Sequoia published a piece recently that anyone leading an organisation should read. "From Hierarchy to Intelligence" argues, with a genuinely compelling 2,000-year history, that organisational hierarchy has always been an information routing protocol.
The Roman contubernium. The Prussian General Staff. The railroads. The modern corporation.
Each one solved the same problem: a human can manage three to eight people, so you add layers. Layers add latency. The more people, the more layers, the slower the information flows.
Block is now building a company around the premise that AI can carry the information load that hierarchy was designed to carry — making permanent middle management unnecessary. They're reorganising into three roles: individual contributors, directly responsible individuals, and player-coaches. No management layer. A "world model" replaces the context that managers used to carry.
It's a seductive thesis. It's also largely correct — about the part it chooses to examine.
I co-lead a company navigating this exact question. We deploy AI across our operations, and the pressure to flatten is real and rational. Competitors who route information faster will move faster. AI can route information faster than any chain of human managers.
The diagnosis is right: much of what hierarchy does is relay, summarise, and align. AI does all three without the latency, without the distortion, without the meetings.
But hierarchy doesn't only route information. It carries load that the Sequoia thesis doesn't account for — and if you remove the structure without understanding the full weight it bears, you will break things that are far harder to rebuild than they were to maintain.
I wrote recently about what happens when AI becomes load-bearing infrastructure — the dependencies you accumulate, the architectural choices that determine whether you have leverage or whether your vendor does. The same pattern applies to human infrastructure, but in reverse.
AI becoming load-bearing is a risk you're choosing to take.
Hierarchy being load-bearing is a fact you're choosing to ignore.
Here's what hierarchy actually carries, beyond the information it routes.
Judgment development. People learn judgment by proximity to someone exercising it under uncertainty. Not from documentation, not from training modules — from watching a senior colleague decide between two imperfect options and explain why.
That manager is running an apprenticeship that doesn't appear on any org chart. Flatten the hierarchy and the apprenticeship disappears. You can still hire people with judgment. You can no longer develop it internally. Over time, that's a capability you can only buy, never build — and it gets more expensive every year.
Accountability. When AI routes a decision, who owns the outcome? Block's "directly responsible individual" model is a thoughtful answer for well-scoped problems: one person owns merchant churn in a specific segment for ninety days, with full authority to pull resources across teams.
But the decisions that actually matter are the ones where the problem is ambiguous, the stakes are existential, and someone needs to say: this is my call. Hierarchy provides default accountability. Remove it and you need an explicit replacement — which most organisations won't build until after something has already gone wrong.
Organisational memory. The manager who says "we tried that in 2019 and it failed because the regulatory environment wasn't ready" is carrying knowledge that exists in no document, no system, no world model. It was never written down because it never needed to be — it lived in a person who was present when the decision was made.
A world model can only contain what was captured. The knowledge most likely to be lost in a restructure is precisely the knowledge that was never captured: institutional scar tissue, context, the reasons behind decisions that now look arbitrary.
Translation. Strategy doesn't execute itself. "We're pivoting to enterprise" needs someone to interpret that into "your team stops building feature Y and starts building feature Z, and here's why that matters for you."
Translation requires understanding both the strategic intent and the human reality of the people executing it. AI can transmit strategy perfectly. It cannot yet do the translation — which involves persuasion, political awareness, and the credibility that comes from shared history.
Culture. Culture is not a document. It's transmitted through relationships: the way a senior person handles a difficult client call, the standards they hold in a review, how they respond when someone fails.
Block's "player-coach" role — someone who both builds and develops people — acknowledges this. It's also a concession that proves the point. You cannot remove the human layer entirely, because culture requires human carriers.
These five functions are entangled with information routing in the current system. You can't surgically extract the routing and leave the rest intact. They travel together.
Removing hierarchy to eliminate latency is like removing load-bearing walls to open up the floor plan — it works, but only if you've identified which walls are structural and installed alternative support before you start swinging.
So what does a successful transition actually look like?
Software engineers have a discipline for this. When you refactor a legacy codebase, you follow principles learned through decades of painful experience — and every one of them maps directly onto what organisations are attempting right now.
Understand before you change. The most dangerous refactors are the ones where the team doesn't fully understand what the existing code does — especially the parts that appear to do nothing. Before you flatten, map what your middle managers actually do. Not their job descriptions — their real functions, including the invisible ones.
If you can't articulate what a role carries beyond information routing, you are not ready to remove it.
Strangle, don't rewrite. The strangler fig pattern: build the new system alongside the old, route new work through it, and retire the old components only as the new ones prove reliable.
You don't announce a restructure. You build the AI coordination layer, demonstrate it can carry the routing load, and then redirect human roles toward the functions AI can't carry — judgment, accountability, translation, culture. The transition is gradual and evidence-based, not a reorg memo.
Write the tests first. You never refactor without tests that tell you if you've broken something. What are your leading indicators of organisational capability?
If you flatten and six months later your best junior people are leaving because they're not developing, your incident response is slower because nobody owns the ambiguous cases, and your strategy execution is drifting because the translation layer is gone — you've broken something you didn't test for. Define the tests before you start.
Preserve the interfaces. The best refactors don't change what a system does from the outside. Your customers, partners, regulators, and people shouldn't experience the transition as confusion about who's responsible. They should experience it as the organisation getting faster and more responsive.
If they experience it as chaos, you've failed — regardless of how elegant the internal architecture looks.
The Sequoia thesis is a venture capital argument dressed in a 2,000-year history lesson. It's well-constructed and directionally correct about the information routing problem. But it's a thesis about what hierarchy costs, not about what hierarchy carries.
The difference matters — because the costs are visible and the carrying capacity is not. Until it's gone.
The organisations that get this right will treat it as a genuine engineering problem: understand the current system, identify the full load, build alternatives before removing structure, test relentlessly.
The ones that don't — the ones that read Block's playbook and start removing middle management because AI can "carry the context" — will discover what every experienced engineer already knows.
The most expensive failures come from removing code you didn't understand.
The question isn't whether to restructure. The question is whether you know what survives — and whether you've built what's needed to carry the rest.
Sequoia's "From Hierarchy to Intelligence" is the piece that prompted this essay. It's worth reading in full.