The Earth Should Be a Garden, Not a Mine

In March 2024, the author Joanna Maciejewska posted a line that has been repeated ever since: “I want AI to do my laundry and dishes”, so that she could do her art and writing, not the other way around.

Two and a half years later, I’d stretch her complaint a little further.

Why are the robots writing emails instead of working in mines and fields, handling toxic chemicals, or pulling plastic out of the ocean?

I don’t think the answer is technology. I think it’s incentives. And I think fixing them starts with a rule I came close to writing down in Part 3: Point the Machines at What’s Scarce, but never stated outright.

Here it is:

Be abundant in flows. Conserve the stocks.

Build, recycle and regenerate as freely as we like. Never run down an aquifer, a layer of topsoil or an ore body faster than it’s replaced.

And when growth finally bumps into the planet’s stocks, turn it outward: heavy mining goes to the asteroids, and the Earth stays a garden.

This opens Book Two, Living in it. Part 8 closed Book One by asking what people are for once machines do the work. Before we get to how we’ll live, we need to settle how we treat the planet we live on.

The rule I almost wrote down

In Part 3, I said we should direct machine intelligence at every physical bottleneck “in the most sustainable way possible.” Then I explained why:

“Why the ‘sustainable’ way? Not for greenwashing or virtue-signaling; to literally sustain supply.”

And then: “If you’re able to increase a steady state of supply for all the constraints in a system, you can keep the stocks and flows at a level that creates abundance without burning out.”

That was the right instinct. But it was a mood, not a rule. So this piece states it as a rule, credits the people it comes from, and runs every real scarcity we face through it.

Stocks, flows, and who named them

The language of stocks and flows isn’t mine. It belongs to Donella Meadows, the systems scientist who co-wrote The Limits to Growth in 1972, and whose primer Thinking in Systems was published in 2008, after her death.

Her favorite picture was a bathtub. The water in the tub is a stock: something that accumulates over time. The faucet and the drain are flows: the rates that fill it and empty it.

Sunlight, harvests and recycled metal are flows. Aquifers, topsoil, ore bodies and fossil fuels are stocks that refill slowly, if at all. An economy gets into trouble when it quietly lives off its stocks and calls the result income.

The economist Herman Daly turned that idea into rules. In his 1990 paper “Toward some operational principles of sustainable development” in Ecological Economics (a free summary is here), he laid out what became the standard three:

1. Harvest renewable resources no faster than they regenerate.
2. Emit wastes no faster than the environment can absorb them.
3. Use up non-renewables no faster than we develop renewable substitutes for them. Daly called this a “quasi-sustainable” way to use them.

This isn’t abstract. A 2024 study in Nature analyzed 170,000 wells across 1,693 aquifer systems and found that groundwater declines had accelerated over the past four decades in 30% of the world’s regional aquifers.

But it also found recoveries. In Bangkok, pumping fees and the licensing of wells are credited with reversing the decline of the city’s aquifer. The catch: recovery is much slower than draining.

Stocks drain fast and refill slowly. And a rule, not a miracle, turned the faucet.

So here’s my version, in three parts:

1. Be abundant in flows. Build, recycle and regenerate freely: energy from the sun, materials that loop, food from living soil.
2. Conserve the stocks. Draw them down no faster than they’re replaced, and when you must draw on one, pay into its substitute.
3. Turn growth outward. When we hit the planet’s stocks, the next ore body should be in space, not under a forest.

Four kinds of scarcity, sorted again

In Part 3, I sorted scarcity into four kinds. The rule sorts them again.

Labor is a flow. Doctors, builders and caregivers are scarce because there aren’t enough hours. Machine labor on demand adds hours.

Production bottlenecks are flows stuck behind a narrow pipe: chips, power plants, housing, refining. Widen the pipe, and they flow.

Physically finite things are stocks: land, minerals, fresh water in a given place. These are the ones the rule protects. As I put it then, “Those need better extraction, substitution, and recycling, and eventually mining materials from beyond Earth.”

Scarcity by human nature, like beachfront or “the attention of one particular person,” doesn’t respond to either. That’s where money stays.

So here’s what’s actually scarce as of fall 2026: where machines already help, and what’s still in the way.

What’s actually scarce in 2026

Critical minerals and rare earths

The stock: ore bodies. The real bottleneck: refining.

Rare earths aren’t especially rare in the ground. What’s scarce is refining capacity, and most of it sits in China. Beijing’s April 2025 licensing rules on seven medium and heavy rare earths are still in force, and a broader October 2025 package is only suspended until November 10, 2026.

In May 2026, China’s exports of controlled rare-earth compounds and metals to the US fell to zero. In June, Beijing added MP Materials and USA Rare Earth to its export-control list.

Where machines help: substitution and search. A University of New Hampshire team used AI to read the scientific literature and build a database of 67,573 magnetic compounds, flagging 25 overlooked high-temperature magnets that could someday stand in for rare-earth ones.

Niron Magnetics is building a Minnesota plant for 1,500 tonnes a year of magnets made without rare earths, due in 2027.

And KoBold Metals, the AI explorer I mentioned in Part 3, broke ground at Mingomba in Zambia on April 29, 2026: a mine costing more than $2 billion, aiming at roughly 300,000 tonnes of copper a year, with output in the early 2030s.

What’s left: building refineries and mines outside one country. Finding got faster. Building didn’t.

Recycling

The flow we’re not running. Recycling is how a stock becomes a flow. We barely do it.

The world generated a record 62 million tonnes of e-waste in 2022, and only 22.3% of it was documented as collected and recycled. The UN projects that share will fall to 20% by 2030. For plastics, the OECD’s figure is still about 9%.

Where machines help: chemistry. In 2025, David Baker’s lab used generative AI to design working ester-cutting enzymes from scratch, in shapes nature never made. That’s the family of enzymes that can break down PET plastic.

What’s left: the economics. Li-Cycle, once a leading battery recycler, went into creditor protection and sold its assets to Glencore in 2025. Redwood Materials cut 10% of its staff in April 2026 to chase energy storage.

Carbios, whose enzymatic PET plant is the flagship test of this whole idea, missed its September 2026 financing target, pushing production back again.

None of them stalled on chemistry. They stalled because virgin material is cheap and the build-out is expensive.

Water

The stock: aquifers. The flows: the ocean, and the air.

In February 2026, a Saudi plant at Yanbu set a Guinness record of 1.55 kWh per cubic meter for its reverse-osmosis step. Dubai’s Hassyan plant was awarded at $0.365 per cubic meter, which its developer called the first plant of its scale under 40 cents.

And the 2025 Nobel Prize in Chemistry I wrote about in Part 3, for metal-organic frameworks, is already becoming a product.

Atoco, founded by laureate Omar Yaghi, has a shipping-container unit that pulls 2,000 to 4,000 liters of water a day out of the air on grid power, and up to 1,000 off-grid. No price yet.

What’s left: energy, brine disposal, and pumping water inland and uphill. Water from air works, but at the scale of a building, not a city.

Food

The stocks: topsoil, and groundwater, most of which is pumped to irrigate farms.

Vertical farming was supposed to be the answer. Instead, one industry CEO estimates investors lost about $2.7 billion on just four failures: Plenty, AppHarvest, Bowery and AeroFarms.

Indoor farms tried to replace a free flow, the sun, with electricity. The sun won on price.

Where machines help: in the field. Carbon Robotics’ LaserWeeder uses computer vision and lasers to kill weeds without herbicide, and the company passed $100 million in revenue in the year to March 2026. Installations of agricultural service robots grew 2.5-fold in 2024.

Regenerative farming looks promising but uneven. A 2025 conference analysis of 5,709 paired observations found that half showed gains in yield or soil carbon, 16% gained both, and 7.5% showed losses. It’s an abstract, not a final paper, so treat it as a signal.

What’s left: labor in the field, and measuring soil well enough to know whether we’re spending the stock or growing it. Both are jobs machines are good at.

Energy and storage

The flow: the sun, the biggest flow we have.

Batteries are what let a flow act like a stock. BloombergNEF found that stationary storage packs fell 45% in a single year, to $70 per kWh in 2025. The IEA counted a record 108 GW of batteries added in 2025, more than gas power added in its best year, 2002.

On September 24, 2026, Fervo Energy’s Cape Station in Utah sent its first power to the grid, the first time a greenfield enhanced geothermal project has done so.

Fusion isn’t here yet. The National Ignition Facility hit a record 8.6 megajoules and a target gain of 4.13 in April 2025, and Google DeepMind is applying AI to plasma control for Commonwealth Fusion Systems’ SPARC reactor.

But target gain isn’t wall-plug gain. No fusion machine has delivered electricity to a grid yet.

What’s left: grid connections, transmission and permits. Making power is no longer the hardest part.

Moving it is. Which is one more argument for microgrids for everything.

Manufacturing

The flow: making things close to where they’re used.

3D printing is maturing, not exploding. Global additive manufacturing revenue reached $24.2 billion in 2025, up 10.9%, according to the Wohlers Report. Printing services grew 15.5%; printer sales grew 3.6%.

Meanwhile, automation keeps spreading in big factories: a record 5 million industrial robots are now at work, with more than 600,000 installed in 2025.

What’s left: scale. Big factories still win on cost today, so “every region its own fab” is a bet on the cost curve, not a result.

I play that bet out in its own section below. It’s exactly where machine intelligence should be pointed.

Mining

The stock itself.

The good news: this is where robots already work. The number of autonomous haul trucks jumped from 2,080 to 3,832 between July 2024 and July 2025, with most of the growth in China.

The bad news: the rock is getting worse. A 2016 study found the average grade of copper ore had fallen about 25% in ten years, while energy use grew 46% against a 30% rise in production. Lower grades mean more rock, energy and tailings for every tonne of metal.

What’s left: that’s a stock running down in real time. Robots make digging safer and cheaper. They don’t put the ore back.

In Part 5, I wrote, “Mining is the exception to all of this, since materials are where they are.” The rule says: recycle first, substitute second, and over time, send the heaviest digging off-world.

Space

The next stock.

AstroForge’s first mission, Brokkr-1, failed in 2023. Its second, Odin, was declared lost in March 2025.

Its third, Vestri, will ride along on Intuitive Machines’ IM-3 mission, which NASA now expects in the first part of 2027. No private spacecraft has reached an asteroid yet.

NASA’s OSIRIS-REx returned 121.6 grams from Bennu, the largest sample ever collected. China’s Tianwen-2 reached the asteroid Kamoʻoalewa in July 2026 and plans to bring its sample home in late 2027.

Where machines help: everywhere. Signals can take minutes to reach an asteroid, so anything that mines one will have to think for itself.

What’s left: launch, reliability and economics. My own estimate is that meaningful tonnage is decades away.

That’s fine. It’s the right long-term destination.

The pattern

Line these up, and a pattern jumps out. The science is moving almost everywhere. AI is finding magnets, designing enzymes, steering plasma, exploring for copper and driving haul trucks.

What’s stuck is the build-out: refineries, recycling plants, transmission lines, field robots and new mines. Slow, physical, capital-hungry work. Which brings me back to the question.

Why are robots writing emails?

Follow the money. Stanford’s AI Index counted $581.7 billion in corporate AI investment in 2025. Of the $344.7 billion in private investment, the largest focus area by far, AI infrastructure, models, research and governance, took $143.2 billion.

Sorted another way, by what the companies make, generative AI companies drew $170.9 billion. The two counts overlap, but they point the same way. Robotics and energy management trailed far behind.

Now follow the attention. In the latest earnings season, “AI” came up on 331 of 493 S&P 500 earnings calls, or 67%, the third straight quarter above 65%.

FactSet also found that the companies citing AI were up 15.7% this year through mid-September, versus 8.1% for those that didn’t.

That’s correlation, not proof. But you can see the incentive.

If you run a public company, you need an AI story for your investors. The fastest AI story to tell is software you can roll out this quarter: an assistant in the inbox, a summarizer in the slide deck, a chatbot on the help page. So that’s where AI diffuses.

Whether it pays off is a separate question. One much-debated MIT report claimed that 95% of organizations got no measurable return from their generative AI spending.

It rests on 52 interviews, 153 survey responses and a review of public deployments, so treat it as one data point, not a verdict.

Meanwhile, the hard physical jobs pay back slowly. A recycling plant needs years of capital and permits before it earns a dollar. Software ships on Friday.

It’s not that nobody points AI at the physical world. The Ocean Cleanup removed a record 25 million kilograms of trash from oceans and rivers in 2025, and it’s working with AWS on AI that finds plastic.

It’s that the market rewards the quick story more than the slow build.

So the robots write emails. Not because emails matter more than mines, but because the rules reward the story you can tell fastest.

That’s backwards from the human-will rule. The jobs the fewest people want, the toxic, dangerous and backbreaking ones, are exactly the jobs machine labor on demand should take first.

Change the rules, not the numbers

Meadows has an answer for this, too. In her 1999 essay Leverage Points: Places to Intervene in a System, she ranked twelve places to push on a system, from weakest to strongest.

Dead last: parameters, the numbers, like subsidies, taxes and standards. That’s where most of our political fights happen.

Much higher up: the rules of the system, its incentives and constraints. Higher still: its goals. As she put it, “Power over the rules is real power.”

Applied here, I’d do four things.

1. Change what we tax. In Part 18: Every Spending Bill Should Show Up on Your Paycheck, my first tax sketch argued for shifting the tax base off human labor and onto “things that stay scarce no matter how productive machines get: land values, natural resources, and permits for scarce compute.” That’s the stock rule, written into the tax code. Charge for drawing down a stock, the way Bangkok charged for its groundwater, and recycled flows can win on cost. Changing a tax rate is a parameter. Changing what gets taxed is a rule.
2. Reward building, not announcing. Back the companies doing hard physical work now: the laser weeders, the magnet makers, the recyclers, the geothermal drillers. Public money can help the way it already does, with loans and long-term purchase commitments that take risk out of the build-out. Every one of these companies is building the machines and supply chains that machine labor on demand will run on later.
3. Change the goal by changing the scoreboard. Part 19: Plan for the Bad Transition lists signposts to track the transition. Add the stocks: aquifer levels, soil carbon, the recycled share of our metals, and energy per tonne of ore. If an AI story has to show up in those numbers, the story changes.
4. Point machine intelligence at the bottleneck. In Part 3, I said where I think machine intelligence may matter most: “Not as a worker, but as a scientist.” Every section above names a bottleneck a scientist could attack: separation chemistry, cheaper enzymes, better membranes, smarter grids. That’s the scarcity flywheel: better materials make better chips and robots, which make more labs, which find better materials. Email doesn’t turn it.

Where the trusts come in

The rule also tells us what local ownership is for.

In Part 5, I proposed that each town’s essential infrastructure, including the local machine fleet that delivers the floor, belong to a trust: one person, one vote, membership by residency, and it can never be sold. Your membership in it is your share of universal basic capital.

A trust can write the rule into its charter.

Run the flows as hard as you like: the solar, the batteries, the recycling line, the farm robots. Meter the stocks: the aquifer under town, the soil in its fields. Draw them down no faster than they refill.

That’s microgrids for everything, with a conservation rule built in. And because a trust can’t be sold, nobody can buy it to strip the stocks.

The gift-back fits, too. The builders who plant the garden earn their capped return. Then the town tends it.

Gigafactories should build microfactories

Earlier, I called “every region its own fab” a bet on the cost curve. So here’s how I think the cost math could flip, and where it won’t.

Start with why big wins. There are three reasons, and only one of them is people.

Labor. One crew can run a huge line. The bigger the line, the less human work goes into each unit.

Physics. A tank’s surface grows with the square of its size, but its volume grows with the cube. So a bigger vessel needs less material, and loses less heat, for every tonne it holds.

Chemical engineers call this the six-tenths rule, from Roger Williams Jr. in 1947: double a plant’s capacity, and its cost rises only about half again.

Precision. Some processes only work big and nonstop. A furnace for window or bottle glass runs around the clock at about 1,550°C, on a run that can last 16 years or more.

TSMC makes the most advanced computer chips in the world. Its new chip complex in Arizona is a $165 billion “gigafab” of six individual chip fabs.

So the three reasons big wins: labor, physics, and precision.

Most obviously, machine labor on demand erases the first reason.

Think about what happens when you scale out. Open a second coffee shop across town, and the first one serves fewer people. Spread a region’s demand over many town nodes, and each one needs to make far less.

That weakens the case for huge, nonstop lines. A machine crew doesn’t need a full shift to be worth calling in, so a small plant can run short batches and sit idle in between.

And small-batch tools keep getting better. Electric-arc mini-mills, which I’ll come back to, melt scrap one batch at a time. 3D printing needs no molds, so one part is as easy to set up as a hundred.

Even chips have a small-batch version. Japan’s national research institute, AIST, developed the Minimal Fab: small, standard tools that work on half-inch wafers without a cleanroom, designed to cost about a thousandth as much to set up as a big fab.

It’s built for chips needed in small batches, not for leading-edge processors.

As you can see, for a lot of goods, the third reason big wins (precision) can shrink right along with demand.

Not for all of them. Some processes have a minimum size set by physics and precision. Splitting the demand doesn’t make a tiny version efficient. It just leaves a big one half-idle.

Remember that glass furnace? Restarting one after an abrupt shutdown takes about a year and at least $60 million, according to the National Glass Association.

And a single EUV scanner, the machine that prints the finest features on leading-edge chips, costs around $180 million. Only one company, ASML, makes them. Plants like those stay regional, or bigger, serving many towns at once.

The second reason, physics, doesn’t go away either. A small tank still loses more heat per liter than a big one.

But that’s a penalty on each unit, and other savings can outweigh it: shorter trips, a town that keeps running when a distant plant goes down, and scrap that comes straight back in, in a closed loop.

There’s research behind this shift, too. In 2013, Columbia’s Eric Dahlgren, Klaus Lackner and colleagues argued that cheap automation lets economies of numbers, many small, standard units, save as much capital as one giant unit.

Why small wins. In a 2020 paper in Science, Charlie Wilson and colleagues compared energy technologies of every size. They call the small ones granular: “small in size, low in cost, many in number.”

Those spread faster, got cheaper faster, and carried less investment risk.

Bent Flyvbjerg’s database of about 16,000 big projects agrees. Solar went over budget by 1% on average, and nuclear by 120%. In How Big Things Get Done (2023), with Dan Gardner, he credits Lego-style modularity.

Repetition is how costs fall. In 1936, the aircraft engineer Theodore Wright found that every doubling of total production cut costs by a steady share. That’s Wright’s law.

Solar modules have fallen about 20% per doubling, from $132 a watt in 1975 to 27 cents in 2024, in today’s dollars.

Here’s the catch, in Wilson’s own paper: small things got cheap partly through “large production runs.” Small units, made in big factories.

The factory factory. That’s the model I’d bet on. At Tesla’s 2016 shareholder meeting, Elon Musk said the greatest potential was in “building the machine that makes the machine.”

Robots already build robots. Back in 2003, Fanuc’s factory near Mount Fuji was making about 50 robots per 24-hour shift, running unsupervised for up to 30 days.

The dream is older: John von Neumann sketched a self-copying machine on paper in the 1940s. In 2005, Adrian Bowyer’s RepRap set out to build a 3D printer that prints itself.

By 2008, it copied all of its printed parts, which made up 48% of the machine. The motors, sensors and chips still came from elsewhere.

Now imagine today’s gigafactories turning out tomorrow’s microfactories. My bet is that they start arriving 10 to 15 years from now.

A gigafactory for making microfactories. One big factory to make a thousand or a million smaller ones.

Picture how this plays out: the kit for the factory ships, the town assembles it, machine labor maintains it, and the great gift-back hands it over.

Regional gigafactories can still keep dibs on the leading-edge chips, the float glass, the motors, and the rush jobs: when a town needs ten thousand of something by Friday, the network sends its order to one of the big plants.

Steel already made the move. Mini-mills melt scrap in electric furnaces. They started with bars, like the rebar inside concrete, and worked their way up.

Clayton Christensen made them his classic case of disruption in The Innovator’s Dilemma (1997).

Today, America has 102 minimills run by 47 companies. The integrated mills, giants that make iron from ore, are down to two companies at eight sites, with about 28% of production.

Mini-mills won because their input is a local flow: scrap. But “mini” is relative: a new one in California will make 450,000 tons of rebar a year. That’s a regional plant, not a town workshop.

Small needs to be many. A while back, a company called Arrival promised electric vans from compact “microfactories” and hit a $13 billion valuation.

And yet it never delivered a working production van to customers like UPS, and its UK arm entered administration in February 2024.

NuScale’s small modular reactors in Idaho saw their target price rise to $89 per megawatt-hour, even after subsidies. The project was cancelled in November 2023 for lack of buyers.

The lesson: small isn’t the same as many. You need small and numerous, not small and a pilot program. Costs fall with the thousandth identical unit, not the first.

What “cost” means without a price

Once a microfactory is built, gifted back to its regional community, and maintained by machines, nobody pays for it in the usual sense.

But remember: a missing price isn’t a missing cost. Every unit still uses real resources: materials, energy, machine-hours, land, maintenance, and the work of recovering it at the end of its life.

The second ledger from Part 17: Keep Money, Add a Second Ledger is meant to count and account for exactly those things, in physical units.

Therefore, the real cost is resources per unit delivered: everything it took to put one finished thing in someone’s hands.

Here’s what that looks like for one wooden dining chair. The following numbers are made up and rounded, for illustration only, but they’re the right kinds of numbers.

Chair A, from a gigafactory across an ocean:

• Materials: 8 kg of wood in, a 6 kg chair out. Big plants cut efficiently.
• Energy to make it: 5 kWh.
• Machine-hours: 0.2.
• Transport: 12,000 km by ship and truck, burning roughly 9 kWh of fuel.
• End of life: a landfill. Nothing recovered.

Chair B, from the town’s microfactory:

• Materials: 10 kg of wood from the region’s sustainable timberland in, the same 6 kg chair out. The offcuts heat the shop.
• Energy to make it: 8 kWh, because small machines and a half-full drying kiln waste more.
• Machine-hours: 1.
• Transport: 150 km by electric truck, under 1 kWh.
• End of life: taken back, repaired, and the wood reused. 5 kg recovered.

On making the chair, the gigafactory wins. It uses less wood, less energy and a fifth of the machine time.

But the microfactory spends far less on transport, and it recovers most of the chair at the end of its life.

So which chair costs less? That’s where the garden rule does the scoring.

Chair B’s extra energy comes from the sun, and its extra machine-hours come from machine labor on demand. Both are flows. Chair A’s fuel is a stock, and so is the timber that ends up in a landfill.

Weigh the stocks heavily and the flows lightly, and Chair B wins. Change the inputs, say a coal-powered town or a forest that gets cut faster than it grows, and Chair A could win instead. The ledger would show that, too.

Notice what the example never needed: a price.

The factory itself counts, too. One that sits idle most of the week spreads its steel, chips and energy over fewer things.

On the face of that comparison, I’d bet small-scale wins wherever goods are heavy, perishable or custom, or made from a local flow: food, furniture, building materials, clothes, spare parts and recycling.

Trucks and warehouses use resources too, and so does making a thousand of something when all you needed was one. In a world of abundance, I imagine many more things will be made custom and in small batch sizes.

Large-scale plants will still win for things like leading-edge chips, float glass, and the motors needed inside every microfactory.

So the goal should be for every town to make its own essentials, not everything. Become resilient first, and sometime between 2040 and 2050, self-sufficient enough to trade surplus freely, with the regional gigafactories as the factory factories to produce microfactory inputs and outputs at scale.

At the limit, self-sufficiency is a regional goal, not a town-by-town one. Not every town needs every resource. A region can come close to having them all, and its borders can ebb and flow with what it trades.

We should then aim to size each plant to the region it serves, with the end goal of production surplus built in by design.

And if we make everything to be recycled, a discarded product flows back in as tomorrow’s raw material.

That closes the loop in the way we want: abundant in flows, conserving the stocks.

If I’ve drawn any of these lines in the wrong place, the second ledger will show it, and we’ll be able to adjust.

Scale of production is only half the picture. The other half is information: how a shared record of what exists, and where, can shrink the booms and busts of today’s supply chains. That’s also in Part 17.

Objections

Isn’t this just degrowth?

No. Degrowth says shrink the economy. I’m saying shrink the drawdown and grow everything else: energy, food, goods, recycling, science.

To be fair, Daly, the founder of steady-state economics, whose ideas helped inspire degrowth, went further than I do. In that 1990 paper, he called “sustainable growth” a “bad oxymoron,” because he saw the economy as a subsystem of a finite planet. He separated growth, defined as getting physically bigger, from development, defined as getting better.

On this planet’s stocks, I agree with him. There’s no question we have finite resources. Where I’d push back is on the boundary. The sun’s flow is enormous, materials can loop, and there are vast amounts of ore in space that nobody lives on top of.

Abundant flows, conserved stocks and an outward frontier don’t make a smaller economy. They make a differently shaped one.

Won’t asteroid mining take forever?

Probably decades, yes. Humanity has brought home about a quarter pound of asteroid, total.

That’s why turning outward is the third part of this vision, not the first. Recycling and substitution can carry us for a long time.

My roadmap in Part 19 puts the outward turn in Stage 5, 2040 and beyond, and for real tonnage I’d bet on later. Still, SpaceX’s Starship has made incredible progress in carrying payloads to space.

Naming it as a target now makes sure the long-term plan for heavy industry considers it vital, and doesn’t just default to “dig deeper here on Earth, forever.”

Why not let the market sort it out?

It is sorting it out. Toward emails.

Markets are very good at following prices. The trouble is that most stocks aren’t priced at all. Pumping an aquifer costs you the electricity, not the water you’ll never get back.

In Meadows’ terms, that’s a problem with the rules, not with markets. Bangkok didn’t ban pumping. It licensed and priced it, and the aquifer began to recover.

Set the rules right, and the market does the rest. That’s not central planning. It’s steering.

Tend it like a garden, not a mine

I’ve started calling the people who think this way Worldbuilders. Doomers want a brake. Accelerationists want a gas pedal. Worldbuilders want a steering wheel.

This rule is one way to steer. Run the flows as fast as we can build them.

Keep the stocks whole. Send the heavy digging up and out.

And put the machines where the work is: in the mines, the fields, the recycling plants and the ocean. Not in your inbox.

AI should be mining, not writing your emails. And one day, it should be mining somewhere other than here.

The Earth should be treated like a garden, not a mine. Don’t get me wrong, we need to mine. But we should mine as a means to an end that takes us to the stars, not binds us to a barren wasteland.

Next, I’ll lay out what Worldbuilders actually believe, in Part 10: A Floor for Everyone, a Frontier for Anyone.