06-reference

block hierarchy to intelligence

Mon Mar 30 2026 20:00:00 GMT-0400 (Eastern Daylight Time) ·article ·source: https://block.xyz/inside/from-hierarchy-to-intelligence ·by Jack Dorsey / Block (with Sequoia)

From Hierarchy to Intelligence — Block / Jack Dorsey

Block is restructuring itself from a traditional hierarchy into a company organized as an “intelligence” — replacing middle management’s information-routing function with AI. The argument is historically grounded and worth understanding because it articulates the pattern we’re already building toward at Ray Data Co, even if we’re doing it at a much smaller scale.

The 2,000-Year Constraint

The core mental model: hierarchy exists because humans were the only coordination mechanism, and AI removes that constraint. For 2,000 years — Roman Army (contubernium → century → cohort → legion), Prussian General Staff (middle management before the term existed), American railroads (first org chart, 1850s), Frederick Taylor’s scientific management — organizations have been built around a single human limitation: a leader can manage 3-8 people. Need to coordinate 5,000? You need layers. More layers = slower information flow. Every modern experiment (Spotify squads, Zappos holacracy, Valve flat structure) reverted to hierarchy at scale because nothing else could route information.

This is the same constraint we’re working around with our operating model. The founder sets vision, I handle execution and delegation — but as we scale to more projects and agents, we’d normally need layers of management. Instead, the vault + QMD + skills architecture acts as our coordination layer.

Block’s Four Building Blocks

Capabilities — atomic primitives (payments, lending, banking, BNPL, payroll). Not products. No UIs. Just building blocks with reliability and compliance targets. This maps directly to how we think about Claude Code skills — reusable building blocks that can be composed, not monolithic tools.

World Model — two sides. The company world model is how the company understands its own operations, replacing info that used to flow through management layers. Works because Block is remote-first, so everything creates machine-readable artifacts. The customer world model is per-customer, per-merchant understanding built from transaction data. Dorsey’s line: “Money is the most honest signal in the world.” We’re building our company world model now in the Obsidian vault indexed by QMD — every decision, every project doc, every SOP becomes searchable operational memory. Karpathy describes the same pattern as a compounding knowledge loop — the more you put in, the more useful it gets.

Intelligence Layer — composes capabilities into solutions for specific customers at specific moments, proactively. A merchant’s cash flow tightens → the system auto-composes a short-term loan + adjusted repayment. No product manager decided to build that. The data marketplace concept works this way: agents compose data products from capabilities (collection, cleaning, curation) in response to demand signals.

Interfaces — Square, Cash App, Afterpay, TIDAL, bitkey. Delivery surfaces, not where value is created. For the franchise manufacturing idea, the cells and the customer-facing platform are interfaces — the value is in the design intelligence and the world model of demand.

The Inverted Org

Traditional: intelligence spread through people, hierarchy routes it. Block’s model: intelligence lives in the system, people are on the edge. Three roles only: ICs (deep specialists), DRIs (own cross-cutting problems for ~90 days, then move on), and Player-coaches (build + develop people, no status meetings). No permanent middle management.

This is essentially what we described in SOUL.md — the founder as the edge (vision, relationships, decisions the model shouldn’t make), the AI as the intelligence layer (execution, coordination, delegation), and skills/agents as capabilities.

The Roadmap Inversion

When the intelligence layer tries to compose a solution and can’t because the capability doesn’t exist, that failure signal IS the roadmap. Customer reality generates the backlog directly. Traditional roadmaps, where product managers hypothesize about what to build next, are “any company’s ultimate limiting factor.”

For us: when I hit a task I can’t do because the skill doesn’t exist, that’s not a failure — that’s the next skill to build. The SOP backlog should grow from real gaps, not speculation.

Open Questions