What if robots stopped doing just the simple, repetitive work and instead took on the most important, life-changing jobs in the world?
I talk with Fred Parietti, Co-founder and CEO of Multiply Labs, about how advanced robotics can transform pharmaceutical manufacturing by automating the complex and high-value tasks that save lives.
You'll like this interview if you're interested in robotics, automation in healthcare, and how technology can tackle some of the toughest challenges in next-generation medicine.
Episode Links
Transcript
[00:00:00] Fred Parietti: Try to convince a scientist to bring a teleoperation rig into a clean room, right? and then use, you know, not their own hands, but a teleoperation rig to manipulate the cells. They're going to tell you you're crazy, right? Yeah. And I can tell you how many robots I saw that make pizzas. And I'm like, guys, right? Like, you know, there are two kinds of pizzas. Either pizzas for which minimum wage labor is fine and there are pizza for which you really want the human experience of someone.
[00:00:32] Audrow Nash: like an incredible chef. Yeah.
[00:00:35] Fred Parietti: And in both cases, the value per pizza is very low. The same amount of mass in terms of like gene edited cells for actually saving people's life, has a value of a few million dollars. Not like the $20 of a pizza, right?
[00:00:53] Audrow Nash: Alright. Hi, Fred. Would you introduce yourself?
[00:00:56] Fred Parietti: Yes. Hi Audrow. It’s great to be, you know, here. Yeah. My name is Fred. I'm the co-founder and CEO at Multiply Labs, which is a robotics company that develops, automation technology for manufacturing next generation pharmaceuticals.
[00:01:15] Audrow Nash: Oh, go ahead.
[00:01:17] Fred Parietti: My background is in mechanical engineer engineering, and so I've always been a roboticist. And so, you know, it's very exciting to be working in the space where, you know, you started and where all the things that you always wanted to build are.
[00:01:34] Audrow Nash: Hell, yeah. Yeah. It's so cool. And we met. And, like. I don't even know, it might have been like 2014 or something. And I came to MIT while. You were doing your PhD. You gave me a tour of your lab, and you were working on supernumerary limbs, so, like, extra arms for people in manufacturing. How did you get from here? From there to here. Like, what was your path and how did. You discover pharmaceuticals as an interesting area?
[00:01:59] Fred Parietti: Yeah, that's a very, very interesting, maybe unusual story. But, it's meet me. My, my dream has always been to to to build robots. And that's all I always wanted to do. And I guess I knew I was not patient enough to be a professor. So that was like, off the table.
[00:02:17] Audrow Nash: Yeah.
[00:02:19] Fred Parietti: But, you know, a robot, I want to build the, like, the most advanced possible robots. And so I always knew that was very expensive. Right. And so. Okay. Like, they must be very useful. They must do something that's extremely valuable. I must be absolutely sure that the robots I'm building are doing something super valuable, because that would justify their cost. And the cost of all the amazing engineers that need to hire to work with me and build the robots. And. And so during their PhD, I met our co-founder, Alice, and she was not a robotics person. She is a now she's a professor. At that time, she was also a PhD student, in pharmaceutical technologies. And and she told me how, manual pharma manufacturing is, and I could not believe it. I just assumed that making drugs was like making chips. So I assumed it had already been automated.
[00:03:17] Audrow Nash: You mean like, silicon chips? Like computer chips kind of thing?
[00:03:20] Fred Parietti: Exactly, exactly, exactly. Right. And so, but it turns out that factoring is even more valuable than silicon, you know, semiconductor manufacturing, it's probably the only thing that's more available on a per month or per volume basis. And yet, the from the point of view of manufacturing it, it's still entirely manual, especially for the most advanced drugs, which are the most, you know, difficult to make. And so in other words bio biological drugs. So therapy is gene therapy is RNA. Those are just very hard to make. Most of them, not all of them, but most of them. And I just could not believe it. And so I was like, wow, this is definitely justifies developing extremely advanced robots. You know, it's a huge need, for the technology. There's a huge value because nothing is more valuable than saving on improving lives. Right. And and and and ever quires precision. It requires sterility. It sounds like a perfect application for robotics. It also looked at that time like a pretty empty market meaning that pharma companies didn't have and still don't have engineering departments. They're really focused on approving the biology to the FDA and then selling the drugs, but they don't have engineers. And most robotics companies at the time, like we're talking about like 7 or 8 years ago, right? Most robotics companies then, were focusing on lab automation. That was something that was non-regulated. So they were focusing on non FDA stuff because they were afraid, you know, of the FDA. And we thought, wait a minute. Like FDA compliant manufacturing is where the value is because those are the drugs that go to patients. And those are the drugs are you really need to be sterile. So they are the ones where robots are the most useful. So that's how you know, you know, it came to be. And so we would start in multiple labs to address that need through the most cutting edge possible robotics. So very different from my PhD, but I have to say, the PhD research on super supernumerary robotic limbs. You know, it told me, you know, everything I know about robotics. And so, it was just about. Okay, how do I apply that? In a way that can help, the most people and quite frankly, let us develop the most advanced possible robotic technology. I like to us, you know, I'll be building robots anyways. So it's about what are the most useful robots I can build. Provided that I'd we also building them, you know, for free on my own. Right. So, so that's how it started.
[00:06:11] Audrow Nash: Totally. And yeah, by going for the highest value area, then. You have the most justification for the most advanced robot and the best people like, so you can pay their salaries and all the things like. This.
[00:06:24] Fred Parietti: Yes I, I just hate to be constrained by all. There's no budget for that sensor. For that material or I hate to be constrained by oh I somehow can not work with the best people and you know in these, you know, environment with these application both constraints are gone. Like I know I can you know, do a very convincing pitch to the best engineers to be here. And I know everybody in our team is just, you know.
[00:06:53] Audrow Nash: Super good.
[00:06:54] Fred Parietti: And, you know, at that level and then at the same time, I never need to tell them, oh, you know what? You know, I would love to develop that technology, but we can't. It's too expensive. Like it's never the problem in such a high value field. If the robots work, pretty much any price will do because the robots are, you know, basically making such valuable things.
[00:07:17] Audrow Nash: Yeah. Yeah. That's that's so interesting. And it's interesting to me because I have never talked. To another robotics company in the pharmaceutical industry for this kind of thing. I've talked to, like, I don't even know, like maybe it's 100 or 100 and 50 or 200 startups at this point. And it's like, I don't think I've ever met another one that's working in the pharmaceutical space. Why do you think that is? Just they're scared. Away by the very large frustrations of dealing with the FDA, or they're just not aware of. It, or why do you think that is?
[00:07:52] Fred Parietti: I think there is maybe three things. One, yes. Regulatory. There's a regulatory fear. Be, biology and, biology. It's not like engineering yet. And so it's very scary. Scary to engineers.
[00:08:11] Audrow Nash: Why is that? What do you mean?
[00:08:12] Fred Parietti: And three. Yeah. So, and the third one was, is just very different from the day to day experience of most people, right? So, yeah. So, but let's let's go back. So first regulatory, right? Yes. Traditionally, robotics people are in engineers in general, mechanical software engineers. They're afraid of the FDA. They feel like it's something that they cannot control. And investors are also terrified by the FDA and.
[00:08:41] Audrow Nash: Song cycles and. Everything.
[00:08:43] Fred Parietti: So the venture world is divided in two, and there is people that invest in FDA stuff, life science investors, and then every everything else is lumped on and they're tech investor. So. A tech investor can invest in nuclear fusion or in financial services. And that's totally fine. But they will not touch medical drugs.
[00:09:03] Audrow Nash: Yeah.
[00:09:04] Fred Parietti: Going to the FDA. So like that's how scary that is. So that's was the first obstacle. Luckily from our point of view we actually provide the technology to the pharma companies. So it is the pharma companies that still approve the drugs and around their facilities. But at least we were not afraid to operate in this era, in this market, which I think a lot of people are still held back by that fear. Yeah. The second part was by essentially biology. So, yeah, to an engineer, biology is like black magic, right?
[00:09:42] Audrow Nash: I agree. Yes.
[00:09:44] Fred Parietti: It's in honestly, even to a scientist is just from a scientist point of view. It's all they know. So they I think they think it's normal that everything is black magic. But but.
[00:09:53] Audrow Nash: You know, it's different than your point of view.
[00:09:55] Fred Parietti: Right? Right. Like, I can, you know, honestly, like it is easier to go to mass like we know we can go to mass is just how many million dollars it would take, right? Yes. But we know we can do it right? Finding a single drug that treats Alzheimer's. Nobody on the planet can guarantee that.
[00:10:16] Audrow Nash: Right. You don't know how. Big of a sinkhole that is needed.
[00:10:20] Fred Parietti: And I would say more money, as many more billions have been invested in to find a cure for Alzheimer's than going on to Mars. And I think we're, we're going on Mars, probably. Wow. And so yeah. So like, biology is hard. It's it's basically alien technology that we don't fully understand. And, and so it is extremely scary for an engineer to put the success of their company and their technology in the, in the hands of substantially opaque, mysterious alien technology, i.e. biology. Right. So people much rather do. Hey, if I develop these whatever payments you know app I know I can do it is just a, you know, a market. You can build a crank that makes money. I feel like much more easily in biology, you know, you might have the best science in the best current knowledge, and you do the clinical trial and it fails and you're screwed. Right. So I, I don't know if we can say that in the, in the podcast. I will say.
[00:11:22] Audrow Nash: Whatever. It doesn't matter, I swear. Do whatever you like. You know.
[00:11:25] Fred Parietti: You are in trouble. You're in deep, deep trouble. Your company might be worth zero. Like, not -10%. It might be worth zero. Even with the best science behind it. Because the biology is not yet fully understood.
[00:11:40] Audrow Nash: Yes, but so what you're doing. So our number two, what you're doing. Is something different. You're not developing the drugs per se. You're developing the technology that makes the drugs. So like you're selling the shovels to the companies to go build. They're trying to do the gold rush of drug development through. But it's a.
[00:11:59] Fred Parietti: Very it's a very safe shovel that we're selling, meaning that because we know and deeply respect. Right. We are we are humble engineers, meaning that we never walk into their homes and tell the pharma companies, you know what, you're going to change your biology in order to feed it to our robots.
[00:12:19] Audrow Nash: Right. We do literally the opposite.
[00:12:22] Fred Parietti: We're like, hey, guys, we know robotics is easier than biology. Therefore, we will build a robotic system that adapts to your process. Right? So, so that if you spent ten years developing this process, and if the only thing by trial and error that you know works, we are not going to modify it. We will teach our robots to replicate it of course, more efficiently, more with more.
[00:12:46] Audrow Nash: And with robots.
[00:12:47] Fred Parietti: Too. Right, right. They're robots. They do. They do that very well. But we adapt. We adapt the lower risk area robotics to the higher risk area biology. The other way around would not work so effectively. Our technology is baked in. It. The assumption or the knowledge that biology is black magic and you do not want to mess with black magic.
[00:13:12] Audrow Nash: Yeah. Hey, that's so funny. Yeah, like, we. Like, have these spells and we know they work. And therefore we do that over every time versus.
[00:13:19] Fred Parietti: Finding that spell.
[00:13:21] Audrow Nash: Yeah. Took years ever of effort.
[00:13:24] Fred Parietti: From the brightest minds in the space. And in most cases they actually discovered it by accident.
[00:13:30] Audrow Nash: Yeah. It's amazing that that seems to be a frequent medical thing where it's the discoveries by accident where they just like, notice something and they're like, why is that? Okay. And then point three, did you want to touch on for in more detail or.
[00:13:46] Fred Parietti: Yeah. Is the difference with the lived experience. So a lot of robotics people, obviously they are absorbed by robotics. And so they never think so. They're like well what is what are my daily problems except for robotics? I need to follow my clothes, clean my house right hook.
[00:14:02] Audrow Nash: Because I need to, you know, you know, live. Yeah.
[00:14:05] Fred Parietti: You live and I don't have time because I need to work on the robots. And so I think that's very normal for them. And it was normal for me as well. I had absolutely no clue. It was more intuitive for me to develop wearable robots then to develop, you know, robots.
[00:14:21] Audrow Nash: For pharmaceutical.
[00:14:21] Fred Parietti: Processing, because I had to meet our co-founder, and she told me I would have never come up with that idea on my own.
[00:14:29] Audrow Nash: That's awesome though. I mean, it's like you, you stumbled on a massive problem and then you went, That could be really valuable. And then you like that can make it so you can remove those constraints you mentioned. And then you went into this area and it's really awesome. I feel like. The way I've been. Thinking of it for a while is it's like technology or like, I don't know, computers, robots, these kinds of things. And biology are like the two things that the future is built on. Yeah. And I feel pretty knowledgeable about robotics. I know nothing about biology, and it's cool that you sit right in the middle, of this, I don't know.
[00:15:07] Fred Parietti: It's very exciting. We love it. We can learn a lot. We can. We get to be at the edge of, you know, you know, organic technology, robotics, hardware and software. And that, by the way, also enables, you know, you know, AI and advanced, you know, learning techniques and so on. But also, you know, we interface with biology, which is a whole new alien technology that we don't understand. We get to learn basically from the brightest minds in biology. And totally it's great because in their own there is a lot of respect. So in engineers are very intimidated by biologists because, you know, they are magicians, right? And they're they're really like they are like kind of like a wizards, you know, and, and but but what I discovered is that the biology is, are also very intimidated by the robotic engineers.
[00:15:59] Audrow Nash: Yeah.
[00:15:59] Fred Parietti: Yeah. And so as long as nobody else tries to drown out the other voice, you can work together and be very productive and everybody feels like they're learning a ton, and everybody loves it.
[00:16:08] Audrow Nash: Hell, yeah. Yeah, I would imagine, because it's it's super cool. You have all the experts going, oh you could do that. Like, oh. That's super cool. Oh, we could. Do this for that. That'd be a lot of fun. Tell me about it, because I don't know almost. Anything about pharmaceuticals. Tell me about what you mean by next gen pharmaceuticals. Like, what are they? You mentioned them at the beginning. But how do you make them? What are the processes involved, like, conceptually? How should I understand these to understand later what your robots are doing.
[00:16:36] Fred Parietti: Yeah. No. Super important. And you know, keep in mind if there's someone that is scientists or biologists listening, you know, keep in mind that this is a mechanical engineer interpretation and simplification. Right.
[00:16:52] Audrow Nash: So having said that.
[00:16:54] Fred Parietti: You know, effectively, you know, there is a degree of complexity in all medicines, you know, you know, you know, drugs. And so the simplest one are the ones that are they're called small molecule, you know, drugs and they're essentially chemicals that are so simple you can actually synthesize them with chemical reactions okay. Right. So small molecules, just a few, you know, limited number, a series of atoms. Right. And you can just, you know, like it is chemistry. You just mix it together artificially. You can start making, with that metal, some more complicated, molecules that are typical of life, you know, like, you know, like, yeah, I guess some, you know, some biological molecules. Not just simple small molecule chemicals, but what you can synthesize artificially is very limited. When you actually get into, like, antibodies that are, you know, gigantic, you know, from the point of view of just how, how, you know.
[00:17:57] Audrow Nash: Like, like structure, right? Yeah.
[00:17:59] Fred Parietti: Like you actually cannot synthesize that stuff.
[00:18:01] Audrow Nash: You have to.
[00:18:02] Fred Parietti: Use a genetically engineered organism and convince it to make that thing for you. So you're you're using life as a factory because our technology is not advanced enough to synthesize those type of molecules.
[00:18:16] Audrow Nash: Okay. So you said small molecule. So that's the first one. The second one, this is this one that we're synthesizing with life as a factory. What's that one called.
[00:18:26] Fred Parietti: Large molecules. Right. Or or or and and and and in that for example, a typical example is antibodies, right? You you don't synthesize that chemically. You ask a genetic engineer organism right in a, in a that to make as much as possible. And then you blend them, you destroy them and and then you filter out only.
[00:18:46] Audrow Nash: The antibodies and of.
[00:18:47] Fred Parietti: Those that, that target, you know, crazy, entity. And but it gets worse.
[00:18:53] Audrow Nash: Okay.
[00:18:54] Fred Parietti: What if you want what if your drug is not a molecule, however big? What if the drug is a cell?
[00:19:03] Audrow Nash: Okay, so just like leaving cell. Yeah. So my knowledge is super molecules would. Make up cells. Is that true? Like you would have a cell being composed of many molecules. Okay I don't I don't know how they're related. You can see.
[00:19:19] Fred Parietti: That that way. Yeah. Basically many I guess millions of them. And so and so because the cell is an organism, it's now living it can duplicate. Right? Yeah.
[00:19:29] Audrow Nash: So how do you make those for sure.
[00:19:31] Fred Parietti: You know, the outer tone of things. Right. And so so again we cannot synthesize cells from scratch. Right. And so in the case of cell therapy you start from the cells of the patient. Right. Like you know a blood sample or a sample of inner tissue or the sales of a donor. So the raw material is itself living. And then you genetically engineer that right to perform a particular function.
[00:19:56] Audrow Nash: Crazy. Okay. And when we say we are genetically engineered what does that mean? Like you expose it to specific conditions that make it express in a specific way that's desirable. Is it something like that, like you expose it to this new exposure to that and somehow it turns it into whatever it is you want?
[00:20:14] Fred Parietti: Yeah. You know. Exactly. And so, so, for example, cell therapy right now, the most, you know, the most common ones that are already a few of them approved, you know, about a dozen. And so, you know, you gene modified cell therapies, you start from a blood sample of the patient, right? You separate out a subset, typically of immune cells, but it can also be other cells. And then.
[00:20:38] Audrow Nash: You.
[00:20:39] Fred Parietti: Modify them genetically, either using a virus that that delivers the payload to those cells. Wow. And modifies, you know, them or you can also modify them in other ways, like zapping them with electricity and then the so that, you know, if there is that, you know, basically, the modifying agent around them is synapse in. Yeah. And so, and then the cell now, you know, has been modified. So if your cell, if you're the patient or the donor, it has now been modified genetically. So it can maybe get a new receptor.
[00:21:15] Audrow Nash: Or in your body doesn't rejected or see it as a threat. And then.
[00:21:18] Fred Parietti: There's that. Or you need to give it a new sensor to recognize that cancer or that particular tissue that you wanted to target.
[00:21:26] Audrow Nash: Cool. So I see why. You're saying this is black magic. This is crazy. Yeah.
[00:21:31] Fred Parietti: It is. It's happening. Like, I don't know right now.
[00:21:33] Audrow Nash: I have no idea how people would come to these ideas of how to do these things. And I suspect you're about to list even more complex ones, too, which is not.
[00:21:41] Fred Parietti: I mean, these is kind of like the pinnacle. And the idea is, you know what? If you could edit that cell in multiple ways and so giving them multiple, you know, it stores or, you know, so effective you have to in the cell like a little robot. Yeah. But it's a robot we cannot build. It's only a robot we can modify.
[00:22:01] Audrow Nash: We can configure it through exposure to specific.
[00:22:03] Fred Parietti: Yeah. It's almost like, I don't know, like you have a horse and you give that horse, like, night vision goggles or whatever, right? And now I can see. But you, we don't have the technology to actually fabricate that horse on our own, right.
[00:22:15] Audrow Nash: Yeah.
[00:22:16] Fred Parietti: We we but we can modify, I guess, genetic code to develop night vision. And so.
[00:22:22] Audrow Nash: And I love the metaphor and.
[00:22:24] Fred Parietti: Basically after, you know, after you've done that, then you need, a few millions to a few billions of those cells. So you let them duplicate and then you infuse them back into patients. So now the patient has a better immune system if their immune system. But now it sees, you know, the illness and can target it.
[00:22:46] Audrow Nash: Wild. Okay. So those. Are kind of the three types small molecules, large molecules and cell. Therapy.
[00:22:53] Fred Parietti: They said that B two is gene therapy. So in some cases you actually give the virus containing the modification directly to the patient. And that's then when you want to do editing, you know, in, in, in, you know, more broadly. Right. Or if you think the virus can target that part, of, you know, the body of the patient, honestly. And so and then, of course, there is RNA. RNA does not have a permanent modification of, you know, the genetic code. Okay. But it does tell the sales of the patient,
[00:23:31] Audrow Nash: To do some.
[00:23:32] Fred Parietti: Instructions to. Yeah, to make different proteins on the Z. So it's basically kind of like temporarily tricks them into behaving as if they had a different genetic code, but they don't. And so, so those are all like, you know, that's it. There's no chemical reactions. And so if you don't have chemical reactions, but you need to convince or modify, you know, living beings, cells, right, to synthesize this stuff, then your process becomes much less like chemistry and much more like cooking or growing plants. Right. Excel that you don't need to be sterile when you're cook, right? In this case, if your batch of cells has even a single wrong cell, imaging like a bacteria should not be there. Yeah, well, it can be.
[00:24:21] Audrow Nash: In the old.
[00:24:21] Fred Parietti: Fashioned right now. It becomes dangerous as opposed to, you know, healthful and curative. Right. And and it's not like a normal drug. If you're making a bunch of chemicals, you can sterilize it. All right. Or in fact, if you're taken by mouth, you don't even need to sterilize it because your stomach is sterilized, right? Yeah. It's amazing. But you can't sterilize a batch of cells because you are going to kill the bad as well as the good ones.
[00:24:48] Audrow Nash: Yeah. So the only way.
[00:24:50] Fred Parietti: To do it is to do this very manual kind of culture process, in a completely sterile way, from day one to the final day. And without making a single mistake. Single mistake means you need to throw it away.
[00:25:04] Audrow Nash: Wow. And so give me an idea of. The, like, timescale and number of steps for these kinds of things. Like is it like, because if I'm cooking, like, I could be done in an hour, which I assume that these processes do not take an hour, I assume they take a while. How how long. Roughly, or. For some of.
[00:25:23] Fred Parietti: The shortest we've seen, if, you know, take a few days. And so maybe hundreds of individual tasks and steps. Wow. Typical length right now is 1 to 2 weeks and thousands of individual steps.
[00:25:37] Audrow Nash: Wow. And these are currently being. Performed by people, like, in lab suits that are it's kind of like factory work, but for pharmaceuticals in the sense. Like, you can think of it like a factory or it's, it's, it's is to is to.
[00:25:52] Fred Parietti: Qualified, you know, the type of workforce you need to, to consider, like a factory. Yeah. You're in clean rooms, so it's not a lab. It's a clean room. Yeah. And so you need people that are super trained.
[00:26:05] Audrow Nash: And probably, like, all have at least a bachelor's in some sort of biology.
[00:26:09] Fred Parietti: Or minimum. And in many cases, like master masters, PhD PhDs. Wow. I mean, we've seen like MD, PhD who's working on this stuff. And so and obviously it's never their dream to be a manufacturing operator in a clean room. Yeah. I mean, it's a mess. You need to gown up, right? Because you are a risk to do.
[00:26:28] Audrow Nash: You have a lot of bacteria and all sorts of contaminants coming from you all the time? Yeah. For sure. That's wild. So it's an expensive workforce?
[00:26:37] Fred Parietti: That doesn't want to do the job.
[00:26:38] Audrow Nash: That does it? Yeah. That hates it. But they are needed and they're probably paid pretty well. Is it true? Because the it's like undesirable. It's hard work. It's is it repetitive. Because they end up doing the same processes.
[00:26:53] Fred Parietti: Super, super repetitive, but also repetitive and stressful because it's repetitive. But if you mess.
[00:26:58] Audrow Nash: Up at all.
[00:26:59] Fred Parietti: You throw anything, anything like you could touch the wrong thing there on the surface for like half a second.
[00:27:06] Audrow Nash: It's that wow, that's crazy.
[00:27:09] Fred Parietti: And or breathe on it or sneeze or whatever. Scratch your forehead and you didn't realize now your glove is contaminated. And so, so it's expensive as operators, but from the point of view of these people that.
[00:27:22] Audrow Nash: They don't even like doing it. Yeah.
[00:27:24] Fred Parietti: They would like to be in drug discovery as opposed to drug manufacturing or maybe process optimization as a, as opposed to process execution. So from the point of view of the employer, they're expensive. But from the point of view of the employees, undesirable, underpaid with respect to the ideal job.
[00:27:41] Audrow Nash: Okay. Yeah. They don't love it. And I wonder, so that sounds interesting. You have all these people doing this. I was thinking factory is in, like a series of steps where people are doing repetitive work. Not necessarily the skill, because there's some skilled manufacturing also. But so it's, it's kind of like a pharmaceutical factory in the way I kind of defined it right now. So for the.
[00:28:08] Fred Parietti: Simpler drugs can be scaled easily. So you can make like, you know, 5 million tablets of aspirin. But for this new biological next generation therapeutics. Oh, it's so manual. It is so manual. And by the way, if you try to change the process and simplify it, most likely the cells are going to die in or the results are going to be different so that the the FDA now needs to look again at.
[00:28:31] Audrow Nash: 3 to.
[00:28:32] Fred Parietti: 5 approve it. Right. So the best time.
[00:28:35] Audrow Nash: Consuming and expensive everything. So imagine it.
[00:28:39] Fred Parietti: You go to one that your favorite restaurant in the world. And there's only that chef in that location with local ingredients that makes that dish. And now imagine, oh, okay, you want to replicate that restaurant in your backyard in a, in a, in another, in another continent. Yeah. Right. And so it's not just about following the same recipe. You know, you also need first of all, you need to have the same skill of the chef, you know, and you know, in the same implicit knowledge that they had about their process. But the thing is you have that you need the same ingredients. Yeah. And even the same tomato seed. If you primed it outside of your house, as opposed to the other continent where different different sun, different soil, the tomato is going to taste different. So any change the process when life is involved, right? When biology is involved results inevitably in something that is different taste right and that it's good if you're talking about wine or food, that you can appreciate the variation. But if you're talking about a drug that goes inside of.
[00:29:44] Audrow Nash: You, you.
[00:29:45] Fred Parietti: Don't want to see, oh, what's the flavor of the day.
[00:29:47] Audrow Nash: Right? Yes. Yeah.
[00:29:49] Fred Parietti: And so that's why it's so hard to scale biology. Bioprocessing.
[00:29:55] Audrow Nash: Yeah. Because it's like what I'm imagining. Is like a really difficult optimization problem to go from beginning to end process or to the medication being. Developed. And you can introduce uncertainty at every level when you change anything. So it's like yes, you have drugs made in one spot because they've learned how to make it there and they can quality control it a lot more easily. And they've nailed down their supply lines and all sorts of things like this. So you don't make it somewhere different. So it just sounds very. Very it sounds like a very. Thorny set of problems with a. Lot of implicitly. Defined relationships between all the steps that we don't understand if they're important or not. So it makes it very hard to automate, from an engineering perspective where we just go, you kind of like, give this wiggly path that gets from one spot to another, and we just want to make a straight line from there, from start to end. But they need to go through all those wiggly steps because they're not exactly. Sure where. There's the errors that they're avoiding.
[00:30:59] Fred Parietti: Exactly. Actually, that's exactly it. Like it would be simple for an engineer to say, oh, it's a wiggly path. Let me draw a straight line. And but that would assume you know what you're talking about in terms of biology. And because this biology, we need to assume that every wiggles is motivated even when you don't know why.
[00:31:22] Audrow Nash: Yeah, right. And so the best performance fence. Have you ever heard of that idea.
[00:31:27] Fred Parietti: Oh no. No. What's that?
[00:31:29] Audrow Nash: You cannot take down a fence unless you know. Why it's there. This kind of thing. I don't know. I've just one. One colleague mentioned that quite a bit. So, like, you don't take down the thing unless you know why it's there and. That is it.
[00:31:44] Fred Parietti: That is it. And and in biology you know. Yeah. Anybody on the fence looks very irrational. But you just need to have you know a lot of respect honestly for OSI how many scientists year, year went into finding out that things that miraculously work. Yeah it happens to works and so and so our job is okay. How can I learn how to make that fence with a robotic technology? And and and. Yeah. And so it's, it's, it's a honestly like to me, it's actually no different from, you know, the good there's two other good examples, like, like a couture dress or, you know.
[00:32:30] Audrow Nash: I don't know what it is.
[00:32:32] Fred Parietti: Like, you know, like a, like a high fashion. Oh yeah. Right. You know, there's, there's dozens of artisans that do that. Right. And then the stitching and so on. Right. And so you couldn't simplify it a lot. Right. But you not look the same.
[00:32:48] Audrow Nash: Right.
[00:32:49] Fred Parietti: Because you know, like they did all of that to appeal to the taste of the time for that season or whatever. Right. And so and so how can you do that at scale, how that, you know, now it's clearly not possible in high fashion to get to every single person the same quality and scale. You need to teach a robot to be as skill as the artisan. Right? Because any simplification would come at a cost. In terms of, you know, in that case, their, esthetic, I guess.
[00:33:23] Audrow Nash: Yeah. It's not the same thing basically, as that was being made.
[00:33:28] Fred Parietti: Correct. And so we need to do that in biology. And in fact, the price point and the value and the cost of an advanced next generation drug is similar to the top high fashion, you know, dresses. Yeah. And yet our challenge is to make that high fashion available to everybody and affordable to everybody. And so the level of refinement you need in robotics to take that product and not change it, not to ruin it, but still scale it. That's a big robotics challenge, I bet.
[00:34:00] Audrow Nash: Yeah, it sounds very, very hard. Okay. So I feel like I get. That making drugs is really, really complex. And there's a lot of steps and it's very difficult and time consuming. And you need sanitation. The whole way through. What are your robots do? How are they involved in this whole thing? What kinds of tasks are they doing and how how are you working with the pharmaceutical companies for different drugs?
[00:34:30] Fred Parietti: Yeah. So, Essentially the most important thing for us is not messing with the biology. And so instead of designing new instruments or new ways to manipulate, you know, the cells or the biology, we teach the robots to do exactly what the scientists are doing right. And so we teach the robots to use the same instruments, the same reagents. Reagents are typically, you know, liquids containing, you know, you know, reagents or molecules that the cells need, right? And consumables, those consumables are typically sets of boxes, bags, tubes, the, the sales things you go through with right. And you use them once and you throw them away because they need to be sterile. Right. And you can never have the sales of two patients being in contact. Right. So so the most important part for us is, robots that use the same instruments, the same consumables and the same reagents. So we have a very modular system because every pharma company uses a different set of instruments or consumables or reagents. And so we basically surround imaging like a robot, a set of robotic arms surrounded by a set of modules or cubes, kind of like a shed to shelves on the sides. And these shelves are made by modules, and each modules contains an instrument that was used by people originally and now is being operated by a robot. So to teach the robots to do that, and it depends if an instrument is the bottleneck, we can duplicate or triplicate or more that particular module. If the robotic arm is the bottleneck, we can duplicate or triplicate the robotic arms. And in fact, right now in our latest robots, we actually have four robotic arms at the same. Oh. Cool. So I'm back at multiple robotic arms.
[00:36:19] Audrow Nash: You're back? Yeah.
[00:36:21] Fred Parietti: Just in a very, very different context.
[00:36:25] Audrow Nash: Yeah.
[00:36:25] Fred Parietti: So yeah. So yeah, my mom was like, oh, you're back at doing robotics. I'm like, mom, I never stopped doing robotics.
[00:36:33] Audrow Nash: Yeah.
[00:36:34] Fred Parietti: It's just happened. So that now it looks kind of, you know, like before, but obviously they're not wearable anymore. And they are doing something very different.
[00:36:42] Audrow Nash: Yes. Okay. So you have all these robotic arms. And they have all these bins and in these bins they might have tools or things that they'll interact with. And then you have them doing I suppose they have like a little work station or like a work site where they're manipulating something. And how does all that go?
[00:37:02] Fred Parietti: Yeah. So you know what? I also have a quick video here that I wanted to show you and will describe it for the people listening. So here it is. Okay. Okay. And so, so, yeah. So, on the left you can see, the manual process. So scientists doing stuff by hand and on the right you can see the robotic process. And in particular, a robot robotic arm on a rail.
[00:37:30] Audrow Nash: Mat.
[00:37:31] Fred Parietti: Surround is surrounded by these bays or modules, right. And every module, every kind of like space opening on the sides of the robotic arm, contains a particular instrument. And so the robotic arm is handling syringes and vials and so on. And the same way that the person is doing it. But now in a much more repeatable and sterile way, and so, yeah, so, you know, you kind of see like human touch points, people are touching everything, of course, when they need to, you know, handle stuff manually. And, but the robots has zero human touch points and human touch points are the number one source of contamination and errors.
[00:38:11] Audrow Nash: Right.
[00:38:12] Fred Parietti: And so obviously the robot doesn't breathe, right. It's much easier to filter the air, you know, and control the air quality inside the robotic system. So it's effectively a much, you know, more repeatable, and more sterile process. And so these example only contains a single robotic arm. But I have another example with oh four.
[00:38:39] Audrow Nash: So I can see that.
[00:38:41] Fred Parietti: Okay. Here it is.
[00:38:42] Audrow Nash: Oh, they're on the ceiling. I love it.
[00:38:44] Fred Parietti: Yes.
[00:38:44] Audrow Nash: So ceiling is such a good spot for robots.
[00:38:46] Fred Parietti: Yes. We need to cram so many robotic arms that we need to put a few on the ceiling than the few on the floor. The ones on the floor are also on a rail. Sometimes it is also a vertical axis. Yeah, the ones on the ceiling are stationary, but they are still quite helpful. And yeah. So here you can see train moving. Oh, by the way, you see these imperfections that we need to inject to convince people that, you know, this is not a render. So we always add at least that, you know, if a frame or a sequence with people around them.
[00:39:17] Audrow Nash: So people see that this is true. Super cool.
[00:39:22] Fred Parietti: And so yeah. So these are the multiple robotic arms and in this case, you know, a single robotic arm could do the same, right? Yeah. But it would be slower.
[00:39:32] Audrow Nash: Yes for sure. Yeah. You might as. Well make it faster for sure. And then it becomes. A more interesting problem too because you have to parallelize them. And then you get to work on some of that awesome tech that you like for all of this exotic.
[00:39:44] Fred Parietti: So what about, you know, you know, for example, avoidance of collisions, right? What about optimizing the plans for multiple robotic arms that they're facing with, you know, a dozen or two different modules so that if the planned space becomes huge, exponentially complicated. And so there's a lot of optimization there that, you know, our software team, you know, works on. And yeah, so this is this is all very exciting. But it's also very, very valuable and very impactful application. And so we just enjoy that intersection between top robotics and also top, you know, good impact, positive impact in society.
[00:40:28] Audrow Nash: Oh yeah. Love it. That is so cool. Okay. So I see that these. Are doing a bunch of similar tasks to what people would do. The. Thing. Yeah. Go ahead.
[00:40:38] Fred Parietti: Yeah. It's a it's it's even them it's more than just similar. Sorry I stopped no. Go ahead now. It is like they need to be the same equivalent. They need to be statistically equivalent by virtue of measuring the biology and the effects on the cells. If the robots are not statistically equivalent to the scientist, we are messing with the biology and we have a problem.
[00:41:03] Audrow Nash: Cool. So how do you. Evaluate if they're statistically equivalent? I'm imagining like you look at the results and you make a distribution of it in some form, and then you see if the distributions look similar. So it might be like we generate this many cells for this batch. And this many of them are of high quality. And then you compare across many runs. And if the distributions look similar then you say that's the same or good. Is that is it something like that.
[00:41:28] Fred Parietti: It is exactly that. But it is another layer of complexity, which is biology is very random, predictable, even with these, you know, checks in places. And so what you need to do is you need to start from the same cells, donor cells, human cells.
[00:41:45] Audrow Nash: And so you have them do. It and you have you compare a and a kind of thing.
[00:41:50] Fred Parietti: So and we do at the same time from the same donor, from the same cell bag.
[00:41:55] Audrow Nash: And you do the same room and you see that you get similar.
[00:41:58] Fred Parietti: The main process in the robotic process, because if you did it with different starting cells, the variability might be caused by the cells. Right? It it because maybe one patient was different from the other. We know not probably. And so you need to always add a control right there.
[00:42:15] Audrow Nash: So you just split in two. Yeah. Yes. That is super cool. Very clever. I imagine with enough data you wouldn't need to do that approach. But the thing is, if you are. Just starting or if it's expensive or time consuming, say you have like. A. 300 step in process, where you don't want to do it to get there every time. This approach probably makes a lot of sense to kind of converge on the results quickly and tune your systems quickly.
[00:42:46] Fred Parietti: Yeah. And you know, from the engineering point of view, it might look like a lot of work, but from the, biology point of view, like is like, oh my God, we can't believe we just need to prove equivalence. This way. And it only takes two months of testing. So that's really like the.
[00:43:04] Audrow Nash: Huge wind.
[00:43:05] Fred Parietti: Speed of light for pharma. Yes.
[00:43:07] Audrow Nash: And wow, by the way, it's so funny.
[00:43:10] Fred Parietti: Yeah, yeah, yeah. It's crazy. And so we're the only company that has ever published, peer reviewed data, showing equivalence between a manual process and robotic process. Right. So, like, nobody else has a, you know, published is people have, you know, white papers and all this stuff, but like, we went on a peer reviewed, you know, you know, journal publications. And so, so it's very solid data, because that's the kind of credibility you need, you know, I work with companies. Yeah, exactly.
[00:43:43] Audrow Nash: Yeah. They just wouldn't use it. It would be like an R&D thing, versus something used in production.
[00:43:51] Fred Parietti: Correct? Correct. And the whole point for us is the using production.
[00:43:54] Audrow Nash: Yes. Very cool. So I'm imagining this isn't so I guess what's the scale of your. Operation or deployments or. How how many robots or. Cells are there like work cells like this are deployed or. How, I don't know, how large would you say? I don't know exactly what the metric would be. But how big and how much deployment do you have? So.
[00:44:24] Fred Parietti: So I can't be too specific because we only announced two big partners, and I don't want people to reverse engineer, you know? Yeah. The, the details of those collaborations I see. But what I can tell you is that this year, our number of deployments will be, you know, the robot you saw before. That is a pretty big robot, by the way, with a four robotic arms we are talking about, you know, 30 by ten feet, in area. So hopefully so it is a big multi-million dollar, you know, a robotic system. We are going to deploy these here in the 5 to 10 year range. Okay. And we already deployed several of them. And so.
[00:45:12] Audrow Nash: Okay. So you have like. Tens of systems deployed that are of similar size. So that probably.
[00:45:18] Fred Parietti: The idea, yes, is that we are effectively already sold out for the next two years.
[00:45:23] Audrow Nash: So congrats.
[00:45:24] Fred Parietti: We know it's because there is so much demand for these therapies. You can't make them by hand. It's just not possible. So now it's on our problem. Yes. Okay. We need to address how do.
[00:45:34] Audrow Nash: You scale.
[00:45:35] Fred Parietti: Chain right to double in next year and then double again the year after. Right. Wow. Just to get through the backlog. Well, that's and soul.
[00:45:45] Audrow Nash: Gazing, though. I mean, I'm sure that that's a very strong position for investors that want to come in and like, you guys are in great terms, for seeking investment, for finding more, customers, all sorts of things.
[00:46:00] Fred Parietti: Yeah. I mean, right now, the kind of the paradox is that when people reach out and they're like, hey, guys, we would really like some robots right now because they're always like, when they reach out to you, there are oh, no, yesterday, the robots yesterday. Totally. Like, okay, there is about a, you know, two year waiting list right now.
[00:46:19] Audrow Nash: Okay. Wow. Oh, that's so exciting.
[00:46:23] Fred Parietti: So we basically need to, to work very closely with them to extend the supply chain, maintaining the same quality. Right. And then find ways to make it happen. And, yeah. Yeah. And so, it's I think it's very exciting. Is it, it reflects the thing where if you go and look for a available application of robotics, like the market is there, right? And, you know, I really sometimes don't like them. So I don't like wishy washy, argument that sometimes people have. Well, you know, oh, we are robotics company and the robots will eventually be used for everything. So our market is infinite. So you don't need to worry investor about our revenue today because we're doing general purpose robotics. And so our market is infinite, right. And that I think, is too many times using that as an excuse to justify like zero revenue. Yeah. Right. To justify a poor business. You know, it's like the Silicon Valley show, when there was that joke, you know, and it's HBO, there was a joke that, you know, they said, you know, the you must try to stay a zero revenue for as much as possible.
[00:47:40] Audrow Nash: Because zero.
[00:47:41] Fred Parietti: Revenue is better than revenue because it lets you paint a gigantic.
[00:47:45] Audrow Nash: Infinite potential. Yeah.
[00:47:46] Fred Parietti: Right. When you start having revenue, well, you actually measured on how much is it growing. Right. And so we are definitely on the camp that you should have available application measurable. You know, growing real you know, revenue. Right. But but I'm seeing you dating too many times right now in robotics, the sort of like infinite potential kind of like hype approach. And I hope that.
[00:48:11] Audrow Nash: Yeah, yeah.
[00:48:13] Fred Parietti: I just hope that that doesn't create a bad name, you know, for robotics, because when the bubble bursts and it will burst, right? Only the people with real revenue, you know, will have a business will survive.
[00:48:24] Audrow Nash: Yeah.
[00:48:24] Fred Parietti: But I hope that people don't conclude, oh, robotics is like that because, you know, the work we are doing is so real.
[00:48:30] Audrow Nash: Hell, yeah. Yeah. I think my impression, I mean, if you. Look like the.com bust or something, it's like, yeah, there was a bunch of bubbles. But clearly the internet is huge now. It's like, I think that robotics will be similar for this, where there are very clear applications for robotics, where you're making a ton of money for companies. And those will be the ones that carry everything. The hype. I'm a lot more skeptical about. I feel similar to you about the general purpose robots for this kind of thing. Yeah, it feels like when I was in academia, One thing I thought was really funny was that a lot of the research that was very cool and like, you'd see a quadrotor doing awesome things. It's like, well, why do we do this? And it's like search and rescue. Why do we do quadrupeds? It's like search and rescue. Every. Everything that couldn't. Even even, like. Pipe robots or, blow up robots. It was like, oh, search and rescue for sure. It was like the catchall. And now it's it's similar as a general purpose robots when we don't know exactly what the market is we're going into.
[00:49:38] Fred Parietti: Yeah. And I can tell you how many robots I saw that, like, make pizzas. And I'm like, guys, right? Like, you know, there are two kinds of pizzas, either pizzas for which minimum wage labor is fine and there are pizza for which you really want the human experience of someone making, like, an incredible. Right? Yeah. And in both cases, the value per pizza is very low. The same amount of math in terms of like gene edited cells for actually saving people's life, has a value of a few million dollars. Not like the $20 on a pizza. Right. And so like like, like it doesn't make it in the technology we're using is substantially the same, right? Yeah, we're using imitation learning, machine learning and reinforcement learning. We are using only collaborative robots. So it's literally the same technology. In one case it is a business. In the other case there's not a business.
[00:50:33] Audrow Nash: Yes. Yeah. We were talking. Before we started recording like for house cleaning and stuff where it's like the humanoid will clean your house. It's like. But. A person that you pay $25 an hour, we'll also clean your house, and they'll probably do a better job, and they're not going to fall over. These kinds of things I really like your perspective on go right to the high value task.
[00:50:58] Fred Parietti: Versus mean.
[00:51:00] Audrow Nash: Go low end. Yeah.
[00:51:02] Fred Parietti: It's exactly it. And and so these you know, our goals is you know, our goal is to develop a very cool robot. All it is mine is right. And so I'm like okay so I need you know I want to the value of the best possible robotic systems. Right. And so I'm like which is smarter going for $20 an hour all is equal. Or going for $20,000 an hour or maybe $200,000 an hour, which is honestly like the order of magnitude we're talking about for next gen therapy manufacturing. Right. And so if you can have a business that is for order of magnitude better, literally, right, 3 or 4, right? Of course I can develop better robots higher, better engineers. Right. And it makes sense to start from there. Now, I'm not saying that we're never going to get to, you know, humanize it. Clean the house. Of course, I'm looking forward to that I can't wait. Very happy to buy for 30 K, a humanoid that cleans my house. I would love it. Right. But, you know, at the same time, you know, it's way more reasonable to start from high value. And then 20 years from now, get 2 or 10 or whatever. Five, certainly, though not today. Get to those mass manufacturing or mass or sorry, low added value tasks. Right. And we need to also avoid the risk that we promise that the human is for the House already today and two years from now, people get bored of it, right? Oh, of course is not happening. And so they then conclude that all robotics is BS and hype. Right? And I don't want to get into a post .com bubble situation for robotics. Right. And I think with AI, with digital, I like, you know, like large language models, they are useful enough that I am pretty sure there will be some disillusionment, but is now going to go back to zero. Yeah, but the ChatGPT moment for robotics has not arrived yet, and so I just don't want to go back into a robotic winter because people were promising cheap humanoid to sweep the floor and could not deliver businesses that could do that.
[00:53:16] Audrow Nash: Yeah. I wonder, my my perspective on this is thinking. About the autonomous cars and all the hype around that. So in like 2015, it was like next year we're going to have autonomous cars. It's going to change everything. There's going to be no more truck drivers, no more Uber drivers. All transportation will be revolutionized. And then it's like, okay, in 2018, it's probably going to happen. And then in. 2020, whatever. And so now, but then. What you see actually. Is the public just totally. Loses interest. Yeah. It was so happy. But I don't think it's maybe maybe the investors because actually, if you look. At the funding, over the last five years in robotics, it was majority autonomous vehicles. And then that went to zero in the last little bit. But, because talking to some investors who were on reports, they've shown me some of the reports, I had an interview with F Prime Sanjay from it where we talked about this and, but the, the slice of more specialized robotics applications, kind of like what you and I seem to value is just gradually increasing. And you have this big overhead. On top of it for the autonomous cars that then goes to zero when the hype dies.
[00:54:34] Fred Parietti: Yeah. Yeah. No. Absolutely, absolutely. So your your finishing.
[00:54:40] Audrow Nash: Yeah. But the thing. That I think that's interesting about that is it feels like so we have a Tesla and it just drives itself everywhere. It makes some mistakes. But so there was so much hype ten. Years ago for this. And then in. My opinion, it's like Tesla quietly solved it. And there's no attention on it. Now that it's solved. It's not totally solved, but it's really, really good. Where I can drive anywhere in town and I just ding it on to autopilot and it does it almost free. But I wonder if it'll be similar to this. But. Yes. What do you think?
[00:55:15] Fred Parietti: It is? True. But keep in mind. So who survived? All right, I Waymo Waymo and Tesla.
[00:55:23] Audrow Nash: Yeah.
[00:55:23] Fred Parietti: Right. Pretty much right.
[00:55:25] Audrow Nash: And Zoox I don't know much Zoox and a cruise. I think is basically dead. Yeah.
[00:55:33] Fred Parietti: And so I mean I had several friends working there and nobody works there anymore. And so, nobody of my friends. Right. And so, and so, so the idea here is, when, you know, when you hype and overhype all these companies were in. Right, you know, in kind of like hyping this so much, then the public then is already very jaded, you know, because, you know, when you get promises, a lot of things that don't happen, I think it's natural to be skeptical, right? Yes.
[00:56:05] Audrow Nash: For sure.
[00:56:05] Fred Parietti: And so public loses interest, investors, even though they love to say that they are decoupled from public markets and but sentiment. Yeah.
[00:56:14] Audrow Nash: Right.
[00:56:15] Fred Parietti: Like carbon copies. In fact, I would say they are amplifiers.
[00:56:20] Audrow Nash: Yeah, I think so too. Right.
[00:56:22] Fred Parietti: Like they are they try to be as sensitive as possible to the public trends.
[00:56:26] Audrow Nash: Yep.
[00:56:28] Fred Parietti: And so basically the investment dries up and 99% of companies die. The only ones who survived are Tesla. Tesla. Because basically Elon can raise infinitely cheap capital. Right. But that's not a industry thing. That's an know like, oh, he has infinite access to capital. Therefore he can fund says anything forever. Right. And that.
[00:56:52] Audrow Nash: And they just make good. Cars anyways too. So like there's even if they didn't do the self-driving, it's like it's still a good car.
[00:56:59] Fred Parietti: Cool, right? But the self-driving on its own would have not been a yes even in business. Right? It was a sense to say it was saved by infinite access to capital. And the same for Waymo, right. Because it's Google with Google.
[00:57:15] Audrow Nash: Yeah.
[00:57:16] Fred Parietti: Right. In the same for a Zoox that was bought by Amazon. So the only right like and so like so basically when when disillusionment sets in post hype.
[00:57:28] Audrow Nash: Everyone dies if they don't attach to a larger entity that funds them.
[00:57:33] Fred Parietti: Unless you have inflated money. And obviously that's not a reasonable plan.
[00:57:37] Audrow Nash: For a startup.
[00:57:38] Fred Parietti: What I meant to do robotics and hype it up, and then try to attach myself to an infinite source of money, like that's not a business plan. And so I have.
[00:57:46] Audrow Nash: Some businesses. Are bought, made to be bought, which I don't really like, but that it is happening.
[00:57:52] Fred Parietti: But then say that you just want an exit as rapidly as possible because you know you're going to die. And the only way to not die is to be bought. Right? And so and so. But to me, you're much stronger if you're actually trying to build something that matters. So apply that technology in a way that, has a, a sustainable business. That's what we're trying to do. And that's what I worry a lot of people are not trying to do with these infinite market pitches. And, and so when the winter comes. Right, you know, then I think a lot of companies would be caught like that. And if it's too many, the public will conclude that robotics is useless or not ready yet. And I don't want to repeat another self-driving car winter.
[00:58:41] Audrow Nash: Was it that bad? Because from my perspective, a lot of the smart. People from the self-driving car industry went into other robotics startups. They started companies in agriculture and mining and like a bunch of really other good robotics companies. I don't know that it was like. Because as. I mentioned, the investment across. Other like more specific robotics applications. Has just been growing over the last years. It's it's a fraction of what was poured into autonomous cars but is still growing.
[00:59:13] Fred Parietti: It is true. It is true. But but I don't.
[00:59:15] Audrow Nash: Want to winter too.
[00:59:16] Fred Parietti: But I also like, if I want a self-driving car right now, I either need to have a Tesla. Right? But then, you know, with the caveat that you still need behind the wheel.
[00:59:28] Audrow Nash: Yes. And you have to pay attention.
[00:59:29] Fred Parietti: Right. Or I need to jump on a Waymo, which I do all the time here in San Francisco. Right. And, but it's it there's not there's not.
[00:59:38] Audrow Nash: Yeah. You got the big. Players and that's it.
[00:59:40] Fred Parietti: So it's kind of like a little disappointing that I can't get a reliable self-driving taxi everywhere I go. And that's basically I think that was totally within reach. Because if Waymo did it, if Tesla, you know, is almost doing it right, then many other players could have done it technically. And so I think they just messed up AI from the point of view of the hype and the business.
[01:00:04] Audrow Nash: Yeah, that's a good point. I wonder my my thought with. Humanoids and I wonder what you think is that it's going to go kind of like autonomous cars. Where. It will take longer than everyone thinks. And the public will lose interest. And then Tesla will quietly make a 30 K humanoid that can actually do useful stuff. But it might be 15 years from now or something, maybe faster. Who knows? Do you think it'll go similar for this? And it'll be kind of like the companies either get bought on maybe favorable, maybe unfavorable terms by larger companies. And then a lot of them go out of business, but Tesla kind of survives. Or maybe who knows what the other ones will do. But what what is. Your thought on this?
[01:00:50] Fred Parietti: I think it's more likely than not that it would go that way. However, I don't want it to go that way.
[01:00:56] Audrow Nash: And because it's too centralized. Yeah.
[01:00:58] Fred Parietti: We're doing. Yeah yeah yeah. But also like, you know, new winter and so on and you know, and then the winter. Yeah. Ten years later. And so no like we don't want it to go that way. And so we're actually doing everything we can to not go that way. So for example we got a humanoid robot for 30 K. Yeah. Not from Tesla. They're not selling it yet. We got it from unitary. All right. And so I can tell you that the hardware you can already find the hardware for reasonably, you know, a reasonable cost like 30 K. Right. And so I assume that Tesla and figure hopefully will get there at some point.
[01:01:34] Audrow Nash: That they're selling. You're saying or that they get to 30 K.
[01:01:38] Fred Parietti: That they, that they hopefully get to sell them for 30 K. Reasonable. Right now though, the only company can do that is unitary. And so, so that's what we got. So the constrained AI now is actually not the hardware. The constraint is the software that nobody, like. The the humanoid is not very useful even though it's fully functional Uber wise. Yeah. Because you need to, teach it to you useful stuff. And so we are basically experimenting to use humanoids in conjunction with our robotic systems. And the idea is, inside our robotic system is completely autonomous with all those robotic arms. Yeah. And the human outside good enough. Exactly. The outside. So the biggest bottleneck right now is loading and unloading. That's. Yeah, manually.
[01:02:26] Audrow Nash: I guess that that's pretty funny.
[01:02:28] Fred Parietti: Yeah. Right.
[01:02:29] Audrow Nash: And so if.
[01:02:30] Fred Parietti: You now have people kind of bored to death loading and unloading a bunch of carts, you know, and cartridges into a robotic system, well, can I use humanoid? I think the hardware is good enough. And so now I'm not pitching to you. Potential customer potential for my company. I'm not pitching. You know what? I'm going to the humanoid, and I'm going to use it to do a bunch of, like, far fetched stuff that we all know doesn't work yet. Now I'm going to tell you we will use the general purpose humanoid that only costs 30 K to pick and place boxes. So instead of having, you know, 2 or 3 people, you have like ten cheap humanoids doing that slowly. But I can do it with current technology, and that's actually very valuable to you, because now your cleanroom doesn't have people, it only has robots loading other robots. Right.
[01:03:18] Audrow Nash: Yeah.
[01:03:19] Fred Parietti: And so I think we are the current technology can be deployed effectively with a solid business model in those applications. And I believe if there are enough of these applications, we can perhaps avoid the winter and disillusionment where maybe people will be, oh, okay, you mind for sweeping? The floor is still far, but I can see human as being very useful for these high value add, you know, applications. And so and so. Maybe it's not all to throw away. I really want to avoid that, because that sort of public disillusionment slows down the technology a lot. And I don't want the development to slow down.
[01:04:00] Audrow Nash: I wonder though, I mean I, I love the. Perspective and I think it's so cool that you can have a, you. Said do you think, do you think there were. So the humanoid loading and unloading the other robotic system. How much per hour of value do you think they're creating? Just because I think you said a very high number, and I want to I want to see if that's right, because that is crazy and inspiring. If that's the case.
[01:04:26] Fred Parietti: Oh, the loading unloading is not that. So like if we're talking about tens of thousands of dollars per hour, you are considering the throughput of the robotic system. Keep in mind of this single cell therapy right now is like around 400 K all the way to $2 million per dose. Wow. So you're talking about a box of cells like, you know, imagine like a leader or, you know, roughly a third to a quarter of a gallon, right? Yeah. A little milk.
[01:04:53] Audrow Nash: One of those little boxes.
[01:04:55] Fred Parietti: Is half $1 million, and we're making a lot of those things in parallel.
[01:04:59] Audrow Nash: Holy cow. So, so.
[01:05:01] Fred Parietti: Imagine, like, we make, you know, dozens or hundreds of them in parallel, and, you know, they take, you know, a few days to a couple of week, it clicks. Right? So the average throughput per hour is actually tens of thousands of dollars. Wow. Loading is just you know, you could load them manually. So I would say 2 or 2. But you can.
[01:05:19] Audrow Nash: Extend the whole system in some sense. You're making your system have softer boundaries with the rest of the other world. Pharmaceutical company. Yeah, the rest. Of the world. So that's really cool. And that adds a lot of value because your system could just run continuously. Yeah. And you don't have to rely on people and hiring issues. And all sorts of other things that might occur and sanitation risks that come with people for all this.
[01:05:43] Fred Parietti: Exactly, exactly, exactly, exactly. And so and so in and so there's, so there is a place for humanoids. And now you might argue, what right do you really need the legs? Not we don't need legs. We just.
[01:05:55] Audrow Nash: I was going to ask.
[01:05:55] Fred Parietti: But two arms. Right. Yeah. You don't need to face. Those are just added for esthetic purposes. Right? You can put those. They're pretty.
[01:06:03] Audrow Nash: Cheap. So, like, why not? You don't have to worry about manufacturing it. That's actually a big valuable thing.
[01:06:08] Fred Parietti: Honestly. We are also evaluating different form factors, for, you know, general purpose, the loading and loading, general purpose robotic part, and provided value is cheap. And we have two arms because you need two arms to pick and play stuff. We're fine. Right. It's just that technology. We believe we've found a very valuable application that I can sell today. You know, with it.
[01:06:37] Audrow Nash: And in the future. Yeah, yeah, yeah.
[01:06:39] Fred Parietti: And that I know customers will pay for it because it's very available. Right. If I was pitching you. Oh, you know what? I'm gonna ask a humanoid to clean your car. And you would have so many questions, and it would be unclear.
[01:06:53] Audrow Nash: If that would be how much you'd be willing.
[01:06:55] Fred Parietti: To pay me. Right?
[01:06:56] Audrow Nash: Yeah. Yeah, that would be funny. Cleans your car, does an okay job, and it costs five K for this kind of thing or something, I don't know. Yeah, that'd be funny. Okay, so that's really cool. I like that. Perspective a lot. I hope we can avoid any winters from that kind of thing. I think getting robots out there is probably the way to do. That with this. It does seem. Like humanoids in a pharma lab toting things around while it is high value and it does involve humanoids. Maybe your your videos and things like this will expose it more, but like, I didn't know anything about this really before. How do you expose? Because there's so much, like. Humanoids doing side flips, videos and things like this. Versus humanoids doing useful work in. It's like like, how do you how do you. Attach this to the public perception of humanoids so that it doesn't burn when humanoids are not useful and then tomorrow or in 101, one year or whatever, we're promised.
[01:08:06] Fred Parietti: Honestly, to tell the truth, I don't care about the public's perception. Yeah, I don't care about the public's perception. The idea is I don't do these because I want to shape the perception. I will do this because I want to develop the best possible robotics and right now, that means using humanoids for simple stuff that you can simulate, like picking place. And that means using their non-human eyes. The multiple robotic arms inside our system to, carry out the complicated tasks the same way the scientists will carry them out and so, super important for us to learn from expert demonstration. And my point of view is if I build something that works, produces, you know, life saving therapies and is obviously commercially successful, that will itself.
[01:08:59] Audrow Nash: That's a proof of perception.
[01:09:01] Fred Parietti: I am not trying to chase on a line to some, you know, ideas that the public got from, you.
[01:09:07] Audrow Nash: Know, something else and everything.
[01:09:09] Fred Parietti: Yeah, right. I guess.
[01:09:11] Audrow Nash: To fickle innocence.
[01:09:12] Fred Parietti: Or I guess I am the type of chase that idea. But it's in my mind and, you know, and I'm the robotics guy, and, you know, this is I'm busy robots, right? I get to see how they look like and what they do. And, Yeah. And if the public wants to be different robots, you know, they're welcome. I love more robotics companies, but, but like, but but it must work in the end. And sometimes chasing too much. You know, I, for example, I hate when people edit their robotic videos and they have a ton of cuts. Yeah. You know, and I'm like, sure, you know, you're just chasing an idea in the minds of people and it will turn against you the moment they realize it. It doesn't work. Yep. And so, yeah, for me, number one is I need to build something. I'm confident it will work.
[01:10:01] Audrow Nash: Oh yeah. So one thing that I wanted to make sure we talk about. So how do you train your. Robot to do things like the human does. So you mentioned that you can do statistical analysis and you take the you split the material and then you look bad before and after or at the human's version in the robots version. And that's a way of kind of seeing if you've matched it. But how do you, teach your robot to do these different steps?
[01:10:30] Fred Parietti: So thank you so much for asking that question, because it's probably our number one priority right now from the robotics R&D point of view. So the idea is, okay, you know, the robots work, they manufacture the drugs. You know, we're going to, expand the supply chain. So we can make, you know, as many as the market wants. But then those robots, when they are deployed in pharma companies, they need to learn a lot of processes, you know, tens, hundreds of processes. And because biology is black magic, right? You never know what the Weezer's the biologists are doing in their process. And they there tends to be a long tail of ten, 15%, 10 to 10%, let's say, of tasks that we've never seen before. Right? Like they shake a ball in a particular way, right? Or they squish a bag or remove bubbles in a particular way. Right. And, and those are basically artisans. They're almost like artists at work. Right? There's that particular skill scientist that somehow has figured out that if you do that to the cells, the cells will have that reaction, which is sometimes key to the performance of the drug. And so our problem after we deploy enough hardware is how do we make that hardware maximally useful to all these processes and scientists, right of companies. Right, without requiring an infinite number of engineers? Because if we go to ten customers, to a hundred customers, I cannot ten next engineers, right? Yeah, for sure. And send them to learn from the scientists. Right. So this is actually a very Tesla like situation where once you deploy a ton of hardware, there must be a way to use that hardware to collect data and then auto add the new features, right. So that effectively every new deployment increases your database and increases your functionality so that you get better and better by virtue of your scale and the answer to that for us is called imitation learning. Right? So effectively what we do is we collect from the pharma companies, they send it to us, right? We hundreds or thousands of hours of, videos of scientists performing the process, which is very easy to collect in the current manual manufacturing floor. Right. It's just imagine you have a bunch of people doing their job. You just get a lot of video of that plentiful easy chip to collect. And then via imitation learning, we extract from those, you know, long form videos, dozens, hundreds of examples for every single task. And once we have the examples, wow, we can then train a robot to either control policy by imitation learning or other techniques. Yeah. To effectively, replicate that task within the, you know, the bounds of their manufacturing facility. Right? Or, you know, or the bounds of that specific process. And, and effectively, the robots are able to auto learn those tasks. And, and this is super important because you must, like a lot of people ask us, well, can you just simulate, you know, that where you can't because.
[01:13:50] Audrow Nash: It's a black.
[01:13:51] Fred Parietti: Magic thing. Yeah. The biology is absent from the physics based simulations. Yeah. So what? We actually reverse engineering is the implicit model that the scientists have in their mind of the biology, and that informs their action. This is not written sometimes it's not even done at an aware level. Yeah, but the best scientists kind of, like, have figured out what works for the cells.
[01:14:15] Audrow Nash: Yeah.
[01:14:15] Fred Parietti: And we need the robots to figure that out as well. And modern, you know, imitation learning or machine learning or, you know, AI techniques, work very well to discover intrinsic patterns, provided that you have enough data. So that's why the video data, video data is plentiful enough. We can get enough to be able to extract implicit knowledge with any less plentiful data. You probably are not going to have enough to extract the implicit knowledge.
[01:14:45] Audrow Nash: Yeah. So that's such an interesting idea. So video the reason. For video is because it's accessible. It's because it's easy to set up a camera and to record it and to have them do it, or they might already be doing it. And then you can take that, chop it up, label it, do things. I it's really. A very cool thing. And it. Seems like so many. Companies are working on a similar problem, which is how do you teach robots to do things in a general way. So you think video is the good way to do that? How do you how do you map what the human. Is doing to what the robot should be doing in video? Like I'm imagining you run a pose detector that assumes what the person is doing in 3D, and you kind of infer from there, and you have to maybe have human annotators. Slice the video. So you say this. Is when they're doing this step and you kind of file it somewhere so you can group those steps. Similarly or yeah.
[01:15:47] Fred Parietti: Honestly, like, very, very good question. So so yes, we do need to slice the videos. Yeah. We believe that can be done at least in this context automatically. And we already have very very promising results there. Awesome. So the idea is that, luckily a fundamental fat training floor is much more controlled than the general environment. And so it actually possible to train, you know, latest generation video algorithms to auto segment, and in fact, it's even better, like you can actually provide, examples of non tasks so that those are like auto labeled on their own. Yeah. But then the algorithm can also say, oh, actually this thing that was not similar to any of your examples, this thing has been repeated 100 times. So it is perhaps you know.
[01:16:39] Audrow Nash: And so there's similar there's similar. Paths through state space in every demonstration. And so that's you're learning something more general from it.
[01:16:48] Fred Parietti: Exactly. And so so that's super important.
[01:16:51] Audrow Nash: The super cool.
[01:16:52] Fred Parietti: The auto segmentation. And then there is yes trajectory extraction. And then there is the training of a control policy. And we have very specific, you know.
[01:17:01] Audrow Nash: Pipeline for this tools.
[01:17:02] Fred Parietti: And pipelines for all of them. When we can, we use off the shelf, when we can't, we also develop our own stuff. The most important part here is plentiful data. And so the problem is that, so video is good because it is very simple to manipulate. And there are a lot of libraries to do it. But I can't go on YouTube and download 1000 hours of cell therapy manufacturing right here. So so that's the problem for which I think robotics is being slower than, you know, other is that, you know, if I was training, you know, whatever I am for like legal language where there's so much legal material that I can either download or scrape or, I guess, steal that, judging by how a lot of those companies got the data. Yeah. And so but nobody can stop me. Right. And that's very good data. Right. There's no way to get that data. So I need a way to get that level of internal size data cheaply and scale in a scalable way. So ideally I need my partners to send that data to me without me being involved because it would be too time consuming for me to personally generate.
[01:18:12] Audrow Nash: It's like text. Data, as you were saying, like a test, like on a moment where it's it's like you make a pipeline that just aggregates everything and then learns from it. Yeah.
[01:18:20] Fred Parietti: And you find a way for your partners to basically pay you to get the hardware that then gives you the data. And of course, they need to agree to give you the data very confidentially. Right? Much. The difference is that we are regulated and Tesla is not. But but but still like this is no different. And that's basically my one, you know, the main, main, you know, problem with Teleoperation. I just don't believe their operation scales the same way. It's not so cheap, is not so plentiful. And by the way, try to convince a scientist to bring a teleoperation into a clean room right in order and then use, you know, not their own hands, but the robot to manipulate the cells. They're going to tell you you're crazy, right? So like in operation, I would have zero data. So obviously, you know, there's a ton of.
[01:19:10] Audrow Nash: One data bias.
[01:19:12] Fred Parietti: Right? If you're lucky. And then there's a ton of problems using videos. So what you are saying is the change of context, you know, from, you know, a human demonstrating it to what the robot should do to replicate that. But that is a technical challenge. And if I solve it, I have something scalable and that's what I like to have in terms of technical challenges.
[01:19:32] Audrow Nash: Yeah, yeah. Because I guess you can just say. Well, what's the hand of the human doing? And you can probably label it like, oh, they grabbed. This. And then they shook that when it's in their hand. And like these kinds of labeling things, you could have the robot do, you can break it down into actions the robot can do. And then that would let you do this mostly autonomously. Like you can onboard new skills fairly easily, which is quite cool.
[01:19:56] Fred Parietti: Precisely it. In fact, do you really care what the human hand is doing provided you know its effect on the bottle on the bio?
[01:20:05] Audrow Nash: Yeah, not at all. So you can.
[01:20:07] Fred Parietti: Even just track the objects that are in contact with the cells.
[01:20:12] Audrow Nash: Oh yeah, that's way easier. Yeah. For sure. Okay. Yeah. It's it's way. Simpler problem planning too, because you don't need to emulate the human arm. You can just emulate what the is doing or whatever it is. I mean.
[01:20:23] Fred Parietti: Effectively they're related because if you're trying to emulate what the human arm is doing, probably having a structure similar to the human arm lets you do it correctly. So that's why we only use collaborative robotic arms. We're cool. Yeah. Six degrees of freedom because they are close enough. The kinematics were never limited. Yeah, you can always copy people.
[01:20:43] Audrow Nash: Yeah, yeah, I would imagine. If you like. Move it similarly, whatever the object is. But I mean, I don't understand the problem that well then you would get a similar effect. But I don't know. So that's really cool. It's a very clever way of approaching this. I also feel like if you guys nail this, there's a lot of other wonderful areas that you can go and solve to. Yeah. Even though pharmaceutical. Yeah, yeah. Even though pharmaceutical is such a big industry and it's so like you they, they don't have engineering teams generally as you were saying. But yeah, tons and tons of other areas would benefit from this same approach.
[01:21:24] Fred Parietti: I agree with you. And this is a cutting edge, I think, of, robotics research and, we just found what we believe is the highest possible impact and value application. But I agree with you that the ripple effect of this technology industries are huge.
[01:21:39] Audrow Nash: Yeah. For sure. Okay. So let's see, getting. Towards wrapping up. One thing that you've brought up a bunch is high value problems. So as opposed to having like a humanoid robot sweep the floors or something like this, have it do incredibly important pharmaceutical work, in a complex system and. Just high value work. How do you find how high. Value problems for this kind of thing? Because that I think a lot of. People who are looking to start companies want to find high value problems and then it's you can go talk to people and things, but it's hard to understand what are the true. High. Value things, what are the real bottlenecks of the people that you're trying to serve? Do you have any any thoughts on. How to find these high value areas?
[01:22:32] Fred Parietti: I mean, I guess from the manufacturing point of view from a very like first principle, you know, point of view, the most high value problems are the ones that have the largest market value per like cubic centimeter, or the.
[01:22:51] Audrow Nash: Ram or whatever. Yeah.
[01:22:53] Fred Parietti: Of, product. Right. And so by that metric, the two highest value products in the east of mankind are semiconductors and pharmaceuticals.
[01:23:01] Audrow Nash: All right.
[01:23:02] Fred Parietti: Way more valuable in gold. So that's basically how we found it out. And another another way for a high value problem is, would be to kind of evaluate, you know, what is, I guess the size of, of deals you can get, right? But then again, like, you need to divide it by the number or volume or weight of products. Yes. Another area that's, that's very high value is, but it metric is maybe aerospace. But I do believe that even a plane per gram or per unit volume is way lower. Yeah. So probably the only other one would basically be military. But from our point, from my point of view, I was like, you know what? I'm not going to make robots for that application. And I much rather basically, make a robot that save lives than the other way around. Yeah. And in, in in the end even obviously it is all arbitrary. This is just my personal preference. But I do believe that these, this stems from the fact that, saving or extending life is more valuable than taking it away. There is fundamentally, you know, we are alive and we want to be alive and be healthy and be happy, and we're willing to pay way more for that. Yeah. Then to somehow kill someone else. And so fundamentally, I think it's a better business. But this was more, you know, basic decision for me. I want to robots that help people.
[01:24:37] Audrow Nash: Yes. Totally. Yeah. Yeah. It's nice to work on something that adds value to people. And I it's like if you don't have health, you don't have anything. Really. Yeah. So yeah, I think adding making it so that we can extend the frontier of our medical system would be phenomenal. So yeah. What you guys are doing. Hell yeah. Okay, so wrapping up, do you have anything that you like for our. Watchers and listeners. What do you hope they take away. From this whole conversation?
[01:25:12] Fred Parietti: Well, I mean, I think the, the most exciting, thing for me is that, you know, robotics right now is one of the hardest, you know, areas of, you know, technology in general. It is considered like the next frontier of AI because it the application of AI to the physical world. So I'm very happy because it used to not be like that, even just a few years ago.
[01:25:33] Audrow Nash: I know it's true.
[01:25:34] Fred Parietti: Yeah. Right. And so I'm like, yeah, so what is cool? And let's build the best possible robots and just, you know, a so I would just encourage people to build in robotics and be you know, my caveats are be careful because you need to find, you know, an application that justifies that level of technological advancement. Because the worst thing we can do right now is to squander all this attention by making people think robotics is useless or not ready yet. So let's avoid that. The other thing I want to tell people is, yeah, like, you know, there are so many cool applications of robotics that help people. And so I would like, you know, yeah. Like, you know, that I feel like there's so many applications and it's impossible for a single company to address all of them. And so I just encourage people to, you know, just you know, get building.
[01:26:25] Audrow Nash: Oh yeah. Love it. And is there any, any places you'd like to point our listeners to. Any calls to action. For people maybe interested in joining you guys. Or anything you can then go there.
[01:26:40] Fred Parietti: I mean, for me, you know, the you know, the areas where you have the most to learn are always the most interesting. And so for my point, from my point of view, like the intersection between robotics and bio.
[01:26:52] Audrow Nash: Yes, is just so cool. Totally.
[01:26:55] Fred Parietti: And, and there is a lot of need of people that have a kind of an understanding of both areas. Right now, those people are exceedingly rare. In fact, they almost don't exist. It's always a robotics person learns about biology or the other way around. But I would definitely point people. There is so much value and so many new things to discover in that intersection that they feel like people, you know, it's just, very worthwhile, you know? And they were.
[01:27:20] Audrow Nash: Yeah. How do you, as a roboticist, like. How do you learn about biology? Like, do you go back and read a book? Do you? I mean, I'm sure the position you're in now working with pharmaceutical companies is an incredible education. But like. How would you learn. If you're a roboticist or software engineer and you want to learn more about biology but have no idea how to approach that?
[01:27:44] Fred Parietti: Yeah, you need to not be, you know, afraid of asking them questions because obviously you're going to look extremely dumb to an expert. But keep in mind that they are probably thinking that are looking extremely dumb to you ask questions about your field. Right. So everybody is held back by the same fear of like looking like, you know, they have no clue because they have no clue. And so after you recognize that, you know, you can just have those conversations. And now we also helped by the fact that, you know, chat GPT doesn't care if you look dumb and they're pretty good. You know, these new models to do like research and ask basic questions. And it's a very interactive way of learning that, you know, pretty much a lot of, but I would say there is a limit to how much you can learn from an LLM because it's too tailored to please you. And so, I much rather, as soon as I have a question that might be dumb, but I think is very interesting to me. I much rather ask it to a person that I know knows in that space, and in return I can answer their questions about my space, and we both win.
[01:28:51] Audrow Nash: Oh, yeah. That's awesome. All right. Thank you Fred. It's been awesome talking with you.
[01:28:56] Fred Parietti: Thank you very much. I really, really, appreciate it was a great conversation.
[01:29:01] Audrow Nash: Hell yeah. Alright. Bye everyone.