Fifteen years ago, Marc Andreessen wrote a contrarian take at the time that software was eating the world. He was profoundly right. The essay has aged into something very rare: a prediction that came so entirely true that its thesis is now the water we all swim in. Every industry he named got eaten. Most of the ones he didn't got eaten too.

But an essay that good doesn't just close a subject, it also opens the next one. Andreessen described a world where software would consume industry after industry. What he did not fully describe, because the technology did not yet credibly exist, was what happens when the software starts to think.

That is the story now. And to understand where it goes, you have to understand the shape of what is happening, because it is a shape we have seen before.

Information was the first thing to become free.

The internet collapsed the cost of distributing knowledge to zero. Everything humanity knew, previously locked in libraries and priced behind gatekeepers, became a search box. This was as revolutionary as it was incomplete. Because while the internet gave access to facts, it did not give judgment. You could look up every symptom of a disease and still have no idea whether you were sick or not. You could read every case on a legal question and still not know what to do. It turned out that while information was really valuable, it was not the same thing as expertise, and expertise stayed exactly where it had always been: scarce, expensive, rationed and locked inside the heads of a small number of trained people.

For all of human history, this was always the fundamental constraint. Knowledge could be copied while expertise could not. A book costs almost nothing to reproduce but a doctor, a lawyer, a master engineer or an experienced underwriter takes decades to produce and cannot be cloned. It turns out that scarcity of expertise is the original and oldest bottleneck of an economy.

That bottleneck is now breaking.

Consider the phone in your pocket.

When the first modern smartphone shipped in 2007, it cost around $500 and was a luxury object for wealthy consumers in rich countries. Fewer than a couple million people had one. It was, by any measure, elite technology, priced for the few.

Look at what happened next. Not to the premium phone, which is still expensive and still sits at the top of the market but at the category. Within fifteen years, a smartphone with more computing power than that original device could be bought for under fifty dollars in a market stall in any developing country. There are now more than six billion smartphones on earth. A technology that began as a device for the affluent became, in a decade and a half, the most widely distributed and powerful tool in the history of our species.

This is the shape.

It is worth naming the steps precisely, because it is a kind of master plan that profoundly disruptive technologies follow whether or not anyone writes it down:

Start at the top, with a premium product for the few, because that is where the value is dense enough to justify the cost.

Use the proceeds to fund the relentless descent down the cost curve, as the components specialize, the volume scales and the price of the same capability falls, and falls and falls.

End with ubiquity, where the capability is so cheap and so abundant that it reaches nearly everyone and the question is no longer who can afford it but what will they do with it.

The phone did this. And the reason this matters is that intelligence is now doing exactly the same thing, on exactly the same curve, only faster.

The cost of intelligence is collapsing.

The price of a fixed unit of machine intelligence, the cost to accomplish a given cognitive task, has been falling at a rate that makes the smartphone's descent look leisurely. Capability that cost a fortune to access eighteen months ago costs a fraction of that today, and the same capability will cost a fraction of that eighteen months from now. You can already watch commoditization arriving in real time: open source intelligence roughly equivalent to the most expensive closed frontier systems is increasingly available at a fraction of the price. The premium tier still exists, as it does in phones. But the floor is dropping out from underneath it at a speed the physical economy has never seen.

And here is why it goes faster than the phone. The smartphone descended one cost curve, the hardware curve, driven by cheaper chips, cheaper memory, cheaper power, and the enormous scale of global manufacturing. Intelligence descends that same hardware curve, because it too runs on silicon and memory and power that are specializing and cheapening exactly as phone components did. But intelligence rides a second curve stacked on top of the first: the models themselves get more efficient. The same capability takes less computation every year. Two discounts, compounding, one on the hardware and one on the intelligence itself. The phone only ever had one.

When both curves run to their conclusion, intelligence will become abundant and very nearly free. Abundant, in the way that information became abundant, in the way that a networked mobile computer in every pocket became abundant. This is not speculation about some distant future. It is the extrapolation of curves that are already well underway, following a pattern we have already lived through once.

Now think about what that means.

The internet gave everyone access to knowledge. This gives everyone access to expertise and is a categorically larger event.

For the first time, the specialized judgment that used to require a trained professional, a credential, a firm, a salary and a great deal of money will become something anyone can summon at almost no cost. The reasoning of a skilled analyst, the diagnostic intuition of an experienced clinician, the drafting skill of a good lawyer, the design sense of a veteran engineer: not the facts they know, which the internet already democratized, but the judgment they apply. That is the thing that was never before copyable, and it is becoming abundant now.

And here the fear arrives, right on schedule.

If a machine can supply expert judgment for free, what happens to the expert? If intelligence is abundant, what is left for people to do? Every wave of automation has summoned some articulation of this same fear, and it deserves a direct answer rather than a reassuring one.

The fear rests on a hidden assumption: that there is a fixed amount of work, so every task a machine takes is a task a human loses. That assumption has been wrong every single time, and it is wrong for a reason. When a valuable thing becomes radically cheaper, we do not use less of it. We use dramatically more, and we invent uses for it that were unthinkable when it was scarce. Cheap information did not end knowledge work; it created whole categories of work that could not have existed when information was expensive and slow. The scarce resource was never labor. It was the capability to turn judgment into action, and we are about to have an effectively unlimited supply of it.

To be definitive, the work will not disappear.

It moves. When expertise was scarce, the bottleneck was getting access to it. When expertise is abundant, the bottleneck becomes what to do with it: which questions are worth asking, which judgment to trust, which problems are worth solving, and who takes responsibility for the outcome. Machines becoming able to reason does not remove the need for a human to own the consequence. It multiplies that need, because now far more decisions can be made, by far more people, than ever before. The human moves up the stack, from producing the analysis to deciding what the analysis is for and what to do with it. None of this makes the transition painless. Real people in real roles will be dislocated. But the direction, in the aggregate and over time, is not less human work. It is more, because ambition expands to fill the capacity available to it, and it always has.

Which brings us to the trap.

If everyone rents the same intelligence from one vendor, no one has any edge.

The best companies have never won on generic capability. They won on something specific and proprietary, a way of doing things that was theirs alone, a hard-won edge encoded into how they operated. The great retailer's mastery of its own logistics. The great insurer's feel for its own risk. The great manufacturer's control of its own process. This edge was the real asset, and it almost always lived in a frustrating place: in the heads of experienced people, in institutional habits, in tacit knowledge that walked out the door when they retired and could never be fully written down.

The reason it could never be fully captured was that encoding it into working systems required engineering, and engineering was scarce and expensive. So companies encoded a sliver of their edge into software and left the vast majority of it trapped in human memory, unsystematized, unscalable, mortal.

That constraint is now dissolving. When intelligence becomes nearly free, the cost of encoding your edge into living systems collapses along with it. For the first time, a company can take the thing that actually makes it special and build it into foundational documents and software that runs it, scales it and compounds it. You can systematize your own alpha.

But this is exactly where the danger sits. Because if the only thing you do with cheap intelligence is consume it the way your competitors also consume it, off the shelf, generic, identical, then you have not built an edge. You have actually erased one. You have taken the capability that used to differentiate you and replaced it with the same commodity everyone else is buying. The company that pours its proprietary expertise into systems it controls builds a deeper moat every day. The company that rents generic intelligence and pipes it into generic workflows becomes interchangeable with every other company doing the same.

The winners of the last era were not the companies that used software. Everyone used software. The winners were the ones who understood that how they used it, how specifically, how proprietarily, how much of their own hard-won edge they built into it, was the whole game. That lesson is about to be relearned, at higher stakes, by everyone.

Which brings us back to the descent.

The master plan is running again. Intelligence begins at the top, expensive and rationed, deployed first against the highest-stakes, best-resourced problems, because that is where the value is dense enough to justify the cost. That funds the descent. The cost curve does its work, twice as fast this time, and the capability slides down toward everyone. And at the bottom of that curve is something the world has never had: expertise itself, abundant and nearly free, available not just to the largest institutions that could once afford armies of specialists, but to the small company, the single founder, the person with an idea but no capital to encode it. The same arc that took the smartphone from a luxury object to a tool in six billion hands is now running on intelligence, and it ends in the same place. Ubiquity.

Andreessen was right that software would eat the world. What comes next is that intelligence will do the same. The descent, itself, is the whole point. The cost of expertise is falling toward zero, and when it lands there, the ability for anyone to build something extraordinary out of their own edge will no longer belong to the few but to everyone.

That is the opportunity and it is the biggest one I have ever seen.