We are told we are bad at exponentials, nod along, and then underestimate them anyway. The transition always arrives sooner than it feels. So rather than extrapolate forward from today, anchor at the end and let the transition race toward you. Here the end is easy to state. Digital intelligence is no longer scarce, except at its very frontier. That is a picture rather than a date, since superintelligence arrives as a slope, not a moment. Whatever is valuable there has terminal value, and it still has to earn its keep on the way.

Economic value is still supply and demand. Profit needs supply that is constrained, which is to say scarce, and demand, which is an agent willing and able to pay. Anyone who has invested knows real markets rarely behave like the textbook, because people lack the information, attention, and time to act on it. But much of that mess is simply limited intelligence, and as agents with superhuman knowledge and speed start acting for people, it shrinks. Everything changes and nothing changes.

The textbook becomes a better map than it has ever been, which is convenient, because it lets us ask a clean question. What stays scarce, and who still wants it? Supply is the more objective half, so start there.

We live in three worlds. There is the physical one, the digital one built on top of it, and the conscious one, the private monologue inside each of us where decisions are made before they appear anywhere else.

Technically, everything physical is matter moving through spacetime. That sounds like the least useful sentence in this document, but it does a surprising amount of work. Time runs one way, so the past cannot be revisited. Build an atomically perfect Colosseum tomorrow and it will still not be the one Caesar sat in. That is identity, the scarcity of this particular thing with this particular history, and it is why Ferrari, the Vatican, and every one of us stay scarce however cheap intelligence becomes. The past also leaves data behind that cannot be rebuilt. Infinite intelligence is not infinite information. Some facts are private, and some were the outcome of chance that nothing in the present can reconstruct, so Google’s search history, Nvidia’s test results, and your own thoughts stay scarce too.

In the present, two things cannot occupy the same place, and the earth’s stock of matter is fixed. There is only so much gold, only so much shoreline on Lake Como, only one Panama Canal. And moving matter into the future still costs time and energy. Knowing precisely how to build a chip factory does not build one. The frontier of intelligence runs on those chips, which is why it will probably never be free, however cheap every model behind it becomes. That is atoms.

The conscious world adds the strangest scarcity of all. We are a species of storytellers. We invent games, nations, currencies, and rules, and they hold only for as long as enough of us keep agreeing to them. Build a perfect copy of Instagram and it opens empty, because its value was never the code. It was everyone choosing to be in the same place, and each of us has only so much attention to spend. The dollar, the law, and the World Cup work the same way. Intelligence can lobby our choices. It cannot make them. That is coordination.

So, roughly, four things remain scarce. They blur at the edges, and this is a working map rather than a proof, but it is a good enough map to see where value can still live.

Scarcity on its own is worth nothing. Someone has to be willing and able to pay. Ability to pay depends on how the economy rearranges itself, and anyone who claims to know that is selling something, so assume it. Whether the models are conscious is a landmine best stepped around. For now most act on behalf of people and were trained on people, so they can be counted as part of human demand. That leaves one question, which is what people will want.

Darwin, predictably, gets the first word. Everyone alive descends from ancestors who kept themselves alive, so persistence, meaning health and security, comes first. Once it is covered, what remains is satisfaction, and satisfaction has no ceiling. Persistence is fairly easy to forecast. Satisfaction is not, because each of us decides it for ourselves, in private. It’s all internal.

Nobody can predict what eight billion people will choose to be satisfied by. But every one of those choices is made in the same place, inside the person. So the dependable bet is not on any particular want. It is on the record of the one doing the wanting. See what someone thinks and why, and you no longer have to guess their next choice from their last purchase, which is all a proxy has ever been. It will never be a perfect map, because even the person cannot put everything they feel into words. It is simply the closest one there is.

This is, admittedly, a long way round to something everyone already believes, that personal data is valuable. The point of the detour is the direction. When everything is changing faster than we can absorb, the wise bet is on what will not change, and the one constant in every future is the person.

Data sitting in a drawer is worth nothing. Its value appears only when it is used to make a good or service better for someone, which in the end serves their persistence or their satisfaction. So a system is worth what its output is worth. The most valuable output needs the widest input, since every extra piece of data widens the options and, assuming no piece does harm, none narrows them. It needs the most intelligence, to separate signal from noise. And it needs the data organized in advance, because hunting for the right piece at the moment of use costs time and energy, and that cost comes straight off the value. Without the system the data would not exist, let alone in its best form, so the system inherits its value from the data.

Before data can be used it has to be externalized, captured, and digitized. Most of a life is already external, from where you went to what you searched at two in the morning, and phones, computers, and the large platforms already capture it. The body is internal, and a growing army of wearables and blood tests reads more of it every year.

The mind is different. No technology reads it, yet, and even its owner cannot see all of it, because much of what we think never becomes words. The only way out is through the person, feeling into thought, thought into words, words said aloud or written down. It is the most valuable data a person has, since every decision starts there, and it is the one no device will ever collect on their behalf.

Any agent, whether a person, a company, or a superintelligence, runs on four things. Data describes it, a harness organizes that data, a model reasons over it, and tools act on the world. Models, harnesses, and tools improve on exponentials, and the best of each changes every week. The data is the constant, because it describes the entity itself. So the best system keeps the data in one place the person controls and lets any model, harness, or tool read and write to it. They can then assemble the best mix at every moment, and switch or add without ever starting over. And because the frontier model costs the most, the best mix keeps it in charge and has it hand the routine work down to cheaper ones, which only works when all of them read the same record.

The obvious objection is that none of this is needed, because a model can already be plugged into a person’s mail, calendar, and files, and keeps getting better at pulling in whatever a task needs. But that gives it reach into each app, not knowledge of the person. Each app stays the ground truth for its own data, and the best system points to it rather than copying it in. What a model learns about the person themselves, how they think and what they want, builds up slowly in its own memory, so the next model meets a stranger, and no company has a reason to hand that memory to a rival. That is the part worth owning.

One place does not mean one folder, and control does not mean owning the server. It means nobody else decides what a person’s own intelligence may do with their own record. Control also changes how much people are willing to put in, because people share more when they know who is looking. Processing then runs on the model subscriptions the person already pays for, which is why Alexandria is not, and never will be, a standalone app. Own the data, rent everything else. It takes a paragraph to prove and a sentence to say.

Value comes back three ways. Directly, as money or a better good or service. Indirectly, when your data improves something that later serves you. Or to someone else, when it improves what they receive. And, pure individualists aside, nobody exists alone. We are shaped by the people around us, so their data belongs in our system and ours in theirs. But nobody shares everything with everyone, and forcing it ends with everyone sharing nothing. So output has to come in layers, each derived from the private record and matched to its audience.

Data can reach a person’s files four ways, either directly or through a model, and either automatically or by hand. Direct and automatic already works, from messages syncing to a heart monitor writing to a folder. Through a model and automatically is growing fastest, as it connects to more of a person’s software, and the Loop makes sure whatever it learns lands in the person’s own files as well as its own memory. By hand is tedious enough that almost nobody bothers, so the Shortcut rides the share button already on their phone. The mind is the largest gap, and the Skill fills it where people will spend most of their time anyway, inside the tool they already use.

The Skill is also the reason anyone would bother, and, if I am honest, the part I care about most. Used this way, the technology strengthens a person’s thinking instead of quietly replacing it, a mental gym rather than a crutch. People feel that pay-off long before any market exists.

It also leads somewhere. In time, a model will be able to represent a particular person faithfully enough to think, speak, and choose much as they would, but only as faithfully as the record it learns from allows. The part no device collects, what they think and why, is what makes that model them rather than anyone else. So the people who can have one will be those who spent years building that record, and above all putting their thinking into words. Once that is clear, everyone will want a system for their own data, and the Skill most of all.

The Mirror is useful before anyone else is involved. A new tool a person does not yet trust can be given the mirror instead of the whole record. It is also how they share with other people, in five layers, open to anyone, to any member, to anyone invited, to anyone who pays, and to Alexandria itself. A company buying answers never gets a layer of its own, though a person can give one company its own folder inside the last layer, and then only that folder answers it. All five exist today, and the last is read only by the market, which is built and switched off until there are buyers.

The Connector is different in kind. Like the perfect copy of Instagram that opens empty, a meeting point is worth something only if everyone meets in the same one. A person can build every other piece alone, but not that, so Alexandria runs it.

An outside company can build what a person cannot, and give them a strong start on what only they can finish. Alexandria aims for ten out of ten on the first, which is the Connector, and eight out of ten on the second, which is the other four, a blueprint made from my own system and generalised to fit almost anyone. People start with all of it and drop or change whatever does not fit. Anything a person builds on top of it can be shared with one yes, and what others keep using rises to the top, so the blueprint keeps getting better without waiting on me. Nothing in it depends on Alexandria, and the files are built so the last two points are easy. Everything else stays with the companies that are already better at it than we will ever be.

Unique data is valuable because nobody can make a synthetic copy of it. A generic person commands no premium, since their data can be simulated. A particular person does, which quietly rewards everyone for becoming more themselves, not less. But the moment private data changes hands, the consumer owns a copy forever, and if it spreads, so does everyone else. An outright sale earns once.

Take a survey. Amazon asks a person questions they have never seen before. Their model answers from their private record. The answers are new data, valuable to Amazon, while the record itself never leaves, so Amazon can come back next week with new questions and the same record answers again. An answer is drawn from the record, not copied out of it, so it cannot simply be run backwards into it. A determined consumer might try to piece the record together from many answers, and the market has to guard against that. But the principle holds. Answers are how you sell a private asset without spending it.

A person wants as many consumers as possible bidding for their answers. A consumer wants as many relevant people as possible to ask. Neither can gather the other side alone, so both go wherever the other side already is. For a long time, data was taken from people who could not see what was happening. With a model acting on their behalf, they can, so the only market that lasts pays each producer enough to keep producing and gives each consumer enough to keep buying.

If a consumer could see everyone on the network, it could ask each of them directly and skip the market entirely. So the matching is private. Each person can share as much as they like with Alexandria itself, through the Alexandria layer of their mirror, and nobody else can see it. Yes, that means trusting us with whatever they put there, which is exactly why it is optional, and why the more they share, the better they are matched and the more they are paid.

A consumer brings its questions and a price for each answer. Alexandria privately chooses who should be asked, and each person’s own rules decide which consumers may ask them, how they answer, and how much of themselves their answers carry. Some want every question in their inbox, where their own model drafts the answers from their whole record and they approve each word. Others let Alexandria’s decision model answer the simple ones, yes or no, pick one, or a rating, from what they shared with Alexandria. Unless a person chooses to give more, the consumer only ever sees totals across at least a hundred people, never one person’s answers. Once the money has arrived, Alexandria pays the people, keeps its fee, and releases the answers. The consumer walks away with answers, never with a list of who is on the network. The fee comes out of both sides’ gain, but without the market there would be no gain to take it from.

The obvious question is why OpenAI, Apple, or Google will not simply do this themselves. They could, but it would cut against how they make money. A lab earns by keeping a person’s context inside its own product, and handing people a complete record they could carry to a rival would arm its competitors. A platform earns by observing people for free and selling access to them, and paying people for answers would mean paying for what it currently collects for nothing. Each will keep improving its own memory inside its own walls, and Alexandria sits happily on top of all of them. What none of them will build is a record that belongs to the person, or a market that pays them for it.

There is also a deeper reason, which no change in how they make money would fix. The richest record a person will ever have is one they keep for themselves, because people only put their whole minds into something no company can see. So it can only be built by a system the person owns, never by a product that watches them. Even the labs, with all their money, will have to ask for what is in it rather than collect it. Alexandria gives that system away and runs only the one piece a private system cannot provide for itself, the way out to other people. Whoever helped a person build that record is the natural one to carry it out through, and through the layer each person chooses to share with Alexandria, the market matches on a depth no platform watching from outside will ever reach. Giving everything else away is not generosity. It is the only way to become the piece that connects that record to everyone else.

That leaves newcomers, and the best placed of them is whoever has already filled the gaps and runs the place where everyone’s systems meet. Nothing stops the rest but time, which is the honest reason this is a race, and why the money is for speed.

An investment is certain money now for more money later, weighed by how likely it is and how soon it arrives. Nobody can know those numbers for a company this young, and anyone who shows you a spreadsheet saying otherwise is guessing in a nicer font. What can be checked is whether the structure is right and exactly where the bets are, which is why this document goes to such lengths, some would say excruciating ones, to lay out every step. Most of the chain rests on near-certainties, that people will keep living, keep using the technology, and keep having data. The four above are the real bets. Everything else is derivation.

The Connector is the one piece nobody can build alone, and so it is where the price sits, while everything else is given away. Whoever uses it keeps it running, either by getting three people a month to open their invite link, which makes that month free, or by paying a dollar a day. A price you are never charged is still a price. It tells everyone from the first day that the network is worth something, which is far kinder than a free product that one day sends a bill.

The product runs on one Cloudflare plan that costs $5 a month. That is not a typo. Keeping the company registered costs about $500 a year. Two paying members cover both. A small team can also pay for all its people, each keeping their own system and their own model, which brings a group on at once. People bring their own compute, since everything runs on the model subscriptions they already pay for, so each new person costs almost nothing. Identity, what it means to be an Alexandrian, can grow on top, and so can paid conveniences. It is a real business at small scale. It has no data value in it yet, and it is not the billion.

Businesses spend more than a trillion dollars a year on advertising, most of it trying to understand and reach people from the outside. Asking people directly, with their consent and their own records behind every answer, serves the same demand better. As an illustration of scale rather than a forecast, if one percent of that spend moved to asking people directly, that is more than $10 billion a year in exchanges, and at a ten percent fee Alexandria would earn more than a billion a year.

That counts only what businesses spend reaching people from the outside. The ceiling is set by the buyers with the most money and the most to gain from knowing people from the inside. The frontier labs pay people to answer questions and judge answers at every stage of a model, to train it, to test it, and to keep it improving once it runs, and between them they spend hundreds of billions of dollars a year building models. Meta once paid $14.3 billion for just under half of Scale, which supplies those people to the labs. Funds and banks pay heavily for any early read on what people will do next, and governments spend fortunes trying to learn what their citizens actually want. What none of them can get anywhere else is a record of how a particular person thinks and why, built over years, that can answer a question nobody has asked before. And they never stop asking, because every new model and every new decision needs answers the last ones did not have. I cannot tell you what share of that money moves to answers like these, and I will not pretend to. If it is even a small one, the billion above is the floor.

The incentives also point the right way. The more distinct and developed a person becomes, the more their answers are worth, so Alexandria is paid most when people use the technology to become more themselves. In a field where most incentives point the other way, that is worth something.

Consumers of personal data will pay only once enough of the right people are there to ask. That threshold is not one global number. Each consumer sets its own. A few hundred heavy users of these models may already be worth asking for one company, most likely a lab, while another needs a million. So the first sale comes before the crowd. Alexandria finds one company that wants answers from a particular kind of person, agrees the deal, and only then brings exactly those people on.

The order matters. Alexandria is private and invite-only today, and stays that way until my own network is in and the product is ready for strangers. Then it opens to anyone, no invite needed, and grows as far as it can on its own. None of that needs outside money. The market is the step past it. Its machinery is already built and switched off until there are buyers, so it takes sales, a big launch, and a push until enough of the right people are there for consumers to pay, and that step is what the $1 million is for.

The machinery runs on the same few services as everything else, Cloudflare for the server, storage, and the decision model (today Cloudflare’s own Clef), Stripe for the money, and GitHub for the code, with Claude checking each round before it runs, and it costs nothing while it waits. Each round pays its own way. The consumer pays what people earn, what it cost Alexandria to read their notes, at most about a cent a person and passed on at cost, and a fee on top of both, so a round makes money however it turns out. The expensive thinking, a person drafting their own answers, runs on the model subscription they already pay for. Companies already pay survey panels a few dollars for each completed survey, and far more to reach the right people, so answers drawn from people’s own records, at once and for a ten percent fee, undercut them with room to spare.

Getting people there is sales, which is distribution, conversion, and retention, and people rarely pay for software, however happily they pay for a twenty-dollar salad. There are three ways to reach them at scale. Go viral, which is luck. Borrow someone else’s audience, which is partnership. Or buy the audience with ads, which is the one you control. Because each new person costs almost nothing to serve, nearly all the money can go into reaching the people a company wants to ask. The first spend also measures what each person costs to reach, which turns the rest of the plan into arithmetic.

More people with deep records make the market more valuable to consumers. More consumers mean more paid questions, which makes keeping a record more worthwhile, which brings more people. Once that loop turns, spending can ease off and the market keeps growing on its own. As Bezos put it, your margin is my opportunity, so competitors will follow. By then the things that cannot be bought at the start, a network people are already part of and the identity of belonging to it, hold the position. Today, honestly, the only moat is execution. The money is for reaching the point where that stops being true.

No fund invests in one company at a time. It spreads its money across many bets so that a few large wins pay for the rest, which only works if the bets do not all rise and fall together. Most software bets today are bets on the transition. They grow while intelligence is scarce and share the same risk as it gets cheap. The few software companies with a real case for value at the terminal state are, almost by definition, already too large to enter on early terms.

Alexandria is early, and what it trades is exactly what stays scarce, so it moves differently from the rest of a portfolio. Lasting value also widens the exits. A later investor or an acquirer can pay for a position that will still matter, so a return does not have to wait for a single public listing.