11
“There Is No Undo”: Waymo Co-CEO’s Stark Warning to AI Robot Builders
Why Self-Driving Cars Need a Different Rulebook Than Chatbots
How Waymo’s leader is rewriting Silicon Valley’s "move fast" playbook for a world where mistakes can’t be undone
The Big Warning from Waymo’s Boss
Dmitri Dolgov, co-CEO of Alphabet’s Waymo (the self-driving car company owned by Google’s parent company), has a serious message for AI developers: stop treating robots like chatbots.
Speaking at a Y Combinator interview released Monday, Dolgov warned that Silicon Valley’s famous motto — "move fast and break things" — becomes dangerous when AI moves from screens into machines that drive around real people.
His proposed replacement?
"Move fast and ship safely."
"Move Fast and Break Things" Doesn’t Work Here
The Core Problem: No Undo Button
| Digital AI (Chatbots) | Physical AI (Cars/Robots) |
|---|---|
| Makes a mistake? User tries again | Makes a mistake? People can get hurt or die |
| "Oops, let me regenerate that" | No retry button. No undo. |
| Cost = wasted time/tokens | Cost = human lives |
Dolgov put it simply: "There’s simply not an undo and a retry button."
Safety Must Be Built In, Not Bolted On
Developers can’t just:
- Build the AI
- Release it to the public
- Add safety patches later
Instead, safety must be woven into every layer from day one:
- The model design
- The training methods
- The system architecture
Think of it like building a house: You don’t build the walls and roof first, then decide to add a foundation afterward. The foundation is the first thing you build.
Waymo’s Impressive Safety Record (So Far)
Waymo isn’t just talking — they’re backing it up with data:
- 220+ million rider-only miles driven (no human driver)
- 500,000+ fully autonomous trips every week
- 56.7 million miles studied vs. human drivers
- Result: Statistically fewer injury crashes and fewer airbag deployments than human drivers
That’s a strong case that careful, safety-first development works.
But It’s Not Perfect — Recalls and Real Problems
Even the leader in autonomous driving has stumbled:
Recall #1: Flooded Roads (May 2024)
- ~3,800 robotaxis recalled
- Software could direct cars into flooded roads
Recall #2: Construction Zones (Shortly After)
- ~3,900 vehicles recalled
- 12+ incidents of robotaxis entering closed freeway construction zones
- Freeway service temporarily suspended while fixes were installed
Federal Scrutiny: Emergency Scenes
The NHTSA (National Highway Traffic Safety Administration) identified a "clear pattern" of driverless vehicles:
- Entering active emergency scenes
- Blocking ambulances and fire trucks
- Failing to respond to lights, smoke, traffic cones
- Passing stopped school buses
NHTSA Administrator Jonathan Morrison (July 2024):
Any automated vehicle that can’t safely interact with first responders "presents a danger to the public" — companies must prioritize fixes.
Why This Matters for Everyone
KEY TAKEAWAY
Physical AI follows different rules than digital AI.
Waymo’s successes prove cautious deployment works.
Waymo’s recalls prove testing never really ends — not even after launch.
This isn’t just about self-driving cars. It applies to any robot or AI system that moves in the real world:
- Delivery robots on sidewalks
- Factory robots working alongside humans
- Surgical robots in hospitals
- Drones in crowded airspace
The standard has to be higher. The cost of failure is irreversible.
Summary
- Dmitri Dolgov (Waymo co-CEO) warns: "Move fast and break things" fails when AI controls machines around people.
- New mantra: "Move fast and ship safely."
- Safety must be foundational — built into models, training, and architecture from the start.
- Chatbots can retry; cars cost lives — no undo button exists.
- Waymo’s stats are strong: 220M+ miles, 500K+ weekly trips, safer than humans in studies.
- But recalls happen: Flooded roads, construction zones, emergency scene interference.
- NHTSA is watching — pattern of blocking first responders is a serious public danger.
- Lesson for all Physical AI: Test rigorously, deploy cautiously, monitor forever.
FAQ
What does "Physical AI" mean?
Physical AI refers to artificial intelligence that controls real-world machines — self-driving cars, robots, drones — instead of just generating text or images on a screen. When Physical AI makes a mistake, physical consequences follow.
Why can’t we just update self-driving cars like we update apps?
We can and do update them (Waymo pushed fixes for the recalls). But updates happen after a problem is found — and in the physical world, that problem might have already caused a crash. Prevention matters more than patches.
Is Waymo actually safer than human drivers?
According to their data, yes. Over 56.7 million rider-only miles, Waymo had statistically lower rates of injury-reported crashes and airbag deployments than human benchmarks. But "safer on average" ≠ "perfect."
What is NHTSA and why do they care?
NHTSA = National Highway Traffic Safety Administration (U.S. government agency). They regulate vehicle safety. When driverless cars block ambulances or enter emergency scenes, that’s a public safety hazard — so NHTSA steps in.
Does this mean self-driving cars aren’t ready?
Not necessarily. It means they require a different development philosophy — one where safety isn’t a feature you add later, but the foundation you build on. Waymo’s expansion shows progress; its recalls show the work isn’t done.
Disclosure: The original article contains affiliate links for investment platforms (Arrived, Realberry, FarmTogether, etc.). This summary focuses solely on the Waymo/Physical AI safety content.
