In January 2026, Fiverr's CEO Micha Kaufman announced the company would cut staff to become "leaner" and "faster." The interesting part wasn't the layoffs. Tech companies cut headcount every quarter now, and nobody blinks. The interesting part was the specific language: "substantially greater productivity, and far fewer management layers."
Not fewer engineers. Not fewer designers. Fewer management layers. The coordination class.
A month earlier, Workday eliminated roughly 1,750 jobs. CEO Carl Eschenbach cited AI explicitly. Amazon restructured thousands of positions as part of what internal documents called "AI-enabled operational efficiency." The pattern across these announcements is consistent: the roles being cut are the ones that exist to move information between other people.
AI is not primarily replacing workers who make things or workers who decide things. It is replacing workers who coordinate things. That coordination layer, the middle of every organization chart, is where most corporate jobs actually live.
The coordination class exists because 20th century corporations needed humans to route information
The modern corporation is a coordination machine. Peter Drucker recognized this in the 1940s when he studied General Motors. The company's competitive advantage wasn't better cars. It was better coordination. GM could manage the flow of information, decisions, and materials across hundreds of thousands of people and dozens of factories in a way smaller competitors couldn't replicate.
That coordination required humans. Layers of humans whose job was to translate between levels of abstraction. The CEO says "increase market share in the Southeast." A VP translates that into "launch three new dealerships in Georgia and Florida by Q3." A director translates that into "find commercial real estate, hire staff, negotiate with suppliers." A manager converts it into task lists, timelines, and status reports.
This is what most middle management actually does. Not strategy. Not execution. Translation and routing.
Ronald Coase won the Nobel Prize for explaining why firms exist at all. His answer: transaction costs. It's cheaper to coordinate work inside a firm than to negotiate contracts in the open market. But Coase's framework has a corollary people quote less often: when coordination costs drop, firms shrink.
AI just dropped the cost of internal coordination through the floor.
Most of what managers do is information processing, and AI handles information processing well
Consider a typical product manager at a mid-size tech company. Their week includes synthesizing customer feedback from support tickets, sales calls, and surveys. Translating that synthesis into prioritized feature requests. Writing specs engineers can implement. Running standups. Reporting status to leadership. Coordinating between design, engineering, and marketing on timelines.
Every one of those tasks is information processing. Aggregation, synthesis, translation, routing, summarization. None require judgment that resists automation. They require context and pattern-matching, which is exactly what language models do well.
Here's the mechanism in slow motion. A language model ingests the same raw data a product manager would: ticket logs, call transcripts, survey responses. It identifies frequency patterns, clusters complaints by theme, cross-references against the product roadmap, and generates a prioritized list with confidence scores. The PM used to spend two days doing this. The model does it in four minutes. What remains for the human is a review pass, maybe thirty minutes of sanity-checking. The other day and a half of that PM's week just evaporated.
Gartner projects that 15% of day-to-day work decisions will be made autonomously through agentic AI by 2028. That number sounds modest until you realize "day-to-day work decisions" is a precise description of what middle managers do from 9 to 5. When Fiverr said "fewer management layers," they weren't being euphemistic. They tested agentic systems that handle workflow coordination, task assignment, and progress tracking across distributed teams. The AI doesn't manage people. It manages information flow. Managing information flow was what the managers were doing.
The org chart is flattening from a pyramid into a dumbbell
The traditional org chart is a pyramid. One CEO, a handful of executives, a larger band of directors, a much larger band of managers, and a base of individual contributors. Each layer exists because the layer above can't directly coordinate with the layer below at scale.
AI collapses this. An executive with access to an agent that continuously monitors project status, customer sentiment, financial metrics, and team productivity doesn't need three layers of humans to tell them what's happening. The information arrives pre-synthesized, in real time, with anomalies flagged.
A 2024 Harvard Business School study found that managers spend roughly 54% of their time on "administrative coordination" rather than strategic or interpersonal work. That 54% is the kill zone. Not because it's unimportant, but because it's exactly the kind of work AI handles well: scheduling, reporting, routing, tracking, summarizing.
Klarna's CEO Sebastian Siemiatkowski announced in late 2025 that the company had reduced its workforce from 5,000 to approximately 3,800 while growing revenue, explicitly crediting AI for absorbing coordination work. The remaining employees work more directly with leadership. Fewer intermediary layers. What emerges looks less like a pyramid and more like a dumbbell: decision-makers at the top, skilled individual contributors at the bottom, an AI layer in between.
The best managers are irreplaceable, but most managers aren't doing what the best managers do
The strongest counterargument: real coordination isn't information processing. It's relationship management. A good manager knows that Sarah on the engineering team is dealing with a personal crisis and needs lighter assignments this sprint. They know the sales team's forecast is optimistic because Dave always inflates his pipeline. They read body language. They sense when morale is slipping before anyone says anything.
This argument deserves respect because it's partly true. The best managers are relationship architects who create conditions for people to do their best work.
But this describes exceptional management, not typical management. The proof is in how most managers actually spend their days. That Harvard study again: 54% administrative coordination. The median manager is a process administrator with the title "manager." They run meetings that could be emails. They write status reports nobody reads carefully. They sit between leadership and execution because information systems were too primitive to connect those layers directly.
When the information system gets smart enough to handle the routing, the exceptional managers become more valuable. Organizations need fewer of them and can pay them more. The median manager, whose primary function is coordination rather than leadership, is in the same position as the switchboard operator when direct-dial phones arrived.
The honest answer about what happens to displaced coordinators is not reassuring
The standard response from tech optimists: "people will move to higher-value work." Historically true. ATMs replaced bank tellers' cash-handling duties; banks hired more tellers for relationship-based selling. Spreadsheets eliminated accounting departments; accountants moved into advisory roles.
But middle management displacement is different in one specific way. The "higher-value work" that managers could theoretically move into — strategy, relationship building, creative problem-solving — was already scarce relative to demand. You can't promote everyone into a strategy role. There aren't enough strategy problems.
The 2025 Challenger data tracked over 1.2 million US job cuts announced that year, with more than 31,000 tied directly to automation. That's the explicit number. The implicit number is much larger. Most companies cite "restructuring" rather than "we replaced your job with software."
The coordination class was a transitional phenomenon. It existed because 20th-century organizations needed humans to perform information processing that 21st-century systems handle automatically. The jobs were real. The work was necessary. The work is being absorbed by machines, and the jobs are following.
If you're in the middle, the time to move is before it collapses under you
Today: Honest self-assessment. What percentage of your week is information routing versus genuine judgment? If the answer is more than 50% routing, you're in the zone AI absorbs first.
This month: Move toward what remains human. Deep expertise in a specific domain. Relationships that can't be replicated by a system. The ability to make decisions under genuine ambiguity, where the right answer depends on values rather than data.
This quarter: Build skills that compound personally, not positionally. A management title is a position. The ability to read a room, negotiate a difficult conversation, or make a strategic bet under uncertainty is a skill. Positions get restructured. Skills travel.
If the coordination class is dissolving, the people who know how to coordinate without a team — solo operators, small studios, independent professionals with agent-augmented workflows — have a structural advantage. They never built the middle layer. They can't lose what they never had.
The middle is collapsing. Are you standing in the part that's falling, or the part that remains?
