On a January morning in 2018, police in Leonia, New Jersey, began turning cars away from the town’s own streets. Leonia is a speck of a place — a mile and a half square, pressed against the New Jersey end of the George Washington Bridge — and for years, whenever the bridge approach seized up, navigation apps had quietly poured the overflow through its side streets. By the end of 2017, when the New York Times came to town, the situation had become one a research paper would later summarize in a single clause: people “could not get out of their driveways.” So Leonia did something municipalities are not supposed to have to do. As northjersey.com reported that January, the council restricted dozens of side streets to residents during rush hours, backed the ban with a $200 fine and handed locals bright yellow windshield stickers, the way a beach town issues parking permits. A town was issuing physical credentials so its own taxpayers could use their own roads, because a server somewhere else had reclassified those roads as a shortcut.
What is striking about Leonia is who it didn’t blame. Not the drivers — each one had made a reasonable choice, following the fastest route on offer. Not the bridge. The town blamed the directions, and it was right to, because what descended on it was a traffic plan nobody wrote. No planner proposed it, no council voted on it, no hearing was held. It emerged from millions of individually sensible decisions, coordinated by software that had never been asked to consider the place where those decisions landed.
Road networks are built on hierarchy. Arterials are engineered for throughput — wide lanes, timed signals, turn pockets. Local streets are engineered for access: the school bus stop, the basketball hoop, the driveway. What preserved the hierarchy was partly geometry and partly obscurity. Shortcuts existed, but exploiting them required local knowledge, and local knowledge was scarce. Navigation apps industrialized it. To a routing engine, the city is a weighted graph in which every public street is an edge with a cost, and a quiet residential grid reads as exactly one thing: underused capacity.
Everyone’s shortcut
The industry’s standard defense is that the apps make the whole network more efficient by spreading cars onto pavement that would otherwise sit idle. A 2024 simulation study of Florence, Milan and Rome — a preprint, built on modeling rather than field measurement — suggests the defense has an expiration date. At low adoption, routing apps really do trim emissions. But as usage grows, the routes begin to converge. The researchers call it conformity: traffic and CO2 concentrate on a shrinking share of roads, the emissions gains fade, and past a threshold they describe as city- and service-dependent, they can disappear altogether. The apps work best when few people have them. Their success turns everyone’s private shortcut into the same crowded street — the product’s growth cannibalizes the thing it sells.
That this can happen has been known, in the abstract, for half a century. In 1968 the German mathematician Dietrich Braess showed that adding a road to a network can slow everyone down at equilibrium; in 2002, the computer scientists Tim Roughgarden and Éva Tardos asked “how bad is selfish routing?” and answered that in a standard model, the worst case has everyone traveling a third longer than the best arrangement requires. This is mathematics about static networks — it convicts no app of causing any particular jam. What it dissolves is the comfortable assumption that handing every driver an optimal route adds up to an optimal city. The apps are running that experiment live, at metropolitan scale, and the spillover pools on streets that were never part of the objective.
Arguing with software
Look at what cities actually did, because the responses are the tell. In Los Angeles, after apps began sending drivers up Baxter Street in Echo Park — described in coverage as one of the steepest streets in the city — Councilman David Ryu introduced a motion in August 2018 asking Google, Waze and Apple to stop directing traffic there, and to coordinate with the city. A motion. A request. That same year Fremont, California, posted signs advising drivers not to trust their navigation apps — a city government buying signage to argue with software, curbside, in real time. These are the instruments available when the entity effectively directing traffic is not the entity responsible for the road. A city that wants one new stop sign holds hearings; a routing change that moves comparable volume is a server-side update, pushed silently and revised silently, leaving no public record that it happened at all.
There is a formal channel, of a kind. Waze — acquired by Google in 2013 — runs what it calls the Connected Citizens Program, since rebranded Waze for Cities: governments feed the company road-closure data, and the company feeds traffic data back. It is a genuine exchange. It is also governance conducted as account management — the documented substitute for public review of routing decisions is a partnership agreement with the company making them. Even Europe’s comparatively muscular platform law runs only partway there. The simulation study’s authors note that the major navigation providers qualify as Very Large Online Platforms under the Digital Services Act, which brings algorithmic-transparency duties. Transparency about the platform’s systems, though — not a say for a mayor in where the cars go.
Underneath all of this sits a shape worth recognizing, because it recurs. Every interface defines its constituency, and a navigation app’s constituency is the person holding the phone. The woman trying to back out of her driveway in Leonia is affected by the system without being a user of it. She has no account, no settings screen, no terms of service to decline; she enters the computation, if at all, as a property of the road. Individually rational choices, a collectively worse outcome, and the costs routed to whoever has no representation in the interface — that pattern extends well beyond traffic. Leonia just made it visible at eight in the morning on a Tuesday.
The yellow sticker deserves a second look. It was covered as a curiosity — plucky town outsmarts app — but it is really an artifact of a transfer of authority: a visa a town had to invent so its own residents could remain legible on streets the system had reassigned. Cities spent a century making decisions about roads slowly, clumsily, and in public. The apps make them instantly, elegantly, and in private, and the only visible trace is a sticker, a motion, a sign pleading with drivers not to believe their phones. Leonia’s ordinance reads now like the first draft of a right that still has no name: the right of the non-user to exist in systems built to route around her.