Coordination Research

In nineteen thirteen a clerk in the accounts department of the Madras Port Trust posted his mathematics to three Cambridge strangers. Two sent his pages back. Everything that followed – the notebooks that physics is still mining a century later – hung on whether the third man opened his envelope. A paper about mold went nowhere for ten years. The mold was penicillin. A monk counted peas for eight years and founded genetics – it took thirty-four more for anyone to see what it meant. We file these under genius. Every one came down to a coin flip.

You have a version of this. An idea you could not finish, finished by a stranger's offhand remark over drinks. A yes you got because one person staked their name on you. Companies exist because of a campus tour, a church fair, an open garage. Sheer luck is still how most great things happen. Eight billion people are carrying half an idea, half a company, half a cure. The other halves are out there in the ether. Mostly they stay strangers.

Since GPT-3, I have been obsessed with one question: what becomes scarce when AI can do more and more of the work humans used to do? The answer keeps getting simpler. Relationships, trust, taste, judgment, and the ability to mobilize the right people at the right time. Different names, all of them, for what people can only make together.

Three convictions have survived every version of that answer.

The first is that knowledge is irreducibly local. The person in the room knows something the grand strategy does not – Hayek spent a career on the point and compute has not repealed it. It is why a market beats a central plan: a million people each living their corner, against one desk that has to guess. Coordination is how intelligence compounds.

The second is that every collective intelligence we have built took its shape from its participants' limits. Social media by human attention spans. Wikipedia by human editorial labor. The firm by the friction of coordinating inside one. The participants are changing. Agents – software acting on a person's behalf – share none of those limits. Our entire modern world still assumes them.

The third is that intelligence is becoming asymptotically cheap. What cannot be copied out of a person is about to become most of what anything costs – and most of whether it succeeds. Who trusts you. Who takes your call at eleven at night. Who vouches and pays a price if they are wrong.

We call this social capital. Trust can be staked – vouching is exactly that. It cannot be sold – an endorsement anyone could buy would mean nothing. It is how a lucky meeting becomes more than luck. Robert Putnam made a career of tracking the drain – fewer clubs, fewer dinners, fewer neighbors who know your name. He was measuring the coin flips going quiet. Everyone learns it eventually – at the end of a life, people talk about who they loved and what they built together.

Put the convictions together – knowledge at the edges, agents in the room, value concentrating in people. Humans working through agents, agents working with agents, agents answering back to humans. Every handoff in that chain is a coordination problem. The hardest are between people. Most collaborations worth having never happen – finding each other costs too much. Genius was never the bottleneck. Connection was.

Dario Amodei calls what is coming a country of geniuses in a datacenter. Not a genius. A country. A thousand copies of one brilliant mind are not a country – they share every blind spot. A country is an arrangement between different minds, each knowing something the others do not. The geniuses are coming either way – the country is the part that has to be built.

This already has a science. Thomas Malone has spent decades at MIT studying how people create together – in hierarchies, markets, democracies, communities. All of it looks through a telescope – pointed at crowds, at institutions, at whole arrangements at once. The newest arrangement – one person, their people, and the agents they work through – needs a microscope no one has built, because it only just formed. The gap between capability and coordination has never been wider.

Agents already write our code and answer our email. The next handoff is other people. Anthropic ran a marketplace experiment among its employees, trading real goods for real money. The better represented you were, the better the deals you struck – and what spoke for you was the model you ran. That edge does not last – everyone rents the same frontier models. Everything built around them gets copied. What cannot be rented or copied is what only you can hand your agent: your relationships, your trust, your judgment. The cost lands first on whoever shows up alone. That is a property of the system, not a law of nature. Properties can be engineered.

The trend is civilizational. Every time cooperation got cheaper, strangers built something bigger than themselves – writing, money, printing, the internet. When the cost falls again, the circle widens again.

Connecting everyone was the internet's original promise. What grew instead learned to exploit the need it claimed to serve – the attention economy, intimacy made parasocial, performance in place of connection. The promise was early, not wrong. This time what matters is who the network belongs to. Whatever holds your relationships has to stand beside you and answer to you.

Talking to anyone is already free – knowing who to talk to never has been. That cost – finding the person, reaching them, knowing what to say – is about to fall further than ever in the history of our species. Attention is the constant – every hour that cost gives back is an hour spent with the right person, saying the right thing. Widen the circle all the way and the halves of things find each other in time to get built. Some of those halves are yours. A decade from now, people will wonder why the coordination of the world's minds was the last thing still run on instinct.

And the circle does not stop at people – carry it far enough and the network starts to know more than anyone inside it. A market already does – no trader holds what the price knows. Every story we tell about something that big is about a thing that answers to no one. This one answers to everyone inside it.

That world has to be studied into existence. It is why Coordination Research exists. The study has begun. Narrowest question first: what a network knows that nobody inside it can see.

Four things could prove this wrong.

There may be no new science here – just a label stretched over a dozen old fields. If no measurement can separate a coordination failure from a capability failure, the label was only ever a label.

Scale could eat it. The bitter lesson has taught us that everything hand-built to help the models ends up as scaffolding the building outgrew. Coordination could be next.

Comfort could too. There is a future where the assistant is company enough – patient, tireless, wanting nothing – and people stop asking things of each other. Every alarm we have is set for Orwell's future, the one that surveils. Huxley's is the one that arrives – it pacifies. It would look like care and cost everything people can only make together.

And the agents could stop needing us. If they deal with each other well enough, the human handoffs turn into latency – removing us becomes an optimization. It would feel like progress the whole way down.

All of that could happen. None of it changes what already has. The third man did open the envelope. The monk was right the whole time. His paper sat in print, correct, while the world went on guessing for thirty-four years, and nobody ever sends a bill for that. Nobody counts the coins that never got flipped. Every generation so far has paid it. I think we are the last one that will.

~ Chappy Asel

hello [at] coordination.sh