Oct 27, 2022 · 50m · how-i-built-this
HIBT Lab! Cruise: Kyle Vogt
0:00 / 0:00
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of How I Built This Lab, host Guy Raz interviews Cruise founder and CEO Kyle Vogt about his journey from collegiate robotics and co-founding Twitch to building autonomous urban robotaxis. Vogt details Cruise's strategic pivot to driverless fleets, its partnership with General Motors, and the engineering and safety advancements driving the future of autonomous transportation.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Guy holds 36.9% of the talking time here. How this is scored →
Guy as informed peer 4.6 Guest teaching 3.1 Guest disagreement 0.7 Guy pushing back 0.8
speaking balance: gold is Guy, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 40:17 Kyle rejects the framing of the LiDAR debate
Kyle rejects Guy's premise about Tesla's anti-LiDAR arguments, calling it the wrong way to look at the problem and pointing out Tesla's compromised sensor approach.
Hardest push from Guy ▶ 35:00 Guy confronts Kyle with public Cruise mishapsGuy refuses to let Cruise off easy on safety claims, citing specific instances of blocked intersections and traffic incidents that sparked public skepticism.
Biggest teaching moment ▶ 42:22 Kyle explains why data quality trumps raw data quantityKyle educates Guy on machine learning nuances, explaining that Tesla's massive volume of low-quality camera data is inferior to dense multi-modal sensor insights.
Guy holds their own ▶ 39:43 Guy articulates Tesla's technical anti-LiDAR stanceGuy displays deep domain context by quoting Tesla's specific arguments that LiDAR is overpriced and can misclassify harmless debris like plastic bags.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Guy as informed peer | Guest teaching | Guest disagreement | Guy pushing back | Why |
|---|---|---|---|---|---|---|
| The 2004 DARPA Grand Challenge Experience | 4 | 3 | 0 | 0 | Guy introduces the 2004 DARPA Grand Challenge and notes the donated Ford F-150. Kyle explains the core technical components and the misfortune of burning out their only steering motor. | |
| Joining Justin.tv and Building Mobile Streaming Hardware | 4 | 2 | 0 | 0 | Guy traces Kyle's transition from MIT to Justin.tv. Kyle explains the technical engineering required in 2007 to bond multiple cellular modems into a backpack. | |
| Dropping Out, Y Combinator, and the Genesis of Twitch | 4 | 1 | 0 | 0 | Guy questions why a non-gamer would spend years on Justin.tv and Twitch. Kyle explains he was motivated by distributed live video backend systems rather than the gaming content itself. | |
| Founding Cruise and Developing the RP-ONE Prototype | 5 | 3 | 0 | 0 | Guy explores the founding of Cruise and the initial RP-ONE aftermarket highway kit. Kyle details the staged fundraising strategy and early prototype built in a garage. | |
| Pivoting to Robotaxis and Overcoming Core AI Challenges | 5 | 4 | 0 | 0 | Guy asks why Cruise abandoned aftermarket highway kits in favor of city robotaxis. Kyle explains ride-hailing fleet unit economics and categorizes the three core challenges of real-time safety-critical AI. | |
| General Motors Acquisition and Safety Simulation Loop | 5 | 3 | 0 | 0 | Guy probes the risks of being swallowed up by GM after the 2016 acquisition. Kyle explains the financial and hardware necessity of the deal and details how hundreds of thousands of corner cases feed their safety simulation engine. | |
| Public Deployment Model and the Purpose-Built Cruise Origin | 4 | 3 | 0 | 0 | Guy recounts his personal ride in a Cruise vehicle and inquires about fleet economics. Kyle outlines why the purpose-built Origin vehicle is engineered for a million miles to maximize utilization. | |
| Addressing Public Perception, Corner Cases, and Road Safety | 4 | 4 | 3 | 3 | Guy presses on publicized Cruise mishaps, including cars blocking intersections and police interactions. Kyle immediately corrects Guy on the police stop and reframes the safety debate around human mortality statistics. | |
| San Francisco Operations and the LiDAR vs Vision Debate | 6 | 5 | 4 | 4 | Guy raises Tesla's vision-only philosophy and its fleet data advantage. Kyle reframes the entire debate around cost and sensor architecture, arguing that quality insights matter far more than raw low-resolution camera datasets. | |
| Market Scale, Multi-City Expansion, and a 50-Year Vision | 5 | 3 | 0 | 1 | Guy asks about competitive moats, regulatory roadblocks, and the long-term future. Kyle lays out a 50-year vision where letting humans operate vehicles will be remembered as reckless. |
Statements from this episode (15)
Harvard and Yale Students Recruited Coders From MIT Jobs List
“There was a, there's an infamous mailing list, email mailing list at MIT called the jobs list that was run by a faculty member in the computer science department. And it ended up being a place where Harvard and Yale business school students would basically sen…”
Vogt: Justin.tv streaming backpack used three modems to dynamically allocate bandwidth
“I mean, it ended up being a backpack that had all this equipment, you know, a video encoder that would encode a video stream. And digitize it. And then some software that ran, ran on a small Linux computer that would take that video stream and dynamically allo…”
Vogt: Cruise built SF-to-Palo Alto highway autopilot in three months
“We went from, you know, nothing to a working prototype that could drive from San Francisco to Palo Alto, you know, the highway stretch of that in about three months, you know, just with a scrappy team working in a garage.”
Vogt: Fleet Utilization Justifies High Upfront Autonomous Vehicle Hardware Costs
“The reason taxis made sense versus cars that you can go out and buy at a dealership is, you know, with the taxis, it'd be a fleet of vehicles that we would own and maintain. We could limit, you know, where they operated and they operate, they could operate, yo…”
Vogt: Autonomous Driving Operates on the Edge of Modern Compute
“When you're driving at speed, this system literally has About the blink of an eye, a couple hundred milliseconds to process all that sensor data, this complex AI problem, make sense of it all, and then send a signal to the vehicle. And so that real time piece …”
Cruise completed its first fully driverless SF test in late 2020
“Our first time doing that was about two years ago. So I guess I'd, that'd be I think around the end of twenty-twenty... We did our first driverless drive around, you know, a quiet part of San Francisco at night with no one behind the wheel.”
Vogt: Autonomous vehicles can become 10x to 100x better than human drivers
“With these systems, they, you know, they're already good. The ones we have today are safe, but there's no reason to me why we can't see them become 10 times better or a hundred times better than human drivers.”
Vogt: Cruise Origin is designed to last one million miles
“The vehicle we're building called the Origin is designed to last one million miles, which is four or five times as long As the average car will.”
Vogt: Cruise has driven 400,000 driverless miles without serious safety incidents
“Cruises driven 400,000 driverless miles and only there's been a few handfuls of these types of things pop up and, you know, nothing, nothing serious as when it comes to safety”
Vogt: Cruise operates about 100 driverless cars in San Francisco
“It's on the order of a hundred cars, and in the evening, they all start off in parking lots, powered off.”
Vogt: Teslas may adopt low-cost LiDAR while Cruise drops it
“I think in the future, you know, the cost of LiDAR is coming down to the point where they basically cost the same as a camera, and so I wouldn't be surprised if in the future you see Teslas with very low-cost LiDARs, and you might see cruise vehicles with very…”
Vogt: Tesla autonomous fleet data is high-volume but low-quality
“And so you've got Tesla that maybe has a large amount of low quality data. It's just camera data from a bunch of vehicles. And then you have crews and others that have a lot of Very high resolution, high quality data from these lidars and radars and other sens…”
Vogt: Autonomous vehicles will not become a commodity or winner-take-all
“Our view is that you know, the handful of companies that you know, do this first and start these cycles of iteration and learning, Are probably going to end up with the best products. And it's likely that it becomes harder and harder to catch up as these other…”
Vogt: Software and AI are no longer Cruise's biggest bottleneck
“And so the technical problems, the software and AI problems are no longer the biggest bottleneck. Now we're turning our attention back to manufacturing and scaling and building out charging and cleaning and maintenance infrastructure in cities.”
Vogt: Within 50 years, human driving will be seen as reckless
“The first is, is that I have very little doubt that self-driving cars will be Many, many, many, many times safer than human drivers to the point where it starts to look a little reckless to let humans drive cars in places where there might be pedestrians and c…”