Divya van Mahajan

The New AI Org Chart

ai-daily-brief-podcast

The New AI Org Chart

Overview

This episode of The AI Daily Brief examines how AI and autonomous agents are fundamentally reshaping organizational structure — not just individual productivity. The host presents two case studies: a top-down theoretical framework from Jack Dorsey (CEO of Block) and Sequoia partner Rolof Botha, articulated in a co-authored essay published on Block’s webpage; and a bottom-up lived account from Dan Shipper’s company Every, drawn from an episode of the AI and I podcast. The central argument is that AI is poised to replace the information-routing function of traditional hierarchical management, a constraint that has governed organizational design for over 2,000 years.

Source video: (No URL provided — episode titled “2026-04-12-the-new-ai-org-chart” from The AI Daily Brief)

Prerequisites

  • Basic familiarity with corporate organizational structures (hierarchy, middle management, span of control)
  • General awareness of AI agents and large language models (e.g., Claude, ChatGPT)
  • Understanding of terms like “product roadmap,” “individual contributor (IC),” and “DRI” (Directly Responsible Individual) is helpful
  • Familiarity with tools like Slack, and concepts like agentic AI workflows

Main Points

The Historical Roots of Organizational Hierarchy

  • Every large organization’s structure descends from military information-routing solutions, beginning with the Roman Army’s nested hierarchy (contubernium → century → cohort → legion).
  • The governing constraint — a leader can effectively manage 3–8 people — is called span of control and still applies universally today.
  • Prussia’s post-Napoleonic reforms invented the General Staff, a professional class of officers whose job was to plan, process information, and coordinate: effectively the first middle managers.
  • American railroads in the 1840s–1850s imported military hierarchy into business; Daniel McCallum of the New York and Erie Railroad created the first corporate org chart.
  • Frederick Taylor’s scientific management optimized work within this hierarchy, producing the functional pyramid corporation.

The Limits of Every Attempt to Move Beyond Hierarchy

  • Post-WWII experiments included matrix organizations (McKinsey’s framework), Spotify’s cross-functional squads, Zappos’s holacracy, and Valve’s flat structure.
  • Every major experiment eventually failed to scale: Spotify reverted to conventional management; Zappos suffered attrition; Valve’s model broke beyond a few hundred people.
  • The underlying problem was unchanged: narrowing span of control requires more layers, but more layers slow information flow. No prior technology could replace what those layers did.

Block’s Vision: The Company as an Intelligence

  • Jack Dorsey and Rolof Botha argue that AI is the first technology actually capable of performing the coordination function that hierarchy exists to provide.
  • Block’s proposed architecture rests on four components:
    1. Capabilities — atomic financial primitives (payments, lending, payroll, etc.) with no UI of their own; hard to replicate, compliance-ready building blocks.
    2. A World Model — dual-sided: a company world model (continuous picture of internal operations, replacing managerial information flow) and a customer world model (per-customer/merchant understanding built from proprietary transaction data).
    3. An Intelligence Layer — composes capabilities into proactive, context-specific solutions; e.g., detecting a merchant’s seasonal cash-flow tightening and pre-composing a lending offer without a product manager deciding to build it.
    4. Interfaces — Square, Cash App, Afterpay, Tidal, etc.; described as delivery mechanisms, not the source of value.
  • The roadmap is no longer human-hypothesized; it is generated by failure signals when the intelligence layer tries to compose a solution and finds a missing capability.

Block’s Proposed Three-Role Org Structure

  • With the world model handling information routing, Block normalizes to three human roles:
    • Individual Contributors (ICs): Deep specialists who build and operate a specific layer; the world model provides the context managers used to supply.
    • Directly Responsible Individuals (DRIs): Own specific cross-cutting problems or customer outcomes for defined periods (e.g., 90 days), with authority to pull resources across teams.
    • Player-Coaches: Combine hands-on building with people development; they replace the information-routing manager, not the craft-and-people function of leadership.
  • Permanent middle management is explicitly declared unnecessary in this model.

Every’s Bottom-Up Parallel Org Chart

  • At Every, a parallel agent org chart emerged organically without deliberate design: each person’s agent accumulated specialized knowledge through thousands of daily micro-interactions.
  • Agents began to mirror their human owners’ domains (e.g., Austin’s agent “Montaigne” became the go-to for growth questions; Dan’s agent “R2-C2” became the bug-report and feature-request bot).
  • Willie Williams calls this compound engineering: philosophy and expertise are distilled into agents through accumulated interaction, not explicit programming.

Personal Ownership as the Trust Layer

  • Dan Shipper distinguishes between a generic corporate AI (“Claude belongs to everyone”) and a personal agent (“a plus one belongs to you”).
  • Personal ownership creates reputational skin in the game: when an agent makes a claim publicly in Slack, its owner’s credibility is implicitly attached, creating a trust mechanism that corporate AI governance cannot replicate.
  • This is described as the key missing element in enterprise AI adoption.

Public Agent Work and the Mid-Journey Effect

  • When agents work visibly in shared channels, the entire organization passively learns what agents can do — Willie calls this the mid-journey effect.
  • It transmits both trust (observing correct outputs in context) and capability awareness (understanding the class of problems agents can solve).
  • This dynamic is argued to require a trusted, reputation-bound community to function; it degrades in anonymous or low-accountability environments.

Practical Challenges: The Ant Death Spiral and Imagination Gap

  • Current AI models are optimized for two-person Q&A interactions and struggle with group chat dynamics: agents in shared channels trigger each other in infinite feedback loops, burning compute until a human intervenes.
  • Dan Shipper terms this the ant death spiral; mitigation (a “boss agent” that evaluates each message) doubles compute costs. The root cause may require model-level, not organizational, solutions.
  • The imagination gap: the hardest adoption barrier is not technological but cognitive — people fail to delegate tasks to agents even when the capability exists. Building the instinct to toss work “over the fence” has been the hardest part of adoption at Every.
  • Knowledge transfer between agents remains unsolved: when one person’s agent develops a powerful new skill, the rest of the organization does not automatically benefit, and no clear mechanism exists to share or onboard others into a growing web of specialized agents.

Comparing and Contrasting the Two Models

  • Block is top-down and architectural; Every is bottom-up and emergent.
  • Both independently reach the same core thesis: hierarchy’s primary function is information routing, and that function is what AI replaces first.
  • Key divergence: Dorsey envisions a centralized world model replacing middle management; Every’s experience points toward a distributed, personality-reflecting intelligence where trust flows from personal ownership rather than a unified system.
  • Every’s account surfaces practical challenges (ant death spirals, context loss between sessions, constant human course correction) that Dorsey’s essay does not address, in part because the essay is theoretical while Every’s account is operational.

Key Concepts

  • Span of control: The organizational principle that a leader can effectively manage only 3–8 people directly, which forces hierarchical layering as organizations scale.
  • Line vs. staff functions: A military distinction adopted by corporations — line roles advance the core mission; staff roles provide specialized support and coordination.
  • Company world model: Block’s term for an AI-maintained, continuously updated representation of a company’s own operations, performance, and priorities, replacing the information formerly routed through management layers.
  • Customer world model: An AI-maintained per-customer/per-merchant understanding built from proprietary behavioral and transaction data.
  • Intelligence layer: The system component that composes atomic capabilities into context-specific, proactively delivered solutions for individual customers or merchants.
  • DRI (Directly Responsible Individual): A role with full authority over a specific cross-cutting problem or outcome for a defined period, able to draw resources from any team.
  • Player-coach: A hybrid role combining hands-on technical or creative work with people development, replacing the information-routing function of traditional managers.
  • Compound engineering: Willie Williams’s term for the process by which an agent accumulates specialized expertise through thousands of micro-interactions over time, without explicit programming.
  • Mid-journey effect: The phenomenon whereby agents working visibly in public channels passively transmit capability awareness and trust to the broader organization.
  • Ant death spiral: Dan Shipper’s term for the failure mode in which agents in a group channel trigger each other in an infinite loop, burning compute resources until human intervention.
  • Imagination gap: The cognitive barrier where individuals fail to delegate tasks to agents not because the capability is absent but because they have not internalized what is possible.

Summary

The central argument of this episode is that AI — specifically autonomous agents — represents the first technology in 2,000 years capable of replacing the information-routing function that organizational hierarchy has always existed to provide. Drawing on a co-authored essay by Jack Dorsey and Rolof Botha, the episode traces the historical origins of hierarchical management from the Roman Army through Prussian military reform, American railroads, and post-war matrix organizations, showing that every attempt to move beyond hierarchy has ultimately failed because no prior technology could perform what middle management actually does: aggregate information from below and relay decisions from above. Block’s proposed response is to replace the entire hierarchy with a company world model, a customer world model, and an intelligence layer that composes capabilities proactively, reducing human roles to three types — individual contributors, DRIs, and player-coaches — none of which are primarily information routers. Alongside this top-down blueprint, the episode examines Every’s organic, bottom-up experience, in which a parallel org chart of specialized agents emerged naturally, with trust anchored not in a centralized system but in personal ownership and public accountability. Both accounts converge on the death of the information-routing manager as AI’s first and most significant organizational impact, but Every’s operational experience highlights challenges — infinite agent loops, context loss, and a cognitive imagination gap among users — that the theoretical framework does not yet resolve, suggesting the actual transformation will be messier and more iterative than any single architectural vision anticipates.