Seven Bets

Those are all wagers on what the next ten years look like. The only choice is whether you place them deliberately. Most people never write their bets down, so they never see how their own assumptions are driving their decisions.

These are mine.

Seven predictions about where AI takes software, work, and the physical world, and why each one matters. They aren't guesses about flying cars or curing death. Predictions like that are entertainment. Whether they land in 2050 or never, nothing about your Monday changes. These are bets on the next ten years, and every one of them has already started.

They're also costing me something. They decide what I build, what I study, and who I spend my time with.

I'm an optimist, but not the passive kind. I grew up organizing in nonprofits and progressive campaigns. My version of hope is the kind you have to work hard as hell for.

I want the version of this future that works for everyone, not just the people already positioned to win. The future isn't destined for abundance or destruction. It gets built by the people who plan for it.

What I'm treating as settled

Before the bets, four things I consider ground truth. Not everyone accepts them yet, but I'm no longer arguing about them.

1. AI is going to write most of the code. People will still decide what software should do, but machines will do most of the actual building.
2. It is getting much easier to make software. It is not getting easier to know what people actually need. The hard part is shifting from can we build this? to is this worth building at all?
3. For a lot of our biggest problems, we already know the answer. We just can't get the work done. We know how to build housing, clean water systems, power plants, roads, and schools. The hard part is permits, money, coordination, politics, and getting dozens of people and institutions to move together.
4. We are starting from a deeply unequal world, and inequality does not undo itself. Wealth, power, and access to resources are already distributed extraordinarily unevenly. Some comes from geography and natural resources; much of it is the accumulated result of millennia of slavery, caste systems, colonization, and exploitation. Those forces concentrated wealth and power, and their effects compound long after the original systems change. Correcting that inequity requires an equal and opposite force: deliberate, sustained collective action to build systems that intentionally direct prosperity and abundance toward the people and places from which they have historically been withheld.

For most of software history, a product changed when a person decided to change it.

Someone noticed a problem, wrote a ticket, designed the fix, built it, tested it, and shipped it. Then the product sat there until people did the whole thing again.

AI is making that cycle dramatically faster, but we are still mostly using it to accelerate the same model. Humans decide what the product should become. AI helps us get there faster.

My bet is that this changes.

A product continuously watches how it is being used and how the world around it is changing. It notices where people struggle, writes and deploys its own updates, measures what happens, keeps what works, and removes what doesn't. It's like a plant, constantly growing and adapting to the world around it.

People still set the boundaries: what the product can change, how it judges success, and most importantly, what it is trying to become.

For most of modern employment, the bargain was simple. A company figured out what needed to be done. It broke that work into roles. You took one of them, did the work, and got paid for your time.

That bargain depends on somebody being able to specify the work in advance. AI is eating that part first.

If a task has a knowable answer, a repeatable process, or a clear definition of done, it is becoming easier to hand to a machine. What remains for people is the work nobody knows how to assign.

Something is broken, but nobody knows why. A new market might exist. Costs are rising somewhere in the system. A technology appeared six months ago and nobody knows what it changes.

There is no ticket for that.

You have to notice it, decide whether it matters, form a theory, test something, read the result, and decide what to do next.

That is entrepreneurship.

Companies start looking less like a few executives making bets while everyone else executes them, and more like portfolios of people hunting for value.

One person finds a way to remove a million dollars of cost. Another discovers a customer nobody was selling to. Someone else finds a new business hiding inside the old one.

Then the old compensation model starts to strain. A salary pays you roughly the same whether the thing you discover is worth $20,000 or $20 million. That made sense when the company defined the problem and told you what success looked like. It makes less sense when you're asked to find the opportunity yourself.

If I'm expected to think like an owner, take risks like an owner, and create asymmetric upside like an owner, eventually I'm going to ask why I don't own any of it.

Bet B says the valuable human work becomes more entrepreneurial. Bet C is how that work gets leverage.

The Cynefin framework makes the split clear: simple and complicated work have knowable answers; complex and chaotic work do not. AI is rapidly absorbing the first two. That leaves people the harder half: living in the complex and chaotic while supervising machines that handle the rest.

That is why I think almost everyone becomes a manager. Not necessarily a manager of people. A manager of AI agents.

You will assign work, give context, set boundaries, review output, catch mistakes, and decide when an agent is ready to operate with less supervision. Some agents will be reliable. Some will need constant correction. Some will be good at one narrow thing and terrible outside it.

That is management.

Span of control used to mean how many people report to you? Now it means:

A few people, managing dozens of capable agents, will be able to produce what once required a department or even an entire company.

Robots eventually do the same thing to physical work. They just arrive on a different clock. That's Bet E.

Software used to be valuable partly because it was hard to build. AI is changing that fast: code is getting cheaper, features are easier to reproduce, and being well built is no longer enough to make a product defensible.

So the moat moves.

To memory. The long-term kind, that outlives any single conversation.

An agent you meet today can be brilliant and still feel like someone you just met. It does not really know what you care about, how you make decisions, which people matter to you, what you tried two years ago, or which old ideas are suddenly relevant again.

Now imagine using that same agent for ten years.

It remembers the people you've worked with, the decisions you've made, the patterns you've missed, the arguments that changed your mind, the projects that failed, and why. More importantly, it knows when to bring those memories back.

You mention a new deal and it remembers someone you met two years ago who could unlock it. You are struggling with a decision and it says, You were in almost this exact situation four years ago. Here's what happened.

You didn't search for those memories. The system knew when they mattered.

At that point, it stops feeling like a brilliant stranger. It starts feeling like a mentor who has known you your whole life.

Software got fast because failure is cheap. You can try a thousand approaches before lunch, keep the one that works, and throw the rest away. AI makes that loop even faster.

Atoms don't work like that.

A bad software experiment costs minutes. A bad robotics experiment can break hardware. A materials candidate still has to survive a real furnace. A new power line still has to be permitted, financed, built, and connected. Every self-driving mile is still an actual mile.

For most of the last fifteen years, electricity was boring. Demand grew slowly, especially in the US, and utilities planned around incremental growth.

That era is over.

AI, electrification, industrial growth, cooling, and rising incomes are pushing electricity demand up at once. We will build enormous amounts of new generation: solar, wind, nuclear, gas.

Building new generation, especially clean generation, is still hard. But the bottleneck is moving downstream.

The constraint is increasingly the physical system that gets electricity where it needs to go.

Transmission lines. Substations. Transformers. Storage. Interconnection. Local distribution. Permitting. Construction.

This may become one of the largest physical construction projects of our lifetimes.

Energy is the biggest victim of Bet E: you cannot prompt a transmission line into existence. Someone has to finance it, permit it, engineer it, manufacture the equipment, negotiate the land, train the crews, and put steel in the ground.

We are short power, and shorter still on the capacity to build the system around it.

There is a version of the next ten years that is extraordinary.

Intelligence becomes abundant. Energy becomes cheaper. Machines expand what people can accomplish. We learn faster, build faster, cure more, waste less.

I believe that version is possible. I do not believe it arrives evenly without an enormous effort to make it so.

New abundance almost always reaches the people with money, infrastructure, and power first. They get the solar panels, the batteries, the AI assistants, the personalized medicine, the schools and companies capable of adapting immediately.

And maybe that would be fine if everyone else simply got the same things a few years later.

That is not what I am worried about.

I am worried that the wealthy get the abundance and everyone else gets the bill.

A wealthy family buys an education combining expert teachers with personalized AI; an underfunded school is handed a chatbot and told it can manage with fewer teachers. A company captures the productivity gains from automation; the displaced worker is told that society is getting richer. The countries that contributed least to industrial emissions pay for their hardest consequences.

That is not a technological inevitability. It is a political choice.

We know because people have changed that distribution before. HIV treatment once cost about $14,000 a year. Today, in much of sub-Saharan Africa, it can cost under $100.

That did not happen because the market grew more generous. It took four decades and millions of people: activists, scientists, manufacturers, health workers, public officials, funders, and organizers around the world. Many dedicated their lives to the fight; some died before the treatments they demanded reached them. Activists organized across borders, generic manufacturers challenged patents and pricing, governments changed the rules, and new institutions financed and delivered treatment at scale.

Millions gained access because people organized until institutions changed and the rules were rewritten.

Equitable distribution has to be fought for and built.

The same will be true of AI, but on a generation-defining scale.

Intelligence may become nearly free. The things that turn intelligence into a better life will not.

A model can design a power grid, diagnose an illness, or explain how to purify water. It cannot cross the distance between knowing what to do and having the capital, infrastructure, institutions, and political power to do it.

A model can cross an ocean instantly. A power grid cannot.

Neither can a hospital, a road, a water system, a factory, a school, or a home. They have to be built in particular places, by particular people. Where they get built, whom they serve, and who benefits from them are decisions made by institutions with power.

And concentrated power rarely volunteers to dilute itself.

The people promising a world where money and jobs stop mattering already live remarkably close to one. They are describing their own lives and calling it everyone's future.