The pitch is seductive: AI agents will replace middle managers, flatten corporate hierarchies, and let individual contributors run teams of bots instead of people. Business Insider calls it the “Great Flattening,” a coming restructuring where AI handles coordination, scheduling, reporting, and oversight tasks that currently justify thousands of management positions. It’s a bold claim. It’s also mostly wrong — at least on the timeline and scale being promised.
Let’s start with what the argument actually says. The thesis, advanced by consultants and AI vendors with obvious incentives, holds that agentic AI — software that can autonomously execute multi-step tasks — will absorb the coordination work that defines middle management. Instead of a team lead assigning tasks, tracking progress, and synthesizing status reports, an AI agent handles all of it. The human worker talks to the agent. The agent talks to other agents. Layers disappear.
There’s a kernel of truth here. But the gap between automating a status report and replacing the person who decides what the team should build next is enormous.
McKinsey estimated in mid-2024 that about 60% of management activities could theoretically be automated with current technology, but that only 25% of those activities would be cost-effective to automate within five years, as reported by McKinsey Global Institute. That’s a meaningful distinction the hype cycle consistently ignores. Can an AI draft a project timeline? Yes. Can it decide whether to kill a project that’s politically sensitive but strategically dead? Not without a human who understands the organizational dynamics that no training dataset captures.
The companies actually deploying AI agents at scale tell a more nuanced story. Klarna, the Swedish fintech, announced in early 2024 that its AI assistant was doing the work of 700 customer service agents. By late 2024, as Wired reported, the company acknowledged it still needed humans for complex cases and had begun rehiring for certain roles. The initial numbers were real. So was the correction.
Same pattern at other firms. Salesforce launched its Agentforce platform with considerable fanfare, promising autonomous AI agents for sales, service, and marketing functions. CEO Marc Benioff told analysts on the company’s Q3 2025 earnings call that adoption was accelerating. But Reuters noted that enterprise customers were still in pilot phases, with full deployment timelines stretching well into 2026 and beyond. Pilots aren’t transformations.
And here’s the structural problem nobody selling this vision wants to discuss: middle managers don’t just coordinate. They translate. They absorb ambiguity from senior leadership and convert it into actionable direction for teams. They handle the employee who’s disengaged, the cross-functional dispute over priorities, the client relationship that’s souring for reasons that don’t appear in any CRM. These are judgment-intensive, context-heavy tasks that current AI systems handle poorly.
A February 2025 survey by Gartner found that 64% of enterprise leaders said their biggest barrier to AI agent deployment wasn’t technology — it was organizational trust. Employees didn’t trust AI agents to make decisions affecting their work. Managers didn’t trust agents to escalate the right problems. Gartner concluded that the “trust deficit” would slow adoption far more than technical limitations.
That finding aligns with what’s happening on the ground. When Microsoft rolled out Copilot across its enterprise customers, internal data showed productivity gains of 14% on specific tasks like email drafting and meeting summarization, according to Microsoft’s Work Trend Index. Those are real, measurable improvements. They’re also incremental. Nobody eliminated a management layer because emails got written faster.
The “Great Flattening” narrative also ignores history. Companies have been flattening hierarchies since the 1990s, driven first by business process reengineering, then by enterprise software, then by cloud-based collaboration tools. Each wave produced genuine efficiency gains. None eliminated middle management as a function. Spans of control widened — managers oversaw more people — but the role persisted because organizations kept discovering they needed someone accountable for translating strategy into execution.
So what will actually happen? AI agents will absorb administrative overhead. Some management positions will disappear, particularly in organizations with bloated reporting structures. The managers who survive will oversee hybrid teams of humans and AI agents, spending less time on coordination and more on judgment calls. That’s meaningful change. It’s not an organizational revolution.
The numbers suggest a more measured trajectory. According to the Bureau of Labor Statistics, the U.S. has roughly 8.4 million management occupations. Even aggressive estimates from firms like Forrester project that AI could displace 10-15% of these roles by 2030 — significant in absolute terms, but far from the wholesale elimination the “flattening” thesis implies.
There’s also a creation effect that gets systematically undercounted. Someone has to design, train, monitor, and correct AI agents. Someone has to decide which processes are appropriate for automation and which aren’t. Early evidence from companies like ServiceNow and Workday suggests these “agent management” roles are being filled by — wait for it — former middle managers who understand the workflows being automated.
The vendor community has every incentive to oversell this. Salesforce, Microsoft, Google, and a constellation of startups are competing for enterprise AI budgets that IDC projects will exceed $630 billion globally by 2028. “AI will replace your managers” is a better sales pitch than “AI will make your managers slightly more efficient.” But the second version is closer to the truth.
None of this means AI agents are trivial. They aren’t. The ability to automate multi-step workflows, synthesize information across systems, and handle routine decisions without human intervention represents genuine progress. Companies that deploy them well will gain real advantages in speed and cost.
But flattening entire organizational structures? That requires AI systems that can handle ambiguity, politics, morale, and strategic judgment — capabilities that remain years away, if they’re achievable at all. The managers reading this shouldn’t panic. They should learn to work with AI agents, because that hybrid model is what’s actually coming. Not a great flattening. A gradual reshaping. Less dramatic, more real.
WebProNews is an iEntry Publication