The Job AI Is Actually Eliminating
Most conversations about AI and jobs focus on tasks — the writing, the summarising, the data entry. That framing misses the more consequential shift. The deeper disruption is not to individual tasks. It is to the structural role of the manager as an information router.
Think about what a middle manager actually spends their time on. A significant portion — in most organisations, the majority — is receiving information from below, filtering it, translating it, and passing it up. Or receiving direction from above, interpreting it, and pushing it down. That work is coordination work. It is valuable when information is scarce or hard to move. It becomes redundant when information flows freely, in real time, with minimal friction.
That is exactly what AI enables.
What Dorsey and Sequoia Are Actually Arguing
Jack Dorsey, in the context of Block's ongoing restructuring, has been direct: AI reduces the information asymmetry that traditionally justifies hierarchy. When any individual contributor can access synthesised context, generate analysis, and communicate across the organisation without a human intermediary, the middle layer loses its primary function.
Sequoia Capital's Roelof Botha has made a similar argument from the investor side. AI-native companies are structurally different — built without the assumption that information needs to be managed by humans at every level. The result is faster decisions, fewer political bottlenecks, and a clearer line between strategy and execution.
A 2025 Writer survey found that only 35% of employees view their leadership as genuine AI champions — meaning nearly two-thirds see their managers as behind the curve on the technology most likely to reshape their roles. That credibility gap is itself a structural problem.
What This Means for Companies of 20 to 100 People
You probably don't have five layers of management to trim. But you almost certainly have informal information bottlenecks — the one person everything flows through, the approval loop that exists out of habit, the weekly check-in that exists to keep a manager informed rather than to move work forward.
AI exposes these. When your team members can research independently, draft communications, synthesise meetings, and track project status in real time, the coordination work that justifies the bottleneck shrinks. The organisations getting this right are moving toward what we call judgment-forward structures — where the work that reaches senior people is genuinely the work requiring their judgment, not their availability.
The Risk of Removing Layers Too Fast
The case for flattening is real, but the execution risk is significant. Middle managers do more than route information. They carry context. They maintain relationships. They notice when someone is struggling before it becomes a performance issue. AI does not do those things yet.
The organisations that have flattened poorly share a common failure pattern: they removed people without redesigning the information flows those people were managing. Decisions stalled, accountability diffused, and execution slowed even as headcount costs dropped.
Removing hierarchy without redesigning information flow is not transformation. It is just subtraction.
Flow First, Structure Second
The practical approach: audit your information flows before you touch your org chart. Map where information originates, how it moves, where it slows down, and who adds value at each step versus who simply handles it in transit.
That audit almost always reveals two things. First, there are legitimate coordination roles that are genuinely hard to replace — people who hold important context and institutional relationships. Second, there are other roles that have become primarily administrative, where the value-add is thin and the routing function could be handled by a well-designed AI workflow.
Once you have that map, you can make deliberate decisions. Which flows can you automate? Where can individual contributors access context directly? What decisions are being pushed up unnecessarily? Only after answering those questions should you consider structural changes — gradually, with honest communication about what is changing and why.
The flat organisation is not a new idea. Holacracy, self-managing teams, radical decentralisation — these have been tried, celebrated, and often quietly abandoned because the coordination costs of flatness were too high. What is genuinely new is the infrastructure. AI-powered context delivery, real-time synthesis, automated tracking — these reduce the coordination cost of flat structures in ways earlier experiments couldn't rely on.
Start with the information. The structure follows.