Mar 24, 2026 · 1h 2m · cheeky-pint
The 20-year journey to fully autonomous cars with Dmitri Dolgov of Waymo
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
Waymo co-CEO Dmitri Dolgov details the 20-year technical and operational evolution of autonomous vehicles, explaining Waymo's foundation AI models, custom robotaxi hardware, and the rigorous engineering required to safely scale commercial Level 4 driverless fleets globally.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. John holds 24.8% of the talking time here. How this is scored →
John as informed peer 4.9 Guest teaching 4.7 Guest disagreement 1.5 John pushing back 1.5
speaking balance: gold is John, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 48:43 Rejection of the L2 to L4 continuum
Dolgov flatly rejects Collison's premise that consumer driver-assist systems and robotaxis will converge on autonomy from both ends, stating they are fundamentally two different problems.
Hardest push from John ▶ 49:45 Collison presses on bridging driver assist to full autonomyCollison refuses to let Dolgov's dismissal pass without clarification, directly challenging whether an engineering team truly cannot work its way up incrementally from L2 to full self-driving.
Biggest teaching moment ▶ 44:50 Emergent under-bus pedestrian trackingDolgov walks Collison through an unexpected real-world case where peripheral LiDAR pulses bounced under a bus chassis to detect foot motion, revealing non-obvious emergent multimodal AI capabilities.
John holds their own ▶ 45:33 Collison articulates intermediate representation theoryCollison demonstrates sharp technical grasp by linking Dolgov's empirical pedestrian detection anecdote directly to the theoretical need for intermediate world representations over raw pixel-space models.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | John as informed peer | Guest teaching | Guest disagreement | John pushing back | Why |
|---|---|---|---|---|---|---|
| Dmitri Dolgov's Background and Journey from Russia to Google | 4 | 2 | 1 | 1 | John Collison opens with conversational biographical questions regarding Dmitri Dolgov's upbringing in Russia and his father's academic career. Collison demonstrates cultural fluency by referencing other prominent Russian tech founders in the UK. | |
| Technical Architecture and Real-Time Sensor Processing in Waymo | 5 | 4 | 1 | 1 | Collison poses a classic systems architecture question about real-time inference in an autonomous vehicle. Dolgov explains the three core sensing modalities and on-board versus off-board compute workloads. | |
| Waymo's Foundation Model, Simulation, and Teacher-Student Framework | 4 | 5 | 2 | 1 | Collison brings up online debates surrounding end-to-end AI vs modular sensor architectures. Dolgov dismisses the binary framing and explains Waymo's three-teacher framework (driver, simulator, critic) derived from a single foundation model. | |
| Technological Evolution, Scaling Laws, and End-to-End AI Architecture | 6 | 5 | 1 | 2 | Collison explores whether autonomy is purely governed by compute scaling laws or distinct algorithmic breakthroughs. He draws thoughtful analogies between tokenization in LLMs and physical world representations. | |
| Evaluating VLMs, Closed-Loop RLFT, and Intermediate World Representations | 5 | 6 | 1 | 1 | Dolgov breaks down why naive end-to-end VLM pipelines work in nominal driving scenarios but fail in long-tail safety, necessitating closed-loop RLFT and intermediate structured representations. Collison follows the simulation complexity closely. | |
| Driving Optimization Objectives: Safety, Nuanced Drop-Offs, and Freeway Complexity | 5 | 4 | 1 | 2 | Collison probes the optimization criteria beyond primary safety, highlighting the complex human-interactive edge cases of curbside drop-offs versus high-speed freeways. Dolgov elaborates on the asymmetrical risks involved. | |
| Accelerated Global Scaling, Environmental Generalization, and Driver Generations | 5 | 5 | 2 | 2 | Collison asks whether driving technology is effectively solved and now strictly a scaling problem. Dolgov distinguishes between core technology readiness and deployment engineering, detailing the architectural leap between Gen 4 and Gen 5. | |
| Sponsor Break: Stripe Agent & Commerce Suite | 5 | 4 | 2 | 2 | Following a sponsor read, Collison questions why the industry still relies on retrofitted passenger vehicles rather than custom autonomous pods. Dolgov defends the iterative de-risking approach and introduces the sixth-generation custom platform. | |
| Comparative Sensor Physics: The Complementary Roles of LiDAR and Radar | 4 | 6 | 1 | 1 | Collison asks an earnest question comparing the practical utility of LiDAR versus Radar. Dolgov provides a clear physics breakdown of photon wavelengths, spatial resolution, and degradation across weather conditions. | |
| Emergent AI Reasoning and the Under-Bus Pedestrian Detection Case Study | 6 | 6 | 1 | 1 | Dolgov illustrates emergent AI capability through a case where peripheral LiDAR sensed foot reflections underneath a bus. Collison expertly synthesizes the anecdote into the broader debate on sensor fusion and world models. | |
| Waymo Commercial Milestones and the Fundamental Divide Between L2 and L4 Autonomy | 5 | 6 | 4 | 3 | Collison suggests autonomy will be solved from both directions—scaling L2/L3 driver assist and expanding L4 robotaxis. Dolgov firmly rejects this continuum, arguing full autonomy is an entirely distinct qualitative problem. | |
| Operational Depot Automation, Fleet Management, and Universal Geographic Access | 4 | 5 | 1 | 1 | Collison asks about the intensive physical logistics behind fleet operations, including depot orchestration and maintenance. Dolgov explains the progressive automation of depots and fleet turnaround. | |
| Macro Urban Impacts: Eliminating Phantom Traffic Jams and Reclaiming Parking Real Estate | 5 | 3 | 1 | 1 | Collison and Dolgov explore the macroeconomic and urban design ramifications of widespread autonomy, focusing on phantom traffic standing waves and the reclaiming of real estate from parking minimums. | |
| Google's Moonshot Stamina, Conquering the Engineering Nines, and Final Thoughts | 5 | 5 | 2 | 2 | Collison asks whether Google began its self-driving project too early given subsequent AI breakthroughs. Dolgov highlights Alphabet's long-term conviction and explains the exponential difficulty of solving the engineering nines. |
Statements from this episode (21)
Dolgov: Waymo vehicles rely on cameras, lidars, and radars for 360-degree vision
“We use three different sensing modalities. There's cameras, there's lighters or lasers, and there are radars. You know, those are the primary ones. There are, you know, also microphones, directional, you know, microphone arrays, but those are the primary three…”
Dolgov: Waymo requires zero real-time cloud connectivity for driving inference
“Nothing real-time in the cloud, and there are some things that can happen, you know, in the cloud, But they're not required.”
Dolgov: Waymo distills driver, simulator, and critic from one foundation model
“The way we think about the, you know, building the Waymo driver it starts with a large off-board foundation model. I can imagine, you know, building a big model that understands how the physical world works and understands the, Important properties of what it …”
Dolgov: Autonomous AI relies more on metrics and data than architecture
“And of course, you know, people like to talk about architectures, but architecture, you know, is important, but really a lot of it comes down primarily to your metrics, to your evaluation mechanisms, to, you know, all of the training recipes, and of course, ne…”
Dolgov: Pure pixels-in end-to-end models cannot easily achieve scale and safety
“If you think about this entire ecosystem of the driver, the simulator, the critic, if that's all you do, pixels in, trajectories out, it becomes very difficult to do all of those three and achieve the high level of safety and performance that we require, and i…”
Dolgov: Autonomous vehicles require closed-loop RLFT, not just imitation learning
“And if all you did was kind of observing how other people drive when you trained the system, maybe observing, you know, just passively how people drive and how they interact maybe also, you know, driving the car yourself and then using imitative learning to tr…”
Dolgov: Waymo augments end-to-end models with structured world representations
“So this is where augmenting that learned representation, those learned embeddings from the encoder decoder with that, you know, more, you know, structured representation is what we do, and we find that this kind of gives us additional knobs to simulate, you kn…”
Dolgov: Freeway autonomous driving is dominated by high-speed edge-case severity
“Freeways, for most of the time, not much happens. They're very well structured, because we designed them that way. But there is still that long tail of really complicated stuff that happens where the consequences Of, you know, bed event are much more severe, r…”
Dolgov: Waymo core driving technology has no remaining foundational gaps
“I would say that we've clearly moved past the stage of scientific research and kind of deep core technology development to this new phase of accelerated global scaling and deployment. So, you know, we still have work to do, right? But I don't see today any lim…”
Waymo plans to start operating in London and Tokyo in 2026
“You know, we are planning to start operating in London and in Tokyo this year.”
Waymo Co-CEO: Gen-6 Custom Robotaxi Is Launching in 2026
“It's, you know, we're not, it's not open to the, you know, the public yet. But, you know, I took a ride in it the other day, fully autonomously, and that's coming this year.”
Dolgov: Waymo Gen-6 Hardware Costs a Fraction of Gen-5
“It is very different from the fifth generation. It is simpler. It is more capable. It is much lower cost. It's thinking a fraction of the cost is, you know, comparable to what you would get like a fancy ADA system.”
Dolgov: Waymo Gen-6 Driver Will Expand to Hyundai Ioniq in 2026
“And then we're gonna put the six generation Waymo driver on other vehicle forms like the Hyundai Ionic that's coming, you know later in the year.”
Waymo Fuses Camera, LiDAR, and Radar Encoders Jointly Rather Than Comparing Outputs
“It's not like, you know, we estimate, you know, what's happening with the world through cameras, and through radars, and through lighters, and then we compare. No, they're like, there's an encoder for camera, there's an encoder for lighters, and they all go in…”
Dolgov: Waymo LiDAR bounced under bus to detect occluded pedestrian
“What actually turned out was happening is that our peripheral lighters bounce under the bus, and there was just a little bit of very, very noisy reflection of the movement of the person's feet. That was enough for the AI models that, hey, I likely there's a pe…”
Waymo runs 3,000 cars delivering 500k rides and 4M miles weekly
“We have about 3000 cars on the roads. We're doing about half a million rides per week. It, that translates to about, you know, over four million fully autonomous miles per week. We are operating in a fully autonomous mode in 11 cities in the US and 10 of those…”
Dolgov: Driver assist systems and full autonomy are fundamentally different problems
“And talk about the technology, and I see it just as fundamentally two different problems. There's driver assist systems, and then there is full autonomy. And I think it's deceptive to think of them as kind of incremental you know, on one spectrum of complexity…”
Waymo co-CEO: Cleaning and charging at depots remain manual processes
“If it is, so cleaning today is a manual process, right? So it'll get flagged in the car, you know, we have fleet management systems, say, hey, you know, car, you know, number, you know, 300, So many needs cleaning and we'll actually on the sensor dome, we're a…”
Dolgov: Waymo driver tech will eventually reach every US USPS address
“Eventually it will. Absolutely. There's no doubt in my mind. I think it's just a matter of when.”
Dolgov: Waymo may serve low-density areas via privately owned vehicles
“And this is where, you know, a personally owned vehicle that is equipped with the Waymo driver is maybe, you know, how you will see it materialized.”
Dolgov: Each additional nine of AV reliability requires 10x more effort
“It's deceptively easy to get started, but it is super hard to go, you know, the full distance and, you know, the number of knives, right? That you have to, like, there's, The standard, you know, engineering rule of thumb that, you know, every next nine takes, …”