Generalist co-founder Pete Florence and Joe Lonsdale discuss why humanoid robots are arriving in the late 2020s rather than the 2030s, and how fleets of general-purpose robots could accelerate science, reindustrialization, and skilled trades.
Joe Lonsdale is an American entrepreneur, investor & philanthropist. He has been a part of founding more than a dozen mission-driven companies, including Palantir, Addepar, OpenGov, Affinity, Epirus, Resilience Bio, and 8VC.
Timestamps
00:00 Faster Than Expected Progress
00:33 Competence Vs Mastery Frontiers
02:11 Making Goods Cheap And New
03:23 Robot...
Questions asked in this interview
2
0:33You can basically just try it on the task a couple minutes later, you have a sense: have we achieved competence?
2:38... progress in things like science or other fields where there are physical limitations on how quickly we can iterate and progress based on people or other types of things in the world needing to physically interact with the world, right?
If you would have asked me like last year, frankly, I would have said, "Well, yes, I do think these things are going to be really important to the world in the 2030s." And now, just understanding what's going on, it seems like they're going to be important to the world in the late 2020s, which is surprising to me. And I don't know what we could say about this, but how do you quantify—these things took months to train, and then they took weeks and days and hours, and I guess they'll take hours or minutes for different types of tasks. How should we be thinking about how these are kind of moving towards a point where they're clearly going to be useful?
I think there's kind of two different capability frontiers that we're mostly focusing our research efforts on today, Joe. One is in what we use the phrase competence, which means that we get a robot that's kind of good at a task, which is actually a really important level of capability, right? And so what we talk about internally is rapidly getting to levels of competence. So call that like 50%, 70%, 80% type of success rate—so not production levels. But the faster we can get to competence levels for taking a general model and having it just very quickly learn a new task, we think that that is a very important type of frontier to work on. It also is we can very quickly iterate on that, in part because it actually doesn't take that much time to measure in the real world if the robot is competent or not. You can basically just try it on the task a couple minutes later, you have a sense: have we achieved competence? But the other really important frontier is what we say these mastery levels of performance. So really actually getting the right numbers of nines of reliability, like 99.99, however many more nines percent reliability, and these levels of speed and improvisational intelligence so that we could actually deploy these models at scale in production doing all these things that we—
That's fair. I do like the idea of just being able to create goods—you have energy and you have materials, and then you have goods for people, if you think about for the middle class or for poor people. Suddenly they should be able to afford a lot of stuff they couldn't before. That's one side of it, but then the other side of it you're saying is there's actually things we could never have made that are really important for humanity. What are examples of that?
But I think the things that we can't even make yet or things that we can't even dream about yet, those types of things. I see that very much happening in part where the ability to just fundamentally advance progress in things like science or other fields where there are physical limitations on how quickly we can iterate and progress based on people or other types of things in the world needing to physically interact with the world, right? Science, I think, is a great example and one that's close to home for me. Actually, before I started working on robotics, I was doing research in sustainable energy. I was trying to make solar cells that could be orders of magnitude cheaper to scale than silicon. This is what I was working on over a decade ago before—
Just as one example. As I would try and think about how could I make my research work in that field 10 or 100 times more efficient, I just needed 10 or 100 times more me, physically running all these experiments in the lab. Of course, different types of scientific automation have existed for decades, but very generalized capabilities to be in a science lab and use your hands to pull off all of the manipulations that are needed to run all these scientific experiments, that has always been actually one of the bottlenecks of just fundamentally advancing science. And that's one where it's like, "Okay, now we can just do 10 times more science per year than we could before." And then the breakthroughs that could come out of this, it's hard to predict.
It's historically been post-doc student bottlenecked or PhD student bottlenecked, right? Part of it here could be like you could turn every lab into a high-throughput lab, and you don't have to buy specialized—like a high-throughput furnace or a high-throughput pipetting machine. You wouldn't have to spend hundreds of thousands of dollars in CAPEX for these high-throughput machines, and have a general purpose embodied intelligence with cheaper arms and cheaper hands just doing all these things.
Lonsdale, J. (2026, September 3). Robotics Genius Predicts the Next Decade [Interview transcript]. Joe Lonsdale. CEOInterviews.AI. https://ceointerviews.ai/interview/1303530/
MLA
Joseph Lonsdale. "Robotics Genius Predicts the Next Decade." Joe Lonsdale, 3 Sep. 2026. Transcript, CEOInterviews.AI, https://ceointerviews.ai/interview/1303530/.
BibTeX
@misc{lonsdale2026_1303530,
author = {Joseph Lonsdale},
title = {Robotics Genius Predicts the Next Decade},
howpublished = {Interview transcript, Joe Lonsdale. CEOInterviews.AI},
year = {2026},
month = {sep},
url = {https://ceointerviews.ai/interview/1303530/},
note = {Speaker-attributed transcript with timestamps}
}
