Ex-Uber AI Safety Lead: Your Home Is Already Exposed. Almost Nobody Knows It | Ep. 419 with John Lunsford Founder of Tethral

0:00Everyone right now is talking about agents communicating with agents. The AI agents created their own social media platform and they're going online and they're buying things. I think people are making up a lot of stories because I don't know if what a lot of is being said is even true. However, you're talking about agents and maybe communication with IoT. Why is no one talking about that? Well, I think one of the reasons that no one really talks about IoT is that it's been a big disappointment for maybe 20 years. It's been promised to be a lot of things, but it never really fulfilled that promise.

0:35And I think what's really cool, but also a bit of a cautionary tale is like right now with AI agents and with agentic actions, the ability for AI to kind of act on its own, like purchase stuff or move things in your house. it actually affords this really unique opportunity for for connected devices not just iot but everything in your ecosystem to kind of act in a coordinated way but what is both an opportunity but a little intimidating about that is like what are what are agents saying to the other machines in your environment the things that you live with your refrigerator your door whatever this is some of the stuff that I think about.

1:16But partly is like, how do we make AI do that better? How do we have it do it safely? But another thing is like, how does it present itself so that we know the thing coming into our home is safe? And that's a big concern of ours with Tetherill is that like, when we give people the tools to create routines, to create agents for their household to help manage the kids or let the dog out. Let's say letting the dog out in the back seems like a simple automation when you're like, hey, AI, do that. But sometimes when AI gets all mixed up, maybe it opens the front door instead of the back door. And maybe it's not your dog, maybe it's your kid.

1:55And so now your kid is in the street. And that is not a world we want. So when we think about the power of AI agents, and they are powerful, we need to focus on the capability that mixing these two systems offers, but the new things that it also introduces. What's the scariest thing that if we knew right now, we would freak out around the communication between AI agents and IoT? What do you think would freak people out right now? So, I mean, the car on autopilot, I think there's already a TV show about that one. It's called upload. It was pretty good. It's fun to watch. But anyway, you know, OK, the scariest thing, and this goes back to my security engineering days, is that consumer connected devices are very insecure.

2:47And for the most part, the fact that there are so many of them, there's like, you know, 15 on average per home now, not just IoT, but connected things. There's so many of them. And the fact that there's so many of them meant it wasn't a very practical surface for adversarial attacks for hackers to come in and mess with your system and lock you in or lock you out or whatever um what i think is is amazing and also incredibly scary about uh ai right now is that things like anthropics mythos anthropic is very much building that in public and it's great because we as the public get to see that capability, that scary capability right up front.

3:32But if they're building in public, you can be sure there are many people building similar things in private. And what happens is that, let's say there are 2 billion connected devices in American homes right now. And the only thing stopping consumer connected tech from being a coordinated attack surface was scale. It used to take hackers one or two at a time to break into one or two devices at a time. And the practicality of that versus the effort was not really there. Didn't make sense. But now we're seeing that AI can scale its capacity, what we call super linearly. And so let's say one agent can be in a thousand homes across 20 devices each.

4:21And the only thing holding it back is the amount of compute you throw at it. And so instead of, you know, a impossibly large surface that protects the consumer front, essentially, now it is it can be orchestrated into a single attack surface. And then if you think about that, moving, moving forward a little bit is not just like messing with people's homes or spying on them or whatever. And that certainly is intelligence. But if you think about that as like a way to subtly increase, let's say the stress of the american consumer by messing with their house their lights turn on in the middle of the night whatever great okay but that's also like a way to disrupt populations before a large attack and so if there was a huge like government attack or attack on infrastructure you can do things like pull demand pool demand where it overloads energy grids because all the houses activate everything at once.

5:20And even though the energy grid itself is secure, consumer IoT, which this energy grid services and sees as legitimate, can then effectively disrupt these critical systems from the consumer standpoint. Wow. I've heard of this before where you do small things to get people scared to drive fear before you do something larger. But the goal is over time fear, which is why I've heard of government hacks into things like power grids, water systems. Can you imagine how many people will think they have ghosts in their house? Stranger things. Stranger things. They're stranger. Maybe stranger things was based on this.

6:03I built FounderStory from a$50 microphone. And the most important thing is I didn't do it alone. For years, I've been using Upwork to hire marketing, editing, branding, you name it. In fact, the editor who cut this very episode and the team behind all of Founder's Story branding found them on Upwork. The quality of people is top-notch and paying people is simple. Upwork is a one-stop platform to find, hire, and pay expert freelancers across development, data, marketing, operations, and more. With Business Plus, you can access the top 1 % of talent on Upwork. And with AI-powered shortlisting, you'll get matched to the right freelancer in under 6 hours.

6:51No endless searching required. It's free to sign up and posting a job is easy. Visit Upwork.com right now and post your job for free. That is Upwork.com to connect with top talent ready to help your business grow. that's u p w o r k dot com upwork dot com but it got me wondering though but what about ai agents do you think ai agents at some point could be the ones doing it without any human interaction absolutely like and and not as a way to like stoke fear but you think about open claw like open claw is i don't know if you've heard of that or not but it's it's this kind of really really fun kind of cobbled together system to allow agents to actually try and do something useful on your computer.

7:39Because my understanding is that like the builder of it was frustrated that agents were restrained from like helping. And that's great. But helping is really double edged because giving it and there's I think this is a great micro example because OpenClaw is also known for being like a terror inside computer systems, deleting entire like histories of information or projects or years of work because you kind of set the wrong or you allowed the wrong parameter. And part of one of the tricky things is like you have to have the technical knowledge to be able to set it correctly. What you're talking about is a step further.

8:16It's like, you know, I'm creating a self-learning system, which is coming out more and more. And it can learn in these data loops to be more effective or more efficient, but without being able to monitor that, you know, the AI you started with is the AI you end with, then it can end up looping itself straight into being something else. And something that you didn't plan for or didn't design for. And already what we talked about earlier, you know, they have tremendous capability and the more unfettered access they become accustomed to, you know, the more critically we need to examine how much access we're really willing to give over, but also not stifling innovation at the same time.

9:04And I think that's really the tricky part. Man, that's, like you said, double-edged sword. I will not use OpenClaw. I have a lot of friends that are doing it. And I'm like, why would I give it access? This open source thing, access to my computer, doing everything, making decisions. I don't really know anything about where's the information going, who has access. I do use Victor, which lives my slack and does a lot of different things. Let's go to when you were at Uber and you created Uber teens on a napkin. Everyone dismissed the idea. What did that teach you, though, about when everyone dismissed your idea, yet you continue to move forward?

9:43Yeah, one other thing. So I started doing napkins of Uber teens. What I didn't know at the time is somebody else had tried it a couple of years before and it failed. um and so one of the one of the really interesting you know i i would love to be able to have a lesson about perseverance here and sometimes like people not believing you just kind of sucks it can be really hard and um yeah it does and then looking back you're like and then those people are like oh i always knew he'd succeed whatever guys like support me then not now um But, you know, I think there's there is that there's there is that like repeated phrase of like, if you're really if you have the conviction to follow through with it.

10:28But in a place like Uber, you know, there is there's conviction, but it needs to be done with an understanding of the environment that you're operating in, because that environment has constraints. it's not like it wouldn't have been successful if i tried to like change all of uber at the same time what you had to do was like take this thing that did have a demonstrated need like i had uh respondents or participants that were sending like kids as eight as eight years old um not on an uber but like in a taxi and a whatever during some research because they had to not because they like had the privilege of not having to look at their kids but because they had to make a really terrible choice every day about like which kid they had to escort to a different part of the city and which kid had to go by themselves and so there was this like deep-seated need and for me that's what really drove the persistence is like a conviction about that there was this problem that needed addressing and it's that conviction that then led to like okay how do i make it how do i push the boundaries of this system that i'm in but also work within the necessary parts of it in order to make this a success.

11:43And it's that balance that is sometimes like really tedious and frustrating, but you can't break every system, but then also expect to be able to use it for deployment. And so, you know, it's something you need to think about when proposing contrarian or controversial ideas is like, how do you get it? How do you keep the core of the idea, the spirit of it, but still make it work within whatever system you're going to be dependent on to deploy it? Yeah. I am probably more motivated by the people that dismissed me than the people that believed in me, which is probably not a good thing. My wife always tells me, you need to let that go.

12:20Stop being so concerned with those people. But I can't help it, John. You love butterflies, monarch butterflies. And I'm trying to understand what lesson in life I can learn from a monarch butterfly. Or is it just the experience? Well, you know, there are two stories that help me think about machines. because of monarchs and um and it's kind of strange to think about monarchs and machines uh no i'm not building a robot monarch butterfly but that'd be kind of cool one of it is is like memory i i think a lot about how machines communicate and how they remember things outside of like the memory we assign them or the memory we allow uh sometimes like you can shape a model's instability, like you're standing in a line at a grocery store with somebody who's really stressed.

13:13And you have to stand in that line next to this stressed person who is kind of acting in a weird way, and they're really stressed out, and they're jittery and whatever. And you can just feel it. And you walk away from that moment impacted by that information transfer, by feeling that they were stressed out. Now, you may be able to throw it away. you may be able to dismiss it you won't let it roll off you great but most people get activated in some way by the communication that's happening in their environment that isn't verbal anyway back to butterflies so uh i was thinking about like butterflies do this weird thing when they're caterpillars and then they turn into goop to be to turn into butterflies which i imagine has got to be painful um but they retain memory as this goop which i don't which i which baffles me i think it's awesome and weird and like alien at the same time but they turn into this sludge sort of and then they still remember the things from when they were a caterpillar but now they're a butterfly um and it and so what it what it really caused me to think about was like okay well well there's this like weird contextual inter uh information transfer happening let me describe a meal I once ate at 2 47 p.m standing at the pantry a sleeve of crackers six chocolate chips straight from a bag a spoonful of peanut butter which is one of my favorites and then because I felt guilty three baby carrots that wasn't a snack that was a crime scene now when this hits I grabbed boom Huel black edition and look it's a complete meal you have instead of food not a shake on top of food.

15:0040 grams of protein mixes with water or milk in about 30 seconds, 27 essential vitamins and minerals, no artificial sweeteners, colors, or flavors. If I'm out the door, the ready to drink bottle is 35 grams of protein, seven grams of fiber under$5 a meal. That is un-freaking-beatable. That's H-U-E-L. New customers get 15 % off online code founder at Huel.com slash founder. Thank you to Huel for partnering and supporting our show. In nature all the time. And I feel it sometimes when I'm around people who are excited and you feel the electricity of a room or an argument that just happened. Like we, so what does that mean for machines?

15:47And one of the things that I figured out is that like, if you have one unstable system, maybe it's deliberately unstable, maybe it's just cheap, and a normally operating system, they can actually infect each other. You can make a perfectly operating, let's say AI agent queuing up to be, you know, the next thing that's coming to purchase an item from the store, but it's having to stand adjacent or exist adjacent to something that has been put there or something that is unstable or something that is there to like cause a little havoc. And that bleeds into yours and that bleeds into these things.

16:23And so not only did we have that first problem of AI being able to act across a lot of spaces simultaneously, now you have the issue of instability being able to infect other AI. And so infection may be a strong word, but it's about information transfer at its core. There's some information that we don't have the ability to reject. It's still imposed on us. We still feel that electricity in the room. And we still feel that awkward tension of a fight that just happened. And so like, it's not, you can ignore it, sure. But it's still there, whether you like it or not. And I think in a way, agents experience that same thing.

17:04I don't know if that's like the Matrix, you know, like the Matrix. Remember, they could infect each other. Or if that's like Terminator, like Liquid Metal Man going, I'm not sure which one. When you hear things online, because there's so many people talking about AI right now. everyone has become an AI expert and they started using chat GPT like three minutes ago. What is something that you hear frequently or even just when you hear it that you just have to say, like, that's not really true or that's not really accurate? Yeah. The thing I think I see, well, not I think, I see all the time where people are like, I automated my entire insert, whatever, stack.

17:43I automated my entire go to market. I automated my entire whatever user base probably i was just writing that linkedin post about my automation my entire go-to-market with fable that continues with yeah yep um so no but and and i love that i love the the ability for it to like partially get us there i was reading something the other day that's like uh it was it was about a vc funder who was like you know once i started recognizing at first ai written content was sounded really eloquent to me once i started recognizing it as ai I started hating it. And so now every time he gets an application that is that very AI feeling cadence that we can't really put words to, but it feels like it, even if it's not like I am AI or whatever, then he outright rejects them.

18:34And I think there's this really interesting adaptation that we always experience with new technology is people get super excited and then it kind of plummets a little bit and then we level out. And we're like, okay, to what degree are we really going to tolerate this? And I think that's the trap with the just got on board with AI folks who are now just experiencing that early excitement of being able to like automate your stack, but really take a look at that output because all those em dashes or dramatic pauses and whatever that it adds is like clear AI tell that everyone else already knows. And so you want this capacity, but then people are still expecting this personalization from like us as people, that effort that we put into writing that email for 20 minutes.

19:27They want that, but then the writer wants that at scale and it doesn't really meet up quite yet. Yeah, I always like to think who whoever the information I'm seeing, are they selling something like if they're selling me a course about something or something, then they're going to be very positive and they're going to replace their whole workflow because they're trying to sell me on their workflow or whatever idea they have. Now, I always tell people that for me, one of the best things I ever did was take anthropology. It was the study of women and men in different cultures was the first anthropology class I took.

20:04Then I took another one. And I think those are the only two classes in college that I ever did not fail when I first went because I end up failing everything else and dropping out. But those two classes shaped a lot of how I sell to people and my perspective of different cultures and humans and human behavior. What did anthropology do for you? One of the things, the biggest thing I think it did for me was demonstrate how we all come at the same information differently. Me, like you talk about selling, I mean, in a little bit of a cynical way, we're all selling something all the time. our opinions our products or whatever our presence our legitimacy things like that and um and it's really interesting to think that like even though you're hearing the words that i'm saying i'm hearing them and and ideally people listening to this are hearing them we're all gonna get something slightly different from that now it's the speaker and it's the context where we Try to control how much that difference kind of oscillates, like moves back and forth or the range of it, because you don't want to be misinterpreted.

21:18But the reality is like those things, those positions, those cultures, how great a night of sleep you had beforehand or not impact how you then approach the information you're getting right now. And some days it'll be great. Some days you're open to the world and you want to talk to everybody and there's only opportunity in front of you. And that's also why those same kind of days with the same emails feel like it's the end of everything. Like you should just quit and crawl into your bed. And part of that is because like we're contextual creatures. Like there's we can't help but be influenced by our past, by our experiences, the ways that we were trained or what we assign value to.

22:00And knowing that and knowing that variance is in part what provoked me to build Tethro the way that it is now. because one of the things that products do and that IoT or connected devices have done for a long time is like, I'm going to ship you this one routine that you can only do if you have this device and this service and this other device because those connections had to be hard-coded and it was a lot of work. But the reality is, is like people had to shape their lives if they wanted to engage with their connection, like that connection. They had to change how they behave to fit around that thing.

22:41But that's not how people want to use things. That's not how people want to adopt AI. People want to figure out how it works in their lives without having to sit in front of like a cloud code session or without having to adopt a workflow that is so hard coded. If anything moves, then it all breaks. And so the system we've been building is one that lets any user talk to it and it will craft and it will repair those connections as your needs change, as your behaviors change, as you have a really bad morning and you're like, I can't, I just need to, I need something relaxing or whatever it is. Like we change and I think our environments need to adopt and adapt to that and not the other way around.

23:27And so, I mean, it's a long answer to a short question, but I feel like that's where I sit on it. What fascinates you most about human behavior in general? I think that people, while personally feeling unique, like I feel different and I don't know how you feel, but we look similar, but we have different experiences, whatever. But there is a common thread in the way that humans have to experience life. like we have to go through things and we there's a there's a variance of like the way that different people experience different things but it's always bounded like within relatively within like a box and so there's what i what's the most interesting is like how do you build stuff to allow for that variability that feeling of uniqueness that kind of celebrates it but then also understands that as a population, you know, we aren't so unique.

24:35People have had similar thoughts or similar ideas that you're having right now, and maybe across the world in a different circumstance. I mean, I have a like a really nerdy reference, if we want to go into that of Darwin. But, you know, Darwin and what is it, Herbert Spencer, back to anthropology, came both came out with ideas around evolution that orbited the same context survival of the fittest actually wasn't darwin survival of the fittest was herbert spencer and um and but but they were two people forming different ideas in different places around similar topics at the same time and so you know we are unique in our experiences and our sense of self and things like that but the way we behave and the way we interact in the environment is like, you know, there's something the same about us.

25:28You and I have a lot of similarities. If I had my glasses on right now, people might think that we're cousins. So talking about survival of the fittest, which, by the way, I never even heard of the other person that you mentioned. I already forgot his name. That's how bad it is. I need to look him up. Do you think the reason why we haven't solved all the world's problems right now, Because it feels like AI is advancing, technology, IoT, 5G, 10G, whatever's next. All these things now have come together where we are in a place like, why do we even have poverty, hunger, disease, all these things?

26:05If we all came together in the world right now, we could solve everything. Is it because of survival of the fittest? you know this actually gets at a at a great point of why those two folks differed so darwin was about reproductive success being able differential reproductive success being able to effectively transmit the successful genes onward it was about reproduction um spencer was about dominance and survival now both both ladder up to survival but they go about the mechanisms very differently um which makes it really interesting but but the point of that to your question of like why haven't we solved is is that even though many of us are interested in survival the mechanism that we want to get there differs because let's say not everybody gets to have champagne but for champagne to be something that is made in small batches for people to enjoy on special occasions or like all the time if you're wealthy or whatever um you know there it only really starts to matter because because of its scarcity um it's fun it's great i love it but uh but it's like you know not everybody could have everything all the time and in order to make a world i think that would be so balanced that we solved things.

27:40And to me, that's what solving the things are, is like, is it just raising the minimum threshold, like raising minimum wage? Or is it balancing out so everyone has the equivalent mechanisms for living and surviving and enjoying life? And I think there's a tension in the reality of people that there is this kind of dominance that is bred into many of us that we exist with. We want to be the first in line. We want to be the fastest. We want to be the whatever. We want to be the builder of companies. And then there's also this element of like where we want to be remembered. We want to have a legacy.

28:20We want to be a part of a community that values what we were. And sometimes I feel like those are in conflict. And I don't know that. And it's also like part of being messy in people so i don't know that there's a clear good path to like technology equals everyone is uh fed and happy and whatever um because i think there's this for some people this kind of fundamental draw or pull to lead and and that's not a bad thing but you know to do so responsibly can be a difficult conversation. So I recently had on the founder of Equity Bee. And what Equity Bee does is they actually allow employees of startups the ability for them to make money from their shares before they leave.

29:15Because many of them after leaving, they never even exercise the right. And this was something that happened to me before. And I found it to be very interesting because right now you have all of these tech startups in San Francisco, like OpenAI, Anthropic, SpaceX, the list goes on, where they're creating millionaires. These are lower level employees, but are becoming millionaires. These are life changing amounts. And it got me thinking about, is it even worth being a founder anymore? Oh, man, that's a, gosh, that's a great question, because there are a couple of things and the number of them we've touched on a little bit.

29:53Is that like conviction about that? It's a real problem that you want to solve. I think people deserve environments that adapt to them. And so I think that is the future of being able to mold our cities, not just our homes in, in the way that makes humanity more successful. And my conviction to that is really strong, but you're right. Like there are, there are job postings for these, you know, soon to IPO or whatever companies that that offer the equivalent of like a million in hiring bonuses, just in equity. Like, it's amazing, like the potential for that. And so, you know, to people considering whether or not to start their own thing or try and get a position, if you're really excellent at your job, try and get a position at one of these companies.

30:49I think that it already tells you what you want to do.

30:57Because in the world we're in now, like the baseline is crazy person conviction. Like it's you have to you have to be willing to sacrifice because your conviction is that strong. And so if you are on the fence of like, I want to make a bunch of money, go be awesome at a startup and support that, grow that until you get to a point where you can't help but make a company. You can't control yourself. You're doing it. And I think that's the point that I got to and why I left Uber. And what do they call that? like golden handcuffs of like being a part of one of those, one of the big companies and you're taken care of in a way.

31:48And sure the benefits that pay awesome. They're great. But it's, it's about like what you're willing to give up for the vision. And does the vision matter that much to you? And it needs to, because if it's just money, cool, great. There are ways to do that. But it might be more effective to go to go join SpaceX. If it's just money you're after. Now, not to say that I'm not like, yes, I would like money. But, you know, they're saying we already know the probability of you making money as a founder is low because majority will go out of business in year two, three, four. As it continues, right, the failure rate goes higher and higher and higher.