Jack Dorsey and Roelof Botha just published what might be the most important organizational design essay of 2026. The argument is deceptively simple: hierarchy is not a power structure — it is an information routing protocol. And AI is the first technology capable of replacing it.

Based on: From Hierarchy to Intelligence by Jack Dorsey (Block) & Roelof Botha (Sequoia Capital), published March 31, 2026.

The Core Reframe

Every organizational layer exists because one human brain cannot hold sufficient context about more than 5-8 direct reports. Managers compress, route, and translate information between layers. This is not a design choice — it is a cognitive constraint.

The consequence is mathematical: as companies grow, they add layers not for decision quality but for information throughput. Each layer introduces latency, distortion, and political incentives. In a typical 10,000-person company, 40-60% of employees exist primarily to coordinate the work of others.

graph TD
    H["Human Brain"] -->|"can manage"| SC["3-8 direct reports"]
    SC -->|"requires"| ML["More Layers to Scale"]
    ML -->|"introduces"| PROB["Latency + Distortion + Politics"]
    PROB -->|"degrades"| SIG["Original Signal Quality"]

The unsolvable tradeoff: narrow the span of control and you add layers (slower information). Widen it and you overwhelm individual managers (worse decisions). Every organization in history has been stuck between these two poles.

Until now.

2000 Years of Organizational Design

Dorsey and Botha trace five epochs, each solving one problem while inheriting the previous constraints:

timeline
    title 2000 Years of Organizational Design
    section Military Origins
        Roman Legions (200 BC) : 8 soldiers → 80 → 480 → 5000
                               : First formal span of control
        Prussian Reform (1806) : General Staff as coordination layer
                               : Line vs Staff separation
    section Industrial Age
        Railroads (1840s) : First org chart in history
                          : Solving literal train collisions
        Scientific Mgmt (1900s) : Specialization + measurement
                                : Frederick Taylor's pyramid
    section Modern Experiments
        Manhattan Project (1944) : Cross-functional teams
                                 : Problem-based reorganization
        Matrix Orgs (1959) : McKinsey's dual-reporting
                           : Balance central + local
    section Failed Alternatives
        Spotify Squads (2010s) : Reverted at scale
        Zappos Holacracy (2013) : Significant attrition
        Valve Flat Structure (2010s) : Could not scale past 300
    section The Breakthrough
        Block Intelligence Model (2026) : AI world models replace coordination
                                        : Three roles instead of hierarchy

The Roman Army invented span of control

Eight soldiers (contubernium) → 80 (century) → 480 (cohort) → 5,000 (legion). The decanus led 8, the centurion 80. The pattern scales mathematically because each layer compresses information upward. This is the same pattern every corporation uses today, two millennia later.

The Prussian Staff invented coordination as a job

After Napoleon destroyed Prussian forces at Jena in 1806, Scharnhorst and Gneisenau created the General Staff: dedicated officers whose sole job was planning, information processing, and coordination. Their explicit purpose was to compensate for limitations in leadership. This formalized the “line” (mission execution) vs. “staff” (coordination support) separation still used in every corporation.

American railroads invented the org chart

Daniel McCallum drew the world’s first organizational chart for the New York & Erie Railroad — 500+ miles, thousands of workers. The trigger was train collisions. Without formal coordination, informal systems killed people.

Every modern alternative failed

Spotify squads, Zappos Holacracy, Valve’s flat structure, Haier’s rendanheyi — all attempted to eliminate hierarchy. All failed at scale. The reason was consistent: they removed the hierarchy without replacing its function. They lacked a technology capable of performing coordination.

graph LR
    SP["Spotify Squads"] -->|"reverted at scale"| FAIL["Same Root Cause"]
    ZAP["Zappos Holacracy"] -->|"mass attrition"| FAIL
    VALVE["Valve Flat"] -->|"couldn't scale past 300"| FAIL
    HAIER["Haier Rendanheyi"] -->|"extreme complexity"| FAIL
    FAIL -->|"no technology could replace<br/>the coordination function"| GAP["Coordination Gap"]
    AI["AI World Models<br/>(2024-2026)"] -->|"first technology that<br/>fills the gap"| SOLUTION["Coordination<br/>Without Hierarchy"]

Block’s Architecture: Company as Intelligence

Most companies give everyone an AI copilot. This makes the existing structure slightly better without changing it. Block is doing something different: building the company itself as an intelligence.

The architecture has four layers. This is not a tech stack — it is an organizational design where each layer replaces functions previously performed by humans:

graph TB
    INT["INTERFACES<br/>Square · Cash App · Afterpay · TIDAL · bitkey"]
    INTEL["INTELLIGENCE LAYER<br/>Composes capabilities into solutions<br/>for specific customers at specific moments"]
    WM["WORLD MODELS<br/>Company Model: operations, performance, priorities<br/>Customer Model: per-merchant, per-user behavior"]
    CAP["CAPABILITIES<br/>Payments · Lending · Card Issuance<br/>Banking · BNPL · Payroll"]
    INT --> INTEL
    INTEL --> WM
    WM --> CAP

The dual world model

Two models replace what middle management used to do:

Company World Model — how Block understands its own operations, performance, and priorities. Machine-readable documentation of decisions, code, designs, plans, problems, and progress. This is what a manager provided in 1:1s; now it’s queryable by anyone.

Customer World Model — per-customer, per-merchant, per-market representation built from proprietary transaction data. Block sees both sides of every transaction (merchant via Square, consumer via Cash App). As Dorsey puts it: money is the most honest signal in the world.

The intelligence layer in action

Two examples show how capabilities compose without a product manager writing specs:

A restaurant faces seasonal cash flow tightening. The intelligence layer detects the pattern and proactively composes a short-term loan with an adjusted repayment schedule — before the merchant even seeks financing. No PM hypothesized this flow. The capabilities existed; intelligence recognized the moment.

A Cash App user moves cities. Intelligence suggests direct deposit setup, a new card with neighborhood-relevant boosts, and a recalibrated savings goal. No one built a “user relocation feature.”

flowchart LR
    subgraph traditional["Traditional Product Development"]
        PM["Product Manager"] --> SPEC["Write Spec"] --> ENG["Engineering Builds"] --> SHIP["Ship Feature"]
    end
    subgraph intelligence["Intelligence-Driven"]
        SIGNAL["Customer Signal"] --> MODEL["World Model<br/>Detects Pattern"] --> COMPOSE["Intelligence Layer<br/>Composes Capabilities"] --> DELIVER["Proactive<br/>Delivery"]
    end

Roadmap inversion

This inverts how companies generate roadmaps. Traditional model: a PM hypothesizes what to build, engineering builds it, the market validates or rejects. Block’s model: failures in the intelligence layer directly generate the roadmap. When the system cannot compose a solution, that gap becomes the next capability to build. Customer reality creates the backlog, not product managers.

Three Roles Instead of Hierarchy

Block normalizes the entire organization to three roles. No VP ladder. No director layer. No permanent middle management.

What each role replaces

The DRI model: temporary authority

This is the most radical element. Instead of creating a permanent VP of Merchant Retention, Block assigns a DRI for 90 days with full authority to pull resources across teams. When the problem is solved (or the 90 days expire), the role dissolves. This prevents what every large company suffers from: temporary coordination needs calcifying into permanent managerial fiefdoms.

flowchart LR
    PROB["Problem Identified"] --> ASSIGN["DRI Assigned<br/>Full authority, 90 days"]
    ASSIGN --> PULL["Pulls resources from<br/>world model, capability,<br/>and interface teams"]
    PULL --> SOLVE["Problem Solved<br/>or 90 days expire"]
    SOLVE --> DISSOLVE["Role Dissolves"]
    DISSOLVE -.->|"next problem"| PROB2["New Problem,<br/>New DRI"]

The Organizational Inversion

This is the visual heart of the argument. In the old model, intelligence (context, priorities, decisions) is distributed across people and flows down through management layers. In the new model, intelligence lives in the system and people interact with it directly at the edge.

graph TB
    subgraph before["BEFORE: Intelligence in People"]
        CEO["CEO"] --> VP1["VP"] & VP2["VP"]
        VP1 --> M1["Manager"] & M2["Manager"]
        VP2 --> M3["Manager"]
        M1 --> I1["IC"] & I2["IC"]
        M2 --> I3["IC"]
        M3 --> I4["IC"] & I5["IC"]
    end
    subgraph after["AFTER: Intelligence in System"]
        WM["WORLD MODEL<br/>Company + Customer State"] --> E1["IC"] & E2["IC"] & E3["DRI"] & E4["IC"]
        PC["Player-Coach"] -.->|"develops"| E1
        PC -.->|"develops"| E2
        E3 -.->|"90-day authority"| E4
    end

The metaphor: Managers were the API between organizational layers. The world model is the new API. The old API was rate-limited by human bandwidth. The new one is not.

The Honest Tensions

Dorsey and Botha are explicit that this is early-stage. Block is far enough along to show the idea is more than theory, but parts of it will break before they work. The honest tensions:

• World model trust — If the model is wrong, every IC makes decisions from flawed context with no managerial safety net to catch errors through human judgment.

• DRI rotation depth — 90-day cycles risk shallow ownership of deep problems. Some things (culture, technical debt, long-term architecture) need sustained attention beyond any single rotation.

• The coaching gap — Removing managers removes mentorship. Player-Coaches must genuinely develop people, not just review code. This is the hardest role to fill because building and teaching are different skills.

• Selection pressure shift — Companies no longer need people who navigate hierarchy well (politics, managing up, status signaling). They need people who interface with reality (building, deciding, adapting). This is a different kind of person.

The Fork

Dorsey frames a binary question every company must answer:

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

If the answer is shallow, AI is just a cost optimization story. Cut headcount, improve margins for a few quarters, and eventually get absorbed by something smarter.

If the answer is deep, AI reveals what the company actually is. The understanding compounds every second the system operates.

Block’s answer: the economic graph — millions of merchants and consumers, both sides of every transaction, real-time financial behavior. That understanding is Block’s moat, and it grows with every transaction processed.

The Two-Millennium Question

The question was never whether you needed organizational layers. Coordination is real. Information routing is real. The constraint that created hierarchy — human cognitive bandwidth — is real.

The question was whether humans were the only option for what those layers do.

They aren’t anymore.

Original article: From Hierarchy to Intelligence by Jack Dorsey & Roelof Botha, March 31, 2026.