Video transcript
Transcript from available captions; it may contain errors.
Hello and welcome back to Equity Tech Crunch's flagship podcast about the business of startups. I'm Rebecca Balon and this is the episode where we bring on industry experts to help us explore a trend in the tech world and dive deep. As regular listeners will know by now, it seems like every big AI player is coming out with a personal agent. Meta has Muse, Open AI has Dots, and a startup called Instinct just raised $1 billion at a $10 billion valuation. And more are coming out every day. Today's guest, Eugeneia Yakuda, thinks those personal AI agents are missing a few important pieces like multiple threads, shared chats, and an app builder. Eugenia founded Replica, one of the first AI companion apps, and now runs Wabby, which earlier this week launched a personal agent that builds the interface you need in the moment. Vuna, welcome to the show. >> Thank you so much, Rebecca, for inviting me. So, Wabby has officially launched the next evolution of Wabby because you launched last year and now you've got what I'm calling privately Wabby 2.0 and this is coming into the space to play with the personal uh AI agents, AI assistants. Can you walk us through what you launched this week? So, what we're launching today uh is um is an app that allows you to talk to an agent that can do anything for you from booking flights to figuring out your um elementary school tours for kids to organizing any family logistics or your logistics in your life that you have. Um and the difference from some other agents that launched recently is that you can do that easily with other people in your life. So, the way Wabby works, um, it's a it's an app that looks like a messenger app basically where you can text Wabby and Wabby will you connect to your Gmail account and Wabby will proactively start getting stuff done for you and organize it by different threads and any thread um you can invite people that you know are relevant for that. So to give you an example for me um I have kids so we have a kids thread with um my husband me um my our nanny my mom and basically Wabby helps us organize everything around our kids and on top of that makes mini apps so that we can see everything in one place. So, for example, for our kids thread, we have an app that recommends where to take them um to in the South Bay area, like all the cool little events for um three and 5-year-olds. We have an app that pulls a preschool menu every day uh for us for to see what they're eating and a couple other ones. >> Yeah. I mean, I want to get more into some of the use cases because I think that everyone's doing it slightly differently, but I haven't seen anyone do exactly what Wobb is doing in in so far as it could be multiplayer and you have this kind of app function. But before we do that, um, you know, we just came off the back of Disrupt at the time of this recording. It was another banger year. And >> congratulations. >> Thank you. You originally, I mean, you've been a guest on our stage a a number of times last year as well. you came to talk about AI chatbot companions because of your other startup replica, but you pitched 12 years ago um an AI concierge, right, to help you do some of this agent stuff. Can you talk a little bit about how you got from that to what is now WBI? >> Yes. So, exactly 12 years ago, 2014, October 19th, crazy was exactly this day. Um I was on on this on stage for Techrunch Disrupt Battlefield um presenting a company that was then called or product that was then called IO that then eventually pivoted into replica. But back then the vision was really close to what we're doing right now with Bobby. It was really a messenger app just like what we have right now where you were texting with an agent that would that could help you basically deal with anything you have in your life. The first demo was restaurant reservations and recommendations, but we kind of showed the, you know, the the vision behind it. We showed the video that I shared actually on Twitter a little bit earlier, and it was basically the same thing as Wii. It was basically a messenger with agents and friends. So, you could talk to anyone, you know, there, and you could at any point tag an agent that would help you >> find a show to go to or book a restaurant reservation or book a trip. We built it back then without language models which are really interesting because they didn't really exist. So deep learning applied to dialog generation like that the first paper on that came out in 2015. >> So we had to build that had to build everything with rule-based conversational engines that we we've built classifiers you know some machine learning but not not you know end to end language models and interestingly worked pretty well. I mean it didn't work as well as today of course but it's kind of interesting to see how we ended up realizing the same vision we we shared at Techrunch uh disrupt 12 years ago. Beautiful poetry just coming full circle and now the tech has finally caught up. I mean Wabby you know when you launched last year um I believe techrunch had the exclusive on that as well as your 20 million preede. Um, you had, you know, angels like Gary Tan, Sarah Guo, Amjad Msad, who's the CEO of Replet, who was also on our stage at disrupt just a few days ago. Now, this was a it was a slightly different app, right? It was more like a YouTube perhaps kind of a social media platform. Anyone can prompt mini apps into existence and then share them or remix them. What made you pivot in a way or evolve into more something that's more of a personal assistant? We started this company um with the idea of building a better interface, a rich interface for AI and you know to bring personal software to people that could be that could help you deal with every aspect of your daily life. Um and so our first iteration um the first iteration of our product was um an app where you could make any app. We could build any app. And what we learned is that even although people love personal software, they don't necessarily always want to build it and it's pretty hard to help them discover other people's software. And so the evolution of Wabby is basically an agent that still builds your personal software but also does a lot more on top of that. Um, and most importantly, you don't have to think about what app to build. The agent by knowing you so deeply and understanding exactly what you're trying to get done will build an app that you need. and you can always change it or prompt it or build another app, but you don't have to. And I think that's the main difference is that amount of productivity that we've put into it. Um, with this with this evolution of the app, >> you actually had to share your launch day with a couple of other companies. Uh, Tolen, they got 30 million for their AI companion that helped users uh navigate life is the way one of our reporters described it. And an AI assistant town as well also has hit a unicorn evaluation just four months after launch. So, you know, there's the apps that make it different. Maybe walk us through a little bit about what that looks like because you showed me a demo and one thing I thought was really cool, like you showed me your you and your partner's chat and then across the top of the chat there was the different apps that related to, you know, your life and like what you two might want to talk about or whatever long-term goals you might be setting. I think what's interesting about WAB is that we're trying to >> shift away from or not away but shift like basically build an agent that's not just focused on productivity. Um obviously productivity is very important. Getting stuff done is super important and we do that for sure. WB is actually fantastic for that because it doesn't even ask you for inputs when you sign up. It starts working for me from the get- go. But I think one of the bigger insights that we had um over time especially after build having built replica is that most people actually don't care that much about saving time. I mean it's it's nice but a lot of people really want to spend you know are looking to spend time spend quality time figure you know do things ideally >> spend the time you have well >> just to spend the time you have well and you know look at the most popular apps obviously Tik Tok and um Instagram they're not there to save any anyone's time and so to us AI shouldn't just be about saving time it should help people spend it better doing more of what they love with people they love uh we believe that AI agents should bring people together, not give people more reasons to spend time alone with AI. And what I saw that, you know, over and over again like people are launching these AI products that are really focused on the single player experience. So if you just texting your agents, the pitch that we keep hearing is, oh, now you have your agent, you don't need to talk to anyone else. The my agent will just talk to a your agent. And to us, that feels wrong because why would um these AIs make us even more disconnected? Um, so that was kind of the the main reason to build Wobby is to to make something that's really focused not just on productivity, but also on long-term goals you might have like, you know, losing weight, learning something um that will help you feel inspired and discover some new stuff, whether it's a music show you you want to see or um uh a fun trip you want to take, but also just basically helping you engage with other people. um in a different way, maybe like help you feel more connected to other people. My wobbby, for example, tells me if I haven't reached out to a friend for a while and suggests, you know, what what to say. I I showed you today, but we have a a chat with my boyfriend where um we have a few apps that Wybe built for us and one is, you know, all the music shows that we want to go to. One is a bookshelf, a shared bookshelf where we put all the books that we're reading. We read a lot together. uh ones we want to read, ones we're you know and Bobby suggests some other ones we might like. >> Did you guys both come up with those apps yourself? Did you decide like oh these are the apps we want or or is that something that the agent decided or is that something you guys were like this would be fun? >> Mostly it's either us saying what we want to achieve. So more talking about the goal and not the app itself or uh Wabby just looking into our shared accounts and saying hey like you seems like you guys are going on this trip. I made you a little widget to keep track of everything or seems like you have, you know, a lot of things on your calendar that might be interesting to each other. I made a shared calendar for you. >> Yeah. Cool. Okay. So, you can do, you know, these group chats, you can do singular chats based on different ones. Like you showed me that you have like a health chat and then that has its own health apps for tracking your workouts, tracking, you know, your diet and things like that. And so I think the multi-thread is a really interesting uh design choice because you know I've been testing out Muse for example and Muse is everything's in one chat. You know even the long-term goals you know whether it's productivity or long-term goal setting it's all in one chat. And I I'm curious why companies would choose to do that if it was like a a deliberate design choice or if it's you know um because it's maybe prohibitive or or difficult to do the multi-thread. What do you think? I think it's a design choice. I clearly instinct you know wanted to do that all in like just iMessage. So a single thread makes sense because how there it is not possible for one contact to start different threads and messages. What are the downsides? >> I think the downsides are pretty significant actually like if you think about a single threaded experience it gets worse over time. So in the beginning it's easy is you don't have necessarily to like you know think about it. You just start with one. But what I found with most of these agents is that after a while as I used them as I've been using them, the experience start to get worse. You know, now everything is in one thread. I started maybe like 15 different tasks. And for me to see and in order to like actually understand what's going on, I need to constantly scroll and read and try to like, you know, uh fish this information from um in this one in the history of that one chat. Um, and I think it also closes the door on proivity what like broadly because yes, you can be proactive about certain things, but if you only have one single thread, you should be very selective in terms of like what you're actually going to send to that thread. Um, and so for Wabby, for example, we have a bunch of different things that we want to maybe reach out to you about. Maybe there's a cool music show that you want to you want to see. Maybe there's a new restaurant that opened that you might be interested in. maybe some, you know, there's something um maybe there's a birthday coming up for one of your friends and we might recommend you a gift. If all of that was going into one thread, that would not be a great experience. But because we start threads for all important parts of your life, um then all the notifications go to actually all those messages go to a specific thread and then it seems much more organized and much more manageable. >> Yeah. And then I also wonder if you think this is a concern. You know, I've talked to a lot of researchers about um what happens when the context window gets longer. Um, and you know, typically the longer the conversation, the more that guard rails will break down and also there's more chance of hallucination. I'm not we I don't know that if anyone's reporting this with instinct or muse yet, but it wouldn't surprise me if that ended up happening down the line. >> That actually is also a problem. Of course, having things more organized allows you to allows us on, you know, the back end as well to understand better like, okay, well, this is related to um to your kids stuff and this related to your health stuff. So you know things are not being mixed up and you know it's easier for us to follow the context and then the threads share context. So we do understand you know in between each other. So if we need to pull some context from another thread we can. Uh but it definitely allows for a much more organized way of looking at things both from the user perspective but also from the tech side. >> And one other thing I want to uh bring up about the threads that you share with people. Um, so you are able to provide access to your calendar, your email, your bank account, etc. And that's not just for you, that's also what others can potentially access, but only within the chat. Can you explain why that's significant? >> So, one of the um interesting design decisions we made, product decisions, was to be able to was to allow people to share permissions inside a particular chat. So for example, if you're talking to a boyfriend, you probably you probably want to share your partner, you probably want to share um a lot. So for that chat, for example, I give full permission uh for the agent to access anything. So if he wants to ask in that chat again, he can do it separately, but in that chat if he can, he can ask things like when's Eugene free next week or if she bought tickets for that trip we were talking about and so on so on so on. In some other chats, I only give access to, you know, some connectors, like for example, just for to my memory and my calendar. So, it really depends. But I think this idea that you don't give permissions, you know, to everyone, but you give permissions to different chats and you can see because you're also in that chat. So, if someone asks some information about you, you always see what actually is being asked and what what response. I think that's a really cool way to solve for shared context without necessarily breaking user trust. you're obviously coming into a space that's getting increasingly more crowded. Um I I wanted to ask you about what you thought about the hardware angle of it all. So, you know, Meta earlier this month um introduced Muse Gadgets. So, it's an open source project that lets developers build their own hardware that connects to Muse. Like, do you think that hardware will be um an important piece of the AI interface going forward? Are you or do you think like just being on our phones is um is it for the for the near future? >> I think the the phone is really powerful. It's very hard to replace um replace that. However, I do think that you know our phones need to become a lot more AI first so to say that the operating system right now on our phones is really not matching the current AI era. Like if think about my home screen on my iPhone, it's still you know like 300 apps from the app store. Um, and 90% of those I forgot what they even are. I don't even know what they do anymore, nor do I use them. Uh, and so I just think there's something wrong with that paradigm. I think we'll have we'll still our main way to access things will be a phone. It's just going to need to become a lot more AI first, AI native. Uh, whatever the operating system on it should be should look a lot more like Bobby in my in in my view. There should be apps that are generated on the fly for you for by an agent that knows you well. Um everything should be a lot more social and multiplayer again just like Bobby. Um and probably the idea of you know downloading um apps off the app store is the main way to access software is probably not going to be the the most prevalent way um in the future. But I'm not necessarily believer that we'll all buy new gadgets to to access agents. The phone seems to be a perfectly fine way to do that. >> Yeah. Well, uh, you know, clearly it's, you know, fine for a lot of people. I mean, I think Muse has already had 2.8 million installs, um, as of late September. You know, close to 700,000 US daily activives in in terms of estimates. Uh, I don't think we have Instincts numbers yet, and hopefully you'll share lobbies with us soon. But this is, yeah, this is a competitive space. I mean, we're talking uh, Instinct, they raised recently, you know, $1 billion series C at a $10 billion valuation. Wabby raised 20 million last year. I know you said that you're in talks to raise again. You know, you've just opened this up not invite only. This is everyone can download it if they want to, right? So, I can imagine that's a lot of compute. Like that's a pretty expensive um task to keep this running. Like how close are you to new funding or like how are you keeping costs down? >> Well, we're definitely getting a lot of um inboundout interests. Um and right now the you know the focus is on making um also dropping the cost uh more significantly without uh without decreasing quality at all. Uh we do have some of the best engineers working um on this. We have a 15 person team and I would probably say it's the best team in consumer AI that you can find right now in this market. So hopefully we can we can provide a much better experience but also at a lower cost some of these other companies. And in terms of monetization, you don't have a a strategy for that yet. >> For now, it's free. And I think we want to keep it free for a little bit um for a while. >> Free usually means ads or like data or something right? >> For now, we're keeping it free. It's probably going to become premium at some point, but um you know, it's not necessarily it doesn't necessarily have to be ads. So, for example, Instinct mentioned that they want to go the kind of they want to stay free, but they want to maybe take a fee from each transaction. Um, I think the playbook yet for agents hasn't been kind of defined. So, I think we're we're all thinking about a few different options and we'll try a few a few different things. >> There's other companies we talked about on the show a couple weeks ago, Tab, Underdog, Hark, I can't remember which one said that they wanted to do a take rate, right, for each transaction, but I think Sean made the point, one of our uh our reporters that that might change the incentive structure for the company, right? like might they start you know suggesting that you purchase more things >> um just so that they can you know facilitate more transactions and like what if someone doesn't even want to use it for transactions right like half the stuff that you described has nothing to do with transactions I agree and I think the interesting part um the interesting point for for wabby is that not every interaction is actually inference and computer heavy um if you're talking to any of these agents big agents you you mentioned you're just talking to um you're always burning computer. You're always asking them to do something for you. Whereas in Wabby, we're seeing people talking to each other because they're group chats. We're seeing people using apps. Um and you know, when you use an app on Wabby, it's almost always, you know, doesn't require much of compute at all. So, for example, if I'm if in my health thread, I'm I'm tracking my weight with Wabby, I'm tracking my weightlifting, none of that is really computer heavy because I'm not sending Wabby on these complicated complex agent tasks. So I think what that's the interesting uh case for WABI is that like not everything what every interaction is the super heavy um multi multi-step task >> that requires a ton of reasoning. >> Exactly. I'm talking to a bunch of users on Bobby and you know my family and all that. So a lot of that those interactions are free for us. So they're more like a um and I do believe do believe in that a lot. I think we will see more and more of these use cases on Wabi especially the more more people join so you have more of your network that you can talk to. >> You know that reminds me of something that um Amjad Msad from Replet said on stage uh last week. He was talking about and a few people have mentioned this to me that a lot of software development especially like you know disposable software personal software development is actually you know using uh LLMs to build uh non-deterministic or deterministic software right so you know you're using it to build an app but that app isn't like powered by an LLM it's just a normal app right rules-based >> etc and so it actually doesn't necessarily need to use as much compute as you were saying. >> Yes, for sure. And I I do think that apps are critical. Like if you think about it right now, you have, you know, your apps on your phone are separate from your agents or your AI, whether it's GP or cloud or whatever, and then separate from, you know, your messenger apps where you coordinate with other people. I do think that that all needs to, you know, be part of the same app just like what we're doing with Wabby where it's your group chat. You can message other people, your agent, your AI is right there with you. And it makes your apps or your apps are also right there with you. And I think those um apps are pretty critical because not everything can be done just through text or even if it could or even if you can do it, it's just not very convenient. I do think that, you know, we've always talked about currently we we're in this Microsoft DOSs era of AI interfaces and there will be a Mac OS, Windows, so to say for everyone to use and it's going to feel completely archaic that everything just happen in these single thread chat chats. It it already does. >> Absolutely. And it's and I do feel there's just so much to that cannot be done in chat. Well, at least like you can't really dis discovery is very limited. all sorts of repeating use cases that just don't work very well. Like again, like for example, if you think about some of the things I'm doing with Wabby is, you know, tracking my weight daily. How would I do that? Or tracking calories. How would I do that with instinct? Like obviously I can just send it like, okay, this is what, you know, what I saw today on my scale and this is what I'm eating. But then how am I going to like how am I going to find out like for example, how much did I weigh like last week? I need to ask it each time or how many calories have I consumed today? Like it's just stupid. what am I going to every time like ask what's the total after each time I log something? Um, and these are just the simplest use cases that you know technically you could probably figure some way out to do that just in text, but it's just so inconvenient that it's again it's just like using your computer through a terminal. Um, it's not it's just not convenient at all. And I do think it makes >> uh a lot of these use cases almost like um inaccessible for for most users. So if you think about it, I do think there's right now there's just such a big spread between, you know, kind of the early adopters that uh do know how to create really complex workflows with AI and a regular consumer that still uses AI and Chip T for, you know, Chip T for search and just general kind of question as a question answering service. Um, and don't really know how to how to tackle these other aspects of of life. >> Yeah. So I do think that with Wabby what we're trying to do is to really show how this magical agentic experience can be brought to everyone and not just for the backlog of your tedious tasks but for all aspects of life whether it's helping you grow as a partner or connect better with you know your friends then >> um your friends or your network or anything like that. now. Okay. So, you you had sent me uh you know, late last week, you sent me a link to to try out the app. And the first thing I had to do was connect my email and that was like a like I feel like a lot of people who really want to try some of these apps to get the most out of them, these, you know, personal agents, um really still feel uncomfortable about connecting their personal information, their emails, like letting a company read all of their emails. How do you make people feel safe? >> So, we started with making people basically as part of like the onboarding sign up their Gmail and let us read through through it just mostly to create to be able to create a really good experience. I do feel we have a pretty magical uh first-time experience where we pre-create a bunch of threads for you based on what we learn about you. But then do I have like 20 years of my Gmail just like somewhere on a cloud that you guys are paying for? Is that what happens? >> No, we just we just look through it to come up with these topics. We don't store it anywhere. But right now we're working on a new version. The next update you'll be able to skip it and kind of join Wabby without giving access to that. I mean the caveat is of course it's in my view a slightly less great experience because you have to basically tell us what you want for Wobby to work on. But I do like in the end of the day the more data you connect uh Wave can just be a better assistant to you a better agent because we we bet a lot on that proactivity element but in order to be truly proactive we need some of the some of that context. >> Okay. So then let's say I give you, you know, the connectors are are all the specific APIs of each company, right? So if you're connecting your your finances via Plaid, that's Plaid's connector. It's Gmail's connector. It's, you know, you're you're trusting the connectors essentially from these larger companies, these bigger enterprises to know that that's a secure connection. >> Yes. Correct. So you know we're really working with you all the bigger companies and APIs to to do that to bring that experience >> in terms of like data storage data collection. And this is, you know, a lot of people's extremely personal info, right? Like their, you know, medical information, sometimes their kids information. What would be a red flag for you in another company, let's say you weren't doing WY. What would be another what would be a red flag for you in terms of not feeling safe to to make those connections? >> I think there's a ton of um there's definitely, you know, a ton of stuff to look out for, but I do think that, you know, you probably want to make sure that that is legitimate startup. It's not a one person kind of bipoded thing you know that maybe has some funding some bigger backers behind them. For us I think the you know we've built I built company called replica before which is which was the first CI companion and people were sharing very personal information with us. they were, you know, they've they've been having conversations with replica that, you know, we would know things that no other service even >> today's AI agents, uh, you know, would never learn all the things that people would tell their replica. And, you know, knock on wood, we had with, you know, over 40 million users, we had no um, not one accident, actual incident with any of the that data going anywhere. We made a firm commitment to our users from the very beginning with replica that we will never sell their data. We're never moni never going to monetize their data, never share it with third party and um you know we we kept our promise on that front and also in terms of security. So I think we have pretty good track record and you know we're making the same promise uh here with Bobby. I do think that that trust is absolutely critical with your with your users in the end of the day and trusting you with all this information. Um, but I think it's, you know, we're definitely building something that's that people can trust. >> Uh, well, that's about all the time we have today. Thank you, Eugenia, for joining us talking about positioning for WABI and, um, what you think about AI agents in general? Thanks for coming on the show. Is there anywhere that our listeners can connect with you online? >> Yes, of course. Um, uh, probably on X. Um, E Kuy Da is my handle. Just DM me. My ideas are open and happy to walk you through the app, but ideally of course just uh download Wabby and find me on Wabby. My handle is Eugenia and text me anytime. Great. Um well, thanks again our listeners. You can find me on X and LinkedIn. You can find Equity on X and threads at EquityPod. Talk to you next time. Equity is hosted by Techrunch Senior Reporters and produced by Terresa Loans Solo with editing by Kell. Subscribe on YouTube or wherever you get your podcasts and find out what's next at techfrunch.com/events. Thanks so much for listening and we'll talk to you next time.
Assuntos: Agentes de IA, Assistente pessoal, Dots, Muse. Publicado no YouTube em 2026-10-22.
