The organizational structures we’ve built over the past century weren’t designed for efficiency. They were designed for control — and that distinction is about to cost a lot of people their jobs.
Part I: The Promise That Never Arrived
Every major technology wave of the last fifty years arrived with the same promise: this one will cut the red tape. The personal computer was supposed to eliminate the need for large administrative staffs. Email was supposed to kill the memo. Enterprise software was supposed to make reporting seamless. The cloud was supposed to flatten everything.
None of it did. If anything, bureaucracy metastasized. Corporate hierarchies grew taller. Middle management bloomed into its own professional class. The organizational chart at a Fortune 500 company in 2020 often looked more complex than one from 1980 — more layers, more titles, more sign-off chains, more meetings to discuss meetings.
That paradox has a simple explanation: prior technologies made individuals more productive, but they didn’t change the underlying architecture of coordination. An email system makes a middle manager’s job faster, but it doesn’t make the middle manager unnecessary. A CRM makes a sales director’s reporting easier, but it doesn’t eliminate the need for someone to translate strategy into territory plans.
AI is different. And the difference isn’t incremental — it’s structural.
For the first time, a technology is able to perform the coordination function itself — synthesizing information from disparate sources, routing it to decision-makers, tracking execution, flagging exceptions, and generating reports — all without a human intermediary. The middle manager’s core job isn’t just assisted by AI. In many organizations, it’s being replaced by it.
This guide is about what that shift actually means: for organizations, for careers, for the flow of information, and for the fundamental question of who gets to make decisions at work.
Part II: Why Bureaucracy Existed in the First Place
Before we can understand why AI is dismantling bureaucracy, we need to understand why bureaucracy was built. It wasn’t an accident, and it wasn’t malice. It was a rational response to a very real problem: the cost of moving information through large organizations.
Consider a manufacturing company in 1955 with 10,000 employees spread across twelve plants in eight states. The CEO cannot personally know what’s happening on every production floor. The plant floor workers cannot personally understand the company’s strategic direction. Between these two extremes, a layer of humans had to exist — collecting data from below, packaging it for consumption above, and translating directives back downward.
This is what organizational theorists call the information bottleneck problem, and it is the foundational reason hierarchies exist. Managers were, at their core, information routers. They aggregated context from their teams, filtered out noise, and transmitted what mattered to the level above. They received strategy from above, decomposed it into actionable tasks, and transmitted it downward.
That’s not a trivial function. In a world without reliable communication technology, without real-time data systems, and without any ability to synthesize information across thousands of employees simultaneously, you needed human processors at every layer. The organization chart was, essentially, a network architecture diagram — a map of how information moved through a company.
The span of control — how many direct reports any one manager could effectively oversee — was the hardware constraint of this system. Research going back to Graicunas’s early studies in the 1930s suggested that a manager could effectively coordinate between five and eight direct reports.
More than that, and communication complexity grew exponentially. So organizations built tiers: supervisors managed workers, managers managed supervisors, directors managed managers, VPs managed directors, and so on.
By the time you hit eight layers between the CEO and the front-line worker — which was not uncommon at large corporations — the information flowing through that chain had been filtered, translated, summarized, and re-summarized multiple times. Each handoff introduced latency. Each layer introduced political distortion. The telephone game was built into the organizational structure itself.
The deeper problem was that information wasn’t just transmitted through these hierarchies — it was hoarded by them. In a bureaucratic organization, information is power. The person who controls what the executive hears controls the executive’s decisions. Middle managers, consciously or not, became gatekeepers — deciding what data went up, shaping narratives, protecting their domains, and building fiefdoms.
This wasn’t always malicious. Often it was simply human nature operating rationally inside an incentive structure that rewarded information control.
Part III: The Middle Manager as Information Broker
Ask most executives what middle managers actually do, and you’ll get answers about mentorship, development, team culture, and strategy execution. These are real and valuable functions. But they’re not the primary driver of the middle management layer’s size and cost.
The primary driver — the reason large organizations employ so many people in managerial roles — is coordination. Someone has to track who’s doing what, ensure dependencies are managed, flag blockers, write status updates, attend alignment meetings, and communicate decisions across the org.
Chris Williams, a former VP of HR at Microsoft, describes the core function precisely: “A huge portion of what middle management is, is translating requirements from the vague to the specific. And deciding what is noise and what is not.” That filtration process — deciding what reaches your team and what doesn’t — is the central job of the middle management layer.
A Harvard Business School study led by professor Manuel Hoffmann offers some of the most rigorous data on what happens when AI starts absorbing this coordination load. Studying 50,032 global software developers over two years, with half using GitHub Copilot and half not, Hoffmann’s team found that access to AI caused a measurable shift in how work got done: coding activities as a percentage of all work increased by 5%, while project management activities dropped by 10%. Developers using AI were less likely to ask managers or peers for help, worked in smaller groups, and moved faster.
The interpretation is striking: AI allowed individual contributors to work more autonomously, reducing the coordination overhead that justified managerial oversight in the first place.
This is the theoretical mechanism made empirically visible. When individuals can answer their own questions — by querying an AI that has been trained on the company’s documentation, code, processes, and past decisions — they stop needing a human layer to filter and route that information.
When an AI agent can automatically flag a blocker and cross-reference a dependency, the status meeting becomes redundant. When a generative AI can draft the board update, the director of communications has a very different job.
Part IV: The Great Flattening Is Already Happening
This isn’t theory. It’s showing up in layoff announcements, org chart restructurings, and executive memos at the world’s largest companies — and it’s accelerating fast.
In October 2025, Amazon announced it would cut roughly 14,000 corporate positions — about 4% of its white-collar workforce — specifically to “reduce bureaucracy” and “remove organizational layers.” This was not presented as a cost-cutting measure driven by poor performance. Amazon’s generative AI investments were on the rise. The layoffs were reframed as an organizational upgrade: fewer layers, more ownership, faster decisions.
CEO Andy Jassy had been telegraphing this for months. Earlier in 2025, he told employees directly that Amazon “will need fewer people doing some of the jobs that are being done today,” specifically citing generative AI’s growing role in planning, analytics, and forecasting. The Washington Post reported that Amazon was becoming the latest business giant “to tout thinner ranks and flattened hierarchies, and lean into innovations unleashed by artificial intelligence.”
Then, in February 2026, the thesis got its most explicit articulation. Jack Dorsey’s company, Block, cut approximately 4,000 employees — about 40% of its workforce — and Dorsey followed up with a co-authored essay alongside Sequoia Capital’s Roelof Botha titled “From Hierarchy to Intelligence.” The message was unambiguous.
As Fortune reported, Dorsey and Botha argued that corporate hierarchies had always existed to solve a single problem: routing information through organizations too large for any one person to oversee. Managers aggregate context from below, translate strategy from above, and maintain alignment across teams. AI, they argued, can now perform those functions continuously and at scale — making the human intermediary structurally redundant.
“There is no need for a permanent middle management layer,” the essay stated. Their proposed org structure consisted of three roles: individual contributors who build and execute, senior player-coaches who stay hands-on while developing people, and AI systems that handle the coordination layer formerly occupied by middle management.
This is not an isolated CEO’s eccentric vision. It is the direction in which an entire generation of companies is moving — and the data back it up.
According to a Gartner report from October 2024, by 2026, one in five companies will use AI to flatten their organizational structures, eliminating more than half of the middle management positions that currently exist. A Korn Ferry Workforce 2025 survey of 15,000 professionals found that 41% of workers said their employers had already reduced the number of management tiers. Middle managers made up 29% of all layoffs in 2024 — a remarkable figure given they represent a much smaller share of overall headcount.
Fortune’s August 2025 report documented this vividly: at Moderna, the technology and human resources departments were merged into a single function under one Chief People and Digital Technology Officer, with AI handling HR support and some junior roles. At McKinsey, thousands of AI agents are deployed to support consultants — building decks, summarizing research, verifying logic. At one healthcare company, a 10-person software development team was replaced by a three-person unit: a product owner, a software engineer who prompts AI coding tools, and a systems architect.
Rob Levin, a McKinsey senior partner, put the ratio bluntly: “50 to 100 AI agents can be managed by just two or three people.”
Part V: How AI Rewires the Flow of Information
The organizational restructuring described above is just the visible tip. The deeper transformation is happening in how information itself moves through organizations — and this is where the changes are most profound, and most permanent.
In a traditional hierarchy, information flows through people. A customer complaint reaches a front-line agent, gets escalated to a team lead, gets summarized in a weekly report to the manager, gets aggregated in a monthly update to the director, gets packaged into a quarterly review for the VP, and might — might — reach an executive who can actually change the product or policy that caused the complaint in the first place. The journey from signal to action could take weeks or months. By the time the insight reached someone with the authority to act on it, it was already stale.
More insidiously, information changed shape with every handoff. Facts became opinions. Opinions became summaries. Summaries became bullet points. The nuance that lived in the original customer’s words — the specific frustration, the workaround they’d developed, the feature they were asking for — was progressively compressed and distorted as it rose through the hierarchy. This isn’t a failure of individual managers; it’s the inherent physics of human-mediated information relay.
AI collapses this entire chain.
When a customer interaction is logged, an AI can — in real time — categorize it, cross-reference it against ten thousand similar interactions, identify the product defect pattern, surface the relevant engineering ticket, and push a synthesized insight directly to the product manager with the authority to fix it. The signal-to-action journey that once took months takes minutes. The distortion introduced by seven layers of human translation is eliminated.
This is what Dorsey and Botha meant by a company’s “world model” — an AI-maintained, continuously updated picture of everything happening inside and outside the organization. As CoinDesk reported, Block’s model aggregates internal data from code, decisions, workflows, and performance metrics, while a second model maps customer and merchant behavior using real-time transaction data. Together, they replace the contextual awareness that managers traditionally carried in their heads.
The implications for how decisions get made are enormous. In bureaucratic organizations, decision rights are tightly coupled to information access. You can only decide on something if you know about it — and you only know about it if someone in the hierarchy deemed it worth reporting to you. This creates a structural bias toward decisions being made by people who are far from the front lines, working from compressed and filtered information, often weeks after the relevant events occurred.
In an AI-mediated organization, information access can be radically democratized. Every employee, in principle, can query the same underlying data. The front-line worker who notices that customers in a particular region are churning at twice the average rate doesn’t need to write a memo, wait for approval, and watch it die in someone’s inbox. They can flag it directly, with the supporting data surfaced by AI, in a channel where decision-makers are paying attention.
This is what Andrej Karpathy — one of the most respected voices in AI — was gesturing toward in his recently published LLM Wiki concept on GitHub. Rather than using AI merely to retrieve information from document stores at query time (the standard RAG approach), Karpathy describes a model where AI actively builds and maintains a persistent, interlinked knowledge base — continuously integrating new information, flagging contradictions, cross-referencing entities, and keeping the organizational picture current.
The key insight is this: “The tedious part of maintaining a knowledge base is not the reading or the thinking — it’s the bookkeeping.” Updating cross-references, noting when new data contradicts old claims, maintaining consistency across dozens of documents — these are tasks humans abandon because the maintenance burden grows faster than the value. AI doesn’t get bored and doesn’t forget to update a cross-reference. Applied to organizational knowledge, this means the institutional memory that once lived in middle managers’ heads — the context, the history, the relationships — can now be maintained and made accessible to everyone.
The consequences for organizational politics are equally dramatic. In traditional hierarchies, information is hoarded because information is power. The manager who controls the weekly status report controls what the VP believes about the state of the project. The director who owns the data controls the narrative. AI-driven radical transparency doesn’t just make information flow faster — it breaks the gatekeeping model entirely. When everyone in the organization can access the same real-time operational picture, the politics of information control become structurally impossible.
Part VI: What Gets Automated, Layer by Layer
The transition isn’t happening all at once. It’s happening function by function, process by process. Here’s how it’s unfolding across the specific workflows that once justified middle management:
Status reporting and updates. This is the lowest-hanging fruit and the first to go. Generative AI systems can now synthesize updates from project management tools, calendars, code commits, and communication channels into coherent status reports — written in plain English, tailored to the audience, delivered automatically. The manager who spent two hours a week collecting, formatting, and sending status updates no longer needs to.
Approvals and sign-offs. Many routine approvals — time-off requests, expense reports, small purchase orders, access requests — involve applying a rule set to a context. AI can do this faster and more consistently than humans. Amazon’s stated goal of increasing the ratio of individual contributors to managers by 15% is largely enabled by automating these approval workflows.
Performance monitoring and reporting. Rather than a manager manually tracking team metrics, consolidating data from multiple systems, and writing quarterly performance summaries, AI tools can do this continuously. This doesn’t just save time — it changes the nature of the manager’s engagement, shifting it from data collection to interpretation and coaching.
Coordination and scheduling. One of the most labor-intensive aspects of middle management is simply orchestrating the work — who needs to talk to whom, which dependencies need to be resolved, which meetings need to happen. AI scheduling and workflow tools are absorbing this function progressively, routing dependencies automatically and surfacing blockers without requiring a human coordinator.
Customer escalation routing. The traditional escalation path — front-line agent to team lead to specialist to manager — exists because no single agent has the context or authority to resolve complex issues. AI agents can now handle many escalations end-to-end, querying relevant policy, surfacing the customer’s history, and resolving the issue without routing it through multiple human layers.
Budget forecasting and resource allocation. Finance teams in large organizations spend enormous effort producing quarterly forecasts, variance analyses, and budget reports. AI-driven financial planning tools are increasingly handling the heavy lifting of data aggregation and model generation, with human analysts focused on interpretation and decision-making rather than number-crunching.
Edwige Sacco, the Head of Workforce Innovation at KPMG, described the KPMG experience precisely in her HBR interview: “Because of AI, an associate should be able to show up prepared for a meeting or a strategic conversation the way a manager or above used to.” Tasks that required a middle manager’s level of synthesis and preparation are now achievable by individual contributors.
Part VII: The New Organizational Structure
If the middle management layer is being absorbed by AI, what does the resulting organization look like?
The emerging answer, visible in companies that are furthest along in this transition, is a radical flattening combined with an expansion of two remaining human layers: the highly autonomous individual contributor and the senior orchestrator.
Individual contributors become more powerful. When AI handles coordination, status updates, scheduling, and information routing, the individual contributor is freed to focus on actual work. More importantly, they’re empowered to make more decisions independently. They have access to the same organizational data that was previously siloed at the manager level. They can query AI for context, precedent, and policy. They can act without waiting for approval chains that no longer need to exist.
Senior orchestrators replace middle managers. The layer above individual contributors doesn’t disappear — it changes. The “player-coach” model that Dorsey and Botha describe in their essay is one formulation: senior people who do real work themselves (coding, selling, building) while also developing the people around them. The pure management role — spending 100% of your time in meetings, writing reports, and managing up — becomes harder to justify when AI handles the administrative and coordination functions that justified it.
Nick South of Boston Consulting Group articulates this well: the orchestration layer “will need to be bigger than it is today” in terms of skill complexity, even if smaller in headcount. What these people need is “a combination of sufficient AI proficiency to manage a human-agentic workforce, plus the core skills of logic, understanding of ethics, rhetoric, and communication skills.”
Departments blur and merge. As AI breaks down information silos, the rationale for strict departmental separation weakens. Moderna’s merger of HR and technology under a single executive is one example. Fortune’s reporting notes that strict divisions between teams are starting to blur as AI automates tasks that once required handoffs between functions. The cross-functional team — once a special project structure — becomes the default mode.
Span of control explodes. The fundamental constraint of the old hierarchy — that one manager could effectively oversee only five to eight people — was driven by the communication overhead of coordination. When AI handles coordination, that constraint loosens dramatically. Meta’s AI team operates at a 50-to-one employee-to-manager ratio. One senior engineer can supervise and coach fifty people when AI is generating the status summaries, flagging issues, and routing decisions that previously required a human manager’s full-time attention.
Part VIII: The Human Cost and the Real Risks
It would be dishonest to write about this transition as pure progress. The speed and scale of what’s happening carries significant risks — some already visible, some coming into focus.
The knowledge drain problem. Middle managers are not just coordinators. They are repositories of institutional context — the reasons decisions were made, the approaches that failed, the relationships that matter across departments. This knowledge rarely lives in documentation. It lives in people. As researchers at People Managing People note, “when managers leave, their networks dissolve.
The informal relationships that helped navigate bureaucracy, resolve conflicts, and move projects forward disappear.” Organizations that cut aggressively before building systems to capture and transfer this knowledge will pay a price they don’t fully feel for years.
The career pipeline problem. Traditional organizational structures provided a development path: individual contributor → team lead → manager → director → VP. Each transition built different skills. Remove the middle rungs, and the progression breaks. Korn Ferry’s 2025 survey found that 37% of employees who’d experienced management flattening felt directionless — not freed or empowered, but aimless.
The question Kate Barney, Chief People Officer at Smartly, poses cuts deep: “If you have a bunch of experts running around doing stuff, are they going to have the time and patience to be mentoring entry-level college grads?” You cannot skip the developmental stages that create future senior leadership — you just displace the cost into the future.
The surveillance risk. Radical information transparency cuts both ways. An organization where AI continuously monitors every employee’s output, communication patterns, and project contributions is also an organization with unprecedented surveillance infrastructure. The efficiency gains from real-time visibility can easily shade into micromanagement by algorithm — where workers feel watched rather than empowered.
The difference between AI-enabled radical transparency and AI-enabled panopticon management is primarily cultural, and cultures under financial pressure tend to drift toward control.
The concentration of power problem. Flat hierarchies don’t distribute power — they often concentrate it. When you remove five layers between the CEO and the worker, the CEO’s direct influence over organizational culture and decision-making expands dramatically. The middle management layer, for all its inefficiencies, also served as a buffer — absorbing and distributing the impact of executive decisions, and providing a counter-weight to top-down authority. Remove it, and the power dynamic tilts sharply upward.
AI bureaucracy replacing human bureaucracy. History suggests that bureaucracy is self-replicating. Prior waves of organizational change didn’t eliminate bureaucracy — they transformed it. There is already significant evidence that as organizations flatten their management layers, new forms of process-overhead emerge: AI governance frameworks, compliance reviews for automated decision systems, committees to oversee algorithmic performance management. It is plausible — perhaps likely — that AI eliminates one kind of bureaucracy while generating another.
Part IX: The Roles That Survive — and Those That Thrive
Not every function of the middle manager is automatable. Several remain deeply, irreducibly human:
Judgment under genuine ambiguity. AI systems are excellent at applying rules to situations that resemble training data. They are poor at navigating situations that are genuinely novel — where the right answer doesn’t exist in any playbook and requires integrating values, context, and judgment in ways that can’t be reduced to a decision tree. Senior leaders who can navigate that ambiguity become more valuable, not less.
Trust and relationship-building. The KPMG Head of Workforce Innovation stated plainly: “AI can’t do one of the most critical aspects of their job. They know how to speak to our clients. They can read between the lines, and help others articulate a problem.” External relationships — with clients, partners, regulators, and communities — depend on trust that is built through sustained human presence. No AI agent has yet demonstrated the ability to navigate a tense client relationship or rebuild trust after a serious failure the way a skilled human can.
Culture-building and psychological safety. Professor Eva Selenko of Loughborough Business School makes the point simply: “It’s not sustainable to just have a boss as an algorithm, it’s not going to work in the longer term.” Employees still need to be managed with empathy. Teams still need someone who notices when morale is tanking, when a key person is burning out, or when interpersonal conflict is about to derail a critical project. These are signals that dashboards don’t capture.
Creativity and strategic vision. AI can summarize what has happened and model what might happen. It cannot independently imagine what should happen. The executives who define organizational direction — who see market shifts before they’re visible in data, who make bets that require conviction rather than probability — become more central as the analytical and coordinative work gets automated below them.
The new titles emerging from this transition reflect these remaining human functions: AI Workflow Lead, Organizational Architect, Human-AI Integration Specialist, Chief AI Officer. LinkedIn’s research showed that companies with a “Head of AI” position had tripled in five years. These aren’t just rebrands — they reflect genuinely new roles at the intersection of organizational design and AI deployment.
Part X: Lessons from the Industries Furthest Along
Technology. The sector has moved fastest, partly because the tools were developed there. Google eliminated 35% of managers overseeing small teams in 2025. Meta built AI teams operating at 50-to-one ratios. The pattern is consistent: fewer layers, larger spans of control, AI handling the coordination overhead.
Consulting and professional services. McKinsey deploying thousands of AI agents to build decks and verify logic is a remarkable statement from an industry that has historically sold human expertise. Fortune’s reporting notes that 40% of McKinsey’s revenue now comes from advising on AI and related technologies — the firm is, in part, monetizing its own transformation.
Healthcare administration. Healthcare has some of the most bloated administrative structures in any industry, with administrative overhead consuming roughly 34% of total hospital costs in the US. AI is beginning to compress this: automated prior authorizations, AI-assisted clinical documentation, intelligent scheduling, real-time compliance monitoring. The administrative reduction potential here is enormous.
Financial services. Compliance, risk reporting, and advisory layers — all historically heavily staffed with middle managers — are being compressed by AI. Firms are deploying AI models that can generate regulatory reports, monitor transactions in real time, and surface anomalies without requiring human reviewers at every step of the process.
Part XI: The Counterargument — Will Bureaucracy Just Evolve?
Intellectual honesty requires engaging with the strongest version of the skeptical case.
The history of organizational technology is a history of false promises. Every major tool introduced to reduce complexity has eventually been captured by organizational inertia and redeployed as a new source of it. Email created more communication overhead than it eliminated. CRMs created new reporting burdens. Even the flat organization structure of the startup world — heralded as the antidote to corporate hierarchy — tends to accumulate informal power structures and process layers as it scales.
There are several concrete reasons to believe AI might follow the same path:
AI governance generates its own bureaucracy. As AI systems take on more consequential decisions — hiring, firing, credit approval, medical recommendations — regulatory and internal oversight requirements grow. Someone has to audit the models. Someone has to review the edge cases. Someone has to sign off on model deployments. The “AI oversight” function is nascent today, but it’s growing, and it has all the hallmarks of a bureaucratic layer in formation.
Political resistance is real and rational. Those with power in organizations — particularly senior managers and executives who benefit from the current information hierarchy — have strong incentives to resist the kind of radical transparency that AI could enable. “If those at the top’s job might be partly doable by an AI, they will not give up that power,” as Professor Selenko observes. The flattening will happen where executive leadership champions it. Where it doesn’t, new layers will accumulate around AI systems rather than replacing the old ones.
Coordination costs don’t disappear — they shift. Current and former Block employees told The Guardian that roughly 95% of AI-generated code changes still require human modification, and that AI tools cannot yet lead in regulated areas like banking and money transfers. The gap between theoretical capability and practical deployment is still significant. Organizations that cut coordination layers before AI is genuinely ready to fill them don’t get flat, agile structures. They get chaotic ones.
Part XII: The Permission Economy Is Ending
There is a concept worth naming explicitly, because it captures what is fundamentally changing: the permission economy.
For most of the history of modern work, a large fraction of organizational energy was consumed not in doing work but in gaining permission to do work. Approval chains, budget sign-offs, legal reviews, manager sign-offs, leadership alignment meetings — these are all mechanisms by which permission was dispensed downward through hierarchies. The bureaucratic organization was, at its core, a permission-allocation machine.
The permission economy wasn’t just inefficient — it was philosophically corrosive. It communicated, continuously and systemically, that workers could not be trusted to act without authorization. It made the person holding the approval stamp more powerful than the person doing the work. It privileged caution and process over speed and judgment. It selected for organizational survivors who were good at navigating permission structures, not necessarily for people who were good at doing the actual work.
AI dismantles the permission economy by collapsing the information asymmetry that justified it. Approvals existed because the approver was supposed to know more than the requester — more about risk, policy, resource availability, organizational priorities. When AI makes that information available to everyone, the approval becomes a formality. And when automating the approval becomes cheaper and more reliable than routing it through a human, the entire chain becomes redundant.
What replaces the permission economy is something closer to an accountability economy — where workers are empowered to act with greater autonomy, but where AI creates transparent, real-time visibility into outcomes. You don’t need to ask permission for every decision when your decisions are immediately visible in organizational data and reviewable by anyone. The accountability is built into the architecture, not into the approval chain.
Conclusion: What This Means for You
The organizations that will thrive in the next decade are those that grasp the fundamental nature of what AI is doing: not just automating tasks but replacing coordination infrastructure. The companies that are restructuring now — Amazon, Block, Meta, McKinsey, Moderna — are not simply cutting costs. They are rebuilding their organizational nervous systems around AI as the default information router and coordinator.
If you are a worker in a coordination-heavy role — managing status updates, running alignment meetings, maintaining reporting structures, filtering information between organizational layers — you should take this transition seriously. The question is not whether these functions will be automated, but how fast, and whether your organization will help you migrate to higher-value work before the transition hits.
If you are a leader, the most important question you face is not “how do we implement AI?” but “what does our organization actually look like once AI handles coordination?” The answer to that question requires rethinking spans of control, career development pathways, information governance, and culture — not just deploying a new set of tools on top of the existing hierarchy.
And if you are a policymaker or an economist, the shift described in this guide deserves serious attention. Middle management is not the bottom of the labor market — it is the middle. These are well-educated, well-compensated workers whose skills were developed in service of an organizational architecture that is becoming obsolete. Bloomberg and Live Data Technologies found that middle managers made up a third of all corporate layoffs in 2023 — and that was before the current wave accelerated. The reskilling and transition challenge is real, large, and not yet adequately addressed by any public policy framework.
The bureaucratic organization was not a mistake. It was the best available architecture for coordinating human effort at scale, given the information technology of the 20th century. AI is not correcting bureaucracy — it is obsoleting the conditions that created it. The result will be organizations that are faster, flatter, and more transparent than anything we’ve previously built at scale.
Whether they will also be more equitable, more humane, and more innovative is not determined by the technology. It will be determined by the choices leaders make in the next several years — about what to automate, what to preserve, and who gets to benefit from the efficiency that the flattening creates.
The org chart is being redrawn. The question is who holds the pen.
