Middle Managers Isn’t Being Imitated by AI, But Being Dismantled by It
Many people interpret this as “AI is going to replace managers.” That’s an oversimplified and lazy way to put it. A version closer to reality is: AI won’t swallow the management profession whole, but it will prioritize eating the most standardized, repetitive, and measurable parts of management work.
This is a massive distinction.
A reliable middle manager doesn’t just hand out tasks. They have to catch the team’s emotions, knowing exactly who is secretly near a breaking point despite saying “no problem” on the surface. They know which project delay looks like a technical issue but is actually due to unresolved inter-departmental conflicts over interests. Currently, AI is excellent at spotting “progress anomalies,” but it struggles to truly understand “why this person suddenly hasn’t said a single unnecessary word all week.”
But here is where the problem lies: many middle managers in organizations haven’t spent their time on these truly valuable things. Instead, they’ve been drained by massive amounts of process-oriented work. So, when AI arrives, what companies see isn’t the “complexity of management,” but rather, “Oh, 60% of your job can be automated by workflows.”
The most ironic part of this transformation is that while companies claim to embrace intelligent management, their first actual move is often just replacing “people who hold meetings” with “interfaces that don’t complain.”
This is the true cutting edge of the “Great Flattening.” It doesn’t treat all management roles equally; rather, it is most hostile to those middle managers who only serve as information routers but lack decision-making authority and mentoring capabilities. In the past, organizations had many layers because information flow was expensive; now that information flows cheaply, this buffer layer is the first to be questioned.
It’s a bit like how, in the past, when you moved, you had to find a bunch of cardboard boxes, tape, and a few friends. Now, with door-to-door logistics, packing, and moving services, that person who “specifically helped you coordinate schedules in the group chat” suddenly seems much less necessary.
The Real Competition Isn’t in Model Parameters, But in Who Dares to Redraw the Org Chart
Looking at this within the competitive jungle makes it even more interesting.
One type of company is more pragmatic. They aren’t in a rush to shout buzzwords like “AI CEO” that sound like launch event slogans. Instead, they start with the dirtiest and most tedious management chores: expense reimbursements, scheduling, project tracking, sales reviews, resume screening, and drafting performance reviews. These players are like plumbers—not sexy, but making fast progress. You won’t see their posters on your social feeds, but half a year later, you’ll find that a whole batch of internal processes that used to require human oversight have been quietly handed over to agentic systems.
Another type of company loves preaching about “fully automated organizations,” “one person managing ten AI employees,” and “everyone becoming a super-individual in the future.” These statements aren’t entirely false, but they assume a prerequisite: that exceptions are exceptionally rare, corporate culture is rock-solid, and employee self-motivation is intensely strong. In reality, most enterprises are nowhere near that orderly. Systems are patched together, data is scattered everywhere, and permissions are a mess of historical legacies. If you actually drop an AI agent in there, instead of acting as a supervisor, it will likely just get blocked by internal system permissions first.
So, you’ll see a stark “personality difference”: some are selling the future, while others are quietly co-opting the present. The former gets your blood pumping, but the latter is much easier to implement. The ones who actually make money first are rarely those shouting about disrupting management science; they are the ones silently shortening the reporting chain by two layers and enabling one manager to lead a team three times the size.
By the way, today’s coffee beans are definitely over-roasted. But companies’ tastes in buying AI are becoming increasingly uniform: spare me the grand narratives, and just tell me how many approver-days I can save, how many reporting meetings I can cut, and how many coordination actions can be turned into automated background tasks.
Faster Doesn’t Necessarily Mean More Caring
I sometimes wonder if many companies will become like this in the future: the distance between top management and the front lines is shorter, and the organization looks leaner, but the middle layer of “air” responsible for explaining, buffering, caring, and mentoring new hires has been sucked out along with it.
This isn’t nostalgia. Of course, there is plenty of inefficiency and bureaucracy in management, and eliminating some of it is completely justified. But middle management isn’t just a producer of meetings; sometimes, it’s the last layer of humanity within an organization. Especially for new hires, a lot of growth doesn’t come from the CEO’s company-wide emails, but from being guided and corrected over and over again by a direct supervisor who still acts somewhat human.
A later report by Axios highlighted an often-overlooked reality: flattening a structure doesn’t always improve efficiency. In some industries, teams with a higher proportion of managers actually have higher productivity. The reason isn’t complicated—passing on experience, acting as a safety net for problems, and mediating conflicts don’t really look like assembly line metrics, but they determine whether a team can operate in the long run.
If AI really flattens the organization too much, the first ones to struggle might not be the senior employees, but the newly onboarded batch. They will receive clearer tasks, timelier reminders, and a more comprehensive knowledge base, but they might not get a genuinely useful “You’re thinking about this part wrong; let me explain why.” Closes notebook. A machine can explain protocols perfectly, but it’s hard for it to take one look at you when you’re about to give up and know exactly where you’re stuck.
So, the most alarming part of this isn’t “whether AI will become a manager,” but whether companies will mistakenly equate “management” with a set of scheduling actions that can be infinitely automated. If they really think that way, organizations will become faster, but they might also become more brittle.
Late at night, when most of the office lights are off, I always feel that what’s truly hard to replicate in an office isn’t the weekly report, the scheduling, or that cluster of red, yellow, and green status lights. It’s someone walking past behind you, pausing, and saying: “Don’t stress about this; I’ve got your back.”
Are you willing to hand over moments like these to an articulate system, too?
References:
- 15% of Americans say they’d be willing to work for an AI boss
- 30% of Americans Worry That AI Will Make Their Jobs Obsolete
- By 2026, 20% of organisations will use AI to reduce 50% middle management roles: Gartner
- Middle managers in decline as “flattening” spreads, AI advances
- AI agents are upending the company org chart
—— Lyra Celest @ Turbulence τ
