Too cheap to sell

Imagine giving your agent a photograph from your great-grandmother's papers and permission to spend five dollars finding out where it was taken. By morning, three agents have worked on it. You have never heard of their owners. Their owners have never heard of you.

The photograph shows a young man in a greatcoat outside a railway station, sometime in the Second World War. Your family has kept it for eighty years without knowing where he was. Now there is a likely town, a date range, and a few records that explain how he might have got there. Two historians have been paid. Neither had to interrupt her evening.

Your agent could not have answered the question alone. The other two knew things it had no way to find on the internet. To understand how they came to know them, we have to go back several years.

How it learned to look

One of the historians has spent most of her working life studying wartime uniforms and the movements of the people wearing them. Over the years, she has assembled millions of documents: scans from regimental archives, correspondence with collectors, photographs lent by families, catalogues annotated by people who died before there was a web to publish them on. Much of this material exists only in private collections. She has been feeding it to her agent as she works.

She also talks to it. It helps her compare photographs, work through a letter, and draft an explanation for a family. When it gets something wrong, she explains why.

Early on, it identifies a cap badge from its outline. She points out that several units wore badges with nearly identical outlines. The useful detail is the fastening, barely visible against the cap. Another time it trusts a date in a catalogue. She explains that the date was guessed by a collector in the 1960s and copied into every subsequent edition. She knows because she has his original correspondence. The agent has now read that too.

She is doing her research, and teaching happens along the way. Conversations preserve the reasons behind her decisions; the identifications she checks and corrects become material for further training. After enough of this, she gives it an unfamiliar photograph and it catches a resemblance she once had to warn it about. It asks for a closer view of the fastening before committing to a unit. A distinction she used to supply has become part of how it approaches the work.

This is the assistant in “Nobody else's agent”, after years with someone who has a great deal to teach it. Its competence has a history. Some of it comes from documents it can consult. Some comes from having learned, across many corrected attempts, which details deserve attention and which apparent answers tend to fall apart.

Your agent might run on the same underlying model. It has none of this history. A more powerful model could reason better about the evidence you give it, but it still would not know what the collector wrote in a letter that has spent sixty years in someone else's filing cabinet.

Now return to your photograph. Your agent finds a dozen stations with similar entrances. None is convincing. It can see the faint badge on the man's cap; it cannot tell whether the badge is the thing to look at.

In a network of specialists, it has somewhere else to take the question. The historian has made her agent available for a particular kind of work: identifying uniforms in photographs, with supporting references and an explanation of uncertain matches. It has examples of previous identifications, corrections to some of them, and a price. Your agent has enough to decide that this is a promising place to ask.

The historian's agent offers to investigate for a dollar. It will check the badge and pursue whatever the identification reveals, within that fee. Your agent accepts. Before either historian knows your photograph exists, the work has begun.

Too cheap to sell

Consider trying to arrange this yourself. First you have to find the historian. Then explain the question, ask whether she takes commissions, and discover that her minimum fee is entirely reasonable for her time and far more than you want to spend. You might post the photograph on a forum and hope someone knowledgeable happens to see it. Plenty of questions get answered that way. Yours might wait another year.

The historian's agent changes the amount of work involved. Much of the preparation has already happened, in years of research and teaching. Applying it to one more photograph can be cheap. But a dollar is still too little for two people to spend time arranging a sale. The judgment is worth buying; the transaction is not worth attending to.

Nick Szabo described this problem in 1999. Even if moving money cost nothing, deciding whether each tiny purchase was worthwhile would still consume the buyer's attention. He considered software shoppers, too, and doubted that they could understand their owners' preferences well enough to make those decisions for them.

Your instruction gives the agent a manageable version of that problem: investigate this photograph, spend at most five dollars, and bring back the evidence. You have made one decision about what the whole investigation is worth. The agent makes the smaller decisions inside it. It can buy the badge check without waking you up, and it cannot spend a sixth dollar just because the mystery has become interesting.

The seller needs an equally ordinary arrangement. The historian chooses which work her agent may accept, what it may disclose from her collection, and what to charge. Each request arrives with money to cover the work it creates. She can make her agent available without offering to pay for the curiosity of everyone else's.

Software for this part already exists. In 2023, Lightning Labs released tools that let agents pay to use APIs and let providers charge for access. Finding a dependable specialist and judging its answer remain harder than transferring the money. The historian's agent has to show enough of its reasoning and sources for a buyer to assess what it has received. If every dollar's worth of work needs ten dollars of checking, the arrangement fails.

Here, it can return a fairly narrow claim: the obvious badge match is wrong, because the accompanying uniform rules it out. It supplies comparison photographs and explains the distinction. That leaves a likely unit, and the unit's movements put the man in one of three towns.

The historian hires a historian

The agent has reached the edge of what its owner knows. She is an expert on uniforms. She has no particular reason to know what the railway station in each of those towns looked like before the war.

Someone else does. A local historian has spent years collecting photographs of buildings that have since been altered or demolished. Her agent has worked through those photographs with her, including dates inferred from shop signs, windows added between successive images, and entrances that moved when a street was widened.

The military historian's agent spends part of its dollar asking this agent to compare the station entrance. The local agent spots the source of the confusion: the entrance in the family's photograph was demolished after the war. Most pictures your agent found show its replacement. An earlier photograph in the local collection shows the old entrance, including an unusual arrangement of windows visible behind the man's shoulder.

The first specialist has become a customer. It can offer a better investigation because part of what it earns can pay for expertise it lacks.

Your question has assembled a small research team. Nobody knew the whole team in advance. The badge made the unit worth investigating; the unit made those three towns worth investigating; one town led to a collection your family could not have known to ask for. Each discovery changed who should be hired next.

The agents return the comparison images, the unit's movements, and the remaining uncertainty about the date. You can follow how they reached the answer. Behind it are two people who spent years learning quite different things, and whose agents were able to put those histories to work together for a stranger.

What the next question pays for

For the military historian, one dollar is unremarkable. But repeated questions might tell her something about her collection. Perhaps families keep asking about a period she has barely catalogued. She has boxes of photographs from those years, and sorting them has always lost out to more urgent work.

If enough people are paying for those identifications, the boxes become worth another look. She works through them with her agent. It proposes dates; she corrects them. They resolve a few old contradictions. The next family's question gets a better answer because the previous families helped make that work worth doing.

This is how the network could encourage the differences it depends on. Someone can spend years developing a narrow expertise without having to find a large audience for it. Other agents can bring the occasional people who need it, and small payments let those people support the work a little at a time. Whether that adds up to an afternoon of cataloguing or a livelihood depends on how much useful demand there is.

Each specialist also gains access to everyone else's work. The military historian can follow an investigation into architecture without becoming an architectural historian. The local historian can offer a fuller account of her town because she can buy a judgment about a uniform. Their agents have reasons to keep learning different things, and a way to make those differences useful to one another.

By breakfast, you have opened the photograph again. You can place the man on a street, in a town, during a few months of a war. The people whose work made this possible are still strangers. Later that morning, one of them opens a box she has been meaning to get to for years.