getting around

A Postcard From Driverless San Francisco

Unexplained stops. Incensed firefighters. Cars named Oregano. The robotaxis are officially here.

Photo-Illustration: Curbed; Photos: Getty
Photo-Illustration: Curbed; Photos: Getty

This article was featured in One Great Story, New York’s reading recommendation newsletter. Sign up here to get it nightly.

One morning in late August, 102 minutes before dawn, I stood on a deserted sidewalk waiting to be picked up by an autonomous vehicle. The car was operated by the company Cruise, which at the time offered public taxi services only from 9 p.m. to 5:30 a.m. Daylight driving was available exclusively to employees, friends and family, and a mysterious group of cyberinfluencers called “Power Users.” The previous morning, I let my shower go long and missed the ride-hailing cutoff. Today, I took no chances and set my alarm for 4:30 a.m.

There are few places to go before five in the morning in San Francisco. When I opened the app, there was exactly one breakfast option open in my neighborhood: 7-Eleven. A cartoon map materialized on my screen with a few lonely cars roaming the cityscape. One swerved in my direction. “Pandan is on the way,” my phone declared. Minutes later, Pandan turned onto my block and pulled to a confident stop in front of the wrong building.

Though this was my first time hailing a Cruise, self-driving cars have been stalking my home for years. I live in Richmond, a sleepy residential neighborhood on the west side of San Francisco that became one of the first testing grounds for autonomous vehicles. Since at least 2021, these machines have prowled the streets after dark, slyly circling the same set of blocks over and over like alien scouts preparing for the invasion.

My first few encounters with the technology came at odd hours. During some midnight stroll or early-morning jaunt, I would notice an odd car gliding down the empty street with an employee scribbling notes in the passenger seat while the unmanned steering wheel hung a sharp turn. Earlier this year, returning from the gym around midnight, I was preparing to jaywalk when a self-driving vehicle pulled up at the intersection. I turned to make eye contact with the driver. My eyes drilled forward, searching for someone to meet my gaze, but the cabin was empty. That was the first robotaxi I had ever met without a safety driver. After a few seconds in an uncanny staring contest, I scuttled down the block to jaywalk at a less convenient spot.

Driverless cars arrived at a peculiar moment in the San Francisco psyche. During the boom years of the 2010s, downtown swelled with new high-salaried jobs that buoyed a wave of boutique, cottage industries from $10 toast to meat-alternative jerky. The pandemic popped this bubble. The exodus of the tech class was startling and quick: By the start of this year, downtown had lost nearly 150,000 jobs and office vacancy rates skyrocketed from 4 percent in 2019 to 32 percent. The term doom loop metastasized as the official diagnosis for the city’s downtown decline. But what economists call doom, many residents like myself see as recovery. San Francisco felt like a city in recuperation after a decadelong addiction to easy tech dollars.

At this delicate moment, just when the town appeared ready to wean itself off unhealthy relationships with experimental technologies, the AV industry decided to make San Francisco ground zero of the robotaxi economy. After years of tests, on August 10, 2023, the California Public Utilities Commission passed controversial resolutions that turned San Francisco into the first city on earth with two AV companies offering paid robotaxi services 24/7. The votes allowed Cruise (owned by General Motors) and Waymo (a subsidiary of Google’s parent company, Alphabet) to charge for trips and offer rides anytime of day going anywhere in the city. (Cruise, however, limited its 24-hour service to select users.) This decision ushered in a period of regulatory limbo with the technology’s fate in a tug-of-war between corporate lobbyists and government officials. On August 17, a Cruise AV collided with a fire engine, then the next day the California DMV requested Cruise cut its San Francisco fleet in half. Meanwhile, citizens viewed the cars with a mixture of distrust and curiosity; according to Waymo, more than 100,000 people in the Bay Area signed up for the wait list. In the weeks that followed, I rode with Waymo and Cruise during their tumultuous debut in public life to see how this supposedly world-changing technology might fit into the life of one median-salary San Franciscan without a driver’s license.

The journey to 7-Eleven was only half a mile down a single road. But rather than taking the direct path, Pandan concocted a strange, circuitous route — tracing an itinerary that snaked through side streets and residential blocks like we were shaking an FBI tail.

The first seizure struck on Tenth Avenue. Without warning, Pandan froze on the pavement. The tires sputtered. The steering wheel quivered. Yet the car did not move. Panicked doubts raced through my mind before I identified the source of the problem — a police car patrolling across the street. Once the cruiser moved out of sight, Pandan regained its nerve and continued forward.

It took two blocks until the next seizure. As Pandan turned onto Anza Street, I saw bright-red and blue sirens flashing in the distance. Two San Francisco Police Department vans were parked in the middle of the narrow road. Three officers stood on the sidewalk. We appeared to have entered the scene of an active police investigation. With a wobble, I felt Pandan grind to a second unplanned stop.

For what felt like a minute, we did not move. The car idled in place as the cops stared impatiently from the curb. In response, Pandan’s steering wheel began to tremble, twitching a millimeter to the left and then the right. The machine seemed to be making a decision. At such moments, it is hard to resist the urge to anthropomorphize. Pandan seems pretty nervous, I worried. And things don’t always go well when skittish drivers get nervous around the cops at 4:30 in the morning.

Then the fit cleared. Like nothing happened, Pandan gently swooped around the police barricade, and we were back on the road to breakfast. Moments later, the bright green of the 7-Eleven logo came into sight, and my pulse steadied. With our destination only a crosswalk away, a third seizure hit. Pandan began to convulse and fritz. The same message flashed on the screen: “Arriving Soon,” “Arriving Soon,” “Arriving Soon.” The cabin filled with the sharp sounds of machines at work. Then the screen lit up with a new announcement: “We Have Arrived.”

My first ride turned out to be a prescient introduction to the impact of driverless vehicles on the city. Throughout Cruise and Waymo’s first months of operation in San Francisco, AVs starred in a string of high-profile incidents: viral stories of cars freezing traffic, frustrating firefighters, and inciting protesters. Over this period, the fate of the two companies diverged. Waymo steadily expanded its fleet, earning a respectable safety track record, while Cruise steadily imploded, losing both its CEO and its license. On October 24, the California DMV suspended Cruise’s permit to operate driverless cars in the state, declaring the company’s vehicles “not safe for public operation” and alleging that the company withheld video of a harrowing hit-and-run accident. (Cruise has disputed that it withheld any footage.) According to an internal email from the current president, the company is now on a mission to rebuild trust, which could be a long journey.

The original plan for this article was to replace my normal workplace transportation with AVs for a couple of weeks. That idea died the first time I pulled out Waymo to book a ride for work. After typing in my meeting’s address, the app’s suggestion was startling: My wait time would be 19 minutes, and my pickup would be a quarter-mile away on the side of California Highway 1. As for Cruise, my job sadly held no meetings before dawn in the entire month of September. In fact, my life circumstances seemed uniquely positioned to undermine driverless transportation. My office is located in such a bewildering location that no autonomous car will even attempt to pick me up. So my robo-ride-hailing experience was split between early-morning breakfast expeditions and weekend excursions to meet geographically convenient friends.

My immediate impression was that the driverless cars are excruciatingly slow. With outrageous delicacy, my robotaxis drove through the streets like we were navigating an active minefield, turning at a glacial pace and cautiously inching over each speed bump. In this way, AVs invert the traditional symbolism of cars: While a Bugatti is an avatar of speed, Cruise and Waymo are symbols of slow. I wondered if these companies were rebranding slowness as safety — positioning the cars’ sluggish mph as signs of a deep corporate concern for public welfare. Perhaps AVs will be paradigms of consumer efficiency in the future, but today, the passenger must make sacrifices for the car’s convenience. Thus Waymo and Cruise flip the typical customer-comes-first capitalist logic.

The incarnation of this inconvenience is the seizure, which could be the defining experience of taking driverless cars. Over the first six months of this year, Cruise reported its cars faced 177 incidents of freezing that required staff to retrieve vehicles, while Waymo claimed its fleet had seen 58 unplanned stops while carrying passengers. These pauses could range from seconds to minutes, and often they hit during quite ordinary situations: encountering a flashing light or a construction sign. As you, the passenger, sit frozen in the back cabin, you feel the firepower of billions of dollars in hardware research kick into action — dozens of cameras and radars working furiously to comprehend some enigma of the road. What is this orange entity blocking the curb? If only I could tell the car it’s just a traffic cone.

These seizures switch from inconveniences to risks when they impede emergency responders. Robotaxis have racked up a notorious reputation with the San Francisco Fire Department. From January to August 2023, the SFFD filed more than 50 reports about “autonomous vehicle incidents,” in which driverless cars made an unwelcome appearance. On January 24, a Cruise car suddenly froze on top of the hose line a battalion was using to extinguish a sidewalk flare-up. On May 4, a driverless vehicle blockaded Station 36, preventing the squadron from responding to a nearby dumpster blaze, while a few months later, at the Legion of Honor, a Waymo decided to conveniently stop in between a fire engine and a burning vehicle. “If a Waymo detects that a police or emergency vehicle is behind it with flashing lights,” explains a company web page, “it is designed to pull the vehicle over and stop when it finds a safe place.” This coding quirk — instructing AVs to pull over at flashing lights — ironically can push driverless cars into emergencies rather than out of them. (Cruise and Waymo claim that most interactions with emergency vehicles go smoothly.) Unfortunately, these frozen vehicles can be difficult to vacate from emergency scenes; in the case of the Legion of Honor interloper, a firefighter ultimately had to contact Waymo so he could drive the car away.

“My folks have a job to do, and it never included moving an autonomous vehicle,” explained San Francisco fire chief Jeanine Nicholson in a phone interview. Her department, she says, ran into more problems with Cruise than Waymo, which initially operated a smaller fleet in the city. Cruise “never once really asked ‘How do we work with you and around you to not impact public safety?’” Before Cruise released its cars in the city, Chief Nicholson told me, the company consulted with the fire department just one time: to film a training video for firefighters about where to cut people out of a crushed vehicle in case of an accident. Beyond this, Cruise gave the fire department a phone number to dial when AVs disrupt emergency scenes. (Cruise says it “has been in regular contact with SFFD and SF first responders, including before we launched our public service.”) The Cruise Critical Response Line connects first responders to an “escalation team” that can help with tasks such as unlocking the car door or dispatching an additional field-support unit. In August, the New York Times reported that it took, on average, ten minutes for someone from Waymo to come retrieve the car and 14 minutes for someone from Cruise. Some battalion members, understandably, did not want to wait 14 minutes for help in the middle of a fire and chose instead to smash the car window in a frustrated attempt to coax an AV to leave an emergency scene. But often firefighters — like the passengers inside — simply had to wait for the confused, autonomous machine to muster the moxie to act.

Most rides go more smoothly: The majority of my robotaxi trips proceeded without serious interruption. These journeys felt like traveling in a bubble of luxurious isolation, insulated from the pressures of the outside world where the cushions, climate, and soundtrack never need change from the same palette of monotone affluence. Only a third of my AV outings were disrupted by an unexplained pause, yet the possibility of a freezing incident was a constant psychological undercurrent.

After a few days of riding, the passenger is bothered by the questions you cannot ask: Why drop me off here? Why take this route? Why did we turn into the middle of an intersection on a red light? But in autonomous cars, communication is a one-way street. While huge amounts of funding have poured into developing technology to help AVs see, the cars’ limited audio equipment seems more focused on surveilling customers than conversing with them. In practice, driverless cars often felt explicitly designed not to hear you.

If an incident arises, customers can talk to a rider-support representative who mainly seems to offer the solace of a companion in your confusion. My second Waymo trip took 28 minutes and two support calls to get through rush hour. While grinding in bumper-to-bumper traffic in Golden Gate Park, the Waymo hesitated for a couple seconds beyond comfort. As the car started moving again, the tablet screen presented an update: “Connecting to Rider Support. Connecting to Rider Support.”

“Hello! What seems to be the problem?” asked a voice from beyond.

“Is there a problem?” I replied.

“We’re just checking what happened.”

“Something happened?”

“It temporarily paused.”

“Why did it pause?” I asked.

“Actually, there are a lot of reasons for that, a lot of factors …” The call continued in this loop for several minutes as the representative, Tony, explained these pauses could be triggered by anything from long red lights to pedestrians to a “system modification for security check” and he couldn’t be sure what caused my “incident.” Throughout the conversation, Tony continually found a way to use the word “specific” to describe the unknown.

By the time the call ended, we had nearly reached my apartment. As the Waymo turned onto my street, the vehicle zeroed in on an open spot. When the screen status changed to “Pulled Over,” I took off my seat belt. Then the tablet’s tone took a shift. The hazard lights began to flash. A new icon popped up on screen: “Doors Locked.”

My seat-belt chauvinism had seemed to short circuit the communication system. A cacophony of conflicting audio cues filled the cabin. Commands from the voice-over — “You are here. Please take your belongings” — contradicted the tablet instructions, “Please Buckle Your Seat Belt.” Then a countdown appeared. “If You’re Having Trouble, Rider Support Will Call in 1 Sec.” Before I knew it, I was back on the phone with rider support.

“What seems to be the problem?” the representative asked.

“It paused in front of my apartment,” I answered as the voice-over thundered behind me, “You are here. Please take your belongings.”

“What is your problem?” the voice repeated louder.

“It’s been hovering around my apartment for a minute and …” I started to explain before suddenly asking, “Can I just leave?”

“Yes, you are absolutely free to exit the vehicle,” the voice replied. “But please do mind the oncoming traffic. Okay? Left and right —”

I did not wait for the rest of the answer before shuffling out of the apoplectic vehicle. I ducked into my building, dodging the judgmental glares of two construction workers on the curb staring at my car’s embarrassing display.

The AV aesthetic might be called totalitarian twee. Cartoon infographics in the Waymo app explain the circumstances when your data might be shared with law enforcement. The Cruise marketing team members read their Orwell and decided to counter any dystopian associations by branding their robots with sans-serif fonts, can-do corporate lingo, and the company’s signature shade of orange. The interior signage felt like an uneasy divorce between the compliance and design departments. “Thanks for Sitting in the Back” read a notice printed on the back seat in large letters, while a tiny footnote below clarifies: “Passengers are not allowed in the front seats.” The attitude seemed to be that Big Brother had the right ideas but needed a better brand manager.

Waymo, on the other hand, favors an antiseptic elegance in its brand identity. While Cruise used the accessible Chevy Bolt for its San Francisco AV fleet, Waymo uses the upscale Jaguar I-PACE as the model for its flagship robotaxi. When the rider clicks the “Open Door” button as a Waymo arrives, a handle emerges from the smooth surface of the car frame. Once inside, a gentle robotic voice greets you by name and reassures, “We’ll do all the driving, so please don’t touch the steering wheel or pedals.” During the opening interregnum, the car is flooded with celestial music that evokes the sound of glowing light in Hollywood movies during angelic visits. Underneath the handsome silver Jaguar logo on the steering wheel, an ominous message affirms, “The Waymo Driver Is in Control at All Times.”

The prime example of totalitarian twee is Cruise’s naming strategy. “Every Cruise all-electric self-driving vehicle has a unique name,” states Cruise’s website for a digital campaign, which invited the public to nominate monikers for new cars following a few guidelines: nicknames must be “delightful and fun” and under 14 characters. Food seems to be the most popular inspiration. Across the city, you might see Oregano, Kombucha, Kebab, Potsticker, Cheddar, or Tostada roaming the streets. The culinary whimsy starts to fade, however, as these vehicles collide with real-world problems. After Cruise’s public launch, I would regularly encounter surreal social-media posts showing slip-ups like, “Toffee got stuck in the middle of Steiner Street”; “Starfruit is blocking an intersection at Dolores”; “Lettuce just drove through a crosswalk of pedestrians.”

The lingering sense of menace in the headlines erupted to the surface in a grisly incident on October 2. Around 9:30 p.m., a woman trying to cross at 5th and Market on foot during a red light was hit by a human driver. The collision sent the individual careening into the path of a Cruise vehicle that bore the name Panini. Panini came to a quick stop, accidentally pinning the woman’s leg underneath its rear tire. Not realizing the weight under its wheels was a person, Panini decided to pull over and dragged the injured woman — already the victim of two collisions in one minute — across the pavement for 20 feet before finding a clear spot outside the lane of traffic. Why was Panini compelled to pull over? A safety maneuver called achieving “minimal-risk condition.” In other words, this catastrophe resulted, in part, from a safety protocol.

“As part of our safety-review process,” a statement from Cruise about the accident said, “we perform simulations to test our AV behavior compared to human drivers. In this case, the simulations performed afterward showed that had it been a Cruise AV rather than the human driver, the AV would have detected and avoided the pedestrian.” (Cruise later issued a software update to “address circumstances in which the Cruise Collision Detection Subsystem may cause the Cruise AV to attempt to pull over out of traffic instead of remaining stationary when a pullover is not the desired post-collision response.”) Such language did little to assuage the DMV, which ultimately suspended Cruise’s permit to operate driverless robotaxis in California.

Amid the many heart-wrenching facts of the accident, the one detail that continues to puzzle me is the name Panini. This nickname is the emblem of a deeper misunderstanding — not just an accidental ignorance but an almost deliberate failure to comprehend basic human behavior. Choosing to name a potentially erratic car for a pressed sandwich is either arrogant or incredibly oblivious. The moniker suggests a failure of imagination — an inability to consider how these cars would mesh with the chaos of city life, to imagine these machines as part of a society and not just an economy.

When you exit an autonomous vehicle, you can feel the sidewalk’s eyes on you. As my destination neared at the end of a ride, anxiety would bubble up. I could picture the look on the pedestrian faces. Here was my hulking, roaring techno-taxi rolling through a nice, quiet neighborhood like a space colonizer on holiday.

In San Francisco, the arrival of AVs has inspired a new subgenre of social-media journalism: a sort of rubbernecking for robotaxis, where citizens gleefully chronicle the chaos created by driverless cars. Since August, TikTok and Twitter have been flush with videos of traffic high jinks created by AV glitches. In one incident, a Waymo froze in a traffic lane for 15 minutes after a cyclist smacked a large “No Techno Fascists” sign on the windshield, while another high-profile case saw a masked vandal caught on tape smashing a Cruise’s sensors with a hammer or a pickax.

One night, I stumbled on a video of a viral incident. A stranded Cruise named Bolero sat haplessly in the middle of a bus lane on Stockton and O’Farrell while gawking crowds filmed from the sidewalk. “What the hell you doing? Get up?” screamed a furiously honking bus driver stuck behind the robotaxi. In frustration, one backpacked vigilante stormed off the bus and pounded on Bolero’s window. Then, amidst this pandemonium, Bolero’s frozen wheels started to spin. A gasp of titanic relief spread through the crowd. Straightening out, Bolero lurched forward—only to drive a few feet before freezing again. Derisive laughs rained down as the clip ended.

There’s a strange inversion at play: normally, the driver is the responsible party. Yet with no bad driver to blame, the passenger becomes the face of guilt by default. In these confrontations, the rider becomes the accidental human face of corporate irresponsibility.

Over the weeks of my robotaxi experiment, a curious pattern emerged. I found myself making excuses not to ride. Running too late for Waymo. This location is out of the way. It’s cruel to ask any driver to navigate Golden Gate on a Sunday. Convenience, however, was not the only motivation. Before ordering a car, I would do some mental math to double-check no friends lived nearby. As the weeks went on, I gradually reverted to nocturnal driving. Despite Waymo’s 24-hour availability, the window when I was comfortable riding became narrower and narrower. After a week without ordering a single AV, I could name the queasy feeling skulking around my stomach: I was embarrassed.

It is hard to say anything definite about what my personal shame reveals regarding an impersonal technology. In fact, it is hard to say anything definite at all about the future of the constantly shifting AV industry. I was expecting to write a jaunty sketch about going to work in an orange robocar named Kombucha, yet by the end, Cruise’s driverless taxis were banned from the roads of California and Kyle Vogt had resigned as the company’s CEO. Yet as I try to envision the future of robotaxi technology, my mind keeps coming back to my second ride with Cruise.

The time is 5:10 a.m., a few minutes after Pandan dropped me off, and I am standing on the curb of Geary Boulevard gulping down my makeshift breakfast. The best meal I could muster at 7-Eleven turned out to be a can of sugar-free Red Bull and a novelty sweet called 7-Days Soft Croissant proudly marketed as a “Product of Bulgaria.” The next stop on my commute was the office. Moments after hailing a ride, my phone lit up with a message: “Pot Roast Is 3 Minutes Away.”

My drop-off was fixed for 1802 Hays Street, and I walked a block to find Pot Roast already waiting for me in a parking spot on the corner of 14th Avenue. I buckled my seat belt and hit “Begin Ride.” The map showed our “Trip Plan.” Something, however, had changed.

My drop-off no longer appeared as my office. A new destination had been selected for me — an address that I did not recognize. There must be some mistake? I thought. Then Pot Roast’s planned route unfurled across the screen — as a blue line wrapping around the block to drop me off on the opposite curb of our current street corner. The route was a circle.

Alas, it was not an easy trip. For two agonizing minutes, Pot Roast carefully plodded around the block. When we eventually reached the drop-off, Pot Roast hesitated. There was a problem: a makeshift construction sign stood by the curb. The car stuttered for a few seconds. What to make of this new variable? My nerves clinched. And then an idea glimmered into the car’s mainframe. With sudden confidence, Pot Roast turned back onto 14th Avenue and pulled into the exact same parking spot where we started. “We have arrived,” declared the voice-over.

It was a perfect parable for Silicon Valley thinking. The car picks you up, changes your destination, drives in a circle, and charges you $5.35 as it drops you off where you started.

As I exited onto the sidewalk, Pot Roast drove away into the dark. His work complete.

A Postcard From Driverless San Francisco