E395: Russ d'Sa on AI, Agents, and the Future of Work

What happens when computers stop being tools and start behaving like collaborators? In this episode, I sit down with Russ d’Sa, Founder and CEO of LiveKit, to discuss why voice AI may become one of the most important computing platforms of the next decade. Russ explains how LiveKit powers AI experiences for companies including Tesla and xAI, why voice is emerging as the natural interface for AI agents, and what the rise of digital labor means for workers, founders, and society.

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:00Speaker 00

  If you have AI agents that can do basically all the mundane tasks that a human doesn't want to do, right? AI already that does that. So let's imagine that over time you unlock all of that data and it can perform any task in a digital space. And then you take that AI or a version of it and you put it into a physical robot that has arms and legs and can walk around and can fold laundry and take your kids to school and build houses and all of that stuff. If that exists in a 10-year time horizon, what are humans doing? Ultimately, most work will be done by machines. But in that time, I do think that there's going to be new jobs. In the interim, there will be this period of amazing creativity and productivity.

• 0:47David Weisburd?

  My guest today is Rust Asaf, founder and CEO of LiveKit, a unicorn AI company that has raised more than $180 million from investors, including Index, Ultimeter, and Redpoint Ventures. LiveKit powers voice AI for companies, including Tesla and XAI, giving Russ a unique point of view on the future of computing, digital labor, and the next wave of AI. Without further ado, here's my conversation with Russ. So you power voice AI for Tesla and XAI. Five years from now, what percentage of interactions with AI will happen through voice?

• 1:19Speaker 00

  It's hard to pin it down to a percentage.

• 1:24Speaker 00

  The reason why is that voice is not like in isolation the interface or the ultimate interface. There are certainly things that benefit from having a screen. Keyboards aside, like screens are useful for consumption of content, right? Like one screen that people have been using for such a long time is the TV. And I don't think that that's going to go away, right? The lean back experience is still going to have a screen involved. The other thing that I can think of that will have a screen Is creative work. So, you know, if you have to go into like a 3D modeling program, of course, you'll use probably an agent that helps you navigate that program. But for any kind of fine grain type of precision work,

• 2:07Speaker 00

  you're still going to use a screen and maybe even a keyboard and a mouse. But I do think that voice takes over for a lot of tasks that are easy to delegate to agents, right? Being able to have an agent like scan through your email and pick out things that are important you should pay attention to. Or when I get into a Tesla, when I'm going somewhere, I say, hey, Grok, can you take me to this place? And then I turn on self-driving. And so I think there are these contexts where especially voice AI is extremely useful and much faster than typing. But screens are going to be around too. I think the future is ultimately multimodal. It's going to be a combination of voice and touch,

• 2:52Speaker 00

  some screens and computer vision.

• 2:57David Weisburd?

  Is it oversimplified to say that easy tasks will be voice and much more difficult tasks will be interface?

• 3:05Speaker 00

  I do think that's an oversimplification. The reason why is that the way that kind of we're moving in the future is that agents are able to do more and more complicated tasks. So today, a lot of voice AI is a simple assistant on your phone. You can ask it questions almost like a Google search. When is the next World Cup game happening or et cetera? Those are simple tasks that are very

• 3:32Speaker 00

  kind of natural for use your voice. But then there are other types of things that are more complicated tasks. Maybe you want your agent to go and call a restaurant and see if there are tables available and book you a reservation. Right? That's a fairly complicated task that involves multiple turns. And an agent can do that even today. Agents can do that. And you can use your voice to simply fire off that task. So the way I think about it is there are a set of complicated tasks where a very wealthy person might have like a personal assistant, a human personal assistant that the person that hires that assistant says, hey, I go and do this, like I'm going to New York for the weekend. Figure out where I should stay. I want to go from Tuesday through Saturday.

• 4:17Speaker 00

  And I also want to have dinner with my friend David over there on Friday night or something like that. And you know, your human personal system would go through and put together an itinerary and present it to you and ask you what you think of it. And if this is okay with you, maybe it'll ask your preference for flights and things like that. But that human assistant can go and perform all of those things underneath that are required to generate that itinerary. And an AI agent can do the same thing. It's already possible today to do something like that. And people are doing things like that. And so I don't think that voice versus touchscreens or keyboards and mice. I don't think that the split is based on complexity of task.

• 4:56David Weisburd?

  What are voice AI agents doing today?

• 4:59Speaker 00

  I would generalize it to customer experience, which includes customer service. So it's things like, of course, as you said, customer service or support. It's a patient intake at a hospital. It's insurance eligibility checking. It's loan qualification. There are millions of phone calls around the world that happen every day. Where one of the people, the person answering the call or even making a call, you know, for like, say debt collection, making an outbound call, where they're doing it on behalf of a business. I think that's the key thing about kind of where voice AI is being deployed and scaled up today.

• 5:41Speaker 00

  It's where they want to take that person that is answering or making a call on behalf of the business and replace it with an AI. And the reason that that's the place where voice AI comes in now is because the telephony system is default voice. There is no visual component to it. There's no UI.

• 6:01David Weisburd?

  Not multimodal.

• 6:02Speaker 00

  That's where you're seeing voice agents today.

• 6:05David Weisburd?

  It's kind of how coding agents have been an early adopter of Anthropics and Codex and from OpenAI. And the reason for that is that it's very logical. Coding is a very logical use case. It doesn't have all these edge cases that some interaction, some combination of physical and real world application combined with something else.

• 6:27Speaker 00

  Two things. One, coding agents is the most lucrative. AI use case today. After that is actually voice. It's the second most lucrative. The difference between coding agents and voice AI applications is that voice AI applications are synchronous. Right? So you're generating lots of tokens. When you're having this conversation with the AI. So your customer support call, for instance, it might take 30 minutes to fix your router.

• 6:59Speaker 00

  And in that 30 minutes, you're generating lots of tokens through that entire conversation.

• 7:04Speaker 00

  Coding agents are different in that they are asynchronous. And that you are giving direction to the coding agent. And then the agent is going and through reasoning, it's generating code, testing that code, seeing if it works. If it doesn't, going back, editing that code, modifying it. So it's kind of this agentic loop where the AI is building your code, building your code, working, working, and then coming back to you and being like, okay, that's done, check it out. And then give me the next kind of task. And so it's not as synchronous in that way, and it generates a lot more tokens over time. I mean, just because people aren't talking to the customer support person for eight hours a day, but they are building and generating code now for eight hours a day. And so

• 7:48Speaker 00

  coding agents just tend to be more long horizon tasks that generate more tokens because those sessions are longer versus the ones where you're talking to the AI with your voice. I think you're also right, too. The reason why coding agents have become so good and people are so comfortable using them is because you can close the loop with them and that there is a verifiable way to execute the code and see that it works or see that it runs. But for, say, a support call or like a patient intake at a hospital, it's a bit more open ended. There are criteria that are important and that are verifiable, like does the patient get a calendar item created on the doctor's calendar? Right? There are definitely success criteria that you can check and verify

• 8:33Speaker 00

  for correctness, but it is more open ended. It's a conversation. It can go in any different direction versus coding is not exactly the same way.

• 8:41David Weisburd?

  With any new technologies, there's this inflection point that allows them to proliferate. What's enabled voice AI to be so popular today?

• 8:49Speaker 00

  There's two things. It's speed and it's intelligence. I think those are the key kind of innovations over the last couple of years that have unlocked.

• 9:01Speaker 00

  Voice as this new interface to computer and allow it to proliferate in these use cases that I mentioned. Speed is really critical. It's very different than, say, like chatbots. So a chatbot, your prior for texting with a computer, right? Chatting with a computer is chatting with a person. And when you chat with a person, you don't expect a response right away. It's asynchronous communication. You don't even expect the typing indicator to show up right away. You know, the person's likely going to respond to you maybe within the next day or two or a few hours, maybe.

• 9:41Speaker 00

  But your prior for talking to a computer is talking to a person. And for decades, we have been conditioned that when you talk to another person, that they're going to respond very quickly. For English, I think the average latency is 250 milliseconds or less. And so you expect a response from the computer in that time. Otherwise, it's kind of, it falls into the uncanny valley kind of zone.

• 10:13David Weisburd?

  You're constantly reminded that's not a person.

• 10:16Speaker 00

  Correct. Right? I think most people don't realize there are companies that build these like virtual avatars that have like a visual kind of likeness. It's a 3D model or it's actual pixels, like rendered video avatars that they attach to the AI. And that's really cool kind of emerging sort of capability. But avatars have long kind of had this uncanny valley problem where, you know, your visual system like sees something and if it looks a little bit off, like if the mouth doesn't match the words that are being said or something like that, if the AI is looking around in a weird way, your mind like rejects it as real. Voice has the same thing, just maybe not as sensitive.

• 10:59Speaker 00

  Right? Like, our visual systems are very tuned for this stuff because most of our neurons in our brain are actually dedicated to visual processing. But auditory processing is still, we can still detect when things aren't quite real. And so latency is such an important part of that. It's one of the key criteria for making a voice AI feel real. And then I think the second one is, is also attached to this uncanny valley idea. Yeah, it's intelligence. And what I mean by intelligence is I don't necessarily mean that the model is smart. That's part of it.

• 11:38Speaker 00

  But it's also kind of conversational intelligence. So how do you know when I'm talking, when I'm done talking? Like, right now, I see you nodding your head. It's called a back channel.

• 11:50Speaker 00

  We do this in voice as well over the phone. Like, someone says something and you say, uh huh, or okay. You're not actually interrupting that person. And they know you're not interrupting them. You're just acknowledging what they're saying, or that you're jotting it down, or you're hearing them. Right? You're doing this right now.

• 12:05David Weisburd?

  I'm trained as a podcaster.

• 12:06Speaker 00

  There, yeah, that's true. Maybe you're extra good at it. But how do you know? How does an AI understand that? Right? Or how does an AI understand when I am done speaking and sharing my thought? And then it's okay for the AI to start speaking? Or how does AI know that I'm truly interrupting it? Because what it said I don't agree with, or it did the wrong thing? Or it went down a path I didn't intend, and I need to, like, steer it away? How does the AI know that that's happening? And that's what, that's another part of the intelligence, is conversational intelligence. And I think those two things improving greatly. So turn-taking models for understanding when a human is done speaking, interruption detection. And then, of course, like, the speech to text and the LLM and the text to speech in that chain

• 12:50Speaker 00

  for a voice AI application, all of those inference calls getting really fast, too. The network transport that LiveKit provides being super fast, like, that's really shrunk the N10 latency. And so those things coming together and being used in a single package. Is, in my opinion, what I think has led to this proliferation of voice.

• 13:10David Weisburd?

  There's this big philosophical debate on AI, on whether it's gonna create new jobs or whether it's gonna be destructive to human labor. Where do you stand on this debate?

• 13:21Speaker 00

  I think it's going to move through phases. Sometimes at dinners, I have a spicy take here, which is like that all of the foundational model lab leaders saying that there will always be new jobs and that it doesn't, you know, AI doesn't replace the jobs. That's just not true fundamentally. I mean, think of the end state, right? If you have AI agents that can do basically all the mundane tasks that a human doesn't want to do, right? AI already that does that. We have it already. Maybe for some specific tasks and specific workflows, you need some data to fine tune the model to perform that task well at a human level.

• 14:05Speaker 00

  But we know the techniques. It's really a lack of data. So let's imagine that over time, you unlock all of that data and it can perform any task kind of in a digital space. And then you take that AI or a version of it and you put it into a physical robot that has arms and legs and can walk around and can fold laundry and take your kids to school and build houses and all of that stuff. Those are things that I think very few people would dispute will eventually exist. Let's say in a 10-year time horizon, if that exists in a 10-year time horizon, and it will take time to distribute it worldwide, but if you have an AI that can do all the mundane work in the physical world that a human doesn't

• 14:48Speaker 00

  want to do, and it can do all the digital work that a human doesn't want to do, what are humans doing?

• 14:57Speaker 00

  I think that there are some problems that we will need to solve in society around how do people pay for things, how do they find fulfillment in their life, or how do they progress over time. I think that's something that's just very important to the human psychology and happiness. But ultimately, I do think that most work will be done by machines. I think people ignoring that, they're ignoring it for a reason. There's like, you know, they have companies to build and brands to grow and all of that stuff. But I think like, the writing is on the wall. It's just that the wall's pretty far in the future. So that's one thought. But in that time, I do think that there's going to be new jobs,

• 15:41Speaker 00

  right? So I'm not taking the stance that like, it's gonna destroy all jobs right away. Over time, I think that AI will do most of the work. But I think in the interim, there will be this period of amazing creativity and productivity. One, because AI allows you to do so many things faster, right? If you could only build a company with one idea, now you can like actually work on like five different ideas at the same time, leveraging AI to help you do that. So that's the productivity gain. And I think the creativity gain, people often say that for the last 20, 30, 40 years, people say that ideas are cheap, execution

• 16:21Speaker 00

  is everything. But execution is now getting commoditized. And so in a world like that, ideas actually end up becoming the more valuable component of the two. And so I think, you know, another word for idea can be imagination or creation, creativity. And so I think that creative people, more people will become creative, or rather, maybe lots of us are creative, but maybe just don't have the ability to execute. And AI gives us now this ability to execute. And so I think we'll have a period of time where that's this common thing for people

• 17:04Speaker 00

  in the workplace. They'll be doing more creation, more supervision, and then AI will do a lot of the mundane part.

• 17:12David Weisburd?

  For people looking to avoid losing their job over this next 10-year period, as you define it, what are some first principles? Based on your vantage point of seeing a lot of these voice applications being integrated into the market, what should people be doing in order to avoid having their job replaced by AI? Expert calls have always been one of the most powerful ways to build conviction. But today, investors are asked to cover more companies, move faster, and do it with leaner teams. With AlphaSense AI-led expert calls, their TGS call service team sources experts based on your research criteria and lets the AI interviewer get to work. The magic is in the AI interviewer. Purpose-built and knowledgeable-based information to conduct

• 17:55David Weisburd?

  high-quality, context-rich conversations on your behalf, acting as a trusted extension of your team. Then they take it one step further. Your call transcripts flow natively into your AlphaSense experience and become queryable, searchable, and comparable, so your primary insights plug directly into earnings prep, digital work streams, and pitchbooks with zero tool switching. And with AlphaSense expert call services, the AI-led expert calls are just one option, because we know the importance of a hybrid expert research approach. AI for coverage and efficiency, humans for complexity and conviction. It's the institutional edge that scales research without scaling headcount. For hedge funds, that means validating thesis assumptions across dozens of experts before earnings instead of a handful. For

• 18:38David Weisburd?

  private equity, it means faster pre-IOI scans and deeper commercial diligence. For investment banks and asset managers, it means pulling real operator perspectives straight into models and sector positioning without disconnected tools or manual handoffs. All of it lives inside the AlphaSense platform, trusted by 75% of the world's top hedge funds alongside filings, broker research, news, and more than 240,000 expert call transcripts, turning raw conversations into comparable, auditable insight. Take advantage of AlphaSense AI-led expert calls now. The first to see wins, the rest follow. Learn more at alpha-sense.com/HowIInvest.

• 19:19Speaker 00

  I think you have to...

• 19:23Speaker 00

  Look at the ways that you can leverage AI for whatever you want to pursue. It's a hard question to give general advice for. I think that the work that AI is going to do is that humans are doing today. It's all the largely undifferentiated work. You could argue that coding, software engineering, it has been differentiated, but maybe the actual engineering of software is not that differentiated. It's like the design and like the product strategy and the understanding of like the user experience that is really the differentiated part. And then you could argue that maybe the coding of it is not so differentiated. I think there are certain realms where that is not true.

• 20:04Speaker 00

  Like I think for a certain deep infrastructure, that's not true. But I think for the next incremental SaaS application, I think that that is largely true, that it's undifferentiated. And so I think undifferentiated work is where AI is going to shine. But then taste and how you steer that AI and what you leverage it for to create is what is differentiated. And so my advice to people, generally speaking, is look to pursuits that are differentiated

• 20:42Speaker 00

  in a world where AI can effectively perform any repeatable task with precision over and over and over. I think like there's this fear right now that is real. My wife's siblings are considerably younger than we both are. And they just graduated college and they're looking for jobs and it's really hard to find jobs. So I think that there's, there's definitely like an effect in the market right now on kind of more of the entry level work.

• 21:14David Weisburd?

  Because the entry level work tends to be repetitive and tends to not require all the things that you gain through your career, like taste, like judgment.

• 21:22Speaker 00

  Correct, yes. Look in places where that's not the case. The other thing, it kind of sounds a little odd to say back, but I'll be out of a job at some point, but as an engineer. Physical world is still a ways away from being automated by AI. And so my wife's a special needs preschool teacher and that's going to be a really hard job for AI to automate. It's also going to be.

• 21:49Speaker 00

  Challenging for parents to accept like a physical robot or something screen with an avatar that's immobile, working with their children in a classroom setting. And so I think physical world kind of pursuits are also an area that have some more longevity.

• 22:07David Weisburd?

  It's more of a paradox, which is the things that humans, that require humans, AI will be bad at, and the things that AI will be good at, humans are bad at.

• 22:20Speaker 00

  I think that's right.

• 22:21David Weisburd?

  What about reputation as well as relationships? Is that going to become commoditized?

• 22:28Speaker 00

  I don't think that reputation will become commoditized.

• 22:34Speaker 00

  The analogy I draw to this, or the source material that I use to come to this conclusion, that reputation won't be commoditized, is Amazon Prime Video. So I don't know, do you have Prime Video?

• 22:51David Weisburd?

  Yeah.

• 22:51Speaker 00

  Okay. Do you spend much time on it? I don't, because I don't like love.

• 22:55David Weisburd?

  I don't, I think it's the worst branding decision ever to call it Prime Video.

• 22:59Speaker 00

  So if you go on Amazon Prime Video, it's just like a sea of garbage. I don't know if like this stuff is AI generated, or I think actual humans made a lot of the stuff on there. You know, I'm a fan of the company and I have a lot of respect for everything they've built, of course, but they're a titan, right? But Amazon's approach to their video streaming platform is kind of similar to their approach to their store. It's the everything store, like the long tail is there, mixed in with like all the, the good stuff too, or the popular items for people to buy. And it's the same thing with their video collection. It's, there's so much garbage on there. And there are some good shows as well, you know, Fallout's great and The Boys is awesome.

• 23:43Speaker 00

  But there's all this garbage.

• 23:45David Weisburd?

  And one thing that I've learned from talking to hundreds of investors, founders and business leaders, is that planning for the future isn't just about growing assets. It's also about protecting the people who matter most. Life comes with a lot of important milestones. Maybe you're buying. At home, starting a family, planning for college, or simply thinking more intentionally about your financial future. As those responsibilities grow, it's worth making sure you protect those you love. Not sure what kind of life insurance you may need? Concerned about how much it may cost? Think life insurance is too complicated? Your local State Farm agent can help answer those questions. They can help you explore your options for life insurance and choose coverage that you can adjust as your family's needs change over time. Financial protection for your loved ones when it matters most. Learn more about life insurance at statefarm.com. Like a good neighbor, State Farm is there.

• 24:29David Weisburd?

  One thing that I've learned from talking to hundreds of investors, founders, and business leaders is that planning for the future isn't just about growing assets, it's also about protecting the people who matter most. Life comes with a lot of important milestones. Maybe you're buying a home, starting a family, planning for college, or simply thinking more intentionally about your financial future. As those responsibilities grow, it's worth making sure you protect those you love. Not sure what kind of life insurance you may need? Concerned about how much it may cost? Think life insurance is too complicated? Your local State Farm agent can help answer those questions. They can help you explore your options for life insurance and choose coverage that you can adjust as your family's needs change over time. Financial protection for your loved ones when it matters most. Learn more about life insurance at statefarm.com. Like a good neighbor, State Farm is there.

• 25:13David Weisburd?

  One thing that I've learned from talking to hundreds of investors, founders, and business leaders is that planning for the future isn't just about growing assets, it's also about protecting the people who matter most. Life comes with a lot of important milestones. Maybe you're buying a home, starting a family, planning for college, or simply thinking more intentionally about your financial future. As those responsibilities grow, it's worth making sure you protect those you love. Not sure what kind of life insurance you may need? Concerned about how much it may cost? Think life insurance is too complicated? Your local State Farm agent can help answer those questions. They can help you explore your options for life insurance and choose coverage that you can adjust as your family's needs change over time. Financial protection for your loved ones when it matters most. Learn more about life insurance at statefarm.com. Like a good neighbor, State Farm is there.

• 25:57Speaker 00

  What ends up happening is you actually just don't watch any of it. You kind of just become blind to it and ignore it. And you go and you watch the stuff that... Your friends watch or tell you about or from like a director that you respect. And so I think that idea of like relational media as it pertains to to reputation. I think that's going to be the same thing in the world of AI. It's just that there's going to be so much more garbage. But the best stuff will still kind of float to the top. And maybe there will be more best stuff across different categories. But it's today, the case that you cannot consume all of the products and all of the content that exists in the world. And there'll just be like several orders of magnitude more and the best stuff will rise to the top.

• 26:42Speaker 00

  So I don't think that reputation is commoditized.

• 26:46Speaker 00

  What was the second one?

• 26:46David Weisburd?

  Relationship.

• 26:47Speaker 00

  Relationships. Relationships are also not commoditized for two reasons.

• 26:54Speaker 00

  And I'm talking about AI today. I'm not talking about AI in 15 years, if there are some breakthroughs and we train these things differently. But today, L L Ms do not have emotion. That's something that is still innately human and relationships are emotional things. Right? It's a connection between people and some relationships are transactional, but the ones that last for a long time are much deeper than that. And they're these emotional bonds that we have with people. And I just, I don't see current AI replacing that. I think that that actually in a world where AI is doing all of this work and it's like a machine that we've created that can, we can automate to do any task.

• 27:34Speaker 00

  One of the kind of last pieces that define us as humans from the machines is our ability to have relationships and to have emotions. And so I don't think that's commoditized either.

• 27:50David Weisburd?

  How do you incorporate reputation, relationships, advice you would give your wife, siblings?

• 27:57Speaker 00

  It's a tough question. I don't know if I have an answer to that one. It's hard because I think that. The question I get from them is like, oh, well, how should I navigate this world of where it's hard to find a job? And what do you think I should do? Should I go back to school? Should I study something else? I think what's hard is that the question that they're asking me about is work. Right? And work tends to be somewhat transactional in nature. Right? I mean, it's you do work and you get paid. So there's a transactional kind of contract right there embedded in the whole thing. And it's difficult to like reason about relationships and reputation in that lens, just because they're right out of school. They don't have a reputation yet.

• 28:42Speaker 00

  They haven't established themselves. So maybe like the advice that I give there, this is actually the advice I give is go up, try to get your foot in the door somewhere and start to work your way up. Right? Because I still think that there's like this time horizon before AI really does take all the jobs en masse. And in that time, like, can you build up your expertise in a field, your ability to to supervise and define how these problems are solved, not the actual solving of the problem? Can you like get to that point now before kind of AI comes from the bottom for more of the entry level stuff? I think my advice is like to start building the relationships and start building the

• 29:25Speaker 00

  skills to level up before AI comes from underneath and takes it from you.

• 29:29David Weisburd?

  A CEO of LiveKit, a billion dollar company, raised $180 million in venture. What are you looking for from people that you're hiring?

• 29:37Speaker 00

  The two things that we look for, one is like slope. Right? Tying back to like entry level jobs and things like that. I, I don't really think we care about resumes very much. The why intercept doesn't matter so much as the trajectory of a person.

• 29:51David Weisburd?

  Is that really true?

• 29:52Speaker 00

  Yeah, it's totally true.

• 29:53David Weisburd?

  So you would hire an English major that you thought could think like an engineer and train them to be an engineer?

• 30:00Speaker 00

  Uh, well, I don't know if we'd train them to be an engineer. I think maybe they'd start off in a different part of the, the org, just based on their experience.

• 30:09David Weisburd?

  Critical thinking as a skill in and of itself is valuable enough.

• 30:12Speaker 00

  Absolutely.

• 30:13David Weisburd?

  What's an example of that?

• 30:16Speaker 00

  Well, one of the, one of like the strongest.

• 30:21Speaker 00

  People on our team is this engineer who started at LiveKit when he was 17 years old and now he's 22 and his slope is just like insane. And when we first met him, he didn't know much, but he had a high sense, high sense of like agency. If there was something that he didn't know, he would just go in and figure it out and learn it. And this was before all of the AI stuff. Now you like pair someone with an attitude like that, or that is a character trait with AI and it's just kind of insane. So I think that's my second criteria. Well, agency is one, right?

• 31:02Speaker 00

  And the second one is just being like AI-pilled. It's like looking for different places where you can leverage this new type of technology versus rejecting it. I think that's the second thing is if you are able to leverage this maximally, it's kind of wild how much productivity you can unlock for yourself. And maybe that also ties into like young people coming into the workforce. One thing that I've noticed when I talk to young people is that they've never really heard of, for example, they've never heard of Claude Code. I've been on like three or four interviews in the last few months.

• 31:45Speaker 00

  They weren't for engineers, but they were for, you know, folks in different other areas that go to market, et cetera. And I asked them, I'm like, but they don't live in the Bay area. That's the other thing too. So I'll ask someone who this one in particular I can think of is someone who's been working for a well-known tech company, but out in Florida. And I'm like, oh, have you heard of Claude Code? And they're like, no, what is that? And they're like, I've heard of Claude. Well, what's Claude Code? And I think it's illustrative of that a lot of people around the world and even around our country are not aware of the cutting-edge tools that we have at our disposal. And so I think that it's really important that people figure out how to leverage these things for them instead of having them take their jobs.

• 32:29David Weisburd?

  Let's just use a thought experiment. Maybe one of the smartest MIT math majors that have never produced a single line of code, you hire him or her, how long before they could become a great engineer?

• 32:43Speaker 00

  Maybe they never actually become a great engineer.

• 32:48Speaker 00

  And maybe they don't have to. When AI can engineer. It's different for LiveKit than it is for a lot of companies. The reason why is that LiveKit is like critical infrastructure. And so, infrastructure has this property that it is always expected to work. The only time you really pay attention to infrastructure is when it doesn't work. Otherwise, you just take it for granted, like the lights in this room. Like, we'll only really think about PG&E, the power company in California, when there's a power outage. And I think that that property for LiveKit in particular, the fact that it's infrastructure, the fact of reliability is the most important feature of it.

• 33:28Speaker 00

  It requires a different way of thinking about how to use AI and leverage AI than the typical like SaaS application or even a consumer application. Because changes that are made to our codebase can be potentially catastrophic and take down very critical workloads that people around the world, no exaggeration, actually depend on. Very critical. Yeah, it's more than just like revenue. There's like a moral obligation that we have. Someone has an emergency and is trying to call their doctor or trying to get an ambulance to their house. It's, you know, that's something that,

• 34:10Speaker 00

  like, life or death really does matter on this infrastructure working. And so, we can't just allow people to kind of vibe code improvements or changes to the infrastructure. Like, there has to be a human that reviews that code, that takes a look at it and interrogates what it's doing. I think another tricky part about these AI models, they generate so much code that reading code can be a pretty daunting task in this new world. I think, though, that LiveKit accepted. I think that for, let's say, the math major gets hired to build something at the application layer. And I don't want to be reductive, but like, let's say it's like a SaaS product. Let's say it's a CRM, right?

• 34:55Speaker 00

  You could argue that really what the important thing to understand is in this world of coding agents is that you understand what the business rules are, or what the user experience is, or which systems that like the the person needs to interact with, like how do they need to view this information? What is an easy way to get new information into that CRM? It's like thinking through the workflows and then leveraging AI to actually go and build out this workflow that you have designed for that application. And so I think like the the human is now the orchestrator. They don't really need to understand how do I write into this database? And should I use a loop here or whatever for some business logic? I don't think that the math major has to do a lot of engineering anymore. I

• 35:39Speaker 00

  think that they're doing a lot more product designing, though.

• 35:42David Weisburd?

  I was thinking about high agency. There's a term like taste that gets used a lot in Silicon Valley. And I just heard a video on high agency, which was an intern would come to his boss at the end of each day and say, what else could I do? What else do you need help with? Just constantly. And that intern made it throughout the entire organization up to then running this being the CEO of this public company. So this high agency is, what can I do? What more can I do? And said another way, if you think that AI is going to disrupt a lot of people, there's two philosophies to deal with that. One is you could basically almost have a low agency and a defeatist attitude to it. And the second one is kind of like running away from a lion. You don't have to beat the lion. You just

• 36:26David Weisburd?

  have to beat the other person that's running away from the lion. So if you could get closer and closer to creating more and more value for the organization, you may not avoid the inevitable AI disruption of the entire field in 10, 20 years. But you could significantly lengthen your career just by being more and more useful.

• 36:47Speaker 00

  That's right. And I think today.

• 36:50Speaker 00

  Something that I hear a lot when I talk to especially larger enterprises, larger enterprises are starting to hire people that will do kind of AI modernization for them. So, or help them with AI modernization, right? There are these new tools that are available that can unlock more productivity, cost savings for a large organization, more efficiency in places like, how do I use it, right? I think what AI unlocks is that now you can kind of like price things based on an outcome instead of primitives. And so I think a person who comes into a company, right? Like a new employee or an intern or something like that. Yeah.

• 37:32Speaker 00

  I think what's tricky for large companies is they know that this stuff is there. They know that they can leverage it and it will be a huge unlock for them as a business. But how to do it is actually really tricky. You've got OpenClaw, you've got Hermes, you have these like agent harnesses that you can kind of rig up an AI to do a particular task or a workflow or coordinate with other AIs and they can like do work together. This idea of agent factories, Stripe has written about like this thing called minions, which is their version of that and how it runs a lot of Stripe's processes.

• 38:13Speaker 00

  There are definitely certain companies that are kind of AI native and figured out how to do this, but the vast majority of companies in the world, this is new, they're not AI native, they are still trying to figure out, how do I like retrofit what I have, like AI onto what I have? And so people kind of coming up now, starting their careers that are looking at AI through that lens and how you can come into a place and create efficiencies within that place that are, have been operating for decades in this previous era, that's a huge opportunity for people. And maybe they eventually over time put themselves out of a job, but you've been climbing the ranks steadily as you've been kind of demonstrating those wins within an organization.

• 38:57Speaker 00

  So that is a potential angle.

• 38:59David Weisburd?

  Ideally, you get equity in these companies. So you're building your wealth while you're

• 39:03Speaker 00

  Yeah, I mean, absolutely. Yeah.

• 39:05David Weisburd?

  I thought a lot about this. How do companies integrate AI? And the solution that I now give CEOs is you need to put bodies at that problem, meaning it has to be somebody's job, ideally more than one person's job, to integrate AI for the organization. The reason for that is everybody is so busy. If I gave you, Russ, if you're just Fortune 500 CEO and I gave you here are the 50 ways to use AI. I might even give you data on it and give you empirical data. You're just not going to have enough time during the day to add another thing to your plate. Somebody must be in charge of that, and it needs to be somebody's entire KPI. In other words, it's their full-time job to make sure that AI is being integrated within the organization, and then they could push that agenda to the organization.

• 39:50David Weisburd?

  But just telling everybody to use more AI, which is the default standard, does not seem to be working well.

• 39:58Speaker 00

  There's this kind of a pattern that happens across every paradigm shift. It happened with web. It happens with mobile. I don't know if you remember back in the day when like iOS was coming out in the platform, the app platform came out. There were all these products, companies that were trying to figure out, okay, well, how do I use mobile, right? Or how do I take like my website and make it work in a mobile app? And they're almost doing like this direct translation of their website, like shrinking it down. And then, of course, they're leveraging this new platform, this new paradigm. But are they maximally leveraging it or not? I don't think the answer is yes. And then, of course, you have these mobile native companies that that end up

• 40:41Speaker 00

  emerging like Uber and Snap and things like that, that were like purpose-built for that new paradigm. And I think you're seeing something similar now with AI as well, where it's this new capability. Nobody has time to figure out how to weave it into, you know, whatever their team's kind of charter is. And so you create like this horizontal layer within your organization, and they're responsible for figuring out all these different pockets, like where they can use AI or infuse it. And

• 41:13Speaker 00

  I think that that's gonna pay some dividends. But I think ultimately what will end up happening is I do think that there will be these AI native companies where every single team thinks about what their charter is and what they need to get accomplished. They think about it from like the lens of, okay, how do I leverage AI to be able to do?

• 41:33David Weisburd?

  It is then part of their KPIs.

• 41:34Speaker 00

  Yes, it is part of their KPIs. We're not at that stage yet because this AI movement is happening while there's an existing kind of incumbent shape of a company and working style of a company. And there are new companies being born every day that are kind of like born in this world of AI. And, and I think, you know, over the next five to 10 years, you'll start to see those come up and maybe eventually outpace kind of the existing ones.

• 42:04David Weisburd?

  It's like this whole technological innovation where you have now horses turned into cars and everyone's trying to retrofit the horse carriage companies, but probably the easiest way to solve that is just starting Carcom.

• 42:17Speaker 00

  Totally. Yeah. I think it's a natural part of the cycle. I just think there's a life and kind of death cycle to almost everything. And I think that this is, we've seen this movie before, maybe it's just happening faster every time, but we've definitely seen similar kinds of things occur with these paradigm shifts in computing.

• 42:36David Weisburd?

  Well, if that's the case, there's going to be a lot more jobs, a lot more new jobs because essentially every industry in the world must now be disrupted by its AI native self.

• 42:46Speaker 00

  Maybe. I think that there will be a lot more new jobs. It's just, the question is, is like, is it AI doing those jobs or is it a human?

• 42:55David Weisburd?

  You've built one of the most innovative businesses over the last couple of years in Silicon Valley. If you go back before you started LiveGet, what would be one piece of timeless advice that you would have given a younger Russ that would have helped you either accelerate your success or helped you avoid costly mistakes?

• 43:10Speaker 00

  Kind of related to earlier when I said, there's this shifting.

• 43:18Speaker 00

  Relationship between ideas and execution. The advice that I would have given myself because my fifth company, right? So the first four weren't so great. And not, they were fun, but they weren't as great business successes. And the advice I would give myself is focus on building companies that are execution risk, not market risk. My past work has been a lot of like new ideas and trying to create categories. Yeah, I mean, you could argue that LiveKit has kind of helped create a category in Voice AI.

• 44:03Speaker 00

  But broadly speaking, though, the space that LiveKit plays in is pure execution risk, not market risk. It's that, do you feel that the computer is going to become more human like or less? Right? People talk about AGI, right? AGI is here. We're going to build ASI, all these things. Throw all the like acronyms away. What are we actually trying to do? Like, what are humans really dreaming of doing? It's the same thing that sci fi has been dreaming about for, I don't know, 80, 100 years, maybe. And that is, we're trying to recreate ourselves. We're trying to reverse engineer.

• 44:47Speaker 00

  How does the human brain work? And we're trying to recreate that synthetically. And so the bet, even back when we started LiveKit, though more fuzzy, and then of course, solidified as OpenAI, you know, when OpenAI built Voice Mode on us and all of that stuff. The bet has always been that the computer is going to be brought to life. It's going to be more like a person. And if that's true, how will you interact with that kind of computer? You'll interact with it like you interact with a person. And so we're using eyes and ears and a mouth. And the computer is going to use the equivalent sensors for it.

• 45:30Speaker 00

  And it's going to feel very human like to use it, interact with it, do work with it, collaborate. And so there's going to need to be infrastructure around that kind of core LLM.

• 45:46Speaker 00

  To build applications on top that allow you to have those kinds of experiences when you use the computer or use AI underneath. And so we said, okay, well, we don't know if we're gonna build the winning company. We don't know if we're gonna dominate the market. But what we do know is that there's land out there and we need to like sail to that land. And there's definitely land out there. It's not even like this horizon and we're just guessing. And I think that that's emblematic of like market risk versus execution risk is that the market was there. And what we went out to do started as an open source project. So it literally wasn't even trying to build a business at the start. It was just in earnest, trying to build value in a place where we we knew that the world would

• 46:29Speaker 00

  would eventually be. And that would be my advice to most people is I think that there's like short term kind of ideas that you can pursue. And you might even do financially quite well in doing that. But I think for longevity to really build something generational, you have to look at kind of like a market that you believe is inevitable and really massive and just go in that direction. And you make small turns and course corrections and evolutions over time, but just generally go in that direction. Here's an example, right? Do do we feel like with Elon, like, do you feel that the world is going to back

• 47:10Speaker 00

  when he started Tesla or acquired his way into Tesla? Back then, you could ask yourself the question, do you think that one day that we'll go and like colonize other planets? Or do you think that we're just gonna stay confined to Earth forever? And I think these answers are very obvious. Like, you know, most people can say the obvious thing. And so, you know, if you can figure out a sustainable way to keep working on it, just go in that direction and build stuff and build value towards that kind of obvious or inevitable future.

• 47:41David Weisburd?

  The future is inevitable, given a long enough time horizon.

• 47:45Speaker 00

  Yes, I think so.

• 47:47David Weisburd?

  Well, Russ, thanks so much for jumping on.

• 47:49Speaker 00

  Thanks, David. Appreciate it.