Product lessons from Waymo and a decade of PM experience

Executive overview

Building a fully autonomous vehicle forces every standard PM assumption to break — MVP means something different when safety failures are irreversible, and "iterate fast" must coexist with a much higher bar. Shweta Shrivastava, Senior Director of PM at Waymo, draws on stints at Amazon, Cisco, and a safety-AI startup to contrast what changes at the frontier of hard tech and what stays constant.

The constants: working backwards from the customer problem, knowing what you're not building, and disrupting yourself before someone else does.

If there's no conflict and no contention in your listening, something is missing.

What Waymo PMs do differently

• Waymo PMs must go technically deep — more so than typical software roles.
• Comfort with ambiguity is table stakes; this is a multi-decade mission, not a sprint.
• Mission alignment matters: 1.35 million road deaths per year attributable to human error is the "why."
• MVP has a different bar — safety cannot be cut; the minimum viable bar for safety is extremely high.
• The product is fully autonomous (L4) from day one — no human takeover expected, unlike driver-assist systems.

How trust is built into the ride experience

• Deep learning models trained on good human driving data make the car feel natural, not robotic.
• The car reads pedestrian body orientation, road gestures, and city-specific social norms — not just explicit signs.
• Designers learned to slow the car on downhill gradients in San Francisco even without a strict safety need — riders expected it.
• Riders can see what the car sees via an in-car monitor, reinforcing transparency.
• Adhering to the speed limit is a trust signal riders explicitly appreciate.
• The first five minutes feel novel; after that, the experience becomes uneventful — which is the goal.

Metrics: commercial and system behaviour

• Two buckets: commercial/operational metrics (trips per week, active users, funnel, cost to operate) and system behaviour metrics (safety, compliance, progress).
• Safety benchmark: collisions per 100,000 miles versus a human driving baseline — Waymo targets better-than-human performance.
• "Stops and strandings" tracks the opposite failure: being too cautious. A car that never moves is safe but useless.
• Impact on surrounding traffic is also measured — how much does the vehicle slow other road users?

Where full self-driving actually stands

• L4 (fully autonomous, no human at wheel) is already live in Phoenix (open access) and San Francisco.
• L5 (any road, no map, unstructured off-roading) may remain niche — L4 covers most real-world value.
• Remaining technology gap: driving in snow is still unsolved.
• Regulatory and city-by-city expansion is the current pace-setter, not fundamental tech breakthroughs.

Keeping leaders bought in on long-horizon bets

• Show meaningful progress in both technology and commercial deployment — not just engineering milestones.
• Let results speak: accelerating milestones and exceeding internal expectations are the strongest investor signals.
• Don't manufacture short-term wins that don't align with business goals — investors, especially long-term backers, see through it.
• Focus on building the right business and genuine customer value; buy-in follows.

Core PM lessons across Amazon, Cisco, Waymo

• Always work backwards from the customer or user problem — building technology for its own sake rarely succeeds.
• Amazon's PR/FAQ process (write the launch press release before building) forces rigorous thinking about value proposition.
• Know what you're not building. Products that try to please every customer end up serving none.
• Large companies drift toward incrementalism. Disrupt your own model before a startup does it for you.
• Amazon Honeycode (no-code app platform) is a personal example: a first-of-its-kind bet inside an infrastructure-first company.

Developing listening and empathy

• Listening and empathy are harder than influencing without authority or writing PRDs — easier to name than to practice.
• Develop them through repetition across different company cultures, team dynamics, and product types.
• With each engineering leader, understand their constraints and what "impact" means to them specifically.
• Proactively challenge your own assumptions. No conflict in a conversation usually means you're not really listening.

On getting promoted

• Focus on impact, not on the promotion itself — optimising visibly for promotion is a negative signal.
• Make your ambitions known to your manager so they can put you on high-visibility, skill-stretching projects.
• Dedicate fully to work that creates real business value; the rest follows.

The rule of seven (and ten)

• If an email thread hits seven replies without resolution, call or meet in person.
• Long email chains that add more people without converging are a waste of everyone's time.
• At a larger company the limit shifts — around ten — but the principle holds.