What Block Gets Right and Wrong About AI-Driven Organizations

Block recently published an essay arguing that AI will replace organizational hierarchy — that the span-of-control constraint governing every large organization since the Roman legions can finally be broken. The essay, introduced with an endorsement from Sequoia, spends considerable time on military history before arriving at Block's vision: a company organized as "an intelligence" rather than a hierarchy, where AI maintains a "world model" of operations and coordinates work that previously required layers of human management.

The piece is ambitious. It is also roughly 80% historical context, 15% vision, and 5% acknowledgment that none of this exists yet. Let's extract what's actually useful.

The Real Insight — and Its Limits

Strip away the Romans, the Prussians, the railroads, and Frederick Taylor, and you arrive at Block's core claim: the primary job of middle management is information routing, and AI is getting good at information routing.

This is a useful observation wrapped in a misleading reduction. Yes, a manager's calendar is full of status meetings, alignment sessions, and context-shuttling. That coordination work is real, and AI will absorb much of it. But Block's essay treats this slice as the whole job, because routing functions are automatable and the rest isn't.

What the essay doesn't account for: a good manager is the buffer between the people and the machine. They absorb institutional friction so their team doesn't have to. They notice when someone is struggling before it shows up in any metric a world model could track. They translate between what leadership wants and what the team can actually deliver. They advocate for resources, shield people from bad decisions above them, and occasionally break the machine on purpose when the machine needs breaking. A director operates the levers of the company to affect real change. They don't just route information through the hierarchy. They reshape the hierarchy when it stops serving its purpose.

None of this is information routing. It's judgment, advocacy, and care. A "world model" can tell you what's blocked in Jira. It can't tell you that your best engineer is about to quit because she feels undervalued, or that two team leads are in a cold war that's silently killing throughput.

Block's essay needs you to forget this, because if management is primarily a human function with some coordination overhead, then AI optimizes the overhead without replacing the role. That's a much less dramatic story than the end of two thousand years of hierarchy.

The Architecture Worth Examining

Block proposes four layers: capabilities, world model, intelligence layer, and interfaces.

Capabilities are atomic business primitives — payments, lending, card issuance — that have no UI of their own. This is sound platform thinking. Separating capabilities from products means they can be composed flexibly rather than locked into predetermined interfaces. It is also not new. Amazon, Stripe, and Twilio have built businesses on this principle. But Block is right that most companies haven't internalized it.

The world model has two halves — and it's worth noting that the term itself borrows prestige from ML research, where "world model" refers to a learned causal representation of an environment. What Block describes is closer to an internal data platform with AI on top. The naming is doing rhetorical work. That said, the underlying ideas have merit. The company world model — a continuously updated picture of what's being built, what's blocked, and where resources sit — is essentially what good internal tooling already does, made more ambitious with AI. The customer world model — a per-customer, per-merchant understanding built from transaction data — is where Block has a genuine structural advantage. They see both sides of millions of transactions daily, buyer through Cash App and seller through Square. Most companies have to infer customer reality from partial signals. Block observes it directly. That's a real moat, and it's the most underappreciated point in the essay.

The intelligence layer composes capabilities into solutions for specific customers at specific moments. The examples are compelling on paper: a restaurant getting a proactive loan offer because the model detects a seasonal cash flow pattern; a user getting neighborhood-optimized Card boosts after relocating. But these are product features, not evidence of a new organizational paradigm. Good product teams have always tried to anticipate customer needs. AI makes this more feasible at scale. That's meaningful. It's not a civilizational shift.

Interfaces — Square, Cash App, Afterpay — become delivery surfaces rather than the locus of value creation. This is the standard platform argument: the interface is replaceable, the data and intelligence are not.

What the Essay Gets Wrong

Inventing a role to carry the weight it just removed. Block eliminates middle management and then immediately creates the "player-coach" — someone who "still writes code" while also "investing in the growth of the people around them." This is the management role with a new name, minus the organizational authority, plus a full IC workload on top. It just refuses to call that person a manager.

Announcing a vision as if it were a result. The essay hedges — "early stages," "parts of it will likely break" — but the packaging tells a different story: two thousand words of military history, a Sequoia endorsement, and a four-layer architecture presented as the thing that finally breaks a civilizational constraint. Nothing in the essay demonstrates that Block's intelligence layer has coordinated a single team more effectively than a competent manager. The vision may be correct. But no evidence has been offered.

Generalizing from a specific structural advantage. Block's intelligence layer depends on seeing both sides of millions of transactions daily — buyer through Cash App, seller through Square. A law firm, a construction company, a hospital system: none of them have this data, and none of them can build it. The essay claims "the pattern behind this... will reshape how companies of all kinds operate." But the pattern requires the platform, and most companies don't have one. What Block is describing is a competitive strategy for Block, presented as a general theory of organizations — unless this essay is a prelude to a product announcement, offering the world model as a platform others can build on. That would make the universality claim coherent. It would also make this a sales pitch, not an organizational thesis.

The Real Bottleneck

Block's entire thesis rests on a premise that sounds obvious but isn't: that information routing is the bottleneck in organizations. AI can route information faster than middle managers — no argument there. But was information routing the actual constraint, or just the most visible one?

Consider what actually stalls organizations:

• Trust. Google's Project Aristotle found that psychological safety — not structure, not talent, not resources — was the strongest predictor of team effectiveness. A world model can surface metrics. It can't surface the problems people stopped reporting because the last person who did got fired.
• Incentive alignment. A company that hasn't decided whether it's a software business or a services business will have a product team building for scale and a services team building for the client in front of them — and neither is wrong. They just have fundamentally different definitions of success. No amount of information routing resolves this. It's a strategic identity question that requires a human leader to decide.
• Taste. Organizations don't just need more information — they need someone who knows what to do with it. A leader with bad taste builds mediocre products slowly. An intelligence layer with no taste builds them faster. The bottleneck was never speed.

Block looked at the org chart, saw information moving up and down, and concluded the movement was the constraint. But information routing is just the most measurable thing hierarchy does. The actual bottlenecks — trust, incentive alignment, taste — are invisible to any system that can only observe artifacts and metrics.

The fear behind essays like this is real: what if a competitor appears where every employee is an AI agent, offering the same service at a hundredth of the price? That's a legitimate existential concern. But the answer isn't to hollow out your management layer and replace it with a coordination model that doesn't exist yet. The companies that will survive AI disruption are the ones whose value lives in judgment, trust, regulatory position, or proprietary data — not the ones that moved fastest to eliminate the humans who provide those things.

What Matters Here

The most valuable thing in the Block essay isn't the organizational theory — it's the product architecture. Build composable capabilities. Invest in proprietary data signals. Use AI to surface opportunities your human processes would miss. These are concrete, testable moves. Steal them regardless of how your org chart looks.

The organizational argument is less convincing. Every decade produces a new thesis about the end of hierarchy — holacracy, flat organizations, Spotify squads, DAO governance. They all share the same insight (hierarchy is slow) and the same blind spot (hierarchy exists because the alternatives fail at scale). Block may prove to be different. But "we have AI now" is not sufficient evidence.

Block's essay is trying to apply a logic we've seen work elsewhere: the cost of software is now zero because AI automated code production. If AI automated code, maybe it can automate management too. But code production was genuinely mechanical. Management is not. AI is exceptionally good at automating the parts of your business that were always mechanical. It is not good at replacing the parts that require being human.

There's a deeper issue that the Block essay doesn't touch. As AI absorbs more of what people used to do — write code, route information, coordinate projects — people are watching the tasks they built their identity around disappear. The organizational question isn't just "what do people do now?" It's "how do we help people remember that their value was never reducible to the tasks they performed?" Everyone has inherent worth that exists independent of their productivity. The companies that understand this will retain the people who matter. The ones that treat humans as a cost to be optimized away will find themselves, eventually, with no one left worth keeping.