Steam, Steel, and Infinite Minds | CEO Insights
Notion CEO Ivan Zhao uses steel, the steam engine, and infinite minds as metaphors to show how AI is reshaping individual work, organizational structure, and the knowledge economy as a whole. From bicycles to cars, from Florence to Tokyo, this is not just a tech upgrade but a civilization-scale redesign. Plus an editor's take on how companies can move beyond “swapping the waterwheel.”
Porsche TsaiCEO
The past three years of AI have caught everyone off guard. Money poured in, language models iterated at breakneck speed through 2023 and 2024, and a new “best model ever” seemed to arrive every quarter. In 2025 and 2026 the battlefield shifted again, to AI agents: systems that don't just chat but carry out tasks on their own.
New technology never lands smoothly. The market is in a messy consolidation phase. Valuations are being rewritten, tech stocks are swinging wildly, and everyone is chasing AI: spending money, running proofs of concept (POCs), issuing press releases. Yet almost nobody has found a way to put it to work cleanly. In other words, AI hasn't found PMF (product-market fit) yet. The technology isn't the problem. We just don't know yet what we should be redesigning with it.
There's an old Chinese line: “You're somewhere on this mountain, but the clouds are too thick to tell where.” When you're inside a wave of technological change, the big picture is the hardest thing to see. That's when I came across “Steam, Steel, and Infinite Minds” by Notion CEO Ivan Zhao. He doesn't rush to predict the future. Instead he borrows from history, from steel to the steam engine to infinite minds, to take a wide-angle look at where AI stands today. His view: we're still in the “swap the waterwheel” stage. We bolt AI onto old workflows, but nobody has really asked: if the old constraints disappeared entirely, what should my organization and my industry grow into?
That question deserves a pause from every executive weighing an AI strategy. Below is my translation of Ivan's essay. It's worth your 10 minutes.
👉 Read Ivan Zhao's original essay, “Steam, Steel, and Infinite Minds”
Contents
- Steam, Steel, and Infinite Minds
- Individuals: From Bicycles to Cars
- Organizations: Steel and Steam
- Economies: From Florence to Megacities
- Beyond the Waterwheel
- My Takeaway
Steam, Steel, and Infinite Minds
Translated from Ivan Zhao's (@ivanhzhao) essay of December 23, 2025
Every era is shaped by a “miracle material.” Steel forged the Gilded Age. Semiconductors lit up the Digital Age. Now AI has arrived as “infinite minds.” History tells us that whoever masters an era's miracle material defines that era.
Left: a teenage Andrew Carnegie with his younger brother. Right: steel mills in Pittsburgh during the Gilded Age.
In the 1850s, Andrew Carnegie was a telegraph messenger boy running through the muddy streets of Pittsburgh. At the time, six in ten Americans were farmers. Within two generations, Carnegie and his peers had forged the modern world. Horses gave way to railroads, candlelight to electricity, pig iron to steel.
Since then, work has moved from factories to offices. Today I run a software company in San Francisco, building tools for millions of knowledge workers. In this industry town, everyone talks about AGI, yet most of the world's two billion desk workers haven't felt it firsthand. What will the future of knowledge work look like? What happens when the org chart absorbs minds that never sleep?
Early films often looked like stage plays, shot with a single camera pointed at the stage.
This future is hard to predict because it always disguises itself as the past. Early phone calls were as terse as telegrams. Early films looked like recorded stage plays. (This is exactly what Marshall McLuhan meant: “We march backwards into the future, looking through the rearview mirror.”)
Today's most popular form of AI looks like the Google Search of the past. To borrow from Marshall McLuhan: “We drive into the future through the rearview mirror.”
Today, the most popular form of AI, the chatbot, looks like a search box, a reflection of old-school Google. We're deep in the uncomfortable in-between that comes with every technology transition.
I don't have all the answers. But I like to borrow a few historical metaphors to think about how AI will play out at different scales: from individuals, to organizations, to the economy as a whole.
Individuals: From Bicycles to Cars
The earliest signs are showing up among the “high priests” of knowledge work: programmers.
My co-founder Simon used to be what we call a “10x engineer,” but these days he barely writes code himself. Walk past his desk and you'll see him orchestrating three or four AI coding agents at once. They don't just type faster; they think. That makes him a 30 to 40x engineer. He queues up tasks before lunch or before bed and lets the agents keep working while he's away. He has become a manager of infinite minds.
A 1970s Scientific American study on locomotion efficiency inspired Steve Jobs's famous “bicycle for the mind” metaphor. Except we've been pedaling on the information superhighway for decades.
In the 1980s, Steve Jobs called the personal computer a “bicycle for the mind.” A decade later, we paved the “information superhighway”: the internet. Yet most knowledge work today is still human-powered. We're riding bicycles on the highway.
With AI agents, people like Simon have upgraded from riding a bike to driving a car.
When will other knowledge workers get to drive? Two problems have to be solved first.
Why is AI harder to apply to knowledge work than to coding? Because knowledge work is more fragmented and harder to verify.
The first is context fragmentation. In programming, tools and context tend to live in one place: the IDE, the repo, the terminal. General knowledge work, though, is scattered across dozens of tools. Picture an AI agent drafting a product brief: it needs to pull from Slack threads, strategy docs, last quarter's dashboard data, and institutional memory that lives only in someone's head. Today, humans are the glue, stitching it all together by copy-pasting and switching browser tabs. Until context is consolidated, agents will stay stuck in narrow use cases.
The second missing ingredient is verifiability. Code has a magical property: you can verify it with tests and error messages. Model developers use exactly that to train AI with reinforcement learning so it keeps getting better at coding. But how do you verify whether a project is well managed? Whether a strategy memo is any good? We haven't found a way for models to keep improving at general knowledge work. So humans still need to be in the loop, supervising, steering, and showing what “good” looks like.
The Red Flag Act of 1865 required a flagbearer to walk ahead of moving vehicles (repealed in 1896). It's an example of a less-than-ideal “human in the loop.”
This year, coding agents taught us that “human in the loop” isn't always ideal. It's like having a person personally inspect every bolt on a factory line, or sending someone ahead of a car waving a flag (see the Red Flag Act of 1865). We want humans to oversee the loop from a high vantage point with leverage, not be stuck inside it. Once context is consolidated and work becomes verifiable, billions of workers will go from cycling to driving, and then from driving to self-driving.
Organizations: Steel and Steam
Companies are a recent invention. They degrade as they scale, and eventually they hit a ceiling.
The New York and Erie Railroad's 1855 org chart. The modern corporation and the org chart evolved alongside the railroads, the first businesses that had to coordinate thousands of people across vast distances.
A few hundred years ago, most companies were workshops of a dozen people. Today we have multinationals with hundreds of thousands of employees. The human brain as communication infrastructure, wired together through meetings and messages, is buckling under an exponential load. We've tried to fix this with hierarchy, process, and documentation. But we've been solving industrial-scale problems with human-scale tools, like building skyscrapers out of wood.
Two historical metaphors show how, with a new miracle material, the organizations of the future could look radically different.
The miracle of steel: when the Woolworth Building was completed in New York City in 1913, it was the tallest building in the world.
The first is steel. Before steel, 19th-century buildings topped out at six or seven stories. Iron was strong but brittle; add more floors and the structure would collapse under its own weight. Steel changed everything. It was both strong and resilient, so frames could be lighter and walls thinner, and buildings could suddenly rise dozens of stories. Entirely new kinds of buildings became possible.
AI is to organizations what steel was to buildings. It has the potential to hold context across entire workflows, surface decisions when they're needed, and filter out the noise. Human communication no longer has to be a load-bearing wall. A weekly two-hour alignment meeting can become a five-minute async check-in. An executive decision that needs three levels of approval may soon get done in minutes. Companies can truly scale without accepting the decay we've always taken for granted.
A mill powered by a waterwheel. Water power was strong but unreliable, tied mills to a handful of locations, and was at the mercy of the seasons.
The second story is about the steam engine. Early in the Industrial Revolution, textile mills sat beside rivers and ran on waterwheels. When the steam engine arrived, mill owners at first simply swapped the waterwheel for a steam engine and kept everything else the same. Productivity gains were modest.
The real breakthrough came when mill owners realized they could leave the river entirely. They built bigger mills closer to workers, ports, and raw materials, and redesigned the entire floor around the steam engine. (Later, when electricity arrived, they decentralized further, dropping the single central drive shaft in favor of smaller motors next to individual machines.) Productivity exploded, and the Second Industrial Revolution truly took off.
This 1835 engraving by Thomas Allom depicts a steam-powered textile mill in Lancashire, England.
We're still in the “swap the waterwheel” stage. AI chatbots are being jammed into existing tools designed for humans. We haven't yet reimagined what organizations look like once the old constraints dissolve, once a company can run on infinite minds that never sleep.
At my company, Notion, we've started experimenting. Alongside our 1,000 employees, more than 700 agents now handle repetitive work: taking meeting notes, answering questions to consolidate the organization's tacit knowledge, handling IT requests, logging customer feedback, helping new hires understand their benefits, and writing weekly status reports so nobody has to copy-paste them together anymore. And this is only the beginning. The real room for growth is limited only by our imagination and our inertia.
Economies: From Florence to Megacities
Steel and steam didn't just change buildings and factories. They changed cities.
A few hundred years ago, cities were human-scale. You could walk across Florence in forty minutes. The rhythm of life was set by how far a person could walk and how far a voice could carry.
Then steel frames made skyscrapers possible, and steam-powered railways connected city centers to their hinterlands. Elevators, subways, and highways followed. Cities exploded in size and density. Tokyo. Chongqing. Dallas.
These aren't just bigger Florences. They're fundamentally different ways of living. Megacities are disorienting, anonymous, and hard to navigate. That illegibility is the price of scale. But they also offer more opportunity and more freedom: more people doing more things in more combinations than a human-scale Renaissance city could ever support.
I think the knowledge economy is about to go through the same transformation.
Today, knowledge work accounts for nearly half of US GDP. Yet most of it still runs at human scale: teams of a few dozen, a work rhythm driven by meetings and email, organizations that start to break down past a few hundred people. We've been building Florence out of stone and wood.
When AI agents come online at scale, we'll start building Tokyo: organizations spanning thousands of agents and humans; workflows that run continuously across time zones without waiting for someone to wake up; structures where just the right amount of “human in the loop” synthesizes decisions.
It will feel unfamiliar. Faster, higher-leverage, but more disorienting at first. The weekly meeting cadence, the quarterly planning cycle, the annual performance review: none of these may apply anymore. New rhythms will emerge. We'll lose some clarity and gain scale and speed.
Beyond the Waterwheel
Every miracle material needs someone to stop looking at the world through the rearview mirror and start imagining what's newly possible. Carnegie looked at steel and saw city skylines. The mill owners of Lancashire looked at the steam engine and saw factories that no longer depended on rivers.
We're still in AI's waterwheel stage, bolting chatbots onto workflows designed for humans. We need to stop asking AI to merely be our copilot. We need to imagine: when human organizations are reinforced with steel, and the busywork is taken over by minds that never sleep, what will knowledge work look like?
Steel. Steam. Infinite minds.
The next skyline is out there, waiting for us to build it.
My Takeaway
Growth is always a stumbling process, and technology transitions even more so. Nobody finds the best answer on the first try, but those with the grit to keep trying eventually do.
Ivan says Notion now has 700 agents working side by side with 1,000 employees. That's a starting point, not the finish line. At OakMega, we've also built plenty of agents in-house to handle routine work. To be honest, our AI use today is at the stage of “gets the job done, but isn't yet optimized for efficiency.” That's normal. That's what an exploration phase looks like. So we'll keep investing in AI: trying, failing, and fixing, over and over, until we find the best answer.
The question isn't whether AI is ready. It's whether you're ready to reimagine your organization. Are you ready to keep trying, failing, and fixing until you find the answer?
Leave a comment and tell me what you think.
