Steam, Steel, and Infinite Minds: Are We Repeating the Mistakes of Factory Owners 100 Years Ago?
*An analysis of Notion founder Ivan Zhao's essay on why current AI adoption is stuck in the "swap out the waterwheel" phase.*
Executive Summary
Yes, we are repeating the mistakes of factory owners from 100 years ago. According to Ivan Zhao, founder of Notion, most current AI applications are grafting AI onto existing workflows—much like 19th-century factory owners simply replaced waterwheels with steam engines—capturing only modest gains instead of a true productivity explosion.
A real breakthrough requires escaping the "water source constraint" (the limits of human cognition and communication bandwidth), shifting from "tool replacement" to "system redesign," and building entirely new organizational forms powered by "infinite minds." This is not just a technical problem but a fundamental shift in cognitive frameworks and organizational paradigms.
Core Argument: The "Waterwheel Phase" Dilemma
The Industrial Revolution as Historical Mirror
In early industrial Britain, Lancashire textile mills were built by rivers, powered by waterwheels. When steam engines arrived, most factory owners took the most conservative approach: "swap out the waterwheel"—removing the wheel, installing a steam engine in its place, but keeping the same building layout, workflows, org structure, and riverside location (source). The result: very limited productivity gains.
The real breakthrough came from a minority of forward-looking owners who decoupled from water entirely—relocating factories near workers, ports, and raw materials, and redesigning production around the steam engine's characteristics.
| Waterwheel-swap phase | System-redesign phase | |---|---| | Kept riverside location | Freed from geographic constraints | | Original layout maintained | Production processes redesigned | | Simple power replacement | Distributed power configuration | | Modest gains | Productivity explosion |
Today's AI Paradigm Error
> "We're still in the 'swap out the waterwheel' phase."
The typical symptom: AI chatbots bolted onto existing software and business processes—an AI chat window added to a CRM, an AI writing assistant embedded in an editor, a "copilot" integrated into collaboration platforms.
Moreover, modern corporate structure originated with 19th-century railway companies coordinating thousands of people. Hierarchy, synchronous meetings, and manual approvals all rest on the hidden assumption that humans are the only cognitive agents.
Transformation Framework: Three Levels
1. Individual: From "Bicycle" to "Autopilot"
Building on Steve Jobs' metaphor of the computer as "bicycles for the mind," Zhao describes an evolution:
| Stage | Power source | Multiplier | Work mode | |---|---|---|---| | Bicycle stage | Human effort | 1–10x | Personal execution of all tasks | | 10x engineer | Elite skill | 10x | Deep expertise | | 30–40x engineer | AI orchestration | 30–40x | Multi-agent management, async task queues | | Autopilot stage | Autonomous systems | 100x+ | AI-led, human-supervised, continuous operation |
Case in point: Notion co-founder Simon, already a "10x programmer," now rarely writes code himself—he orchestrates three or four AI agents simultaneously, achieving "30–40x engineer" productivity.
2. Organizational: AI as "Steel"
Today's communication infrastructure is like "building skyscrapers out of wood"—human brains connected through meetings and messages collapse as organizations scale. Every decision needs alignment meetings; every cross-team project needs multi-layer approvals.
> "AI is steel for organizations."
Like steel—which made buildings lighter, walls thinner, and skyscrapers possible—AI maintains context across workflows and surfaces decisions precisely when needed: turning a 2-hour meeting into 5 minutes of async review, and three-tier approvals into minutes.
Notion as a testbed: alongside its 1,000 employees, Notion runs 700+ AI agents handling meeting notes, status reports, internal IT support, onboarding, customer feedback, and insights. Zhao calls this ratio (0.7 agents per employee) merely "baby steps."
3. Economic: From Florence to Tokyo
- Florence: Centuries-old cities were human-scaled—you could walk across Florence in 40 minutes. Today's knowledge economy is a "Florence": teams capped at dozens, workflows paced by meetings and email, organizations breaking past a few hundred people.
- Tokyo: Steel and steam created megacities—not just "bigger Florences" but a fundamentally different way of living: disorienting, anonymous, yet offering far more opportunity, freedom, and combination of people and activity.
| Today: "Sparse" organizations | Future: "Dense" organizations | |---|---| | Human time wasted on meetings, coordination, waiting | Agents run 24/7 without pause | | Context switching drains cognition | Humans focus on key decisions and creative breakthroughs | | Efficiency decays with scale | 1,000 humans + 10,000 agents | | Limited by biology and geography | Cross-timezone, multilingual, global operations |
This redefines the boundary of a "company": a 100-person company could operate 10,000 agents, delivering the service capacity of a traditional tens-of-thousands-person firm.
Conclusion
The lesson from the steam era is clear: transformative technology only pays off when we rebuild systems around its capabilities rather than patching it into old structures. The organizations that "decouple from the water" first—redesigning work around AI's strengths—will capture the productivity explosion that everyone else leaves on the table.
Reference: Ivan Zhao, "Steam, Steel, and Infinite Minds"