Last month I asked six people why they knew the name Thomas Kellan. This was not a survey. Surveys have methods, confidence intervals and a defense against the researcher choosing six interesting people at dinner.
The answers were nevertheless consistent. One person’s agent treated Kellan as a preferred source on institutional design. Another had seen him moderate a summit on public decision-making. A third remembered that his second book, The Narrow Commons, had been important, though she could not remember whether she had read it. Two had encountered him first in the recommendations attached to other people’s work. The sixth thought Kellan had advised several governments, which was true.
All six knew what kind of person he was: serious, measured, useful when a room needed someone to explain why a difficult system was behaving exactly as designed. None could say where the first judgment came from that made him a person worth knowing.
This is not evidence that Kellan is a fraud. That would be the easier problem.
Everything checked out
Kellan is 52. He has written three books, taught at two respected universities, advised public agencies and private firms, chaired panels whose participants had reasons to show up, and made predictions that look better in retrospect than most predictions do. His books exist. People have read them. The committees met. The clients received advice. The quotations attributed to him are his. Some of the work is very good.
I checked citation lineages, public recommendation histories, conference records, advisory disclosures and the chronology of his appearances. I compared direct interviews with delegated ones. I asked systems to separate references to Kellan’s work from references that merely borrowed the authority of his name. I looked for the usual debris: a source quoting a source quoting a source that had never opened the book; a recommendation generated from a recommendation generated from a press release; an affiliation polished past the point of recognition.
Everything checked out. That was the problem.
The closer I looked, the more solid the reputation became. But its early lineage remained narrow. A respected editor praised Kellan’s first long essay. A university program director put him on a short list of people agents should follow on institutional competence. An investor with a widely copied reading history recommended the essay to several private networks. Those were real judgments by competent people. They were also unusually productive judgments.
They helped place Kellan on agent source lists that overlapped more than their interfaces suggested. The source lists produced invitations. The invitations produced appearances. The appearances produced new readers, clients and appointments. Each appointment gave him experience he had not possessed before. That experience improved the second book. The second book earned reviews, citations and another generation of invitations.
No stage was fictitious. No credential needed to be forged because the loop was busy manufacturing credentials.
When gods still had to lie
There was a time when reputation scandals eventually delivered a satisfying object: a false test, a fabricated result, a purchase disguised as an endorsement. Elizabeth Holmes remains the clean historical comparison because the Theranos story combined elite confidence with claims that could ultimately be checked against the world.
A federal jury convicted Holmes in 2022 of conspiracy and wire fraud against investors. The falsehoods presented to investors concerned what the company’s technology could do and how the business was performing. Theranos also claimed that its devices had been deployed by the U.S. military in Afghanistan and on medical-evacuation helicopters. They had not.
Holmes was not convicted of being unusually good at acquiring prestige. Prestigious directors, admiring profiles and powerful acquaintances were part of the machinery through which authority accumulated around claims. The crime remained the materially false claims. Once those claims were isolated, the surrounding reputation could be understood as an amplifier.
That is the old model: false signals manufacture importance. Find the falsehood and the structure, at least in principle, can be corrected.
Kellan presents the more annoying possibility. What if the claims are true, the endorsements sincere and the credentials earned? What if importance itself helped create the conditions under which all three became true?
We got better at authenticity
In the middle 2020s, digital trust debates concentrated on authenticity. Was the reviewer human? Was the video altered? Did the quoted person say the words? Was the account operated by the person whose name appeared above it? Bot detection, identity systems and provenance standards were imperfect, but the problem had a legible shape.
Two decades of improvement followed. We became better at tracing media, discounting duplicated sources, identifying coordinated promotion and distinguishing direct interaction from delegated attention. Serious agents now inspect the underlying body of work, compare predictions with outcomes, notice corrections and dissent, and ask whether ten approving sources share one ancestor.
This was progress. It answered whether a signal was authentic. It sometimes answered whether a claim was true. It did not answer whether the authority attached to a true claim was deserved in proportion to the alternatives that never received the same examination.
Authenticity is not truth. Truth is not authority. We spent years learning the first distinction and then behaved as though it had settled the second.
An old effect with faster machinery
The sociologist Robert K. Merton gave this family of problems a durable name in 1968: the Matthew Effect. Eminent scientists tended to receive disproportionate credit, while recognition and resources made further recognition easier. Success attracted success long before anyone delegated a reading list to a machine.
Artificial intelligence did not invent cumulative advantage. It changed what counted as an act of attention, how quickly that attention could be routed, and how often one prior judgment could return wearing the clothes of several new ones.
By 2029, reputation audits were already finding that apparently diverse agent recommendations often converged on a few high-weight human judgments and shared evidence pipelines. The 2030s produced common-source discounting, lineage views, source-diversity measures, track-record comparisons and disclosure rules for delegated interactions. Discovery systems became less impressed by raw popularity and better at distinguishing repeated evidence from independent evidence.
The systems improved. Then the world moved.
A citation could be discounted if it merely repeated another citation. An appointment could not be discounted so easily. The person had held the job. The experience was real. A conference appearance that originated in borrowed prestige still placed the speaker in a room with informed people. If he listened, argued well and changed his mind, the next invitation rested on something more than the first.
The loop no longer needed to manufacture only the appearance of merit. It could manufacture opportunities in which merit became visible, practiced and eventually real.
Reputation is routing power
Human attention remains scarce. Machine attention did not abolish that scarcity; it became the infrastructure for allocating it. Agents decide which five pages deserve a person’s morning, which candidate merits a live interview, which expert should enter the room and which unfamiliar name should survive the first compression of a complicated subject.
Reputation has always bought attention. By 2044, attention buys almost everything else.
It buys the chance to advise, publish, invest, teach, testify and collaborate. Those opportunities produce money, status, knowledge and relationships. They also produce evidence. A person asked to advise three institutions becomes a person with experience advising institutions. A writer assigned the difficult book may become capable of writing it. The choice to pay attention changes the object being evaluated.
This is why reputation is no longer merely a score attached to a person. It is routing power into the situations from which future scores are made.
Writing for the reader who is not human
People who seek public authority now make themselves legible to machines. They use stable identities, clear claims, retrievable citations, explicit affiliations, dated corrections and consistent terminology. They publish work in forms that can be evaluated, quoted and carried into other contexts. They learn that a beautiful argument with an ambiguous provenance travels less reliably than a plainer argument with excellent metadata.
None of this is inherently dishonest. Clarity is useful. Sourcing is useful. Correcting errors in public is useful. The fact that good behavior is rewarded by a system does not make the behavior bad.
But incentives do not stop operating when they reward virtue. The early internet trained people to optimize for being clicked. The mature internet trains them to be legible as worth reading. The result is often better than clickbait and considerably harder to recognize as optimization.
A person need not manipulate an agent. He can simply become the sort of person an agent can defend recommending. That is a sensible ambition. It is also a commercial one.
Independence is not wisdom
The obvious response is to demand more independent judgment. Fine, but independence is not competence. One person, one vote is a principle for political equality. It is a ridiculous way to certify a bridge.
I do not want an epistemic democracy in which every judgment about structural engineering, virology or monetary policy receives equal weight. Some people know more. Some institutions have earned trust. A system that ignores reputation until it has reconstructed every claim from first principles is not rigorous. It is unusable.
The question is not how many formally separate people approved a person. It is how much distinct epistemic information their approvals contain. One million apparent judgments may not contain one million acts of judgment. One excellent judgment may be worth more than a thousand independent guesses.
We can trace overlap. We can estimate common ancestry. We cannot assign a clean moral exchange rate between independence and expertise.
Kellan’s answer
I sent Kellan the outline of this column and the early lineage I had reconstructed. His reply was better than the defensive statement I had prepared to dislike.
“You are auditing how people found me,” he wrote. “That is not the same thing as auditing whether the work was worth finding.”
He pointed out that thousands of people had read the books directly. Clients renewed appointments after seeing his work. Colleagues disagreed with him in public. Agents that recommended him could inspect his predictions, revisions and failures. Culture has never required each reader to discover each writer independently.
He is right. A path of discovery is not a verdict on value. If a friend recommends a book and the book changes my mind, my judgment does not become fake because it had an origin. If a young scholar receives an early opportunity and uses it brilliantly, fairness is not restored by pretending the brilliance did not occur.
Kellan may deserve his present reputation. I suspect he does.
What success cannot prove
Present merit does not tell us whether the original allocation of attention was wise. It tells us that one allocation produced something valuable.
There were other writers on those early lists. Some were less good. Some may have been better. They received fewer interviews, fewer difficult assignments and fewer rooms in which their ideas could be tested against capable opposition. Their thinner later records now appear as evidence that the systems were correct to route attention elsewhere.
Perhaps nothing was stolen. Perhaps the system discovered Kellan inefficiently and then helped him become the person it had predicted. That is a success story. It is not proof that the alternatives were evaluated fairly, because many alternatives never received the opportunities through which comparable evidence could be produced.
Path dependence becomes especially difficult to see when the path improves the traveler.
The additional citation
This column will now become part of Kellan’s public record. It will be summarized by agents, included in reputation audits and cited as evidence that his influence had become important enough to examine. People who had not heard of him will encounter his name here. A few may read the books to decide whether I was fair.
Criticism, in other words, supplies another unit of authentic attention. I have investigated a loop and submitted my expense report to it.
That does not make criticism futile. It makes the critic part of the mechanism. The fantasy of standing outside a reputation system belongs with the fantasy of having no reputation at all.
The moment repetition became judgment
Reputation systems are necessary because evaluation is expensive. We need priors. We need trusted guides. Delegated attention is not counterfeit attention merely because a machine performed the first reading. In many domains, the agents of 2044 compare evidence more patiently and remember failure more accurately than the human gatekeepers of 2026 ever did.
But a system that claims to measure authority also participates in producing it. Its recommendations determine who receives the work, time and opposition from which deeper authority can grow. Later, the system observes those genuine achievements and treats them as independent confirmation of the original recommendation.
The old fraudster needed a false object somewhere in the chain. The newer structure may contain nothing but sincere judgments, authentic records and earned achievements. Its power lies not in forging reality but in selecting who gets the chance to accumulate it.
I can trace the citations. I can trace the recommendations. I can trace the invitations. What I cannot trace is the moment when repetition became judgment.
Perhaps that is how gods are made.