Driving Automation and Transit
Automation could kill transit as we know it, or make it even better
Picture a mid-sized North American city in 2035. Robotaxis were introduced here in 2030 and local deployment proceeded swiftly. Within a year, the service captured 15% of the transit system’s ridership, primarily younger, tech-savvy commuters and off-peak travelers frustrated by infrequent evening and weekend service. For these users, robotaxis offer something transit cannot: direct, door-to-door service at prices comparable to buses, available whenever needed.
The riders gain convenience, at the cost of ever-increasing congestion on the roads. But there’s another downside: the city’s transit agency, already operating with razor-thin margins, watches its ridership and fare revenue decline precipitously, while costs remain fixed. Faced with budget shortfalls, the agency does what it has to: it cuts service. Weekend frequencies drop. Evening routes disappear. The cuts make the system less useful, which drives more riders to robotaxis, which forces more cuts. Wealthier riders exit, and transit reverts to being a welfare service. Political support evaporates.
That’s one future: automation amplifying the inequalities of today’s mobility patterns. Robotaxis cannibalize transit ridership while transit agencies, slow to adapt, watch their systems deteriorate into skeletal services for those without other options. But there is an alternative. In that scenario, transit agencies embrace automation aggressively, transforming themselves from vehicle operators into mobility managers who deploy automated buses, shuttles, and yes, even robotaxis, as part of an integrated system that serves everyone better.
The first scenario is the default. It will happen unless we act deliberately to prevent it. The second scenario is possible, but requires transit agencies to move faster and more decisively than public institutions typically can. The choice between these futures will be made in the next decade, as robotaxi fleets begin to scale. Automation is going to change how we move, and in so doing, determine what kind of cities we live in. The question is whether we choose the city we want, or have circumstances dictate it to us.
Long-time readers of Changing Lanes will recognize many of the ideas that have shaped my thinking on this question. Over the past year, this newsletter has explored fragments of an argument about automation, transit, and urban futures. Today’s piece pulls those threads together, because they now exist in one place. This is the argument that I, and my co-authors Bern Grush and John Niles, make in The End of Driving, which publishes this week.
The book emerged from a shared conviction that the conversation about driving automation has focused relentlessly on technology while ignoring consequences. We wrote it as a blueprint for managing the transition rather than being overwhelmed by it. The End of Driving examines how automation is poised to reshape cities, transit, and everyday mobility, and how we can steer those changes toward the public good. Over the past year, this newsletter has explored how automation reshapes transit economics, why riders will abandon systems that fail to adapt, and which policy interventions actually work versus which merely sound good. The End of Driving synthesizes that work into a usable framework: not predictions about what will happen, but analysis of what different choices make possible, and what they foreclose.
Today in Changing Lanes, let’s distill one of the book’s most urgent themes: how automation could, and should, transform public transit. Automation can improve the transit we have. But it also permits us, if we choose, to build something new.
But before we consider the future, let’s assess where we are today.
On the plus side, driving automation has indisputably arrived. As of mid-2025, Waymo alone has served 10 million total trips, and provides more than 250,000 paid robotaxi rides weekly across San Francisco, Los Angeles, Phoenix, and Austin. It’s easy to forget, but this represents a 25X increase in just two years. Driver compensation—salaries, overtime, benefits, and health care—represents the majority of operating costs for taxi and ride-hail services, and approximately 60% of a transit agency’s total operating expenses. Automation offers a way to dramatically reduce these costs across all for-hire vehicle services. Those reductions offer a host of possibilities: for robotaxis, this creates a path to price trips competitively with transit fares; for public systems, it offers a way to stretch budgets and expand coverage. Depending on the market, a robotaxi with no driver may be able to offer all the convenience of a personal-vehicle trip—point-to-point travel in a private space—at costs approaching a transit fare, or even less for trips that cross agency boundaries.
On the minus side, the Endless Emergency continues. By mid-2025, North American transit ridership has largely recovered from the pandemic collapse, but not from its longer-term decline. Across the United States, riders took about four billion trips in the second quarter: still 18% below 2019 and 30% below the 2014 peak. Canadian systems have fared somewhat better at roughly 72% of pre-pandemic volume. Both remain below pre-COVID trajectories that were already declining, especially in large urban bus systems, which are 15%–20% short of 2019 levels.
What happens if these trends continue, and nothing else changes? Sadly, the scenario writes itself. The financial mathematics grow dire. As we have often discussed, North American transit agencies typically recover only 40% to 50% of their operating costs from fares, relying on public subsidy for the remainder. As ridership collapses, fare revenue follows, but costs don’t. The subsidy per rider doubles, then triples.
That’s the full nightmare scenario, where the centrifugal force of robotaxis pushes transit to the brink. But we would be irresponsible not to acknowledge the centripetal forces that will counteract them.
The Painful New Equilibrium
Urban transit won’t completely collapse, for reasons rooted in the physics of cities. This pattern is most pronounced in car-oriented North American metros; dense transit-first cities like London, Singapore, or Tokyo face different constraints, where capacity-limited streets and high-frequency, high-capacity urban rail systems will continue to outperform robotaxis. But in cities built around the automobile, the equilibrium looks different.
Firstly, congestion. A robotaxi fleet serving even a modest share of current transit ridership would flood streets with vehicles. Consider San Francisco, where BART and Muni together carry roughly 700,000 weekday trips. If robotaxis captured 30% of those trips, that’s 210,000 additional trips daily. Since a single bus carrying 30 passengers could stand in for about 25 robotaxi trips, replacing those buses with robotaxis would increase road-space demand by an order of magnitude.
That congestion would be intolerable, so cities will respond with congestion pricing or robotaxi restrictions.
Secondly, and in the same vein, cities will attempt to capture some robotaxi revenue to prop up failing transit systems. The policy logic seems straightforward: robotaxi operators use public roads, contribute to congestion, and compete with transit; therefore they should pay for the externalities they create. San Francisco already charges TNCs like Uber $3.25 per airport trip and $0.66 per trip elsewhere, generating roughly $30 million annually for transit.
The bitter truth is that both of these measures won’t restore transit ridership to previous levels. They’ll simply make both options worse. Transit service, already degraded by earlier budget cuts, won’t magically improve because robotaxis become more expensive. Riders will face a choice between expensive-but-convenient robotaxis and cheap-but-terrible transit. Some will choose robotaxis, as the low cost and privacy will make the extra time spent seem painless. But many will choose neither; the majority will simply make fewer trips. This deadweight loss—valuable economic and social interactions that don’t happen because transportation has become too costly or too time-consuming—will only increase.
What emerges from these countervailing forces isn’t collapse, but it certainly isn’t utopia. In this middle-ground scenario, transit ridership stabilizes at perhaps 60% of pre-robotaxi levels. Service frequency on all but the busiest corridors drops substantially: where buses once ran every 10 minutes, they now run every 20. Evening and weekend service becomes skeletal. Vehicle maintenance backlogs grow as deferred capital investment compounds. The rider experience deteriorates across every dimension except price, which remains low precisely because the service has become so degraded that few people are willing to pay more to use it. Meanwhile, downtown robotaxi trips during peak hours cost $15–$20, while off-peak trips in outer neighborhoods run $4–$6. Certain corridors ban robotaxis entirely during specified hours. Shared-ride requirements kick in when congestion exceeds designated thresholds.
The result is a system that’s simultaneously more expensive and less convenient than the robotaxi future we were promised, while also being more congested and less reliable than the transit system we once had. In this scenario, everyone, irrespective of their mode of choice, suffers enormous deadweight loss: trips that don’t happen because both available options have become too expensive, too slow, or too unreliable.
The middle-ground scenario delivers neither speed nor reliability in either bus transit or robotaxi options during peak periods, suggesting that the welfare loss exceeds what simple ridership numbers would indicate. Thankfully, transit in its own right-of-way—subways, some light rail, BRT—continues, and remains as reliable as before, absorbing ridership appropriately. But only for some cities, and some trips.
This is where we’re headed if current trends continue without deliberate intervention. Not the dystopia of complete transit collapse, but the deeper frustration of a system that limps along, satisfying no one, while everyone agrees something should change yet no consensus emerges on what. But the frustrating middle ground isn’t inevitable. The same automation technology that threatens to hollow out conventional transit also enables new approaches to public mobility: approaches that could deliver better service at lower cost.
That’s because automation will permit us to decouple the provision of public transit service from the ownership and operation of vehicles.
For a century, ‘public transit’ has meant government agencies buying buses and trains, hiring operators, and running scheduled service on fixed routes. That model made sense when human labor was cheap relative to capital costs and when coordinating large numbers of individual trips required either fixed schedules or expensive human dispatchers.
Neither assumption holds in an era of automated vehicles and real-time digital coordination. The opportunity exists to reimagine public transit not as a fleet of vehicles but as a guarantee of mobility: a commitment that everyone, regardless of income, can reach the destinations that matter for their lives. The question becomes: how might government ensure this guarantee while harnessing rather than fighting automation’s potential?
In The End of Driving, we explore two promising pathways forward. Both require uncomfortable changes. Both involve trade-offs and uncertainties. But both represent deliberate attempts to shape automation’s trajectory toward public benefit rather than watching helplessly as it unfolds according to purely private logic. The book suggests a reimagination of what “public transportation” means—moving from an ‘own and operate’ model to one of ‘”specify and regulate’. That shift unlocks two distinct pathways forward.
Pathway One: Transit Agencies Learning to Orchestrate
The first possibility we explore in The End of Driving is what we call “microtransit evolution”. To be clear up front: I believe automation’s primary value to transit lies in two spaces. These are:
Retaining high-capacity bus service, as we have it today, but in automated buses; and
Enabling agencies to replace low-performing feeder routes with flexible automated shuttles serving defined catchment areas around major stations and high-capacity corridors.
In practice, these automated shuttles would operate as the feeders they replaced did: by restricting themselves to the areas around major stations, offering short, pooled rides that bridge the gap between neighborhoods and mainline routes. Someone traveling from Subdivision A can share a vehicle with someone from Subdivision B when both are heading to the same station.
In my own view, this is not general-purpose microtransit serving arbitrary origin-destination pairs across entire service areas. That model fails for reasons I documented extensively in July, and automation doesn’t overcome the geometric constraints that doom it. My co-authors may interpret the pathways more broadly, seeing potential for agency-operated automated microtransit beyond this constrained feeder role. I’m skeptical, but the book’s framework accommodates both interpretations.
The problem with flexible transit service is geometric, not merely economic. I won’t rehearse the arguments I made already, beyond saying again, with Jarrett Walker, that “ridership is the death of flexible service”. Unlike fixed routes where additional passengers represent pure revenue gain, each new microtransit rider requires roughly proportional additional resources. No amount of algorithmic sophistication overcomes this.
Automation doesn’t solve the geometry problem, but it changes which geometries become economically viable.
General-purpose microtransit tries to serve arbitrary trips across an entire zone, from any address to any other address. This generates terrible pooling because origins and destinations scatter randomly. But feeder service has a constrained geometry: multiple origins converging on a single destination (the station).
Automation makes this model economically sustainable by eliminating the 60% of operating costs represented by driver wages, benefits, and associated overhead. Current paratransit costs agencies roughly $40 per trip. Automated feeders operating in 0.–2.0 kilometre catchments around major stations could deliver comparable service for $15-to-$20 per trip, and with ride pooling to get three to four passengers per trip, the per-passenger subsidy would be lower than the typical per-passenger per-trip subsidy for traditional fixed-route service.
So automation makes new things possible: feeder service without the cost penalty of human drivers. Vehicles sized to actual demand; one expects 4 to 8 passengers, not 40-seat buses running mostly empty. Service responsive to real-time demand patterns. Integration with fixed-route service that neither cannibalizes nor competes. This model doesn’t replace buses; it strengthens them by feeding more riders into the fixed-route spine. The transit agency orchestrates the entire network, feeding into their fixed-route services in their own rights-of-way, which—as per our thought experiment—are still enormously popular, because congestion is drowning any roads that are unrestricted.
Importantly, the flexible service remains subordinate to the fixed-route spine. Agencies must resist the political pressure to expand automated microtransit beyond feeder catchments. The failure mode is predictable: it works well connecting neighbourhoods to stations, generates enthusiasm, demand builds, politicians push to extend the zone, and suddenly you’re attempting general-purpose flexible service across the entire city. At that point you’re back to the geometry problem, burning subsidy at an unsustainable rate despite the lack of drivers.
Several compelling advantages emerge, though all must be understood within the constrained feeder model.
Firstly, it maintains public control. The agency owns vehicles and operations, ensuring direct accountability. This allows prioritizing equity over profit without depending on private sector cooperation. For transit systems with strong public service missions, this matters enormously.
Secondly, it evolves from current operations. Agencies already run paratransit services. Many already operate small-scale microtransit pilots. Automation improves what they already do rather than requiring wholesale reinvention. The institutional muscle memory exists; automation makes it economically sustainable in a broader (but still constrained) geography.
Thirdly, and importantly, it enables genuine transit-oriented development beyond the traditional walkshed. Breaking free from the 800-metre radius around stations means housing and jobs can be sited 1,500-to-2,000 metres from stations while still supporting car-free lifestyles.
But—critically—this only works if the automated feeders remain feeders. If they become general-purpose transportation, they’re competing with private robotaxis on convenience while trying to match them on cost, which they cannot do. Success requires transit leadership willing to say: “Automated microtransit is for feeders. City-wide service is for subsidized robotaxis—that’s Pathway Two.”
Pathway Two: Public Purpose Through Private Operation
The second pathway represents something genuinely unprecedented in North American transit: systematically leveraging private sector operations to achieve public goals. We call it “microsubsidies,” but it’s really about reimagining what government does in transportation.
The fundamental question is this: Must government operate vehicles to ensure equitable mobility? Or can it specify outcomes and pay for their delivery?
Here’s what becomes possible under this model. Private robotaxis serve public purposes. Government pays per trip that meets specified social goals. Operations are optimized through market competition. Equity is achieved through intelligent subsidy design rather than direct service provision.
This approach is newly viable because robotaxis will operate at scale regardless of what transit policy does. Their cost structure, without drivers, approaches genuine affordability. Digital systems enable trip-by-trip payments and real-time tracking. Government can steer behaviour precisely through incentive structures.
The mechanism works like this. Government specifies performance metrics and pays accordingly. Serve a designated mobility desert? Receive a $5 subsidy per trip. Provide off-peak service between 2200h and 0600h? That’s worth $3. Deploy a wheelchair-accessible vehicle? Claim an $8 subsidy. Guarantee pickup within 10 minutes? Earn $2 more. Generate excessive empty repositioning miles? Pay a $2 congestion fee per mile.
Consider a concrete example. A rider in a low-income neighbourhood requests a wheelchair-accessible vehicle at 2300h. The base fare they pay is $4. The operator receives the base fare plus $5 for the mobility desert, $3 for off-peak service, and $8 for wheelchair accessibility. Total operator revenue: $20. This is half the cost of current paratransit service at $40 per trip, yet it’s available to more people, at more times, in more places.
Why might this work when traditional transit struggles?
It leverages proven capability. Private robotaxis are already operating successfully. Companies are solving extraordinarily hard technical problems and iterating rapidly. Scale economics from serving multiple markets will make individual operations more efficient.
It avoids institutional risk. Government doesn’t own technology risk or bear fleet acquisition and maintenance burdens. Subsidy levels can adjust as costs change. Exit strategies exist if operators fail to perform. This is particularly attractive for municipalities and regions with limited capital budgets.
It enables rapid deployment. There are no procurement cycles for vehicles, no training programs for new systems. Operators bring existing infrastructure. Scale is limited only by subsidy budget, not by how fast an agency can acquire and deploy vehicles.
The objections are easy to predict, so let me address them directly.
This is just privatization! No. Government maintains control over outcomes through performance requirements, which ensure accountability. Consider an analogy: roads are public infrastructure, but in many jurisdictions, the government doesn’t own a fleet of steamrollers for maintenance nor snowplows for clearance, but hires these tasks out to private operators. Public purpose doesn’t require public operation.
Private companies won’t serve unprofitable areas! That’s what subsidies ensure, that operators must serve designated areas to participate in the system. Performance metrics make equity delivery profitable. Non-compliance means losing access to the entire contract.
We’ll lose transit jobs! Jobs transform rather than disappear. The shift is from vehicle operation to service orchestration—from driving buses to managing networks, analyzing data, and ensuring equity compliance. Oversight, planning, and data analysis positions grow substantially. Jobs are lost, but others are gained: the churn that typifies any robust economy.
The Window Is Open
The End of Driving doesn’t make a case for one of these paths over another. Different cities will make different choices based on institutional capacity, political culture, and existing infrastructure. What matters is that both pathways require action soon, not after robotaxis already dominate. The worst outcome is drift, that is, waiting until private robotaxis dominate and then scrambling to respond.
What determines the choice? Agencies with strong institutional capacity, political will for public operation, available capital for fleet acquisition, and union relations that support automation transition should lean toward the orchestration pathway. Their priority is maintaining direct public control.
Cities facing institutional constraints, prioritizing faster deployment, seeking to avoid technology risk, or possessing limited capital but strong regulatory capacity should consider microsubsidies. The emphasis shifts to outcomes rather than operations.
In either case, the opportunities are substantial: universal affordable mobility, at scale, in cities where parking loses its salience. The requirements are clear: understanding the possibilities, regulatory frameworks that steer outcomes toward public benefit, and political leadership willing to act.
What should transit professionals do? Well, first, read The End of Driving to engage with the full argument rather than just the summaries I’ve provided here at Changing Lanes. Assess your agency’s institutional capacity honestly. Choose which pathway fits your context, and start building capability now. Above all, don’t wait for perfect clarity: the window of opportunity to act rather than react won’t remain open indefinitely.
For a year, this newsletter has explored pieces of this transformation—microtransit economics, ridership behaviour, the challenges of shared mobility. The book assembles them into a comprehensive argument about automation, transit, and urban futures. The synthesis matters, because the choices we make now will shape cities for generations. The question has never been whether automation will change transit, for that was never in doubt. The matter before us remains how deliberate policy can decide what kind of change automation will bring.
To learn more, please read The End of Driving, available now.
Respect to , , and for feedback on earlier drafts.
Exciting! To me the best part is that in any of these scenarios, the salience of private parking decreases relative to today.
What do you think about the role of micromobility in these two visions?