We are still in the “waterwheel stage” of AI, forcefully fitting chatbots into workflows designed for humans. History shows us that whoever controls the raw materials defines the era. When knowledge work integrates into never-resting intelligence, what will the future look like? This article is based on a piece by Ivan Zhao, CEO of Notion, compiled, translated, and written by TechFlow.

(Previous context: What are crypto users most concerned about in 2025? Different AI large models provide these answers)

(Additional background: a16 article: When AI invades content platforms, how can crypto staking regain trust?)

Table of Contents

• Personal: From Bicycle to Car
• Organization: Steel and Steam
• Economy: From Florence to Megacities
• Beyond Waterwheels

We are still in the “waterwheel stage” of AI, forcefully fitting chatbots into workflows designed for humans. Every era is shaped by its unique technological raw materials. Steel forged the Gilded Age, semiconductors ushered in the Digital Age. Now, artificial intelligence arrives in the form of infinite intelligence. History tells us: those who control the raw materials define the era.

Left image: Young Andrew Carnegie and his brother. Right image: Steel mill in Pittsburgh during the Gilded Age.

In the 1850s, Andrew Carnegie was a telegram operator running through muddy streets in Pittsburgh, at a time when six out of ten Americans were farmers. Just two generations later, Carnegie and his peers forged the modern world—horses gave way to railroads, candlelight to electric lamps, and iron to steel.

Since then, work shifted from factories to offices. Today, I run a software company in San Francisco, building tools for thousands of knowledge workers. In this tech town, everyone talks about Artificial General Intelligence (AGI), but most of the two billion office workers have yet to feel its presence. Soon, what will knowledge work look like? When organizational structures incorporate never-resting intelligence, what will happen?

Early films often resembled stage plays, with a single camera filming the stage.

The future is often unpredictable because it always disguises itself as the past. Early communication was as brief as telegrams; early films resembled recorded stage plays. As Marshall McLuhan said: “We always drive into the future looking in the rearview mirror.”

Today’s most common form of AI still looks like the old Google search. Quoting McLuhan: “We always drive into the future looking in the rearview mirror.” Today, what we see is AI chatbots mimicking Google search boxes. We are deeply immersed in that uncomfortable transition period that always accompanies every technological revolution.

I don’t have all the answers about what the future will be like. But I like to use a few historical metaphors to think about how AI might play different roles at the individual, organizational, and economic levels.

Personal: From Bicycle to Car

The earliest signs can be seen in the “advanced practitioners” of knowledge work—programmers.

My co-founder Simon was once a “tenfold programmer,” but lately he rarely writes code himself. Walking past his desk, you’ll see him coordinating three or four AI coding assistants simultaneously. These assistants not only type faster but also think, making him an engineer whose efficiency has increased by 30 to 40 times. He often pre-sets task queues before lunch or bedtime, letting AI continue working when he leaves. He has become a manager of infinite intelligence.

A 1970s Scientific American study on movement efficiency inspired Steve Jobs to propose the famous metaphor of the “bicycle of thought.” But since then, for decades, we’ve been riding the information superhighway on bicycles.

In the 1980s, Steve Jobs called personal computers “bicycles for the mind.” A decade later, we paved the “information superhighway” called the internet. But today, most knowledge work still relies on human effort. It’s like we’ve been riding bicycles on a highway.

With AI assistants, people like Simon have upgraded from riding bicycles to driving cars.

When will other types of knowledge workers be able to “drive cars”? Two issues need to be addressed.

Why is AI-assisted knowledge work more difficult than programming assistants? Because knowledge work is more fragmented and harder to verify.

First is contextual fragmentation. In programming, tools and contexts are often centralized: integrated development environments, code repositories, terminals. But general knowledge work is scattered across dozens of tools. Imagine an AI assistant trying to draft a product overview: it needs to extract information from Slack discussion threads, strategic documents, dashboards with last quarter’s data, and organizational memories stored only in someone’s mind. Currently, humans are the glue, copying and pasting, switching tabs in browsers. Without integrated contexts, AI assistants can only serve narrow purposes.

The second missing element is verifiability. Code has a magical property: you can verify it through tests and error messages. Model developers leverage this by training AI to code better through reinforcement learning. But how do you verify if a project management process is good, or if a strategic memo is excellent? We haven’t yet found ways to improve general knowledge work models. Therefore, humans still need to supervise, guide, and demonstrate what “good” looks like.

In 1865, the “Red Flag Act” required a person with a flag to walk in front of automobiles on the street (the law was repealed in 1896).

This year’s programming assistant practices tell us that “humans in the loop” are not always ideal. It’s like having people individually check bolts on a production line, or walk in front of cars to clear the way (see the 1865 Red Flag Act). We should let humans oversee cycles from a higher vantage point, rather than being embedded within them. Once contexts are integrated, work becomes verifiable, and hundreds of millions of workers will shift from “riding bicycles” to “driving cars,” and then from “driving” to “autonomous driving.”

Organization: Steel and Steam

Companies are a modern invention; as they scale, efficiency diminishes and eventually hits a limit.

Organization chart of the New York and Erie Railroad in 1855. Modern companies and their organizational structures evolved from railroad companies, which were among the earliest enterprises requiring coordination of thousands over long distances.

Hundreds of years ago, most companies were just workshops with a dozen people. Today, we have multinational corporations with hundreds of thousands of employees. Communication infrastructure relies on meetings and interconnected minds, which become overwhelmed under exponential growth. We try to solve this with hierarchies, processes, and documents, but that’s like building skyscrapers with wood—using human-scale tools to solve industrial-scale problems.

Two historical metaphors show how organizations might look different when equipped with new technological raw materials.

The miracle of steel: The Woolworth Building in New York, completed in 1913, was once the tallest building in the world.

The first is steel. Before steel, 19th-century buildings were limited to six or seven stories. Iron was strong but brittle and heavy; adding floors would cause structures to collapse under their own weight. Steel changed everything. It is strong and flexible, allowing lighter frameworks, thinner walls, and buildings soaring dozens of stories high. This made new architectural possibilities possible.

AI is the “steel” of organizations. It promises to maintain contextual coherence across workflows, presenting decisions when needed without noise. Human communication no longer needs to serve as load-bearing walls. Weekly alignment meetings could become five-minute asynchronous reviews; high-level decisions requiring three layers of approval might be completed in minutes. Companies can truly scale without the efficiency decline we once considered inevitable.

A mill powered by waterwheels. Water power is strong but unstable, limited by location and season.

The second story is about steam engines. In the early industrial revolution, textile factories were built along rivers, driven by waterwheels. When steam engines appeared, factory owners initially replaced waterwheels with steam engines, but everything else remained the same, with limited productivity gains.

The real breakthrough came when factory owners realized they could completely free themselves from water sources. They built larger factories near workers, ports, and raw materials, and redesigned layouts around steam engines (later, with electricity, they further detached from central power shafts, dispersing small engines to power different machines). Productivity exploded, and the Second Industrial Revolution truly began.

Thomas Allom’s 1835 lithograph depicts a steam-powered textile mill in Lancashire, England.

We are still in the “replacing waterwheels” stage. Forcefully fitting AI chatbots into workflows designed for humans, we have yet to reimagine what organizations will look like when old constraints vanish and companies can run on infinite intelligence working even while you sleep.

At my company Notion, we have been experimenting. Besides 1,000 employees, there are now over 700 AI assistants handling repetitive tasks: recording meetings, answering questions to consolidate team knowledge, handling IT requests, recording customer feedback, helping new employees get familiar with benefits, writing weekly status reports to avoid manual copy-pasting… This is just the beginning. The true potential is only limited by our imagination and inertia.

Economy: From Florence to Megacities

Steel and steam changed not only buildings and factories but also cities.

Until a few hundred years ago, cities were on a human scale. You could walk across Florence in forty minutes; the rhythm of life was determined by human walking distances and sound propagation.

Later, steel frameworks made skyscrapers possible; steam-powered railways connected city centers with hinterlands; elevators, subways, and highways followed. The scale and density of cities expanded rapidly—Tokyo, Chongqing, Dallas.

These are not just enlarged Florence; they are entirely new ways of living. Megacities can be overwhelming, anonymous, and hard to control. This “indistinguishability” is the cost of scale. But they also offer more opportunities, more freedom, supporting more people engaging in more activities in more diverse combinations—something that Renaissance cities on a human scale cannot match.

I believe the knowledge economy is about to undergo a similar transformation.

Today, knowledge work accounts for nearly half of US GDP, but its operation still largely remains on a human scale: teams of dozens, workflows dependent on meetings and emails, organizations that struggle beyond a hundred people… We have been building “Florence” with stone and wood.

When AI assistants are widely adopted, we will build “Tokyo”—organizations composed of thousands of AI and human collaborators; workflows that run across time zones continuously, without waiting for someone to wake up and push forward; decisions synthesized with just the right human participation.

It will be a different experience: faster, with stronger leverage, but initially also more dizzying. Weekly meetings, quarterly planning, annual reviews may no longer be suitable; new rhythms will emerge. We may lose some clarity but gain scale and speed.

Beyond Waterwheels

Every technological raw material demands that people stop looking at the world through the rearview mirror and start imagining new worlds. Carnegie gazed at steel and saw the city skyline; Lancashire’s mill owners looked at steam engines and envisioned factories away from rivers.

We are still in the “waterwheel stage” of AI, forcefully fitting chatbots into workflows designed for humans. We should not be content with AI as a co-pilot but must imagine: when human organizations are reinforced with steel, when trivial tasks are entrusted to never-resting intelligence, what will knowledge work become?

Steel, steam, and infinite intelligence. The next skyline is ahead, waiting for us to build it ourselves.