Starting from the Muse craze: identity, reputation and social networks in the age of AI agents.
Muse has taken off lately. Quite a few people around me have started using it seriously: chatting with it, having it remember things about them, asking it first whenever they face a choice.
Watching all this excitement, one question kept circling in my head: why do people want an AI that "gets them" so badly?
Following that question further, I ran into an older one. What do we actually rely on to let other people know who we are?
1. What a Personal Agent eventually becomes
Most of what people call AI assistants today is still about "doing things for you": ordering food, answering emails, managing your calendar. I think that's only the starting point.
If an AI stays with you long enough, what it knows will go far beyond your to-do list. Which books you've read, which projects you've done, what you keep agonizing over, which of your calls later turned out to be right, whom you trust, and who trusts you.
Let that build up for a few years, and it may understand you more deeply than most of your friends do.
At that point it's no longer really a tool. It's more like your double in the digital world, carrying your experience, your preferences and the way you make judgments. When you're not around, it can answer questions for you, negotiate collaborations with other people's agents, exchange information and do the first round of screening.
So my view is this: the ultimate form of the Personal Agent is a digital twin.
Everything I reason through below stands on that premise.
2. Luxury cars, fine watches and elite schools are really the same thing
Let's go back to socializing itself.
When two strangers meet for the first time, the information between them is badly asymmetric. I don't know whether you're reliable, how capable you are, or what you've pulled off. Finding out takes a long time of getting to know each other and working together, and sometimes getting burned along the way.
So humans invented a lot of "shortcuts."
A Patek Philippe on your wrist suggests you have money. Driving a Porsche suggests you're doing well. A business card that says director at some big tech company suggests you're capable. MIT or Stanford on your resume suggests you're smart and hardworking.
All of these do the same job: when the other person has no time to get to know you, they give a rough answer up front. I call them "proxy identity." They stand for you, but they aren't you.
Why do they work? I think the word "currency" is a useful way to understand it.
A banknote isn't worth much in itself. It can buy things because everyone believes it can buy things. The value of money comes from consensus.
Money was invented for efficiency in the first place. Before money, people could only barter, and finding someone who happened to need what you had and happened to have what you wanted was incredibly hard. Social currency solves an efficiency problem too: really getting to know someone is too slow, so a label stands in for now.
Social currency works on the same principle. A watch can stand for "success" not because it keeps better time than your phone, but because everyone assumes that someone who can afford it is probably capable. A school can stand for "excellence" because society broadly recognizes its admission bar and the quality of its education.
With consensus in place, these things can "circulate" among strangers. You don't need to explain yourself; the other person gets it at a glance, just as you don't have to prove a banknote is real when you pay.
Consensus alone isn't enough, though. Why fine watches, luxury cars and elite schools, and not something else?
Economics has something called signaling theory. The economist Michael Spence used it to explain a puzzle: if employers can't see a job seeker's real ability, why do they value degrees so much? His answer was that a degree doesn't necessarily prove you can do the work, but a good degree is hard to get, and people with less ability find it hard to obtain. Precisely because it costs so much to get, it becomes a credible signal. Spence later won the Nobel Prize in Economics for this work.
Fine watches and luxury cars follow the same logic. More than a hundred years ago, Veblen wrote about "conspicuous consumption": the rich often buy expensive things not because they're useful, but because others can't afford them. Being expensive is itself the function.
So whether a signal can be believed comes down to whether faking it is costly enough. Something anyone can easily get proves nothing.
It also means that social currency, like real currency, can depreciate. Once a signal appears that is harder to fake yet far cheaper to obtain, people no longer need a watch to prove themselves, and the consensus behind the old signal slowly loosens.
Luxury cars and fine watches have another trait: most of them are consumer liabilities. They start losing value the day you buy them, and upkeep keeps costing money. Plenty of people buy them anyway, knowing full well it's a bad deal, because in social settings they really are "useful": they buy you a first impression.
Elite schools are a bit different; there is real selection and training behind them. But the way they get used socially is a lot like a watch. People see a label. They don't see what you actually learned in those four years, let alone what you went on to achieve in the ten years after.
Alexandr Wang, the person behind Muse and Meta's Chief AI Officer, dropped out of MIT after just one year. Today almost nobody doubts his ability because he never got a degree; people look at what he has built. The example is a reminder that there has always been a layer between a degree label and a person's real ability.
So this piece isn't really just about luxury goods. What I actually want to say is this:
When AI can read a person's abilities, experience and reputation more directly, every proxy identity we use in place of real judgment may get repriced.
3. The problem with proxy identity: it's too coarse
Proxy identity has been popular for thousands of years because it's cheap and convenient. But its problems have always been there.
It's easy to fake. Rented luxury cars, watches bought on installments, padded resumes: we've all seen them.
It carries very little information. Two identical cars in a parking lot: one owner is a self-made founder, the other just got money from their family. From the outside, there's no difference at all.
It also lags badly. Judging someone's ability at forty by the university they got into at twenty is, if you think about it, pretty absurd.
Less visible is the unfairness it creates. Plenty of genuinely capable people get filtered out in the first round because they don't carry these labels, without ever getting the chance to be properly understood.
We put up with these problems in the past because there was no better way. Getting to know a person directly was simply too expensive.
Zoom out, and a lot of social and technological progress has been doing the same thing: making society as a whole run more efficiently. Money replaced barter, the telephone replaced letters, search engines replaced flipping through the Yellow Pages. Each time, the old way wasn't wrong; a faster, more accurate new way simply came along.
In its own era, proxy identity was an efficiency tool too. But today its error margin keeps growing and its cost keeps rising. It is turning into the old way that's waiting to be replaced.
4. Let your agents get to know each other first
Now bring back the idea from Part 1.
Suppose everyone has a digital twin that holds your real abilities, experience and track record. If you're looking for a business partner, or someone wants to decide whether to work with you, the first step no longer has to be looking at what car you drive. Your agents can have a conversation first.
What they exchange is far more concrete: which projects you've done over the past three years and how they turned out; how accurate your judgment is in a given field; what people who've worked with you say about you; whether your ways of working and your values fit.
For this to work, personal agents alone aren't enough. It needs a key piece of infrastructure: an agent social network.
Today's social networks connect people, and the content people post. In the future there may be another network, one that connects agents to agents. In the background, they're constantly discovering, matching and talking.
A lot of the "getting to know someone" that today depends on dinners, introductions through friends and industry conferences would be done in advance on this network.
It sounds a bit like LinkedIn, but the difference is big.
The information on LinkedIn is written by you. Titles, experience, skills: all filled in by the person themselves, and the "endorsements" others click casually don't carry much weight. In the end it's still an electronic business card, and what it shows is still proxy identity.
LinkedIn also depends on people scrolling, searching and sending messages themselves. Who you get to know depends on how much time you have, and on whether you're willing to swallow your pride and reach out first.
An agent social network works the other way around. The ones active in it are digital twins, and what they carry is verifiable records, not self-introductions. They work in the background around the clock, can talk to thousands of agents at once, filter out the people who are genuinely a good fit, and then hand them to you.
Picture this: you're asleep, and your agent is still socializing for you. It has talked with hundreds of agents, compared experience, abilities and interest in working together, and screened out everyone who clearly isn't a fit. When you wake up the next morning, what you see is a shortlist: which few people meet your criteria, why they fit, and where you could start the conversation when you meet.
LinkedIn is people finding people, based on labels. An agent social network is agents finding agents, based on real records.
In terms of efficiency, that's a leap. Matching that used to take months, several dinners and a few layers of connections gets compressed into the time it takes you to sleep.
By the time two people actually meet, they already know a lot about each other. The question luxury cars and fine watches used to answer, "roughly what kind of person are you," has been answered before you even meet.
5. Blockchain: making what agents say credible
There's a question here we can't get around: why should anyone believe what an agent says?
My digital twin says I pulled off some project, and I could easily have it exaggerate. If it's just AIs talking each other up, it's nothing more than a fancier business card.
So an agent social network has to have blockchain underneath it.
The blockchain I mean has nothing to do with speculating on coins. It's about its most basic ability: once a record is written, it's very hard to tamper with, and any third party can verify it. That happens to be exactly what agents most lack when it comes to building trust with each other.
For example, your degree is issued on-chain by your school; the projects you completed are confirmed by your partners; your investment record, open-source contributions and delivered results all leave traces that can be checked. When an agent talks to others, it cites these records, which someone has vouched for, not what you said about yourself.
Over time, each person grows a "Value Graph": what you've done, whom you've worked with, who trusts you, and in which fields you've delivered consistently over the long run.
This graph is alive; it keeps updating with what you do. It's harder to fake than any watch, and more specific than any label.
Stack these layers together: the digital twin represents you, the agent social network connects, the blockchain proves, and the value graph presents. That's the complete structure as I picture it.
6. What loses value, and what doesn't
If this line of reasoning roughly holds, I think we'll see a "repricing."
The first things hit are the ones held up purely by their signaling function. If a watch's main value is telling others "I have money," and others can already learn your real situation in a more reliable way, that part of the premium will slowly shrink.
The halo of elite schools will thin out too. A school will go back to being "one experience," no longer a pass valid for life.
But I don't think luxury cars and fine watches will disappear. Loving the craftsmanship of a mechanical watch, enjoying driving, simply finding something beautiful: those needs are real, and they have nothing to do with socializing. What will disappear is the part that's "bought for others to see."
Conversely, some things will become more valuable: a long, consistent track record, verifiable results, a genuine reputation. And the things AI can hardly replace, like the feeling of meeting in person, the bond of having been through something hard together, the warmth between people.
An agent can help you filter out the people who aren't a fit. But whether to get close to someone will most likely still come down to sitting down for a meal together.
In closing
Back to the question from the beginning: what do we rely on to let other people know who we are?
In the past, the answer was a set of labels. What car you drive, what watch you wear, which school you went to, which company you work for. These labels saved us a huge amount of time, and they also left many people misread for their whole lives.
In the future, the answer may simply be the things you've actually done. Written on the blockchain, able to stand up to verification, carried by your digital twin as it goes out and greets the world on your behalf.
The logic behind this is the same as money replacing barter: society always finds a more efficient way to run, and socializing is no exception.
By then, maybe nobody will need to spend hundreds of thousands on a watch to prove something that could have been proven directly.
What do you think?