Source: Unsplash
Waymo now completes 500,000 paid rides every week. It just recalled nearly 4,000 robotaxis for driving into construction zones. A passenger in San Antonio watched their car detect floodwater, slow down, and then drive into it anyway. Another passenger sprinted after a Waymo at the airport while their luggage was locked in the trunk, driving away without them. The robotaxi is here. The interface that explains what it's doing, and why, is not.
By March 2026, Waymo had achieved something genuinely historic: half a million paid robotaxi trips every single week, across 10 cities in the United States. A 6th generation vehicle called Ojai just launched, designed from the ground up with a flat floor, gondola-style doors, and three large adaptive screens so passengers can control their route, music, climate, and leave a tip. The company is expanding to 20 new cities this year and running its first international tests in Tokyo and London. From a business and engineering perspective, this is a spectacular achievement. From a UX perspective, 2026 has been a year of red flags that the industry is not taking seriously enough.
"I pressed the trunk open button, tried to get my luggage, but it doesn't do anything, and it drives away immediately."
— Jin, Waymo passenger stranded at San Jose Airport, NBC News report
The Incidents Tell a Design Story
The individual incidents that have made headlines in 2026 are not just bad PR problems. They are diagnostic data. Each one reveals a gap in how the car communicates with the people inside it, and the people around it.
Take the floodwater incident. A Waymo in San Antonio detected standing water ahead, slowed down, assessed the situation, and then drove into the water anyway, where it was eventually swept into a creek. The car had the sensor data. It made the wrong decision. But here is the design problem nobody is talking about: the passenger had no way of knowing that the car was even processing a decision at that moment. There was no signal saying "I see something ahead." No indication that the car was choosing between two options. No way for the passenger to intervene before the car committed. The driver's seat was empty, and the communication layer that would have replaced the driver's judgment was absent.
Or the 8-lane turn incident. Passengers inside a Waymo realized, mid-maneuver, that their car might cross directly into the path of oncoming traffic. They started shouting. Not because shouting helps, but because there was nothing else to do. When a human driver does something alarming, you can grab the wheel, say "stop," make eye contact, or at minimum understand that a person made a decision you can argue with. In a driverless car, you are a passenger in a sealed box watching decisions happen to you.
And then there is the trunk. A passenger arrived at San Jose Airport, opened the app, pressed the trunk release, and the car drove away with their luggage still locked inside. The interface had a button. The button did not work. There was no confirmation that the trunk had opened. No secondary check. No "are you ready to end your ride?" There was just a car that left.
Trust Is a Design Problem, Not an Engineering Problem
A Forrester study found that 61% of potential AV users will not ride in a driverless car unless it gives them clear visual or audio feedback about what it is doing and why. That number is not about fear of the technology in the abstract. It is about a specific gap: people need to understand the machine's intent before they can trust it.
This is something I think about a lot in my work on enterprise AI products. Trust in an automated system is not binary. It is calibrated through thousands of small moments of communication. The system that says "I'm running your payroll, here's what I'm doing" gets trusted faster than the one that just runs it quietly. The AI that shows its reasoning gets adopted faster than the black box. The same principle applies at 65 mph on a freeway.
Waymo's Ojai vehicle does have screens that show intent. Passengers can see lane changes, object detection, and upcoming maneuvers visualized in real time. That is genuinely good design thinking. But screens inside the car solve the communication problem for the passenger, not for the pedestrian wondering if the car has seen them, not for the cyclist in the blind spot, and not for the construction worker trying to wave the vehicle into the correct lane. The interface has to exist in all directions, not just inward.
The Gemini Experiment: Promising and Limited
Waymo is reportedly developing a Gemini-powered AI assistant for the in-car experience. The idea is sound: a conversational layer that can answer passenger questions, explain what the car is doing, and manage some in-cabin functions like climate control. That is exactly the kind of communication bridge that would help close the trust gap.
But look at what the assistant reportedly cannot do: volume control, window control, seat adjustment, and route changes. These are the basic functions passengers expect to control. When someone is in a car and feels uncomfortable, the first things they reach for are the window and the ability to change where they're going. Blocking those actions from the conversational interface and routing them to the screen, or the app, is a fragmentation of the experience at the exact moment passengers need seamless control.
This is a pattern I have written about before at reloadux.com in the context of enterprise AI products. When you design an AI assistant with an incomplete action set, users lose trust in it faster than if you had no assistant at all. The assistant creates an expectation of capability. When it can't do what you're asking, the frustration is higher than if you had just found the button yourself. The gap between what the assistant implies it can do and what it actually can do is a UX debt that compounds with every failed interaction.
What the Recall Tells Us
In June 2026, Waymo recalled 3,871 robotaxis after they were found to be driving into freeway construction zones. The recall itself is not the story. Car companies recall vehicles all the time. The story is what the recall reveals about the system architecture: nearly 4,000 cars were making the same wrong decision, in similar situations, before anyone caught it. The feedback loop between real-world behavior and system correction is slower than the pace of deployment.
At scale, that gap becomes dangerous. Waymo is targeting 20 new cities this year. Tokyo testing is underway. London is preparing for deployment. Each new city brings new road configurations, new traffic norms, new edge cases the system has never seen. The construction zone recall suggests the system is not always gracefully handling the unexpected. And when it handles the unexpected badly, passengers have very limited options for intervening.
What Good Robotaxi UX Actually Looks Like
Here's what I think the design teams working on these vehicles should be solving, right now:
- Continuous intent broadcasting: The car should always be communicating what it is about to do, not just what it is currently doing. A gentle visual or audio cue that says "preparing to change lanes" or "slowing for traffic ahead" closes the surprise gap that drives passenger anxiety. This is the equivalent of a good driver narrating their decisions to a new passenger.
- Clear intervention affordances: When something seems wrong, passengers need a real escape path. Not a phone number to call after the fact. A clearly labeled button that triggers a safe stop and connects them to a human operator within seconds. Emergency situations should not require the passenger to figure out the app.
- External signaling design: The car needs to communicate with people outside it. Pedestrians, cyclists, and emergency responders all need to understand what a driverless car is about to do. This is a whole design discipline, external HMI, that the automotive industry has barely started working on. Waymo uses some LED indicators, but the language is not standardized, not intuitive, and not sufficient.
- Error recovery design: The trunk incident should be impossible by design. If a trunk release is pressed and the trunk does not open, the ride should not end. There should be a confirmation state, a retry, and a fallback to human support. Every critical interaction in the passenger flow needs a defined error state with a clear recovery path.
- Uncertainty transparency: When the car encounters something it is not sure about, it should say so. Not in a way that panics passengers, but in a way that builds honest calibrated trust. "Unusual traffic ahead, taking a different route" is better than silence followed by an unexpected turn. Transparency about uncertainty is one of the most powerful trust-building tools in AI product design.
The Stakes Are Higher Than a Bad App Experience
I have spent years pointing out that bad enterprise UX costs companies money and productivity. When the UX is bad in a SaaS dashboard, users get frustrated, adoption drops, and someone loses a budget renewal. The consequences are real but recoverable.
When the UX is bad in a vehicle moving at highway speed, an ambulance gets blocked at a mass shooting. A passenger loses their luggage at an airport and has no recourse. A car drives into a flash flood because no one could stop it in time. The stakes of robotaxi UX failures are measured in something more serious than churn rates.
Waymo is not a bad company. The engineering behind its vehicles is genuinely impressive. 500,000 weekly trips is not a small thing. But engineering excellence and UX excellence are not the same, and right now the gap between them is visible in every incident report. The vehicle can navigate San Francisco traffic. It cannot always explain itself to the human beings it is responsible for carrying.
The robotaxi era is not going to be stopped by any of these incidents. The market is too large, the investment is too deep, and the convenience is too real. But the design work that should have been done in parallel with the engineering work is now running behind. And in a vehicle, catching up takes longer than shipping a software update.
The car is ready. The interface isn't. That's the design problem of the decade, and it's time product and UX teams started treating it like one.
Are you watching the robotaxi rollout from a design perspective? I'd love to hear what patterns you're noticing or what UX problems you think are being under-discussed. Drop it in the comments.
Sources:
1. TechCrunch — Waymo's newest robotaxi is Chinese-made, accepting riders: https://techcrunch.com/2026/05/28/waymos-newest-robotaxi-is-chinese-made-built-to-make-money-and-now-accepting-riders/
2. Electric Cars Report — Waymo Ojai Robotaxi Debuts With Sixth-Gen Autonomous Tech: https://electriccarsreport.com/2026/05/waymo-ojai-robotaxi-debuts-with-roomier-design-and-sixth-gen-autonomous-tech/
3. eWeek — Waymo Pulls Nearly 4,000 Robotaxis From Freeways After Recall: https://www.eweek.com/news/waymo-robotaxi-recall-june-2026/
4. TechSpot — Waymo Robotaxi Drives Off With Passenger's Luggage at Airport: https://www.techspot.com/news/112283-waymo-robotaxi-drives-off-passenger-luggage-after-airport.html
5. Yahoo Autos — Waymo Robotaxi Eight-Lane Turn Sparks Passenger Panic: https://autos.yahoo.com/policy-and-environment/articles/waymo-robotaxi-eight-lane-turn-054245292.html
6. SFist — Report: Waymo Vehicles Create a Whole New Set of Traffic Safety Problems: https://sfist.com/2026/06/04/report-as-waymo-works-to-solve-problems-it-creates-new-ones-in-the-process/
7. wongmjane.com — Waymo Is Working on a Gemini AI Assistant: https://wongmjane.com/blog/waymo-gemini