The Product Discipline Behind a Driverless Ride

Shweta Shrivastava, senior director of product management at Waymo, says her teams are responsible for three areas: a large part of the software that runs on board the fully autonomous vehicle and determines its behavior and trajectory; the simulation tools and technologies used to validate system performance; and commercial scaling of the ride-hailing business, one of Waymo's key go-to-market applications. She describes the work as highly complex and technically demanding, but also as a product with unusually direct consequences for safety and trust. The interview begins with Lenny's recent ride in a Waymo on a rainy day in San Francisco, where no driver sat in the front and the app experience felt like Google Maps that sends a car instead of directions.

Lenny describes the ride as mind-blowing because he had never been in a self-driving car with no driver in front; he had used Tesla's self-driving features but had never sat in the back while the car drove. He notes that the Waymo app felt familiar, like Google Maps, except a car shows up and takes the rider to the destination, and riders can change course while driving. He says the experience quickly became normal, and Shweta thanks him for taking the ride and notes that the rainy day was a bonus because the technology performed well.

Lenny noticed that when the Waymo needed to merge into a lane with moving cars, it inched out in a subtle body-language gesture that communicated 'can someone let me in?' without eye contact. Shweta explains that Waymo trains deep-learning models on human driving data because a behavioral car must not seem robotic or unnatural. The goal is familiarity and trust: riders should not feel daunted by the technology or as if they are sitting in a robot, even though the system is intended to be safer than human driving.

The system uses deep-learned models to infer other road users' intent, including where a pedestrian is looking, body orientation, road signs, and hand gestures such as someone trying to stop the vehicle. Shweta says Waymo uses human driving data but discards bad human driving data, so the car can mimic human behavior in a good way. That is why Lenny saw the subtle lane-merge behavior during his ride.

Explicit gestures and signs are not enough because much of driving is governed by social norms. Shweta gives the example that in a particular San Francisco section it may be acceptable for pedestrians to cross even without a walk sign, while another city or intersection may have different norms at a four-way stop or crosswalk. The car therefore has to learn those social norms and react to them; driving is more sophisticated, interactive, and social than people often realize.

Shweta says the biggest difference from traditional software is the technical complexity and the difficulty of the problem. Waymo PMs have to be technically deep and get into details as much as needed, and they have to be comfortable with uncertainty and ambiguity. The work is a long game, so PMs need tenacity to keep improving the product and making autonomous driving a broad reality.

She says people at Waymo tend to be self-selected by the mission of making roads safer. She cites about 1.35 million traffic deaths worldwide every year, most attributable to driving errors and driver distraction, and notes that she has been guilty of checking text messages while driving. With fully autonomous technology, Waymo is trying to minimize that risk. The concept of MVP still applies, but safety is top of mind and the safety bar for an MVP is extremely high; Waymo cannot cut corners on safety.

Shweta distinguishes Waymo from driver-assist or ADAS systems such as Tesla's, which rely on a human driver to take over in complex situations. Waymo started by solving fully autonomous driving without a human driver at the wheel, so the expectation is not that a person will intervene. From the beginning, the design philosophy had to make the system feel predictable and trustworthy to riders. She says riders often describe the first five minutes as 'wow,' after which the ride starts to feel natural and even uneventful, but that naturalness is designed into the system.

Trust comes from many small design choices rather than one feature. Shweta says the car never exceeds the speed limit, which riders appreciate, but the team also learned that strictly adhering to the speed limit is not always the most natural experience. On San Francisco's inclines, human drivers subconsciously slow down going downhill even when staying below the limit, so Waymo modified the behavior to match that expectation. Riders also get visibility into what is happening: rider support can call if a seat belt is not buckled, and riders can look at an in-car monitor to see what the car sees.

Shweta divides Waymo's metrics into commercial and operational metrics and system-behavior metrics. Waymo is no longer in a proof-of-concept or pilot phase: it offers a paid service in Phoenix and is open to the public in San Francisco. On the commercial side, the team tracks trips per week, daily or weekly active users, funnel metrics, and the cost to operate the service.

On driver performance, Waymo measures safety, compliance with road rules, and the ability to make adequate progress without getting unduly stopped or stranded in dense traffic. The goal is to drive safer than humans, but there is no single standard human-driving safety benchmark that everyone uses. Waymo gathers enough data to form its own view of human driving, such as how many collisions a human driver would have per 100,000 miles, and then works to perform better than that benchmark.

Safety alone is not enough because a car that never moves would be safe but useless. Waymo also tracks whether the vehicle slows down unduly or has to rely on rescue help in a given week, and how much it slows down traffic for other road users. The team benchmarks those outcomes against priors and an adequate-performance standard. In other words, the product has to be appropriately assertive and make progress while remaining safe.

Shweta explains that L4 is fully autonomous without a human driver at the wheel and with no expectation of a human driver, which is Waymo's focus. L5 would mean driving on any road, including completely unstructured off-road environments, without maps or priors. She believes Waymo can realize the dream of fully autonomous driving in a big way through L4 service in cities such as Phoenix and San Francisco rather than waiting for L5, which may become more niche. The technology is already there but still needs improvement, and Waymo cannot yet drive in snow.

Lenny asks how Waymo keeps leaders and investors bought into a long-term, large-scale investment. Shweta says Waymo is fortunate to have backing from Alphabet and other investors, and she notes that the autonomous-vehicle industry has seen consolidation over the past year. She says the name of the game is to show meaningful progress, not only in technology but also in commercial deployments, because that is where the product meets the road.

Investors gain confidence when results speak for themselves and when Waymo accelerates milestones and exceeds its own expectations. Shweta says the company focuses on doing what is right for the business and creating value for customers and riders rather than chasing short-term brownie points with investors. She believes investors see through unnatural moves that do not align with business goals. The lesson is to build the right business and do right by users.

Across Amazon, Cisco, Waymo, and startups, Shweta says the core product-management tenet is the same: work backwards from the customer or user problem. Building technology for its own sake does not go far; teams must know what they are building, who they are building it for, and what problem they are solving. She cites Amazon's practice of writing a press release for the finished product before building it, which forces PMs to think through the value proposition. Waymo has its own version of that focus on the customer problem, adapted to a highly integrated product with different flavors of product management.

Shweta says it is also important to know what you are not building. That applies at large companies and startups because teams can be swayed by one customer asking for one thing and another customer asking for something else. A product that tries to be all things to all people usually does not go anywhere, so focus, prioritization, and clarity about what is out of scope are essential.

At large companies, Shweta points to the innovator's dilemma: market-share leaders can become incremental and then be disrupted by an upstart. She says companies need to disrupt themselves before someone else does, and the ones that succeed constantly challenge their own models and product capabilities to produce something more transformational. She cites her experience as the first PM at Amazon for Honeycode, a no-code application development platform, as an example of entering a new space; she built the team around it.

When asked about underrated PM skills, Shweta names listening and empathy. She says product management often emphasizes influencing without authority, prioritization, and writing, but listening and empathy are core to influence and are easier said than done. She advises PMs to come in with a growth mindset and a beginner's mindset, absorb and learn, and not jump to ideas immediately. They should take time to formulate an opinion and stay true to working backwards from the customer problem rather than treating it as a platitude.

Shweta developed those skills by doing the work repeatedly in different environments, product launches, startups, and large companies. She says team dynamics and cultures differ, so PMs need to understand an engineering leader's constraints and where that person is coming from, then see where they are aligned and where they are not. She also says PMs should challenge their own assumptions; if there is no conflict or contention, something may be missing. Listening well means picking up cues and testing whether your assumptions hold.

On promotions, Shweta offers advice that sounds cheeky: the way to get promoted is not to want it too badly. PMs should focus on impact and doing what is right for the business rather than optimizing for the promotion. Ambition is fine, and people should make their ambitions known to their manager or leader, but they should also work on challenging, high-visibility projects that stretch their skills. If they maneuver too much, it becomes visible and distracts from the work; focusing on impact makes the contribution visible when the right opportunity comes.

Waymo is already operating in the Phoenix metro area and San Francisco, where people can download the app and use the service. It has initial fully autonomous testing in Los Angeles and plans to expand there through the rest of the year. Waymo has a list of future cities but is not sharing it yet; in San Francisco and Phoenix the service is open, and Shweta says Phoenix may not even have a waitlist.

In the lightning round, Shweta recommends Crossing the Chasm and Clayton Christensen's The Innovator's Dilemma as books she quotes and recommends often. She enjoyed Top Gun: Maverick, especially its visuals, and finds it inspiring to see what Tom Cruise accomplished at his age. Her favorite interview question, especially for senior candidates, is to ask about a time they failed and what they learned; she sees red flags when candidates claim they have never failed or turn a success story into a failure story.

Shweta is shopping for a foldable e-bike for sustainable mobility, which she frames as doing more mountain biking without doing mountain biking. Her process tip is the 'rule of seven': if seven emails in a thread have not resolved an issue, call the person or get in a room and resolve it; at larger companies the limit might be closer to ten. For riders, her pro tip is to bring a favorite playlist and enjoy the ride. She says people can find her on LinkedIn, look at Waymo's open positions, and be useful by signing up for a ride in Phoenix, San Francisco, or Los Angeles when it opens and giving feedback.