Hierarchy was never the goal. It was the workaround.

Two thousand years from the Roman contubernium to your current org chart, and the constraint has not moved. A human can manage three to eight people directly. Every layer of every modern company exists to route information through that bottleneck. In April 2026, Block published the first public-company manifesto for what comes after — and the implication for everyone smaller than Block matters more than the headline did.

This is a piece about why org charts look the way they do, why every attempt to flatten them in the last twenty years quietly reverted, and what changes when the coordination problem stops being a human problem.

The 2,000-year constraint

In 100 BC the Roman army organized fighting units in groups of eight: the contubernium — the soldiers who shared one tent. Eighty men sat under a centurion. Five thousand under a legate. The structure was not a preference. It was a workaround for a hard limit on what one human could see, decide, and direct.

The pattern repeated every time information volume jumped. The Prussian General Staff in 1806 invented what we would now call middle management — coordinators whose job was to translate field reality into command intent and back again. Daniel McCallum drew the first modern org chart for the Erie Railroad in 1854 because the railroad was the first business big enough that no one human could hold it in their head. Frederick Taylor industrialized the same idea. The Manhattan Project scaled it. McKinsey gave it a matrix. Spotify renamed the boxes "squads." Different vocabulary every century. Same architectural problem underneath: humans cannot directly coordinate other humans past a small number, so we built scaffolding to fake it.

Jack Dorsey and Roelof Botha put it directly in their joint essay: "Two thousand years of organizational innovation has been an attempt to work around this tradeoff without breaking it."

Why every flat-org experiment reverted

If hierarchy were a preference rather than a workaround, the experiments to move past it would have stuck. They did not.

Holacracy launched at Zappos in 2014 with hundreds of "circles" replacing managers. By 2020 it was effectively dismantled. Valve's flat structure produced famous internal politics — without formal hierarchy, informal hierarchy filled the vacuum. Spotify's squads-and-tribes model became the canonical example of progressive org design and was publicly walked back by the engineers who built it less than a decade later.

The pattern across all of them is the same. Removing managers without replacing what managers actually do — coordinating information across people — produces less coordination, not more. The work the manager did still has to happen. Without scaffolding, it gets done worse, by everyone, ad hoc.

So the question that matters is not "can we remove the layers?" It is: "is there a coordination mechanism that can do what those layers were doing?" For 2,000 years the answer was no. That is what changed.

What actually changed

The standard narrative on AI in 2025 was a productivity story. Each individual moves faster. Same org, more output. That story is real but it is small. The Block essay is making a different claim, and the claim is the one that matters.

The claim is that AI can finally do the coordinating itself. Not summarize the meeting after it happens. Not draft the brief faster. Actually carry information across the function and make decisions inside it the way a middle layer of human managers used to. When Block cut roughly 40% of its workforce in February 2026 — about 4,000 people — the stock reacted +18% on the announcement, VentureBeat covered it as the first major company to publicly say "yes, this is because of AI". The market was not pricing a productivity gain. It was pricing the removal of an entire architectural layer.

What replaces the layer, in Block's framing, is a four-part system:

• Capabilities: the things the company can do, exposed as composable services
• A world model: a continuously updating representation of the customer, the product, the market — the company's actual understanding of its situation
• An intelligence layer: AI that reads the world model and decides what to do
• Interfaces: surfaces where human operators set direction, approve consequential decisions, and intervene

The interesting word in that list is world model. A traditional company stores transactional data and pulls reports off it. A company built around a world model stores its own evolving understanding — and the understanding gets deeper every day, not because someone wrote it down but because the system is designed to keep it current. That is the qualifier nothing else replaces.

The qualifier question every company now has to answer

Diana Hu at Y Combinator frames it as three pillars: closed-loop coverage, queryable data, and an intelligence layer on top. It is the same architecture as Block's, scaled down to something a 50-person company can actually build.

The qualifier question all three of those frameworks point at is the same one:

What does your company understand that is genuinely hard to understand, and is that understanding getting deeper every day?

If the answer is nothing — if the company is just running plays anyone can run with the same tools — then AI is a cost-cut and not much else. Whatever margin AI saves, a competitor with the same tooling will compete it away within a quarter.

If the answer is something deep — a customer behavior pattern, a sourcing edge, a creative angle, a renewal signal nobody else can see — then AI is what reveals what the company actually is. The understanding compounds. The competitor with the same tools cannot replicate it because the moat is not the tools.

This is the test most companies fail today, and it is also the test that explains the gap between AI as feature and AI as architecture. Emergence Capital's PMF test for AI-native services is the cleanest version: the only proof you are AI-native is "AI doing a material share of the work at high gross margin." Not revenue. Not logo retention. Margin. Margin is what shows whether the system is actually compounding or just delivering the same service with different production tools.

What this looks like at smaller scale

Block has 4,000 customer-data engineers and a market cap to fund a multi-year platform build. Most companies reading this do not.

That does not change the qualifier question. It changes the answer's shape.

A consumer brand at $5M to $50M in revenue does not need a customer world model the size of Square plus Cash App. It needs a clear answer to what is our honest signal — the equivalent of what Block calls "money is the most honest signal in the world" applied to its own customer base. The signal might be retention behavior. It might be content engagement. It might be a sourcing pattern. The point is to find it, make it queryable, and build the intelligence layer on top of it.

A B2B SaaS company at the same scale does not need a 100-engineer ML team. It needs a closed loop on the metric that defines whether the product is working — usually some combination of activation, expansion, and renewal. The same architecture: closed-loop coverage, queryable data, intelligence layer on top, operator at the interface.

The form factor scales down. The architecture does not.

The honest counter

Not everyone reads Block's restructure as a bellwether. Josh Bersin published a careful skeptic piece arguing that Block's situation is specific to its product and balance sheet, and that broad extrapolation is premature. The counter is worth taking seriously. One company's restructure is not a category-defining moment.

What makes the bellwether read defensible is that Block is not the only company shipping this architecture. The 4As is publishing the "Agency as Marketing Purveyor" thesis — productized services with licensed IP rather than bespoke retainers, which is the same architectural shift inside the agency category. Sequoia keeps publishing AI-native services case reads — Pace, Crosby, Harper, Mercor — every one of them is the same shape. The vocabulary varies. The architecture is consistent.

The bet is not that Block is right about timing. It is that the architecture is right about direction.

What we are doing about it

Leanboat exists because most companies cannot answer the qualifier question on their own — and most consultants pitching them on AI cannot either, because they are bolting AI onto delivery models that pre-date it. Our work starts with the audit: closed-loop coverage, queryable data, intelligence layer. Where each of those is in your company today. Where the gaps are. What the 90-day path to closing them looks like.

If you are about to commit a meaningful chunk of next year's budget to AI tooling, or to a retainer with an agency that says "we use AI" without saying how, the audit is the thing that decides whether that money compounds or evaporates.

Book a strategy call →