A rider-experience teardown of Waymo, built from one real ride across Phoenix on a 106 degree day, one twelve-minute support call, and everything the public record revealed afterward. It builds to three concepts a team could actually ship, so they come first.
An opt-in fee that keeps your priority on a nearby car after a stop, instead of dropping you back into the queue. The fee prices the utilization the fleet gives up, so continuity pays for itself.
See the concept →Above a temperature threshold, the app flips to comfort-first: wait indoors, get pinged three minutes out, step into a pre-cooled cabin.
See the concept →Show riders their ETA next to the fastest possible one, then let them opt into freeways with guardrails as service returns.
See the concept →A mid-trip stop quietly dropped me back into the dispatch queue. In triple-digit heat, a two-minute errand turned into a long, sweaty wait on a curb. The product worked exactly as built. That is the problem.
Pressing “I’m ready” does not call your car back. Your car left to keep earning, so you rejoin the line for whatever car is free. Routing comes from the same instinct: the car takes the route that suits the system, not the one that gets you there fastest.
Both complaints trace to a single choice: optimize for fleet utilization over rider time. As the robotaxi market gets crowded, closing that experience gap stops being a nicety and becomes the next place the category is won or lost.
We took a Waymo across Phoenix and, partway through, I added a stop. Nothing dramatic, just a quick run into a smoke shop for a cold Diet Coke. We got out, the doors closed, and the car slipped back into traffic without us. That part felt normal. I assumed it would loop the block and wait.
When we came back out, I pressed the big blue “I’m ready” button. The app thought for a second and returned a number I was not expecting: a twenty-minute wait. It was 106 degrees. The can was already getting warm. There was no shade worth standing in, and twenty minutes in that heat is a long time to be standing on a curb.
So I did the thing most people do not do. I called customer care. A support agent picked up, and I just told her the situation plainly. She checked and said there were no cars available nearby. Then, while we were still talking, the estimate quietly dropped from twenty minutes to six, and a car was suddenly on its way. I stayed on the line the whole ride home.
Somewhere in those twelve minutes I asked her two things I could not help asking. First, how do you get a job at Waymo. Second, and more useful, what are the real product problems the team is wrestling with right now, because I wanted to study them properly.
What I heard on that call is the reason this teardown exists.
What I took away from that call: the cars sometimes take longer routes than they need to, it is a known issue the team is still working through, and I did not get a clear answer on when it gets fixed.My recollection of a 12-minute call with Waymo support, paraphrased in my own words
It stuck with me, not because it was some dramatic admission, but because it matched what I had just felt on the ride. So I went looking. I rode again, I called support again, and I read everything I could find. Two patterns kept showing up, and they turned out to be two versions of the same problem.
Here is the part that reframed it for me. When you add a stop and step out, Waymo holds your itinerary for up to thirty minutes, but the car itself drives away to serve other passengers. When you tap “I’m ready,” a different car is dispatched to come get you.
So that button is not a recall. It is a fresh request. You are not waiting for your car, you are back in line for whatever car is free. On a hot, busy afternoon with a thin local fleet, that line was twenty minutes long.
An Uber driver waits at the curb because a human is already being paid to sit there. A robotaxi has no one to pay, and an idle car earns nothing, so the economically rational move is to send it off to its next fare. So the wait was the system working as intended, not breaking down.
This is a genuinely hard tradeoff, not a lazy oversight. Holding cars for riders torches the unit economics that make robotaxis viable in the first place. The interesting product question is not “why does it leave,” it is “how do you give the rider continuity without giving up the economics.”
Strip away the symptoms and a single principle is doing the work. Waymo, still young and supply-constrained, optimizes for the things a fleet operator must optimize for: utilization, safety margins, and the logic of its own maps. Those are the right priorities for survival. They are not always the right priorities for the person in the back seat with a melting drink.
The re-pickup gap and the long routes are not two separate failures. They are the same decision, expressed twice. Once in how the fleet is dispatched, and once in how a single car drives. Name the root cause and everything else falls into place.
And the pressure behind that decision keeps rising. Since May 2024, Waymo’s paid rides have grown roughly tenfold, while the fleet has stayed near three thousand vehicles. When demand climbs an order of magnitude and the cars do not, every one of them has to be squeezed harder, and a rider’s spare minutes are the cheapest thing to spend.
A list of complaints gets you a list of patches. The root cause is what gives you a strategy, and that is what the rest of this teardown is after.
This is not just my read on one bad ride. Waymo says it plainly: it bases your price on the most direct route, and in its own words, “even if the car needs to re-route or adapt a route in an unexpected way, the cost to you won’t change.” The company already knows the route you ride and the route you pay for are often two different things.
The usual culprits are a deeply conservative driving style and a long aversion to freeways that pushes cars onto slower surface streets. Waymo finally began offering freeway trips to riders in late 2025, the obvious fix for cross-town detours. Then, in May 2026, it suspended freeway service entirely while it reworked how the cars handle construction zones and flooded roads. So as I write this, riders are back on surface streets, and the slow-route problem is very much live. The car is not lost. It is optimizing for its own maps and its own safety thresholds, not for your fastest arrival.
That same caution is a big part of why Waymo’s safety story is so strong. The fix here is not “drive recklessly.” It is honesty and choice: tell riders the tradeoff, and let them opt in when minutes matter.
My twenty-minute wait was not a fluke of one bad afternoon. Across nearly 95,000 rides studied in late 2025, Waymo averaged a 5.7-minute wait, against 3.2 for Uber, and the gap widens at the worst possible time: a late-afternoon capacity crunch between roughly 4 and 6pm, when demand outruns the fleet. Most of the time it is fine. But a routine stop drops you back into that same queue, and in triple-digit heat the wait stops being an inconvenience and becomes a welfare issue. And recall Waymo’s own wording: it “can’t keep the same car for you.” Here is exactly how my own wait played out.
The number did not improve because the system self-corrected. It improved because I called, waited on the line, and a person intervened. Most riders will not do that. They will just stand in the heat, or stop adding stops at all.
Routing and re-pickup are the headline. But pull the thread and a few more rider-experience gaps show up in the public record. None are fatal. Together they sketch the edge of where a still-maturing product frays.
The car often cannot stop exactly where you asked, leaving riders with a surprise walk at the end. The journey is autonomous, the final approach is still a negotiation.
Riders report slow, hard-to-find support. I got lucky with a helpful agent, but luck is not a support strategy, especially in the moments that scare people.
Waymo’s longest waits cluster in the late afternoon, when ride demand outruns a roughly fixed fleet. The product is most fragile exactly when people most need it.
Waymo still prices above Uber, but the premium fell from 30 to 40% in mid-2025 to about 13% by early 2026. When riders who have tried it choose, Waymo is the most preferred robotaxi brand (around 40%). Goodwill is real, but it is a shrinking moat, which makes experience matter more, not less.
A rider-experience teardown that stops at convenience would be missing the bigger story. The last year handed Waymo a string of harder problems, the kind that decide whether a city lets you keep operating at all. A serious product manager has to hold these in view, because they shape the same rider trust that my twenty-minute wait chipped at.
After Waymos were recorded passing stopped school buses (one Austin district logged 19 instances), Waymo issued a voluntary software recall of 3,067 vehicles and NHTSA opened a probe. A separate January 2026 case, where a Waymo struck a child with minor injuries near a Santa Monica school, triggered another federal investigation.
In San Francisco’s December 2025 blackout, Waymos stalled in intersections more than 1,500 times and one delayed an ambulance by roughly 40 minutes. The city later learned Waymo had only about 75 staff able to intervene across a fleet of thousands.
Waymo recalled 3,791 vehicles in May 2026 after cars drove into flooded roads they could not cross. The same week, it paused freeway service to rework construction-zone and flood handling. Robustness in bad conditions is still being built.
There is still no autonomous wheelchair-accessible Waymo, and under Senate questioning in early 2026 the company offered no timeline. Accessible rides are handled by human-driven partner cars. For a product that markets itself as the future of mobility, that is a real gap.
Driverless cars have been torched in protests, and labor groups are pushing human-driver mandates city by city. Police have sought Waymo footage, raising surveillance concerns. Public goodwill is not guaranteed, and it shapes where Waymo is allowed to grow.
None of this is a safety indictment. By Waymo’s own analyses its cars are in far fewer injury crashes than human drivers, and an independent Swiss Re insurance study found roughly 88% fewer property-damage claims and 92% fewer injury claims per mile. The point is narrower and, I think, more interesting: a product can be statistically safe and still strand you in the heat, skip the people who most need a ride, and lose a city’s trust in a crisis. Those are experience and operations problems, and a rider-focused PM owns them.
A teardown that only diagnoses is half a teardown. The job is to choose. Plotted against rider impact and build effort, the picture is clear: the re-pickup gap and routing honesty sit in the high-impact band and are addressable without re-architecting the autonomy stack. They are mostly dispatch, pricing, and interface problems, which is exactly the territory a product team can move on quickly.
Dropoff precision and crisis-grade reliability matter, but they are heavier lifts tied to perception, mapping, and operations. They belong on the roadmap, not at the front of it.
Ship rider-facing fixes for the two headline symptoms first. They are the cheapest path to the biggest felt improvement, and they directly answer the complaint a frontline agent already admitted to.
An opt-in toggle on any stop. For a small fee, the rider keeps priority on a nearby car instead of rejoining the queue. The fee is the point: it prices the utilization you are asking the fleet to give up, so continuity pays for itself. Today Waymo does the exact opposite (“we can’t keep the same car for you”), so this is open territory, not a feature that already exists.
Above a temperature threshold, the app flips to a comfort-first flow: wait indoors, get pinged three minutes out, and step into a pre-cooled cabin. In a market like Phoenix, this is not a luxury, it is the difference between a delightful product and a punishing one.
Start with pure transparency: show the rider their ETA next to the fastest possible one, since Waymo already knows both. Then, as freeway service returns after its May 2026 pause, let riders opt into it with clear safety guardrails. The opt-in data also reveals exactly which segments are worth the engineering to speed up.
| Area | Waymo ships today | What I’d add |
|---|---|---|
| Multi-stop | Saves your itinerary for 30 min, but a different car returns | Hold My Ride: keep priority on a nearby car |
| Routing | Direct-route pricing; freeways paused May 2026 | Route honesty: ETA vs fastest upfront, guardrailed freeway opt-in |
| Comfort | Minimize-walking pickup; cabin temperature control | Heat mode: wait indoors, pre-cool, ping 3 min out |
Coming up with ideas is the easy part. What turns three concepts into a product plan is two questions a leader will ask in the first five minutes: does it make money, or at least protect it, and can we build it without a multi-year detour. Here is the honest answer for each.
All three concepts live in dispatch, pricing, and the app, not in the self-driving stack, which is deliberate. They never ask the AV team to change how the car perceives or drives, which is the slow, expensive, safety-critical work. A product team can ship these in quarters, not years.
| Bet | Reach | Impact | Confidence | Effort | Call |
|---|---|---|---|---|---|
| Route honesty (ETA vs fastest) | High | Med | High | Low | Ship now |
| Hold My Ride | Med | High | Med | Med | Ship now (MVP) |
| Heat-Aware Pickups | Med | High | Med | Low | Fast follow |
| Fast Lane (freeway opt-in) | High | High | Low | High | Later |
Route transparency (the data already exists) and a Hold My Ride MVP that runs only off-peak, where holding a car is essentially free. Cheap to build, fast to learn from.
Heat-Aware Pickups in the hot markets (Phoenix, Vegas, Miami), and dynamic peak pricing for Hold My Ride once the off-peak version proves the demand.
Fast Lane, the freeway opt-in, which only makes sense once freeway service returns from its May 2026 pause and clears safety review. High value, but not in our hands yet.
That a premium product should not nickel-and-dime: keep the hold optional and cheap or free off-peak, never the default. That it could cannibalize utilization: it is priced and gated by local supply, so it never starves other riders. That showing the fastest ETA might just make people angrier: only surface it when the gap is real, pair it with a way to act on it, and test the wording. And the honest one, that all of this rests on a single ride and public reporting: exactly right, so none of it ships on faith. Each gets a geo holdout experiment with a clear success bar before any full rollout.
Waymo has already won the hardest argument: it runs a real, fully driverless service at meaningful scale, across more than ten cities, with a strong safety story. That lead is enormous. It is also, increasingly, table stakes.
The moat is shifting. Waymo’s premium over Uber has already fallen from 30 to 40% down to about 13%, and its new lower-cost vehicle shows the team understands the economics half of the equation. The experience half, the rider’s time and continuity, is the part still left on the table. And the window to own it is closing, because the field is filling up fast.
When a rider has a genuine choice between robotaxis, they will not pick the one with the best lidar. They will pick the one that does not strand them in the heat.
| Waymo | Uber | Lyft | Tesla | |
|---|---|---|---|---|
| Avg fare | $19.69 | $17.47 | $15.47 | $8.17 |
| Avg wait | 5.7 min | 3.2 min | 5.1 min | 15.3 min |
This started as one ride and one phone call, on one very hot day, in one city. That is a sample size of one, and I will not pretend otherwise. To check whether my experience generalized, I read widely across public reporting, rider accounts, and Waymo’s own documentation, and the same patterns kept reappearing. That gives me confidence in the diagnosis, not certainty.
I do not have Waymo’s internal data. A real product manager on the team could confirm or kill these ideas in an afternoon with the dashboards I cannot see. So treat the concepts here as well-formed hypotheses, ready for the data to judge them, which is exactly how I would bring them into a room. The point was never to be right on the first pass. It was to show how I find a problem, frame it, and turn it into something a team can actually act on.
A word on the numbers, since honesty cuts both ways. The price and wait comparisons come from Obi, a commercial ride-price aggregator, so read them as directional rather than gospel. Most of the safety statistics are Waymo’s own analyses, with the independent Swiss Re insurance study as the strongest outside check. I found no solid evidence that Phoenix heat reduces car availability, so I have treated heat as a comfort problem, not an availability one. And the freeway pause is current as of mid-2026, exactly the kind of fact that can shift between when I write this and when you read it.