However, as density increases, the supporting infrastructure also can get significantly more expensive to create. Case in point: NYC's Second Avenue subway line that's been in the works for somewhere between a century and a few decades, depending upon how you count, and will end up costing something like $20 billion.

Before railroads, the Erie Canal was one of the few ways to ship bulk goods over the Appalachian divide. This caused the mouth of the Hudson to be a center of trade, particularly shipping goods to Europe. To make money on the return trip, they would put passengers in steerage and so lots of immigrants landed in New York City.

At least in London: Every employer moves to London because that's where the employees are; and every employee moves to London because that's where the employers are.

Of course I don't mean /literally/ every employer or employee. But I've certainly been offered jobs in remote locations and turned them down because the locations only had one employer who needed someone with my skill set.

Because a service and investment economy is based on proximity to those with money as opposed to proximity to factories or resources.

Not Manhattan, or most other dense cities. There is simply not enough room for every human being who wants to visit to have 50 square feet of steel, rubber, and plastic around them.

Leaving aside timeframes (which I'd argue for cities are decades out) for purposes of discussion, Manhattan is perhaps notable in how relatively little I'd expect autonomous vehicles to affect it. After all, it already has vast fleets of self-driving cars. They just have (admittedly often rather flawed) organic computers in control.

I'd be much less surprised to see impacts in more distributed areas where I could easily imagine people becoming much more tolerant of very long commutes. Especially if, at the same time, there's a greater overall acceptance of sometimes-remote work in some professions, it's not hard to imagine people doing a 2-3 hour commute a couple of days a week.