Ten Waymo Rides in One SF Week

This is a first-person piece, written after a week of ten Waymo rides across San Francisco in May 2026. Some of what follows is highly specific to the city. Some of it generalizes. We will be clear about which is which.

For the broader status of the self-driving industry (Waymo, Tesla FSD, Mercedes Drive Pilot, Aurora’s trucking program), the Real State of Self-Driving in 2026 Special Report is the right place to start. This piece is the texture of what it feels like to actually ride in one for a week.

How does a ride start?

You open the Waymo One app. You set your pickup and drop-off. The app shows an estimated wait, an estimated drive time, and a price that looks like an Uber quote with a slightly different vibe. You confirm. The app shows the car moving toward you on the map, the way every rideshare app has done for ten years.

The difference starts when the car arrives. There is no driver. There is no “where are you going” small talk. There is no rating-anxiety. There is a car, a chime, an unlock prompt on your phone, and a back seat that is already exactly the temperature you wanted because the climate-control inherits from your last ride. You get in. You shut the door. You hit Start Ride on the in-cabin screen. The car pulls into traffic.

Wait times across our ten rides averaged 5 to 12 minutes. The fastest was 3 minutes (a Sunday morning pickup in the Inner Sunset, the car was already two blocks away). The slowest was 18 minutes (Friday 6:30 PM in SoMa, demand was high enough that the app warned about delays before we even confirmed). Waymo prices surge by demand, the same way Uber does. Three of our ten rides cost more than the equivalent Uber. Five cost less. Two were within a dollar.

What is it like inside the cabin?

Quiet. That is the first thing every Waymo rider notices. The Jaguar I-PACE chassis is electric, the cabin insulation is good, and there is no driver radio or driver phone conversation. You can hear yourself think. You can have a real phone call without worrying about being overheard. The conversational privacy alone is worth the price of admission for anyone who has tried to take a sensitive call in a regular rideshare.

The screens are the second thing. There is a screen in the headrest of the front passenger seat, and a smaller screen on the dashboard visible from the back. Both show what the car is “seeing”: other vehicles as gray boxes, pedestrians as orange figures, cyclists as blue. When the car does something unexpected, the screen usually shows you why. Yielding at an intersection? The screen highlights the pedestrian stepping into the crosswalk. Slowing on a hill? The screen shows the cyclist three cars ahead.

The music control is great. The app remembers your last Spotify playlist and starts it automatically. You can change tracks from the in-cabin screen. Volume controls are physical buttons on the door. It is the closest a rideshare has come to feeling like your own car.

The one cabin friction point is the seatbelt reminder. The car will not move until everyone is buckled. Fair. It is also quite insistent. Three of our ten rides had a passenger in the back seat who took an extra moment to buckle, and each time the dashboard chimed politely but persistently. Less annoying than a human driver staring at you. Slightly more annoying than your own car.

How does it actually drive?

Conservative. Patient. Predictable. The car uses a lidar sensor stack with cameras and radar, and operates inside a defined operational design domain covering most of the city. Within that envelope the driving style is the most consistent we have ever seen in a vehicle.

Lane positioning is precise. The car holds its lane center to within centimeters and signals every lane change with the full three-blink minimum. Right-on-red is taken with a complete stop, a careful look, and a careful go. Left turns at four-way stops are handled in strict order. The car does not nudge into a left turn against oncoming traffic the way a human driver might. It waits for a real gap.

The yielding behavior at pedestrian crosswalks is the cleanest version of the law we have ever experienced. A pedestrian within ten feet of a crosswalk in the direction of crossing causes the car to stop and wait. Other San Francisco drivers honk. The Waymo waits anyway. As a passenger this is mildly slower and significantly more virtuous. As a pedestrian on the rest of the week, when we were walking instead of riding, it became something we looked for in traffic.

The most impressive moment of the week was on Fell Street eastbound during a Tuesday afternoon, when a cyclist swerved into our lane to avoid a parked truck door that had just opened. The Waymo saw the door open before the cyclist did, slowed by the right amount, and shifted left in the lane to give the cyclist exactly the space they needed. A good human driver might have done the same. Many would not have. The car did not even seem to consider the alternative.

What were the weird moments?

Out of ten rides, two had moments that a human driver would not have produced. Both are worth describing in detail because the press tends to lump every Waymo edge case into “robot car gets confused,” which does not accurately describe what we observed.

Moment one: the loading-zone dance. On the Mission ride, we asked to be dropped at a restaurant whose front door is on a one-way street with no legal pickup zone. The Waymo drove past the restaurant, made a right at the next intersection, made another right, came back to the same block from the other direction, and stopped fifteen feet past the door in a legal loading zone. A human driver would have stopped briefly in front of the restaurant and accepted the social cost of a brief illegal stop. The Waymo did not consider that. The detour added about ninety seconds. Both behaviors are defensible. The car’s behavior is the more lawful one.

Moment two: the firetruck. On the FiDi ride, a firetruck approached from behind with lights and sirens. The Waymo pulled to the right immediately, much more aggressively than the cars around it, and held still until the truck had passed. Then it sat for another five seconds. Then it resumed. The first behavior was textbook correct. The five-second pause after the truck cleared was not what a human driver would have done. We assume the car was waiting to confirm no follow-on emergency vehicles. It is a fine instinct. It looked slightly odd in the moment.

Neither moment was unsafe. Neither moment was even particularly inconvenient. Both moments showed a car driving by the exact rules and tolerating the social cost of doing so. Whether that is the right tradeoff depends on what you value in a driver.

What about price?

Across our ten rides, the total Waymo cost was within about five percent of what the equivalent Uber rides would have been. The breakdown was less even than that average suggests. Off-peak rides were noticeably cheaper than Uber. Peak rides were more expensive. Surge pricing on the Waymo One app is real and tracks demand the same way every rideshare service does.

The harder pricing question is what Waymo is actually costing Alphabet to operate. The fares we paid almost certainly do not cover the per-ride economics of the vehicle, the lidar stack, the remote-assistance team, the depot, the cleaning, and the depreciation. Public reporting suggests Waymo is operating at a per-ride loss in 2026 that narrows every year. As a rider you do not see that. As a citizen interested in the long-term economics of autonomy, it is the most important fact about the price.

Where it impressed us, where it broke down

Impressed us:

• Cyclist awareness, every time.
• Stopping for pedestrians at corners where most human drivers would not stop.
• Reading the screen and understanding why the car was slowing.
• The cabin privacy and the cabin quiet.
• The complete absence of the social negotiation of a tip.
• Late-night reliability. Saturday 1:15 AM in the Marina, the car arrived in 7 minutes, drove home in 14, and we were asleep in 25.

Broke down:

• The “find a legal pull-over zone” detour, twice in ten rides, when a human would have stopped at the door.
• One drop-off in the Sunset where the car missed the exact address by half a block, stopped, recalculated, and circled. We just got out a block early. Cost us maybe two minutes.
• Weather. We did not get to test the car in fog (it has been clear in SF this week, which is rare), so we cannot tell you how it handles the bay’s actual signature weather. The published Waymo data says fog is fine. We have not personally verified.
• Highway. Within the San Francisco geofence, highway use is limited. Waymo’s recent expansion has added some highway segments, but the majority of our city rides did not need to use them.

Would we do it again?

We already did. After the formal ten-ride week, we kept using Waymo as the default rideshare for the rest of the week without thinking about it. That is probably the strongest endorsement we can give. The technology has crossed the line from “interesting experiment” to “default tool” inside the geofence Waymo serves.

If you live in San Francisco, Los Angeles, Phoenix, or Austin, where Waymo is now operating, our recommendation is to install the app and take a few rides. The price gap to Uber is small enough that the experience itself is the reason. If you live somewhere Waymo has not reached yet, the timeline depends on when the company expands the geofence to your city.

Frequently asked questions

Is Waymo actually fully driverless?

Yes, in the San Francisco service area. No safety driver in the seat. A remote-assistance team can be contacted by the vehicle if it gets confused, but they do not actively drive the car. The car drives itself.

Where does Waymo currently operate?

As of May 2026: San Francisco (most of the city), Phoenix (large geofence), Los Angeles (expanding), Austin (limited), and several smaller deployments. The Waymo One app shows availability by city.

Is it actually a robotaxi?

Yes. The technical term “robotaxi” refers to a fully autonomous vehicle operating commercial passenger service without a driver. Waymo is the largest robotaxi operator in the world by ride volume in 2026.

How safe is it really?

Waymo publishes safety data quarterly. On a per-mile basis, the published rate of police-reported crashes is meaningfully lower than the human-driver benchmark for the same geographies. The data has methodological caveats (geofence-limited, low-speed-heavy, daylight-skewed), and academic researchers continue to debate the right comparison. The fair version is that Waymo is at least as safe as a careful human driver in the conditions it operates, and probably safer.

Does it work in the rain?

Yes. Waymo cars operate in rain, light fog, and standard nighttime conditions in San Francisco. Heavy weather may degrade service or trigger remote-assistance involvement, but the published reliability data covers a meaningful range of conditions.

The Beginners in AI position on Waymo

This is the cleanest example of pro-technology and pro-human-first lining up that we cover at this site. Self-driving done right gives people their commute time back. It gives the visually impaired and the elderly the kind of autonomy that comes from being able to leave the house at any hour and get to wherever they need to go. It removes one of the more dangerous decisions a tired person makes on a Friday night.

It also has to be done in a way that respects the human drivers it is gradually replacing. The right policy answer involves real retraining, real transition support, and real attention to the social fabric of cities whose taxi and rideshare workforce is changing. The technology side of this is moving faster than the labor side. That is the conversation worth having.

If you are reading this on a Monday morning after the weekend, the takeaway is: try one if you can. Pay attention to what surprises you. Notice what your assumptions were before the ride and which of them held up. That is what we did across ten rides this week, and the version of the answer we landed on is at the top of this piece.

Sources

• Waymo official site
• Waymo safety performance hub for the quarterly crash data
• California PUC Autonomous Vehicle Programs, the regulatory backstop for SF operations
• NHTSA Automated Driving Systems
• Kalra, N., and Paddock, S. M. (2016). “Driving to Safety: How Many Miles of Driving Would It Take to Demonstrate Autonomous Vehicle Reliability?” RAND Corporation
• IIHS Advanced Driver Assistance topic page

You might also like

• The Mobility and Transportation Frontier, our master hub for EVs, self-driving, and the broader transportation shift
• The Real State of Self-Driving in 2026, our Special Report on Waymo, Tesla FSD, Mercedes, and Aurora
• What is a robotaxi? the plain-English glossary entry
• What is an Operational Design Domain (ODD)? the term every self-driving press release uses
• Fusion in 2026: A Status Report, our parallel energy-frontier piece
• Figure 02 vs Optimus vs Digit, the humanoid race head-to-head