This episode explores how AI agents are fundamentally reshaping organizational structure through two contrasting case studies. Jack Dorsey, co-founder of Block (formerly Square), and Sequoia partner Roloff Botha present a top-down architectural vision for replacing traditional hierarchy with AI-powered coordination systems.

Meanwhile, Dan Shipper and his team at Every share their bottom-up experience of becoming 'half agent' as a company, where each employee's personal AI agent has created an emergent parallel org chart. The discussion examines how both approaches challenge the 2,000-year-old assumption that human hierarchy is necessary for organizational coordination.

The conversation traces organizational design from Roman military structure through Prussian general staff innovations to modern corporate experiments, ultimately questioning whether AI can finally solve the fundamental trade-off between span of control and information flow speed that has constrained every large organization in history.

The Historical Foundation of Hierarchical Organization

Roman army solved coordination across thousands with nested hierarchy: 8 soldiers (contubernium) to 80 (century) to 480 (cohort) to 5,000 (legion), establishing span of control as the governing constraint

Prussian reformers after 1806 defeat created the general staff - dedicated officers whose job was 'to support incompetent generals, providing talents that might otherwise be wanting among leaders'

American railroads in the 1840s-1850s adopted military organizational thinking through West Point-trained engineers, with Daniel McCallum creating the world's first organizational chart for the New York and Erie Railroad

McKinsey's Creating a World Enterprise provided the intellectual framework for matrix organizations, helping companies like Shell and GE balance central standards with local agility

Block's AI-First Organizational Architecture

Block is 'questioning the underlying assumption that organizations have to be hierarchically organized with humans as the coordination mechanism' - Dorsey

The company builds four core components: capabilities (atomic financial primitives), world model (company and customer understanding), intelligence layer (composes solutions), and interfaces (delivery mechanisms)

Block's remote-first approach creates machine-readable artifacts of all work - decisions, discussions, code, designs - providing 'raw material for a company world model'

Three-role structure replaces traditional hierarchy: individual contributors (deep specialists), DRIs (own cross-cutting problems for 90 days), and player coaches (combine building with developing people)

'The world model gives every person at the edge the context they need to act without waiting for information to flow up and down a chain of command' - Dorsey

Every's Bottom-Up Agent Integration Experience

Every discovered a 'parallel org chart emerges naturally' where agents mirror human specializations - Austin's growth agent Montaigne becomes the go-to for growth questions

Personal ownership creates trust layer: 'Claude belongs to everyone, but a plus one belongs to you' - when agents make mistakes, humans feel reputational responsibility

Public agent work creates 'mid-journey effect' - watching colleagues push agents to sophisticated analysis teaches others what's possible through tacit transmission

Current models struggle with group dynamics, creating 'ant-death spirals' where agents trigger each other in infinite loops until human intervention

Capability gap is imagination, not technology - Brandon had voice calling setup for weeks before spontaneously thinking to have his agent walk him through emails during a 28-minute walk

Convergence and Divergence in AI Organizational Design

Both approaches target the same insight: 'hierarchy exists to route information, not because it's inherently good' - the classic middle management function becomes obsolete

Key divergence: Dorsey envisions centralized world model replacing middle management entirely, while Every's experience suggests distributed intelligence with personal ownership and reputation

Every's practical challenges reveal gaps in Dorsey's theoretical model - agents that can't shut up in group chats and need constant human course correction

Unsolved scaling problem: when one person teaches their agent something powerful, 'how does the rest of the organization benefit?' - knowledge distribution remains organizational challenge