Machines capable of lifting people and things vertically were not new. The problem was what happened when the rope broke. Otis had devised a safety mechanism that caught the platform instead of allowing it to crash to the ground, and the rather theatrical demonstration at New York’s Crystal Palace was intended to prove it. The platform dropped, the mechanism caught it, and Otis reportedly reassured the crowd that all was safe.

Three years later, the first commercial Otis passenger elevator was installed in E.V. Haughwout’s department store in Manhattan. It travelled at about 40 feet a minute, which would feel almost comically slow today, but speed was hardly the most extraordinary thing about it. People were being asked to step into a machine, leave the ground and trust that the machine would bring them back down again.

The problem, at this point, was fairly easy to understand. Before elevators could change buildings, people had to be willing to get into them.

I think about this surprisingly often because of a question I like asking people we’re considering hiring at Sparklin.

What is the most important thing to think about when designing an autonomous vehicle?

Almost everyone says trust.

It is difficult to argue with them. A car without a human driver asks for a similar surrender of control, except that the world around it is considerably less predictable than an elevator shaft. Other cars can cut across it. A child can run onto the road. A police officer can hold up a hand. Construction can turn a familiar street into something the car has never encountered before. The passenger has to believe that the vehicle has seen what they have seen, and perhaps things they haven’t.

Driverless taxis themselves are no longer new. Waymo has been carrying paying passengers without anyone behind the wheel for years, and what once looked experimental is now a way people get around parts of several American cities. Its vehicles, however, still retain much of the physical architecture of cars designed for people to drive.

On September 3, 2026, Tesla began offering rides in Cybercab in limited areas of Austin. Cybercab makes a different proposition visible. It has no steering wheel or pedals. There is no provision for the passenger to become the driver.

The following day, the National Highway Traffic Safety Administration opened an Audit Query into Tesla’s self-certification that Cybercab complies with applicable federal motor vehicle safety standards. The agency said the inquiry would examine the basis for certifying a vehicle that lacks traditional human controls.

But this is usually where the conversation about autonomous vehicle design stays. Can we trust them? Are they safe? What happens in an emergency? Will passengers understand what the car is doing?

They are important questions, and in 1854 they would have been important questions about elevators too.

Stay with the elevator for another thirty-six years, though, and something curious happens.

In 1890, the Pulitzer Building opened in New York. At 309 feet to the top of its dome, it was the tallest office building in the world. Its publishers promoted the new offices by asking prospective tenants how many valuable hours they had lost waiting for elevators elsewhere. In the Pulitzer Building, they were told, they would never have to wait a minute.

An elevator had become something people could complain was taking too long to arrive.

Thirty-six years earlier, Otis had stood above a crowd proving that an elevator would not fall when its rope was cut. Now New Yorkers were sufficiently accustomed to being carried through buildings that one of their concerns was having to stand around for a few minutes before the machine arrived.

The technology had created a problem that would have been rather difficult to discover through design research in 1854.

And elevators had already begun doing something stranger than that to buildings.

We will get to it later.

For now, there is a car sitting outside your office.

Parking is one of the largest potential urban consequences of autonomous vehicles, although the reason has less to do with parking technology than with why the car is there in the first place.

Suppose you drove to work this morning.

Your car is now three floors below you, under fluorescent lights, between two concrete pillars, doing absolutely nothing. It will remain there while you answer emails, have lunch and sit through meetings. Eight or nine hours later, you will return and ask it to move again.

There isn’t just one rectangle waiting for it.

One recent analysis, derived from USGS data on parking land, estimated about 1.31 billion parking spaces for 276 million registered vehicles in the United States. That is nearly five spaces for every vehicle.

Which sounds ridiculous. A car can only be parked in one place at a time.

But the city doesn’t know which place you will need.

Your car needs somewhere near your home. Another space near work. Perhaps another at the supermarket or restaurant. They are shared by thousands of people, but enough parking has to exist in enough places for driving to remain convenient.

In Los Angeles County, that redundancy becomes physical. Researchers have estimated about 3.3 parking spaces for every car. Together, parking occupies roughly 14 per cent of the county’s incorporated land, about 1.4 times the area used by roads.

And yet your car doesn’t actually need to be at work.

It is there because you drove it there.

A car with no driver can drop you at the office and leave. Perhaps it goes home, finds somewhere cheaper to wait or goes to charge. If it belongs to a fleet, it could be carrying someone else a few minutes later.

Autonomy separates the destination of the passenger from the destination of the vehicle.

What happens next depends on whose car it is.

Your own autonomous car might drop you at work, return home and come back to collect you that evening. The office has lost a parked car, but the city has gained two empty journeys. A shared robotaxi could instead leave you and go looking for its next passenger. A pooled service could carry people travelling in the same direction.

From the back seat, these all look like cars that drive themselves. From the street, they produce quite different cities.

Singapore is already exploring one end of that possibility. Autonomous shuttles in Punggol connect residential areas with MRT stations and neighbourhood amenities, complementing the public transport network. An autonomous vehicle doesn’t necessarily have to become a better taxi. It may also change what we mean by public transport.

Parking still doesn’t disappear when the driver does. Cars have to exist somewhere when nobody needs them. What changes is the old requirement that the car wait wherever the person who drove it happens to be.

And once the car doesn’t have to wait with you, another question becomes surprisingly important.

Where will you meet it?

Seattle encountered a version of this problem in 2019 without needing a single autonomous vehicle.

Late at night in Capitol Hill, thousands of people would leave bars, restaurants, clubs and theatres and request Uber or Lyft rides at roughly the same time. Cars converged on the neighbourhood, passengers tried to find them and the resulting congestion interfered with traffic, police and emergency access.

The scarce resource at midnight wasn’t somewhere to store all those cars. It was somewhere for cars and people to meet.

Seattle created four designated late-night pickup zones. People requesting rides within a geofenced area were directed towards these locations instead of having every vehicle collect every passenger from wherever they happened to be standing. A few months later, the city reported smoother traffic circulation, faster crowd dispersal and better police access.

Ride-hailing had changed the job of a small piece of street.

Autonomous fleets could make that distinction between storage space and exchange space much more important. An office may need fewer places for vehicles to sit for eight hours while needing more room for vehicles arriving and leaving throughout the day. Restaurants, schools, theatres and apartment buildings could face the same problem.

Then the passenger gets out.

Where does the empty car go?

A 2026 study modelled a hypothetical fleet of 1,700 shared autonomous vehicles making about 68,000 daily trips in San Francisco. When researchers prevented empty vehicles from waiting at the kerb between trips, daily vehicle travel increased by more than 200,000 kilometres, nearly 60 per cent.

The cars had to go somewhere.

That complicates the cheerful idea that autonomous vehicles will free cities from parking. Moving storage somewhere else creates traffic getting there. Preventing vehicles from waiting can encourage them to circulate. A fleet may require fewer vehicles overall while putting more kilometres on each one.

Autonomy doesn’t eliminate the question of where a car waits. It separates that question from where its passenger is.

Garages, office basements, parking lots and kerbs can all be reconsidered once the passenger and vehicle no longer need to remain in the same place.

And parking isn’t the only part of the city that assumed there would be a person behind the wheel.

Imagine you’re about to cross a narrow street and a car is approaching.

You look at the car. Perhaps it begins slowing early enough that you step forward. Perhaps the driver waves you across. Perhaps the vehicle creeps forward slightly and you decide to wait. Sometimes there is eye contact. Often there isn’t. You infer intention from speed, distance, position and small behaviours you have learnt to interpret without consciously naming them.

Now think about how much of a journey works this way.

A delivery van is blocking half a narrow street. One driver edges forward, another waits, and somehow they decide who goes first.

A cyclist looks over their shoulder and begins moving across the lane.

A construction worker turns a temporary sign and waves traffic through.

A police officer raises a hand even though the traffic light above the intersection is green.

A football rolls into the road and the driver begins slowing before the child chasing it appears.

None of these interactions requires the driver to consciously describe what they are doing. Much of driving is a continuous process of interpreting people, movement and exceptions, then communicating your own intentions back through speed, position, gesture and sometimes eye contact.

The human behind the windscreen wasn’t just controlling the car.

They were helping the car understand the city.

Autonomous vehicles have to replace at least some of that work. This has created an entire area of research into external human-machine interfaces, or eHMIs. Researchers and carmakers have experimented with displays, coloured lights, text, symbols and other signals intended to tell pedestrians that a vehicle has seen them or intends to yield. Beijing has taken autonomous driving beyond the vehicle, equipping 600 square kilometres of its demonstration zone with smart roadside infrastructure.

We removed the driver, then discovered that the driver had been doing something we hadn’t necessarily thought of as part of driving.

From inside the car, the person was the operator. From outside it, that same person was part of the interface between the vehicle and everyone around it.

The driver is part of the city’s interface.

Not all of this communication needs a technological replacement. Research on pedestrian behaviour suggests that vehicle movement, particularly speed and deceleration, can be especially important in people’s decisions about whether it is safe to cross. An autonomous vehicle may therefore communicate partly by behaving predictably rather than by growing a digital pair of eyes on its bonnet.

And a machine may eventually become better than a human at noticing some things. Cameras don’t get distracted by a phone. Sensors can look in several directions at once. Vehicles can potentially exchange information before the people inside them would even know there was something to communicate about.

But the change is larger than replacing a wave with a light strip.

The human driver has spent more than a century sitting between the car and the city, reading one and operating the other.

Remove that person and both sides may eventually change.

Which brings me back to the question I ask people we’re considering hiring.

What is the most important thing to think about when designing an autonomous vehicle?

Trust is still a very good answer.

Otis had to solve the problem of people believing an elevator would not fall before anyone could discover what elevators would do to buildings. If people don’t trust autonomous vehicles enough to use them, much of what we’ve just imagined may never matter either.

But trust is a question about whether the new thing can enter the world.

The stranger questions begin once it does.

We followed the missing driver into a parking basement and discovered that the passenger and vehicle no longer need the same destination. Follow that change farther and parking becomes pickup and the kerb changes jobs.

Then we looked outside the car and discovered something else. The person we removed wasn’t only driving. They had also been interpreting the city and communicating with it.

The steering wheel was only the most visible assumption.

There are others all over the city.

One is hanging above the next intersection.

Why does a traffic light need to be a light?

AV & City is a Sparklin research series exploring autonomous mobility through design, technology, behaviour and strategy, and what emerging technologies make possible for products, infrastructure and cities.