What Companies Learn When They Fire All Their Managers

Nearly half of US companies eliminated management layers in the past year. The pitch was always the same: flatter, leaner, faster, powered by AI. Then people had to figure out how to actually work.

If your workplace has felt off since the restructuring, you're not imagining it. Something load-bearing got removed along with those manager titles. The question is what, exactly.

Nate B Jones, who studies AI strategy and organizational change, spent time examining three companies running different versions of this experiment at scale. What he found suggests the "flatten everything" approach misses something fundamental about what managers actually do.

The Three Jobs Hidden in One Title

Management, Jones argues, isn't a single function but three distinct jobs bundled together. The first is information routing—deciding who needs to know what and when. This goes back to Roman centurions and Prussian general staffs. It's logistics.

"If you have ever sat in a meeting and thought this could have been an email, you are suffering from bad routing management," Jones notes.

AI handles this piece well. An agent can scan 3,000 customer feedback items, interpret sentiment across languages, monitor competitors, and generate a requirements document before lunch. What used to take multiple people and multiple days happens in hours.

The second job is sensemaking. Not just moving information around, but understanding what it means. Why did the product ship late three times in six months, and why does it keep involving the same team? That requires context AI doesn't have—years of experience in a particular company, product area, or domain.

"The problem is not a shortage of information," Jones says. "The problem is a shortage of signal."

The third job is accountability and feedback. Someone has to own outcomes over time, develop attachment to work, and provide coaching that lands because it's timed right and comes from someone who understands the full picture.

AI can help with parts of this. It can't simulate the feeling of ownership that develops when a product manager tends the same territory for two years.

Three Companies, Three Approaches

Moonshot AI, the company behind the Kimi K2 model, took the most radical path. Zero hierarchy, zero titles, zero formal accountability structures. Five co-founders handle all sensemaking for 300 employees—about 50 direct reports each. Agents route information. Self-reflection handles accountability.

A Chinese journalist embedded with the team for 100 hours described what this produces: extraordinary speed paired with real casualties. At least three senior hires from major tech firms left, one abandoning the industry entirely. Employees reported crying more at Kimi than at any previous company.

"Some mornings you walk in and you just don't know what you should do," a former employee who returned to big tech said. "No one tells you whether you're doing well."

The company screens for people who can self-direct without external validation. That's a specific psychological profile, not a universal one. The journalist used the word "weightlessness" to describe working there—liberating in theory, anxiety-inducing in practice for anyone without the right makeup.

Block and Meta are running different experiments, though the transcript cuts off before detailing them fully. What matters is that each company made different bets about which parts of the management bundle to keep, eliminate, or automate.

What Actually Scales

Here's the pattern: information routing scales with AI immediately. Sensemaking partially scales—AI becomes a useful partner, but can't replace the human judgment that comes from deep context. Accountability doesn't scale at all. You can't simulate someone feeling ownership in their bones.

Even if AI gets 10 times better, Jones argues, that distribution holds. Routing is already solved. Sensemaking might become "AI-assisted human judgment" instead of purely human. Accountability remains fundamentally about humans owning outcomes over time.

The companies that simply cut management without understanding this distinction are removing structure they need. The companies that decompose management into its component parts—then decide which pieces AI handles and which require humans—have a chance at building something that actually works.

This matters because the current wave of management eliminations isn't slowing down. Mark Zuckerberg is actively removing management layers at Meta right now. The question isn't whether more companies will flatten their hierarchies. It's whether they'll do it intelligently.

The evidence so far suggests most won't. They'll cut managers, hand employees AI tools, and expect everything to work itself out. Some percentage of workers will thrive in the weightlessness. Others will drift, anxious and isolated, unclear about whether they're succeeding or failing.

We've seen this movie before with every major technological shift. The tools arrive faster than our understanding of how to use them well. Eventually we figure it out, but the transition costs are real and they land on real people.

The interesting question isn't whether AI will change how companies organize themselves. It obviously will. The question is whether we're paying attention to what we're actually removing when we eliminate management layers, or whether we're just cutting costs and calling it innovation.

Bob Reynolds is Senior Technology Correspondent for Buzzrag.