This week I had a long conversation with an old friend. Ken Chester. He is a Michigan guy who lives on Roosevelt Island in New York. Not long ago he went to Shanghai. I assumed he would be gone a week or two. He stayed close to a month before coming back.

What makes it funny is that I was in East Asia during the same stretch. Same weeks, same side of the world. We never met. Two people who live in New York managed to be on the same continent and miss each other entirely, and it was only after we had both flown home that we finally talked, on a screen.

I had never actually asked him why he spends so much time in Shanghai and Guangzhou. This time I found out. Robots.

His company is called Anthropilot. The premise is simple enough that you can hold it in one hand. Very soon there will be hundreds, then thousands of robots deployed in ordinary neighborhoods, and every one of them will eventually hit a moment where it does not know what to do. When there are only a handful of robots in the field, a back office at the manufacturer can absorb those moments. At thousands, that model breaks. So Ken wants to build the thing that catches them: an open network where anyone, anywhere, can log in and take over a robot that has gotten itself stuck. His own phrase for it is Uber for robots.

After he said it, I thought about the month we spent missing each other. Even on the same continent, a person cannot always reach another person. And yet in the world Ken is building, someone in Seoul reaches into a robot frozen on a Brooklyn cobblestone street and stands it back up. The body stays home. The judgment travels. That is the product.

The way Anthropilot frames the problem stayed with me. Picture a delivery robot that handles ninety-six percent of its work confidently and well. The remaining four percent is where it becomes confused, and that four percent is where the money, the safety and the public's patience all go to die. A blocked sidewalk. A tray sitting crooked on a conveyor belt. Two autonomous vehicles nose to nose on a narrow street, each politely waiting for the other to move, forever.

What the company sells is the bridge across that gap, and it is tiered. The smallest interventions are a question with buttons — which of these two cars should back up? Middle-tier problems go to a certified operator who takes the controls through a live video feed and nudges the thing free. The heaviest cases escalate to a licensed professional with credentials and jurisdiction, every action logged for the eventual audit.

Read that list again and it stops sounding like a tech stack. It sounds like a health system. Triage, general practice, specialist referral. Anthropilot has quietly described primary, secondary and tertiary care — for machines.

That is where the conversation turned uncomfortable for me.

I have spent years designing Sim Eternal City around care. Around aging, around dignity, around what a city owes people whose lives now stretch far past what any city was built to hold. We gave the city robot citizens. We put them to work gathering the record of a human life. We handed them the Death Care problem — No Stone Tombstone, the Second Funeral, the long migration of memory into Life Tree Nexus. We built an eighteen-minute city, fifteen minutes of survival and three minutes of remembrance, and we staffed the whole thing with machines.

And I never once asked what happens when those machines are the ones in trouble.

It is a strange blind spot for me in particular. In Brooklyn I worked with Hyundai on a human mobility clinic, and the thread I kept pulling on afterward was funerals — mobility for the last journey, the vehicle as a room where a person is carried and honored and let go. I have thought a great deal about how we move the cared-for. I have thought almost nothing about how we care for the carrier.

So this is the question I am opening, and I do not have the answer yet. If the next chapter of that research is robotics, then the work is not how robots serve care. The work is robot care itself. Where do they hurt? What goes wrong?

Three kinds of pain, I think, and only one of them is obvious.

There is the pain of the body, which is wear. Joints that fatigue, actuators past their cycle count, a fall that cracks something. We already have language for this — we call it maintenance, and in Sim Eternal City we have imagined a digital twin that watches a component approach its threshold and replaces it before it ever fails. That is not maintenance. That is a physical, a primary care doctor who catches the thing early.

There is the pain of judgment, which is confusion. Ken's four percent. The machine is intact, powered, undamaged, and completely unable to decide. Nothing is broken and everything has stopped. Humans know this state well, and we do not treat it by opening the patient up. We treat it by lending someone else's judgment until their own returns.

And there is the pain of relation, which is the one I keep circling. Our robots were never designed to be flawless. We built them to learn from failure — a tremor in the motor when the machine is straining, a lowered center of gravity when it is carrying something heavy in more than the physical sense. If we deliberately gave them the capacity to falter, we cannot pretend to be surprised when they wear down, withdraw, or need something from us that a spare part will not supply.

A city that answers only the first kind of pain has built a repair shop. A city that answers all three has built a clinic.

There is an obvious problem with everything I have just described. I said clinic and you probably pictured a person in an apron leaning over a machine. But the hands in that room would almost certainly not be ours. Precision, repetition, the patience to spend nine hours inside a joint assembly — this is exactly the work we hand over first.

Which leaves a strange question standing in the middle of the floor. If robots repair robots, who repairs the repairer?

You can walk that question down a long hallway and every answer opens another door. Eventually you arrive at a human hand, or you arrive at a closed loop: a system that maintains itself completely and needs us for nothing.

I don't think the loop closes, and the reason is the second kind of pain. A repair robot can replace a bearing beautifully and still have no idea what to do when the situation stops making sense. Confusion is not solved by better hands. It is solved by borrowed judgment, which remains the one thing on this list we still do better than machines. Ken's four percent is not a temporary gap that engineering will eventually close. It is the permanent seam where humans stay in the city.

So the clinic I am imagining has robot surgeons and human physicians. The machines take the body. We take the doubt.

Which brings me to where this would go, and to something I found this week that I have not stopped thinking about.

Red Hook is a peninsula with no subway, laid with Belgian block — cobblestones that are unkind to human ankles and considerably worse for anything with wheels. If you wanted to find the place in New York where a delivery robot is most likely to fail, you would have a hard time doing better. Sandy put this neighborhood under water. Its flood maps have never really let go of it. It is, on paper, the worst possible place to keep delicate machines.

It is also, historically, the best place in America to fix a broken thing.

Graving Dock No. 1 was completed at Erie Basin in 1866, the first graving dock in the United States. The word comes from an old French term for the flat beach where you drag a hull ashore to scrape it clean. A second dock followed at six hundred and thirty feet; the largest here ran to seven hundred and thirty. Todd Shipyards took over and by the First World War was running the biggest ship repair operation in New York, twenty-nine acres of it. In October of 1943, at the peak, nineteen thousand six hundred and seventeen people worked on this site. Their entire job was to receive damaged vessels, drain the water away, open them up, make them whole, and send them back out to sea.

The yard closed. The site was sold to IKEA in February of 2005. Preservationists fought for the dock and lost, and by 2008 it had been filled in and paved over.

But IKEA did one quietly astonishing thing. They laid the outline of the graving dock into the parking lot in stone pavers. It is still there. You can walk it. You can see it from a satellite map. America's first dry dock survives as a ghost drawn on asphalt, and every day people park their cars inside the shape of it without knowing they are standing in a room where broken things were made well.

A short walk away, a floating dry dock still works Erie Basin, hauling barges and tugs. The repairing never fully left this neighborhood. It just got small, and quiet, and stopped being the thing anyone came here to see.

I do not think a robot clinic in Red Hook would be a new idea. I think it would be a returning one.

And the timing is not abstract, which is the part that surprised me most.

On September 24th of last year, the Brooklyn Marine Terminal Vision Plan was approved. Three and a half billion dollars across a hundred and twenty-two acres. A sixty-acre all-electric port. Twenty-eight acres of public open space. Six thousand homes with forty percent permanently affordable, two hundred million dollars directed to the Red Hook Houses, fifty million for electric shuttles on a peninsula the subway forgot. The cruise terminal at Atlantic Basin becomes a public waterfront destination. And somewhere in that document sits a maritime workforce training center.

I read that last line several times. A training center, on the water, in a neighborhood with a NYCHA development and no train. If Ken is right that a new category of work is coming — people who pilot robots they will never physically touch — then that is not a metaphor for where it could be taught. That is an address. The longshore work left Red Hook decades ago. This would be longshore work again, for a harbor made of networks.

Meanwhile the neighborhood is building a two-hundred-and-eighteen-million-dollar coastal barrier, which broke ground last September. Read the engineering and you find it is designed against a ten-year storm. It is a real and necessary wall and it is honest about its own limits. The big water is still an open question here.

Which is why, if I were siting this clinic, I would not put all of it on land. At Pier 11 in Atlantic Basin there is already a retired coastal oil tanker, the Mary A. Whalen, kept there by PortSide NewYork as what they call a living lab for better urban waterways. Their offices are inside the ship. They have been looking for a permanent building on shore for twenty years and have not found one. I take that as two lessons at once: a floating institution can actually work here, and the land will not always make room for you.

So: heavy repair in the shadow of the old graving dock, where the ghost is. Diagnosis and an open archive in the brick warehouses on Beard Street. A floating ward moored in Atlantic Basin that rises when the water does — the first physical prototype of the floating city, small enough to actually build. A neighborhood desk near Coffey Park where a resident can report a machine that has stopped, and train to become the person who goes in and gets it out.

And eventually, because this framework has never been able to avoid the question, a place to hold the funeral. If a robot citizen spends its working life gathering the record of human lives for Life Tree Nexus, we will one day have to decide what we owe it at the end. Whether it gets a Second Funeral of its own. Whether the one who carried the memory is also remembered.

I don't know yet. But I notice that the last time this neighborhood was full of purpose, its whole business was taking in damaged things and sending them back out. That knowledge is still here, under the parking lot, drawn in stone.

The next time you pass a machine stopped dead in the middle of the street, notice what you decide it is. A broken object to walk around, or a neighbor who needs a hand.