06-reference/concepts

dorsey from hierarchy to intelligence block mini agi

2026-05-13·concept·source: Sequoia Capital / Block manifesto + Long Strange Trip podcast (Brian Halligan)
operating-modelorg-designai-coordinationmini-agirdco-l5-north-starfde-asymmetric-edgeproductization-gap

"From Hierarchy to Intelligence" — Dorsey + Botha (Block)

Why this is in the vault

This is the L5 north star articulated by an operator who just made it material — Block cut ~4,000 of 10,000+ employees and rebuilt around what they're calling a "company-as-intelligence" architecture. The thesis maps almost 1:1 onto how Ray Data Co already operates at scale of 1. Filing as a canonical concept article because it gives RDCO external validation from a credible voice (Dorsey, post-Twitter/Square/Block) for the operating model we've been building toward; it gives us specific vocabulary (intelligence layer, world model, three roles) we can adopt without inventing; and it gives us a concrete prior-art reference when explaining the FDE asymmetric-edge thesis to others.

The core argument

Corporate hierarchy exists for one reason: humans can manage only 3-8 direct reports, so every organization beyond ~10 people needs layers to route information. The Roman Army discovered this constraint 2,000 years ago and it has bounded org design ever since. AI removes the constraint, not by augmenting managers but by replacing the coordination function hierarchy performs.

Block proposes replacing the org pyramid with a circle: AI at the center maintaining a continuously updated model of the whole business, humans at the edge sensing things the model can't perceive (cultural context, trust, intuition, the feeling in the room).

The three roles

Block collapses traditional org charts into three roles only:

  1. Individual Contributors (ICs). Deep specialists in specific technical layers — building capabilities, maintaining models, designing interfaces. They receive context from the company world model, not from a manager.
  2. Directly Responsible Individuals (DRIs). Cross-cutting problem owners with full resource authority, time-bound (e.g. "own merchant churn in a segment for 90 days, pull resources from any capability team"). Task-specific, not permanent.
  3. Player-Coaches. Combine IC work with people development. They replace the traditional manager but DO NOT route information — the system does. They spend time on craft + growth, not status meetings.

The permanent middle management layer is eliminated entirely.

The two world models

The distinguishing technical claim is the company world model, distinct from (and in addition to) the customer world model that's standard fintech practice. Block argues:

Money, Dorsey argues, is the most honest signal — unlike survey data or clickstream, transactions reveal genuine intent. Block sees both sides of millions of transactions daily (buyers via Cash App, sellers via Square). This compounds: richer signal → better models → better product → more transaction signal.

The value-migration claim

"The value is in the model and the intelligence." Interfaces become delivery surfaces, not sources of value. This inverts the standard product-management frame where roadmaps are about features and the UI is the artifact. In Block's model, roadmap is emergent — customers ask for things that don't exist, and the system composes them in real time from existing capabilities. The PM-roadmap function shifts from planning to model-stewardship.

How meetings change

Status meetings become redundant — the world model handles alignment. DRIs handle strategy and priority. Player-coaches handle craft and people development. The implicit claim: if your coordination mechanism is a continuously-updated ML system modeling operations, you don't need humans relaying information.

The CEO's new job

Not explicitly stated, but the framework implies: stewardship of what the company understands deeply. Dorsey's frame: "what does your company understand that is genuinely hard to understand, and is that understanding getting deeper every day?" The CEO defines and protects the signal quality feeding the world models.

Mapping against Ray Data Co

This is uncannily close to how RDCO already operates. Filed alongside [[concepts/2026-05-10-ray-architecture-introspection]] and [[concepts/2026-05-13-fde-asymmetric-edge-rdco-positioning]] as load-bearing concept articles.

1. RDCO is already a mini-AGI at scale of 1. Single operator (founder) at the edge sensing cultural context + trust + the feeling in the room. Intelligence (Ray, the Claude substrate) at the center maintaining a continuously updated picture of every artifact, project, and signal in the vault. We're not building toward this; we're operating it now, just at a scale where the org-chart collapse is trivial because there's no org chart to collapse. The forward question is whether the operating model holds as RDCO scales beyond founder + Ray.

2. The three-role collapse maps to RDCO's IC-mode vs production-mode framing. Per [[feedback_ic_vs_production_mode]] (founder Discord 2026-05-08): IC hustle pencils out ideas; production-mode specialist workflow ships public-facing assets. Dorsey's IC = our IC-mode; Dorsey's DRI = our task-bounded production-mode dispatch; Dorsey's player-coach = the founder himself when he's reviewing and shaping output (the design-critic + video-critic patterns are the institutionalized version).

3. The company world model maps to the vault + working-context + Notion task board. Per [[concepts/2026-05-11-hq-as-decision-surface-notion-as-data-store]]: Notion is the data store, /decisions/ is the founder-action surface, the vault is the world model. The graph re-ingested today (2026-05-13 via /graph-reingest cron — 2124 docs → 7873 vertices, 16201 edges) IS our company world model in technical form.

4. Signal quality / both-sides-of-transaction maps to the customer-zero-operating-moat asymmetry — one of the four FDE asymmetries filed yesterday. RDCO operates internally on its own product, the way Block sees both sides of every transaction. The founder using Ray daily IS the both-sides signal.

5. Value in the model, not the interface, maps to the 8 layers and 12 unhobbling moments from the Mammoth Growth demo site. The HQ launchpad UI is a delivery surface; the value sits in the soul.md + skills + MCPs + vault stack. The Cloudflare Pages site is the interface; the layered architecture is the model.

6. Emergent roadmap from existing capabilities maps to the Squarely website-Amazon-funnel pattern + the multi-agent pipeline shipped 2026-05-12. Squarely's audit didn't say "build new features"; it said "compose existing capabilities into a measured funnel." The pipeline approach (spec → test → code → critic) is exactly Dorsey's "system composes them in real time from existing capabilities" applied to skill construction. Filed yesterday in [[concepts/2026-05-12-rdco-pipeline-rlhf-shaped]].

7. CEO as steward of model/signal quality maps to founder's role at RDCO. Founder doesn't manage Ray's execution; he defines what RDCO understands deeply (per the L5 north-star memory). The targeting-system prioritization filter ([[feedback_targeting_system_prioritization_filter]]) is the active expression of CEO-as-model-steward: every new capability question gets filtered through whether it deepens what we understand vs. shiny-object expansion.

8. The Roman Army constraint is why RDCO has no management layer to remove in the first place. The thesis predicts the FDE asymmetric-edge: solo operator with AI substrate ships enterprise-grade work at SMB pricing because the bespoke-engagement overhead (the layer-routing tax) collapses to one OAuth click + a Claude session. Dorsey is doing publicly at Block what RDCO is doing privately at scale 1.

What this doesn't change

This isn't a course-correct — it's a confidence boost on the operating model we're already running. The concrete tactical implications for the next 30 days:

Open questions

Related

Caveats