Can AI Replace Hierarchies? Inside Block’s AI‑Native Organizational Model

Shift in Organizational Bottlenecks

For a decade, technology managers focused on execution speed—slow development cycles, missed deadlines, and unstable code. AI tools now accelerate execution to the point where coordination latency becomes the dominant constraint. Fred Brooks' observation in *The Mythical Man‑Month* that "adding manpower to a late project makes it later" still holds, but the hidden cost of coordination is now exposed because execution has moved from ~60 km/h to ~300 km/h.

Anthropic’s 2026 Economic Index Report shows experienced practitioners treating LLMs as thinking partners rather than simple search tools, confirming that execution is no longer the primary bottleneck.

Hierarchy as an Information‑Routing Protocol

The core claim of Block’s "From Hierarchy to Intelligence" article is that hierarchical structures exist to route information within the limits of human cognitive bandwidth (3‑8 direct reports). Historical evidence:

Roman legion: each commander aggregated information from 8‑person contubernia, passed decisions upward, and decomposed orders downward.

Modern experiments (Spotify squads, Zappos Holacracy, Valve flat structure) reverted when scaling beyond a few hundred people because no technology replaced the information‑routing function.

AI now offers a technology capable of maintaining a continuously updated "company world model" that can perform this routing.

Block’s Four‑Layer AI‑Native Architecture

Designed for financial services, the stack consists of:

Capabilities Layer : Atomic, API‑driven financial functions (payments, lending, card issuing, payroll). These are high‑value, regulated building blocks with clear reliability and performance metrics.

World Model Layer :

Company World Model – digital artifacts (decisions, code, designs) that describe internal state.

Customer World Model – per‑customer/merchant/market models built from transaction streams, evolving from raw events to causal and predictive insights.

Intelligence Layer : Detects opportunities in real time and composes relevant Capabilities. Examples:

A restaurant approaching a seasonal cash‑flow dip triggers an automatic short‑term loan combined with a repayment plan before the merchant requests financing.

A Cash App user who moves houses receives a bundled offering: direct‑deposit payroll, community‑specific card offers, and updated savings targets.

If a required Capability is missing, the failure signal becomes a backlog entry – "customer reality generates the backlog directly."

Interfaces Layer : Delivery channels (Square, Cash App, Afterpay, TIDAL, bitkey). Value resides in the Model and Intelligence, not the UI.

The architecture inverts traditional roles: the World Model aligns work, Directly Responsible Individuals (DRIs) set priorities, and Player‑Coaches focus on craft and people development, eliminating permanent middle‑management routing functions.

Prerequisites for the Architecture

Atomic capabilities – readily available in finance but difficult to define for consulting, hardware, or creative domains.

High‑density customer signals – transaction data provides clean, real‑time inputs; many SaaS or B2B firms have noisy, low‑resolution data.

Fully digitised work processes – Block’s remote‑first culture generates digital artifacts for every decision; other firms rely on whiteboards or informal meetings.

Reliable Intelligence Layer – real‑time automated decisions must avoid hallucination, causal‑inference errors, and meet strict regulatory compliance.

Core Difficulty: Judgment Gap

Even when the system can aggregate information, nuanced human judgment remains essential. Deloitte’s 2026 State of AI report (3,235 managers) found that 84 % of companies have not redesigned jobs around AI; the primary barrier is skill, not budget. The "taste‑and‑judgment gap" means that while anyone can access AI tools, only a subset can reliably decide what to build, what to test, what to cut, and how to evaluate AI output quality.

Practical Path for a Sub‑100‑Person Team

Make work visible to AI : Document decisions (the "why" and reasoning), adopt asynchronous communication as default, and maintain a single source of truth. The team lead becomes a "Chief Context Officer" responsible for the information bus.

Cultivate judgment, not just tool proficiency :

Context Engineering – manage memory, skills, retrieval contexts for LLMs.

Token Economics – treat token usage as engineering cost and experiment with constrained token budgets.

Judgment standards – define metrics for AI output quality and practice deliberate taste assessment.

Select a bounded subsystem to redesign : Identify atomic Capabilities, required World Model inputs, and how the Intelligence Layer would compose solutions if AI were the infrastructure.

Run a small‑scale experiment : Convert a subset of managers to Player‑Coaches, appoint temporary DRIs for concrete problems, and let the World Model handle alignment. Measure time spent on judgment, craft, and people development versus routine information routing.

Success criteria include faster hypothesis‑driven iteration, clear rollback paths, and observable shift of managerial effort toward irreplaceable contributions.

Implications for Management

If AI can assume information aggregation, resource coordination, and priority setting, the essence of "management" reduces to providing judgment, shaping culture, and developing talent—functions that AI cannot yet replicate reliably.

References

Jack Dorsey & Roelof Botha, "From Hierarchy to Intelligence", Sequoia Capital / Block, March 2026.

Anthropic, "Economic Index Report: Learning Curves", March 2026.

PwC, "2025 Global AI Jobs Barometer", June 2025.

Deloitte, "State of AI in the Enterprise: The Untapped Edge", January 2026.

Nyk (@nyk_builderz), "How to become AI native", 2026.

Dhanji R. Prasanna (Block CTO), "Building the AI‑First Enterprise with Goose", Training Data (Sequoia Capital), September 2025.