Most transit stories are big-city stories: New York’s subway, London’s congestion charge, the trains of Tokyo and Paris. The places where most Americans actually live, the cities of 50,000 or 200,000 or 400,000, rarely make the conversation. They are too spread out for rail and too car-dependent for fixed-route buses.
Arlington, TX is the exception.
Arlington spent twenty-five years saying no to public transit. Three times it asked voters to raise the sales tax and join one of the Dallas-Fort Worth region’s transit authorities, and three times the answer came back no.
Today every one of Arlington’s roughly 400,000 residents can hail a publicly funded ride within a block or two of their door, across all 99 square miles, for about $8 million a year. Joining the regional authority would have cost $40 million. It is a different theory of what transit can be.
I sat down with Dan Ramot, CEO of Via, and Alicia Winkelblech, Arlington’s Director of Transportation, who lived the transition firsthand. (Via now serves more than 800 cities across 30-plus countries, running microtransit, paratransit, school-bus, planning, and autonomous-vehicle networks, and went public on the NYSE in 2025.)
Today’s letter walks Arlington’s route from the inside and discusses the four moves and one discipline any mid-size city can copy: 1. Procure for the Problem, Not the Product; 2. De-Risk the Launch; 3. Match the Mode to the Density; 4. Contract for Control, Not Ownership; and, underneath all four, Don’t Strand the Riders You Already Have. Then it turns to what AI is doing in transit, from network design to the control room.
Listen to our conversation here:
Procure for the Problem, Not the Product
Arlington’s first good decision was to refuse to repeat its old one. Having lost three votes built on buses and rail, it did not draft a fourth.
It spent a year with stakeholders (traditional transit providers, the regional planning agency, consultants, even Uber and Lyft) writing a framework before it built a service. The framework named what the city wanted: cost-effective, flexible, responsive to demand that shifts week to week.
What it did not name was a specific technology, so the city would not be locked into any predefined tools. Only then did Arlington use Federal Transit Administration funding, issue an RFP against that framework, and land on Via. The technology followed the goal.
That posture also protects you from vendor lock-in. A procurement loaded with requirements all but invites a vendor to over-promise, to say yes to everything and deliver little. Ramot’s discipline is the opposite: under-claim, over-deliver, treat it as a shared problem. The test is concrete: will the vendor still work hard once the contract is signed, or only to win it?
De-Risk the Launch So the Public Will Tolerate It
A transportation director’s real job is managing political and financial risk, often more than cost or tech. For Arlington, microtransit was close to risk-free: no permanent infrastructure, no hard capital to defend if it failed.
The city piloted small, saw near-immediate interest, expanded, and only reached all ninety-nine square miles in 2020, three years after the 2017 start. If the community decided it wasn’t working, it could simply go away, and that reversibility is what made it politically absorbable.
Unlike a rail line you can’t unwind, a piloted zone you can grow, shrink, or end. For a public that has said no three times, “we’ll try it small, and you can kill it” is the pitch.
The reversibility has one limit. The city is legally required to serve the paratransit, Handitran, for disabled riders. Arlington moved it onto Via’s software only in 2023, after years of proven trust, and saved it for last on purpose. Guaranteed ADA service is a commitment; you cannot put it on a pilot you might cancel.
Match the Mode to the Density
Arlington’s success invites overreach, so the discipline matters: on-demand vans scale, but only to the right city. Two of Via’s own deployments make the point.
In Wilson, North Carolina (50,000 people, sparse), Via helped the city eliminate the fixed-route bus system entirely and replace it with microtransit, more than doubling ridership on the same budget.
In Sioux Falls, South Dakota (200,000 in the city), removing fixed-route buses would have been a blunder. Enough corridors carry enough demand to keep larger buses on fixed routes, so the city ran fewer, straighter bus routes more often, layered microtransit on top, and grew ridership by nearly fifty percent in under two years.
Carry this into your planning room: microtransit is not the opposite of mass transit but a form of it, and the real question is the best service quality you can buy per dollar, at your density. Manhattan is a subway problem. Much of the US is not dense enough for heavy rail and never will be; some of it wants buses, some is better served by vans.
There is also a floor: if a van carries one rider at a time, you have built a taxi, not transit. Put two to four riders in each van and it is real transit, worth the public money.
Contract for Control, Not Ownership
Every transportation director who outsources operations hears the same objection, and Alicia heard it: this is a private business, so why not just join a transit authority?
The cost line does most of the work (forty million a year for two bus lines and maybe one rail stop, against eight or nine million for citywide microtransit coverage), but the principled answer matters as much. Outsourced operation is the global norm, not a compromise: a large share of US bus service is contracted out, Transport for London runs its network through seven main operators, most of France’s urban transit is contracted to operators such as Keolis and Transdev, and Boston’s own commuter rail is run by Keolis.
What makes a system public is not who provides the service but who sets the strategy and price, monitors the service quality, and owns the data. Arlington's service carries the city's name, and, as Winkelblech put it, the city runs it hands-on; it doesn't just hire a contractor and step back.
On the data specifically, Via’s position is that the city owns it. Make that clause explicit.
Don’t Strand the Riders You Already Have
If a rider builds her morning around the 7:10AM bus on the old Route 22 and finds it gone, you have not modernized anything; you have severed a lifeline.
Routing the new vans is the easy half; keeping faith with the riders who already depend on the service is the hard half. So before removing any existing route, the city should stand up equal-or-better coverage for those same trips first, run the old and new service in parallel through the switch, keep a phone booking option for riders without smartphones, and hold a wait-time standard for transit-dependent trips.
Ridership grows only when coverage expands. In Wilson the old buses served only the streets their routes ran on; the vans now reach nearly the whole city, a short walk and a roughly fifteen-minute pickup away.
The rest is unglamorous but important: months of in-person community engagement, and a call center so residents without smartphones are not locked out. In Muskegon, Michigan, half of surveyed riders said the service helped them get to work; in West Sacramento, Via saw residents’ view of local government improve.
What AI Is Actually Doing
On AI, Ramot drew a line between the AI dust kind and the transformative kind. The genuinely useful AI shows up in everyday operations you already know. Three are already working: designing the optimal network (where fixed route wins, where demand-response does) from data pooled across hundreds of cities; AI agents handling call-center reservations and cancellations; and the matching engine pairing riders to vehicles.
The deeper change is in the control room. A dispatcher today watches everything at once (breakdowns, no-shows, late drivers, traffic) and reacts under pressure. Newer models, reliable only since late last year, can run much of that in the background and surface only what needs a human, turning the dispatcher into a supervisor of agents rather than a person drowning in screens.
There is a corollary for any agency with a graying workforce. The tacit expertise of veteran controllers is lost at retirement. Capturing it before it walks out the door is exactly what these systems can help with, a problem my students are working on inside the Chicago Transit Authority.
Via is making a broader bet with its new AI Labs: that the same approach can help cities with problems beyond transit. What makes it possible now is cost. Fitting software to one city's quirks used to take years and an army of engineers, so it is hard for tech companies to serve municipal government. An AI agent can absorb those quirks far faster and carry them to the next function and the next city. The same holds inside every agency: those that adopt these tools move ahead quickly, and government, long last in line for technology, finally has a chance to catch up.
The Actions for Your City
Stop arguing about modes as identities.
A city is not a “bus city” or a “rail city” or a “microtransit city.” It is a set of riders at a given density, to be served at the best service per dollar you can manage, sometimes with a train, often with a bus, increasingly with a van, and perhaps with all three coordinated.
Arlington was told it wasn’t built for transit. It just needed a different kind of transit.
One metric settles the microtransit argument: the best service quality you can buy per dollar, at your density.
Your playbook is four moves and one discipline: procure for a result, not a vehicle; pilot so you can reverse course; match the mode to your density; contract to keep strategy and data in your hands; and never strand the riders who already depend on the service.
–Jinhua
Listen to our conversation here:
Jinhua Zhao
Professor of Cities and Transportation, Massachusetts Institute of Technology
Connect with me on LinkedIn I Newsletter
You're receiving this because you subscribed to Jinhua Zhao's Mobility Forum Newsletter
Unsubscribe | Update Preferences
