AI Industrial Revolution, where are we now?

Written by: Will Awang

In the past year, I attended several industry conferences focused on AI. On stage, guests took turns demonstrating AI tricks, while people below held up their phones to record the screens, posting to social media before continuing to scroll. But back in the office, it’s the same weekly meetings, the same approvals, the same weekly reports. Big companies have already included token consumption in KPIs, and some rely on scripts to inflate numbers, becoming model employees. The folks on social media today talk about Claude’s revolution, tomorrow about Codex’s prowess, the day after about Gemini’s greatness—are they embracing a revolution, or just rushing to keep up?

These are all noise, not the answers I want.

The real question isn’t whether AI is strong enough—steam engines were already built. It’s who will be the first to tear down the old workshops.

The day the Industrial Revolution truly began wasn’t when Watt improved the steam engine, but when factory owners in Lancashire decided to leave the river and rebuild their workshops around steam power. The same applies to AI—it's not the day the large model was invented, but the day an organization chooses to dismantle old processes and rebuild production around AI. That day hasn’t arrived yet. But it’s on the way.

Two people saw this early on. Notion CEO Ivan Zhao wrote a cold assessment in late 2025 titled “Steam, Steel, and Infinite Minds,” concluding we’re still in the “water wheel replacement” stage—adding AI chatbots to existing tools, but no one is redesigning factories. Former OpenAI employee Leopold Aschenbrenner took a different path: wrote 165 pages of “Situational Awareness,” then built a fund that grew from $225 million to $13.68 billion, all betting on AI infrastructure. One looks inward, the other outward—both bets.

This article isn’t about them. It’s about us—where we stand now, and which part of history we’re repeating.

(Power-loom weaving, engraving by J. Tingle after Thomas Allom, 1835 / Wikimedia Commons)

1. The workshop is still old

Most people’s days look like this: in the morning, they use AI to write an email, saving ten minutes; then spend two hours in a weekly meeting that could have been skipped; in the afternoon, copy-paste the same data across three tools; at night, post on social media “AI is awesome.” The ten minutes saved are eaten back by the old processes unchanged.

Similarly, when the steam engine appeared, factory owners initially just replaced water wheels with steam engines—nothing else changed. Factories still built along rivers, still multi-story buildings, still driven by central shafts powering entire lines. We embed ChatGPT into Slack, add Copilot to Office, embed AI chat windows into workflows—doing the same thing. Tools upgraded, workshop unchanged.

But replacing machines doesn’t mean replacing the workshop. McLuhan said it well:

We drive into the future through the rearview mirror. Using old processes to accommodate new tools is like early films being just stage plays shot on camera. True breakthroughs come when someone completely frees the steam engine from the river and redesigns the entire production around new power.

Compare the timeline of the Industrial Revolution with AI, and you can roughly locate where we are on the map:

Today’s timeline is heavily compressed. It took 60 years from the steam engine to railway mania; AI went from Transformer models to data center booms in just 7 years.

Speed isn’t the problem; the question is where we’re stuck—initial four steps are still upgrading old workshops, installing steam engines, laying railroads, but production methods remain unchanged. The sixth step is the real watershed. We’re probably caught between these two.

Steam engines are in hand, but workshops are still old.

2. All capital is poured into the furthest layer from the factory

Infrastructure is always overbuilt. The ones who go bankrupt are investors, not infrastructure.

In 1846, the UK Parliament passed 263 railway acts, approving 9,500 miles of new railways. At the peak, railway investments accounted for 13% of UK GDP. Railway stocks could be bought with just 10% down, and the middle class flooded in. The bubble burst in 1847. One-third of approved lines were never built, countless investors lost everything. Darwin lost 60% on railway stocks—luckier than most.

But the railways remained.

Today’s AI infrastructure follows the same path. Goldman Sachs estimates that by 2026, global AI infrastructure capital expenditure will reach $765 billion, rising to $1.6 trillion annually by 2031. Capital spending by mega cloud providers will go from about 40% of operating cash flow in 2023 to nearly 70% in 2025. AI-related investments already make up about a quarter of all US investments. Aschenbrenner’s $13.68 billion bet is on this layer—he’s not betting on which application will win, but on the underlying compute power itself.

This capital cycle is isomorphic to real estate development. Building data centers is like building skyscrapers: land is electricity, materials are GPUs and storage, contractors are data center builders, developers are cloud providers, tenants are AI application companies, and rent is API revenue. Cloud companies’ business model is “rent to fund loans”—using API income to cover data center capital expenditure, waiting for AI application explosion to boost valuations.

(Compute real estate: a generation’s infrastructure)

The core risk is the same: does the decline in API prices offset the growth in call volume? If rent falls below the debt service line—this is the nightmare familiar to real estate developers. The lesson from 2008 wasn’t overbuilding houses, but that the structure of the houses didn’t match real demand. The AI equivalent risk: oversupply of general compute power, but scarcity of specialized capabilities for high-value scenarios like finance compliance or medical diagnosis.

Railways, real estate, AI—investment in infrastructure across three eras shares a common rule: overbuilding is normal, material suppliers always lose pricing power, and long-term returns belong to owners holding “key locations.” Just look at Q1 fund holdings on Wall Street—probably 80% are in this infrastructure layer: NVIDIA, data centers, cloud infrastructure. But the railway frenzy taught us: this isn’t the full picture of the AI revolution, nor even the most profitable layer.

What is AI’s core “location”? It’s unique industry data and deeply embedded workflows. For individuals, the real “core” isn’t the stocks they hold, but their irreplaceable judgment and industry knowledge—assuming they’ve rebuilt how they use AI around these assets.

The real returns are in the next layer. But the gap between infrastructure and value creation isn’t seamless. There’s a chasm—historically, this gap has swallowed decades.

3. Who is tearing down the workshop

The people dismantling workshops and those “using AI to improve efficiency” aren’t doing the same thing.

Ivan Zhao’s co-founder Simon used to be a “tenfold programmer,” but now rarely codes himself—he manages three or four AI coding agents, boosting efficiency by 30 to 40 times. Notion now has 1,000 employees and over 700 AI agents. The difference isn’t tools; it’s that Simon tore down his old workshop, while most just replaced a water wheel.

600 million Chinese users have used generative AI tools—142% growth—making it the world’s largest AI demand pool. Yet almost no Chinese company has rebuilt core workflows around AI. The largest demand side, with almost no organizational change on the supply side. This contrast itself signals: it’s not tools that are lacking, but organizations that haven’t caught up. Knowledge work’s context is scattered across dozens of tools and minds, outputs are unverifiable, no one knows how to judge whether a strategic memo is effective.

(Labor market impacts of AI: A new measure and early evidence)

Anthropic has already started on a larger scale. They released the Economic Index, which uses real usage data to map which tasks and industries AI first replaces, then built on that: joint ventures with Goldman Sachs, Blackstone, Hellman & Friedman to create AI-native enterprise service companies; a global alliance with KPMG, 276,000 employees accessing Claude; Accenture’s business group trained 30,000 people, focusing on finance, life sciences, and healthcare.

These consulting firms aren’t AI users—they’re AI railway engineers—they don’t build steam engines or lay tracks, but help companies tear down old factories and rebuild production lines around new power. Without this role, most factory owners wouldn’t know where to start.

Signals are flashing. One of the sharpest comes from the job market.

Young people aged 22-25 entering high-exposure AI careers are 14% less likely to find jobs than their peers in low-exposure roles. Entry-level positions are already being squeezed.

If I were a recent graduate, this number would directly impact my job search. If I were a manager, the next batch of entry-level hires might no longer be people.

Organizations are tearing down workshops—what about individuals? My education, my resume, my industry experience—these are my water wheels. They once powered my entire line, but the steam engine has arrived. Top universities like 985 and 211 are no longer a moat; they’re just proof I once built a good factory by the river.

The question now is: do we have the capacity to leave that river?

Anthropic’s data shows that users who have used AI tools for over six months have a 10% higher success rate on tasks than new users. Those who started six months earlier are already 10% ahead—this gap compounds over time.

But so far, no company has gone bankrupt for not using AI—at least my law firm is still thriving around AI. The winners haven’t been chosen by the market yet. The learning curve is real—early movers are gaining advantages, but most are still at the starting line.

4. My next career doesn’t have a name yet

Will my current job title still exist ten years from now? How many of the tools I used five years ago are still around today? The answer is probably none. But I don’t know what will replace them—because those things don’t exist yet.

Every historical shift has been like this. New things aren’t planned—they grow out of the disappearance of old constraints.

Before the railway, Britain was a collection of isolated local economies. Cotton prices in Manchester and London could differ by 30%. Each city had its own time standard, and no one thought it was a problem. After the railway, everything changed in twenty years. A national unified market emerged for the first time, price differences vanished; standard time was forced by the railway, not invented; stationmasters, telegraph operators, travel agents—these jobs didn’t exist before.

No one foresaw department stores before laying the railway. No one foresaw standard time before building the steam engine.

(The stories of steam, steel, and infinite AI intelligence)

The history of cities tells the same story. Centuries ago, cities were on a human scale—forty-minute walks across Florence. Steel frameworks made skyscrapers possible; railways connected cities to hinterlands; elevators, subways, highways followed. Tokyo, Chongqing, Dallas—not bigger Florence, but entirely new ways of living.

Today’s knowledge work is also on a human scale. Teams of dozens, meetings and emails set the rhythm; beyond a few hundred people, it’s overwhelming. We’re building Florence with stone and wood. AI makes “Tokyo” possible—organizations of thousands of AI agents and personnel, workflows running across time zones. Old weekly meetings, quarterly plans, annual reviews—maybe they no longer make sense.

Simon no longer codes—his job is “managing AI agents.” Two years ago, that role didn’t exist. My next job title might not have a name yet. But someone is already building that future we can’t name.

5. What does the new workshop look like

After tearing down the old workshop, what’s next? YC’s answer: let the company improve itself.

Their internal systems now modify their own code overnight. An employee sends a query during the day, it fails. An overseeing agent detects the failure, deduces the cause, rewrites the code, submits for review, deploys. The next day, the same query runs successfully. The entire process completes while everyone sleeps.

This isn’t AI helping people produce 30% more. It’s the system running a complete closed loop, figuring out how to get better on its own.

YC partner Tom Blomfield calls this “recursive self-improving AI cycle” in an internal talk. His conclusion: most companies are still like the Roman legions—top-down, layered communication, humans as conduits of information. AI breaks not just efficiency in one link, but the entire hierarchical premise.

His new logic: burn tokens, not people. The bottleneck is shifting from manpower to compute power. YC’s data shows that companies reaching Demo Day earn about five times more per person than 18 months ago. Middle management roles are being replaced by AI—“collaboration” no longer needs humans. Everyone should be an IC, builder, operator—each task with a named owner, not a committee.

Another premise: companies must be “AI-readable.” Unrecorded actions are as if they never happened for AI. YC now archives all partner emails, records Slack messages and office hours. One partner used 2,000 hours of recordings over three months to generate a 150-page internal manual—much better than the original. This manual updates automatically each month, becoming a “living brain.”

Tom leaves a question:

If you were to start your company from scratch today, would you build it this way? If your existing organization already has a hierarchy, then the harder question is—would rebuilding it be less painful than continuing to run it like a Roman legion?

People are outside the workshop core—responsible for those areas AI can’t reach yet: offline judgment, new scenarios, high-stakes, high-emotion moments. The company’s “big brain” is assembled from data, records, and industry knowledge. The software running on top is consumable, reproducible. The real assets are in human minds—how the business runs, which steps involve judgment, these understandings are the true wealth.

Ivan Zhao describes this in “Steam, Steel, and Infinite Minds”—an organization of 1,000 employees and over 700 AI agents, where humans handle judgment, agents handle execution. Aschenbrenner bets on compute infrastructure; Ivan Zhao bets on organizational restructuring. Both paths ultimately lead to the same destination: a new production model rebuilt around AI.

6. Epilogue

Between the 1840s and 1850s—railways laid, factories not yet rebuilt.

Where are we? Simon no longer codes. His water wheel was torn down by himself.

The question has never been whether the steam engine is good enough; it’s who will be the first to tear down the old workshop.

I don’t plan to predict the future department store; I just want to do my part—make sure I stand on the railway line, not guarding a drying river.

And you?