The Startup That Crashed Its Own Servers Because of Open Claw

40:40And, you know, like we were like, oh, this is a really cool, you know, like product and all. It was like a huge improvement in the GPT-2 and it could now do all these cool things, but it's like really not reliable enough. And like there's just all of these like, you know, hallucinations, which funny enough, it's like fully gone out of I think most people's like vernacular at this point. right that's true we talked about them a lot and way less and like yeah who's even really thinking about them now so and at the same time you know like the models have only gotten better it's enabled massive companies and new applications and ways that people have before and i think at the time i was like i mean like this is a big step change how like is this like like is this gonna just be the way these models are forever gonna be like are they gonna keep getting better and I remember at the time it really wasn't obvious and then a year came by and there was two generation models that unlocked all these things and another year came by and they kept doing that for like three years straight and then now it's hard to look at that and be like look you can't bet on the models not getting there.

41:41Every time there was a talk or rumor of a wall quote unquote all the model labs and just smash right past that and go on and do something profound. This next question does touch on it matters to your company like do you think we're gonna enter a world where it's like i have my agent and it's like it's my sidekick i want to do everything i want that agent to sort of interact on my behalf or do you think it's more this sort of like fleets of agents different tasks more dispersed than that oh wow that's a really good question what do you think yeah what do i mean i think i want the singular agent like even if the agent is sort of coordinating with obviously like other agents, just something with like all the context.

42:23I mean, maybe it's a terrible security idea, but like, I don't know. It feels, but I will say the opposite of that is I thought with OpenClaw, we saw what was appealing about multi-agent where they're interacting with each other and maybe there's more benefit. I don't, I don't know. I don't, it's, it's chaotic. It's hard to sort of game out the world, but I do think, you know, for, for you guys, if I have this agent with all the context that's sort of my sidekick, you know, maybe it's keeping more of the knowledge itself. It doesn't need to go to sort of external places to understand the world as much or I don't know.

42:56Yeah, that's a really good question. And to be honest, I don't really know if I have the best answer there because my guess is like, well, I mean, even let's take a look at the shape of the world today. Right. And then maybe we can hypothesize right now. I think there's a combination of both because, you know, you're probably using one of two between you know like chat gbt and claw mostly claude right now yeah it was hard to move because i was deeply chat gbt and it had all this context and then and it's like okay it seems like claude smarter uh yeah yeah which do you have a loyalty i'm i'm pretty i'm pretty loyal to the uh to the claude uh team uh you're like it's an ally yeah yeah exactly i mean like in many different ways but i'm also just a bit very personal fan nothing against open ai like you know or any other products at all just my personal preferences um so it has all my memory and by the way i don't know if you have memory turned on for your cloud yeah yeah i didn't realize you can just go in and read your memory oh i need to do that it's like somewhere in the settings i literally was remember i like found the tab i and then i was reading what it knows about me and my goodness it knew a lot i remember i like and i also recommend you asking this question i had a ton of fun with it uh recently it was like based on what you know about me this was exact problem right based on what you know about me do a psycho now a psycho analysis of me and my goodness it was like it was like it knew where i was born it knew you know like uh it was like yeah since like you know because you move from like this rural town in china this specific name you know um and you came to canada when you're about six years old right it like you know this indicates that based on these other messages It's like all of these things.

44:35I was like, whoa. I was like, I don't think anybody knows about me to the depth of that. But Claude does. So I think that Claude, for instance, will be a personal agent. Or there's going to be obviously these providers where it's servicing you as your personal agent. And maybe that'll be the exposure of it. But does that mean that it'll be the only agent interact with? Does that mean that there's not going to be specialized agents for your company handling support questions or writing code or doing all these things? I'd be hard-pressed to believe that's just going to be that singular. Right. Single one.

45:14I think specialization in the same way anything else is, but who knows? All right. Last question. I mean, yeah, sort of fast-growing company, lots of excitement. I mean, what do you want to sort of hold yourself to in a year? Or what would you like to see yourself and the company achieve over the next year? A lot of things. I think the first and foremost thing that comes to mind is really making sure that the knowledge layer and the knowledge infrastructure that is only really talked about, in my opinion, in the broader theory and the broader ether role, really manifests into a product, into a meaningful product.

45:55we truly believe that the biggest inhibitor for a lot of agents and a lot of companies who are investing so deeply into agents right now is just not providing it with the rights and the accurate information the context that it deserves to actually go and do it i personally think the intelligence the models are there i really do right there's a whole debate on whether or not we've achieved agi i won't i won't i won't pick it we're just not good enough prompters you're saying it's like if we could only give it the right information it could be agi uh or close to it. Or at least achieve the vast majority of tasks that a lot of the knowledge work is trying to automate right now.

46:31And we want to be the company to enable that and to kind of build that into an actual form of a product and then enable that into the companies we work with. Amazing. Thank you so much for coming on the show. Thank you so much for having me. Thanks to Han, a fantastic guest. And next up, we have Jesse from Decagon, which ranked number four in late stage. You can think of Decagon as AI agents for your customer experience. So we work mostly with large businesses that have lots of customers and big call centers and contact centers. And we deploy AI agents in front of those customers that have calls with them or chats.

47:06So we almost call it an AI concierge a lot of the time because we think of it as like this new layer around your brand that customers can interact to. And this is the first point of contact that a lot of customers will even talk to in the first place. So we have sort of multiple audiences consuming this. I think from the investor perspective, the space you're in is like one of the two, maybe with legal, where there's like, I guess, almost certainty that there is going to be an AI answer, right? You are fighting with Sierra. There's this great dual legal, like Harvey Ligora, even up there's some others.

47:42But there's so much excitement that these are sort of customer experience and legal, like these applications that really understand what's happening. From the sort of, I guess, layperson view, there's the like, now I'm talking to AI instead of a human. Like, am I happy about that? Am I not? And sort of convincing the sort of regular person that this is like a good experience for them and something that they'll ultimately be happy that AI is answering their question. So I'm excited to sort of get to both like big trends in this conversation today. I guess starting off from the sort of like investor tech industry perspective, like why do you think there is this excitement about like customer experience like what is change what technology has made this possible right now that you think there is going to be this sort of movement in how we handle sort of customer support yeah i would put coding in there as well oh yeah of course yeah coding perhaps the biggest you're right yeah that's almost his own yeah it's a meta sort of process it's changing how we build all these companies and everything yes Yes, coding is almost distinct to me because it's sort of like how the sausage is getting made of it all.

48:50But yes, I agree. Yeah, so I would say those are kind of the three big use cases that have emerged. And it's really just like what are the models good at? And so it turns out that the models are very good at writing code because it's also tokenized. And obviously people have discovered that. And so there's tons of funding and tons of effort going into that. Similar to our space, I think folks have realized that one of the things that AI is very good at is having conversations. And so because our space is inherently conversational, right? Like the product that we build, its purpose is to have conversations with end consumers that has lended itself very well to AI.

49:29And then the market's also massive, right? So I think when you have those combinations of massive market, it is what the AI is good at. And then I would add a third thing, which is it is is actually really easy to see value because you can it's very measurable. Right. There are a lot of use cases where it's like, oh, yeah, I can see this being useful. But, you know, how would I quantify that? Like it is is it like how do I show ROI basically? And in our space, it's very easy because they know companies know I have all this call support. Like I'm having to answer customer questions. So they're they're comparing.

50:01All right. This is what I'm spending now with you. I could be spending less? Like how much is the pitch? This is cheaper. Exactly. Yeah. So if you think about like what the conversations are or what conversations being had in these boardrooms or C-suites, it is like, where can we apply AI? And then the customer experience is one of the biggest ones, right? And it's not just saving costs. And so saving costs is one of the easiest ways to frame it because it's like, oh, we're spending tens of millions or hundreds of millions of dollars a year in our contact center. And we know that a lot of that could be handled by AI.

50:33potentially at an even higher level with higher customer satisfaction. So that's the cost saving side. There's also a whole other side that we focus on, which we would consider like the revenue generating side. And that's pretty large because it's very uncapped. It's like outbound sales. Yeah. So we don't really touch like the cold calling like that. Maybe one day. But there's a lot of sort of revenue generating conversations you could have that are both reactive and proactive, right? So imagine someone reaches out and you resolve their issue, but you also notice certain things in their account that are being underutilized or you're trying to like tell them about this new product that was launched.

51:07We've seen that land very well because you've just solved their issue. And so you've kind of, some people call it like earn the right to, you know, engage them more and like just keep them more engaged as a customer. And then there's also proactive. Upsell is a word that I might use, right? Yeah, upsell, cross-sell. Right. And then you can also be proactive at the right time, right? So imagine, you know, someone signed up, they're going through onboarding flow and you notice that, you know, they've not been active for a week or two. you can reach out at the right time and actually like increase your conversion rates because maybe they were stuck on something and you just unblocked it for them and what's the breakdown voice uh sort of chat email uh we're pretty balanced at this point in terms of voice and chat uh email is much smaller and so i think the rough frame of reference might be like 45 45 10.

51:53and voice people know that it's ai or how close are we to sort of seeming like a human and is that something we want it is pretty close i think there's you're never trying to like hide that it's ai so almost always the first message is hi you know i am jesse your ai you know concierge but But what we've seen is like we just went live last week with a large bank in the US. And when they look through the data, like one of the things both of us, us in the bank were like very impressed by was how just like normal all the conversations were. So there was there was no like, oh, like, are you an AI or anything?

52:36It's like, yes, like you say that you're AI, but because the voice and the conversation flows so smoothly, like no one really cares after that. Like they're just trying to get their issues solved. And so if you can get your issues solved throughout the conversation, they probably just forget. And so we see a lot of nice conversations that end up with, you know, like, this is so helpful. I have a nice day, like that sort of thing, where you would not really say have a nice day to an AI. It's nice. Well, especially if they become our overlords, it's good to thank them enough times along the way.

53:03Exactly. You always want to be nice to them. But yeah, so I think if you can make the experience good enough, then it doesn't really matter. And in fact, we've seen that in a lot of these experiments that where we've deployed AI, the customer satisfaction actually becomes higher. than what do you deal with like the customer just says human human human like how do you treat that how do you do you just resist it because you want to sort of show them well this is actually a great experience or how do you deal with the yeah desire for speaking with a human being the goal really is to make the experience very different very quickly so you have to as soon as possible establish that this is not one of these phone trees that you're used to that are frustrating and you get stuck in loops and you have to press one for this press two for that so the goal is to make that super clear and that means just showing the ai's uh like empathy in the voice and being able to make it super personalized and uh you need to do that in like the first or second message and so we had uh we had a deployment with or a ring where we did a case study and beforehand it was every three people that came in was just agent agent agent like i don't want to bother with this and now about six months in uh from the deployment at that point we measured it and it became one in 20 so it became way smaller because people were willing to give it more of a chance because you could show that it was different and then once they got into the conversation it's like oh yeah this actually can do stuff for me you know how much of the experience and mostly I'm thinking with voice, do you leave to like the bank or whatever that you're a customer versus no, we sort of have a sensibility.

54:46Like for example, like the voice, you have like sort of a menu or they could totally bring in their own voice. Like how do you think about what sort of your special sauce versus stuff you can hand over? Yeah. We see our role as an advisor to the extent that they want it. Right. So we have our unique ability is that we have a ton these deployments where we've uh you know gone live and so we have a lot of experience on how to design conversations and so on our team we have these specialist roles that we call conversation designers which you didn't really like no one really needed them before but like now we have these like really elite folks that are um kind of there to advise the customer on oh you here's the procedure you want to do great that's great we can do that but like in our experience it's better if you you know combine these things or like direct the flow this way because that's what is most natural for a gen ai model and so that we're kind of advisor there and we we partner with them on i'm getting to the fastest uh sort of solution possible same with the voice and so what we found is that everyone has a pretty different opinion on what's a good voice and we'll talk to one customer and like you know they're like oh this is absolutely the best voice and go somewhere else and I don't like that voice.

55:57It's like too happy or something like that. And so, you know, our conversation designers are also sort of like voice experts and they go in and can help people navigate a sample of voices. And sometimes they bring their own too. And so we have the ability to clone a voice. And so they've, sometimes they've spent a bunch of money before like creating a voice. Yeah. And so we can just take those samples and build that into the AI. Do you have customers that deploy different voices based on what they think is customer, trying to match, I don't know, demographics or whatever? of the customer they're interacting with yeah so uh the most common one for sure by country uh like if you pick up the phone in the uk versus here of course you're gonna have different accent you can also even do it by zip code so you can split the us like you can have southern accents or like you know boston accents and does that have an impact um it has i would say like a mild impact um but it's like it's like hard to say if it's like a huge thing but i think it's just more of this general theme of how do we make it as personalized as possible.

56:58And so it's both the sound of the voice, but also what you say, right? So one of the big other advancements in the space in our product recently is this concept of user memory. And like the reason why ChatGPT feels like it gets better over time is that it kind of remembers things about you, right? And so we give our customers, so the businesses that we work with, the option to turn on user memory, and where the AI is creating this dynamic memory profile so that over time it becomes more and more personalized. And the cool thing is that this memory profile doesn't even have to be specific to us.

57:31So you can actually like integrate it into the rest of your experience and have it just keep updating. And so next time they come in, we know exactly like what you've done in the app. We know exactly like what you talked about the last few times you contacted. And it just makes for a much, much better experience. Do you think the business is gonna remain balanced between chat and voice? Or like, do you have a view of where the world's headed? Yeah, of course. So I think prior to Gen AI, the trend was, hey, let's drive everything towards chat because it's more efficient. You can have a human agent that is doing three to five chats at once, whereas over voice, it has to be single-threaded.

58:08I think now that issue is resolved because you can use AI. And I think in our view, it will be pretty balanced in the end because there's just going be different situations different demographics that prefer either one like you know the the common trope is you know younger folks like chatting and older folks like calling but you know if if i'm in the car and i'm like on the go i'd rather call as well and so i think both are very natural very natural means of communicating voice is not going away like if you think about what voice is like voice was like the original ui for humans it's like before we had anything before I keyboards or phones or anything.

58:45It's like the way you communicate is through voice. And so that's not going anywhere. Well, it's funny that we have these almost like putting this voice thing aside for a second. We have like two contradicting trends happening right now. One is obviously the rise of language models and sort of everything in this chat bot. On the other hand, there's TikTok, which is like nobody wants to read everything short form video, like the way people get news information is video. And in some ways, voice is like sort of the synthesis of these two where so you could see sort of cultural movement go back towards conversation if the technology is like able to deliver it um i think it's just a much more natural form of communicating and like now even when i use chat gpt whenever i can if i'm not in like a crowded room or something i use the voice medium just because i don't do that interesting does it get does it get extra information from the emotionality yet or or no it's just like reading it as text right it does So what that model is, it's known as like a voice to voice.

59:41So it's like kind of audio in, audio out. And that is, I think most people would say, including us, that that is like the long-term future of the space. There are a lot of problems with those models right now. Like they're a little bit inconsistent and you have hallucination issues. So in production, when we work with like a bank or a telecom or something, like, of course, you have to be a lot more careful because you can't make any mistakes and so you you can't necessarily use the same technology as you know consumer attached apt for that but in general uh yeah we think that is the future the challenge with voice voices is like harder to sort of audit it in text right or it's like it's harder to audit it's like it lives in this voice to voice sort of so there's like emotionality and stuff that we can't really translate necessarily to language.

1:00:30Is that the right way to explain it? Or like, what's the barrier to voice-to-voice? Yeah. So the main barrier I would describe as the hallucination rate is higher. Why is hallucination rate higher? There's a bunch of reasons. And one of the reasons is that the number of tokens streamed by voice-to-voice models is a lot higher because similar to what you said, right, you're capturing more detail. You're capturing like intonation, et cetera. And so for a sentence for any given sentence if you chop it up in text it would let's say be like you know 10 tokens or something in voice it could be like 80 to 100 and the more tokens you have the more opportunity there is to mess up and so that's why you see higher hallucination rates in these voice to voice models and a lot of the research the labs are doing are is kind of geared towards making that better because i think we would all i think we all want the voice to be very emotive and like to capture our emotions, right?

1:01:26Just like we're talking right now. But until the hallucination rate is cleaned up, you can't really leverage them in these like big production use cases in like in our space. How do you stop like weird edge cases where like, so an agent says something really sort of bad. Obviously a human could do that too. Like what are the systems you have where it's like, oh man, we had one really sort of fire off weirdly. Like you're sort of have another sort of layer of technology monitoring everything or what do you do there to catch when inevitably some weird thing emerges? Yeah. So that is a huge topic.

1:02:01And we think of it as like a three-prong problem. It's not just checking it during the call, which you have to do, but you also can prepare for it beforehand and review it afterwards. So the three prongs I would consider are before the conversation, during the conversation, and after the conversation. Before the conversation, what you can do is we call them simulations. And so you have these AI agents that you've made, and before you release them to any customers, you kind of run them through simulations. So it's almost like a second AI comes in and it's talking to your agent and like trying to get to mess up or testing all the common use cases that it's learned from reading historical transcripts, right?

1:02:39So that's the first thing you do because that really shores it up. And that allows you to iterate faster because let's say next week, I want to change something. Well, before I release I can just run these simulations and I feel good about releasing it. Right. During the conversation, there is what you're saying. So those are like supervisor models. And those have to be like really fast, specialized models that go and detect for certain things. And so I'll give an example. Let's say we'll use the bank example again. Like one of the things that if I'm a bank, I would not want the AI to do necessarily is give financial advice.

1:03:10Right. So I don't want you to give financial advice that's not in the scope of what your job is. And so how do you make sure that it never does that? Even if the user is like trying to get, they're like really determined to get financial advice. The way you do it is you have to have these supervisor models. And so there's these small models that we've, we fine tune ourselves that are really fast and really specialized in detecting things like that. And you run them during the conversation. And if it detects a violation, it can fix it in real time before the, before the response goes out. And then finally the third prong is after the conversation.

1:03:42And so there, what you do is you can review the conversations with a second AI, right? And we were talking a little bit before, but I think this is where a lot of the space is going because if you think about the big advancements in AI right now, it's these slow reasoning models that are really good at coding and really good at reasoning. You're never going to use them in the middle of a call because they could take up to a minute to reply, right? And you're not going to wait there for a minute. It's like, give me a second. And then it's like, thanks for a minute. And it doesn't even like improve those metrics very much anyways.

1:04:16But what they are really good at is kind of this like slow autonomous reasoning. And so one of the things that we've pioneered and we kind of released the first product in our space. What is it called? It's called Duet. So Decagon Duet. And the idea is it's kind of a duet between that like slow reasoning model and like the fast one, right? And so the reasoning model, their job is they can basically run overnight and autonomously just like read every single conversation. and basically figure out what is going well, what's not going well, figure out what you need to do next and actually do it for you.

1:04:50How does it implement what it learns? So let's say it figures out like, oh, there's this one topic that we are not doing really well at. And I've looked at our knowledge base. I've looked at all the procedures. We call them agent operating procedures, AOPs. I've looked at all the AOPs. I've looked at like the coded tools that the AI is able to use. I've looked at, you know, the data that's come in from these tools. And I've realized that the problem in this case is that, you know, in this step of the procedure, we're like sending people off a wrong track because like they don't actually want to do that.

1:05:22And so because I've read 50 ,000 conversations, I'm very confident that the fix is XYZ and I'll go in and suggest the fix. In the morning, someone on the team can come in and just like fix it. Well, so if you think about how the space generally works, like a lot of software, even outside of our space people are building these ai assistants but generally the ai assistants are there to like oh like show me how to find this or like go do this for me you know like go do it but what we're talking about what duet is is it can do that but it's also like a autonomous like agent that can run in the background and that's what a lot of the coding tools are moving towards right now right and obviously i would argue that our the work that our agent has to do is like a lot simpler than coding so they actually do a better job of it um and so that's really what we pioneered and we have some folks in our space who have released similar things but they're they're more focused on that like first use case of just like okay write this for me rather than like actually something that can just like run overnight for several hours and like get a lot of work done i'm sure this is a piece of it but what how would you distill decagons right to win i mean obviously you have very well-known competitor and then obviously an existing industry and i'm sure lots of other startups i've never heard of trying to run at this space like What is your right to win?

1:06:33Yeah. So I think the thing about AI is that the markets are huge and there's a lot of opportunity. So that's a great thing. And then the flip side, of course, opportunity attracts a lot of players. And so in our players, in our space, you have like newer players like Gen AI native companies. Like you mentioned Sierra. And you also have like the older ones, like, you know, the big platforms, Google, Salesforce, etc. And of course, our goal as a team is to destroy all of our competitors. And we hire a lot of killers on the team. You ranked higher than Sierra on this list, which is interesting. Yeah, we continue destroying the competitors.

1:07:16But we have a lot of respect for our competitors. I think that's also something that's very important. When we're building internally, you can't demean anyone or underestimate anyone. That's very careful. But I would say ultimately what it comes down to is like a very different product approach. So I think the reason why in a lot of big spaces there are multiple winners, or in our case, if we want to be the winner, it is because you have to take like kind of a bet on a specific approach in the product. And our bet really is that the biggest differentiator for our products and products in our space over time is going to be sort of the empowerment of non-technical folks and sort of the decrease in the cost of ownership.

1:08:03So you think about how a lot of software normally works. It is the setup phase is like highly consuming. There's like whole like professional services industries based around this because you need technical resources to get Salesforce set up, for example. and there's a lot of benefits to that like you get really locked in afterwards you're they're kind of relying on you but we think that with ai you're going to have to take a very different approach and a lot of our competitors are are very grounded in the older approach which again it's not a big deal to do it for the whole company it's not necessarily bad right like you want like a product team to say oh we're going to use decon for our exactly okay and we want the ops team to be like hey we want to use decon because they're going to allow us to move a lot faster and we're not going to have to rely on an engineering sprint or like call up the vendor every single time.

1:08:48So that's our biggest differentiator. And when a lot of - So it's like Slack where a team can deploy. Exactly. Yeah. And so when a lot of these businesses work with Decagon, it's because of that, right? They feel like, hey, we're a big business. We have a lot of complexity. Getting the AI live is maybe only about 20 % of the work. 80 % of the work is the constant iteration. And we're going to build a lot of new flows. We're going to have to add a lot of new surface areas. And if we're reliant on the vendor and it just kind of feels like a black box, that's not possible. You have tech companies are more likely to adopt Daggergon, right?

1:09:18Or you've been strong in tech? So tech, financial services, airlines, telecom, I think a lot of the older industries, we've seen the same thing where they have less strong engineers overall, potentially, or they still have a very strong engineering team, but like they have a ton of other stuff to do. and building customer experience software is not like one of the things that they want to specialize in, right? And so in the past, they would have had to hire a bunch of professional services or like pay the vendor a ton more. But in our case, it's really, that has become the big differentiator versus the older approach.

1:09:58There are a lot of Americans who hate AI. Like how much do you think that's a barrier to your business succeeding? And like how much do you think there's going to be sort of like a hearts and minds battle or it's just the product has to speak for itself or what do you think sort of ai sentiment translates into like sort of people's willingness to engage with ai voice and customer support oh i think it's very important i definitely don't think that's something you can gloss over because um and so when we're building the product like a big mantra that we try to adhere to is that we're also building for our customers customers because at the end of the day, it's like the goal is to make their experience a lot better.

1:10:39And that of course benefits our customers, right? So if we're working with an airline and, you know, let's say we save them a bunch of money, but then all the customers are pissed because they can't like get what they wanted to do done. Like that's not a win for anyone. Like that might be a short-term win for the airline, but like, you know, they know that like, that's not really what we're going for. And so a lot of the product work that we do is geared towards like, how do you make the end experience better, right? Things like user memory that we talked about earlier. And those are things that are you allowed to do that across customers or is it user memory within a customer?

1:11:11Oh yeah, we don't do that across customers. There's no data sharing across customers. I don't even think people would want that necessarily, but yeah. Yeah. It's hard to say. I mean, if you know me, why do I have to waste all the time? You know what my preferences are, but. So I think what will happen is there will also be consumer agents. So I think what the world looks like in three to five years is all the brands will have their own AI concierge. And, you know, hopefully we're powering and helping them a lot with those. But there's also going to be consumer agents that users use. So, you know, if you and I want to use something like...

1:11:47It's like having a lawyer. It's like their lawyer, my lawyer sort of fighting with each other, right? And then they, yeah, the agents will connect. Right. Isn't that gonna be a nightmare for you like those my i mean in some ways you have to brace for that world right because like my agent you know i assume will be meaner and like i like there's a level of like propriety that i have whereas they're just like going to treat you like a system to be sort of manipulated and and you know try to find every hole to get what we want right um i don't think it makes our life harder i think it is generally good for us and the world overall because it's an arms race you have to fight or i don't think it's necessarily arms race i think it's just it creates much more communication so it's much easier like i think right now a lot of a lot of this communication doesn't even happen because like oh i don't even bother calling that number because like i know i'm gonna get stuck in like some loop but no i'm gonna tell my agent to do it and the agent can actually get it done because like there's a business agent over there and so it just makes like overall like the number of interactions much much healthier but you you're not going to play on sort of my agent side or are you interested in that side of the game no right now we're very focused on working with businesses um and so the the brand agents if you will the consumer agents like i think chat gpt will probably it's like those type of tools will be uh will be the consumer agents you think it'll be chat gpt itself chat gpt i mean uh like perplexity has an agent claude has an agent uh it's like apps like that that are like geared towards mass consumer apps like gemini for example do you think any of your customers will say you're only allowed to talk to us if it's you or like have you seen people say uh no i actually think um i would say the the sentiment in most of our the businesses we work with is is very positive they're like they're excited for agent agent world why uh because i think that is like that will yield more business overall because it's like bringing the barrier to entry a lot lower right so like if you're a hotel or something and like agents can make bookings on your uh on your like in your platform then you presumably you're gonna have a lot more because the buried entry is a lot lower do you think like the rules of like who gets a refund when are going to become more transparent because it's going to be possible to sort of test every customer system and say all right if you say it's broken two days old they'll get like are the rules going to end up being sort of publishable because it's going to be so discoverable or people are going to play this cat and mouse game or they're going to sort of randomize it or like what happens when it's yeah how do you see that playing out i i think the the the steady state is that there are going to be no like games necessarily because like right now there may be some games because friction ads etc right but um in in the in a steady state because it's two ai's working together uh yeah there are transparent rules uh there there also maybe like judgment calls, but the AI is the one making the judgment call.

1:14:38So it's like fairly unbiased. That is, I think that's, that's the goal. And that that's the world that we're driving towards. Right now, I think there's just inefficiencies because you have businesses that have like their leadership might not even like want these games to be there, but it's just over decades, you've kind of built up these things and you know, you're kind of afraid to change it because you know, like it might dramatically increase the number of inquiries coming in or or change your P &L. But part of our job is to really work with them through that process and design these conversations so that you're not losing anything.

1:15:12And in fact, you're both driving up your returns and making the experience better. And I think that's a big reason, back to your question of why the space has puffed off so much. It's really like you have these massive ROI case studies that have already been realized by big businesses where tons of savings, tons of customer experience increases. And that's the goal. What technology do you really want to see improve? And like, where are we on like voice latency? You also are an industry that sort of has to rely on the past generation model. Well, you need sort of smaller models. And I'm sure it's a much more price sensitive customer base.

1:15:55Like, where are you most hopeful that technology improving and the models improving will change your business? A couple of different axes, right? So there's the reasoning models with products like Duet. The reasoning capabilities will continue to help those. Open source models getting better will continue to help our underlying agent pipeline. Because why do you use open source models? It's mostly for performance because you need latency improvements and you need the models to highly specialize. and so we found that by fine-tuning and training those models we're able to get like much faster latency and much higher like comparable to higher performance than the big uh close source models and then it's the voice models and so you know voice-to-voice models getting better and more accurate i think those are probably the three big dimensions that we pay attention to and each one of those improving improves significant things about our product what uh what open source models are you most excited about right now?

1:16:52I mean, you have people all over the world making open source models, right? So China's obviously very good at open source models. The open source models in the US are improving and hopefully even faster. And you also have Mistral. And so I think we found that these models are good at different things. And our customers might also have preferences. And so we have built in a way that's very model agnostic. And you can swap in and out different models. But yeah, we're very hopeful that the smaller parameter models will get better and better because that is just... Are you using like DeepSeek, Minimax, any of those?

1:17:30Mistral, Quinn, we think pretty highly of. And then we're keeping a close eye on the Gemma type models, Halama, things like that. You think you have to stay away from the Chinese models? You don't have to stay away from them. but you want to be fairly diversified so that in situations where you don't want to use them, you can. You just have some customers who don't want to. Yeah. Which then limits how much you can go all in on them. Yeah. But you wouldn't want to go all in on certain models anyways. It's like this. It's like, like you want to, cause if you, if you get over-reliant, it's like an analogy would be like, if you're a country, you don't want to rely on another country for all your oil, you know, it's like anything can happen.

1:18:10So we want to make sure that we're building in a very robust way. Do you build your own models or is there any use in that? We have our own model, but they're not like trained from scratch. So if you're in the application layer, what you're typically doing is you're using one of these open source models. The open source models are not going to actually be that good out of the box, but you can fine tune them. And if you do that correctly, they actually become very performant at the tasks that you want. Are you optimistic across the application layer or where do you think the foundation models will just gobble up applications versus where do you think there's opportunity to build standalone businesses?

1:18:48I'm obviously quite optimistic about the application layer. I think just that you started early enough and you're sort of running alongside them or what? No, I actually think so. One, I think most of the value will accrue to the application layer because you're solving like business problems. I think the labs also agreed with that, which is why they're building a lot into the application layer. And I think the application layer is so vast that there's going to be a lot of different ways to handle it. But generally, I think a good framework is that most of the labs will want to build applications that are fairly broad, like broad and maybe like thin, because they have such a vast surface area that they want to build things that a ton of people can use.

1:19:32So coding obviously is one of those where it's It's mostly like an app that people can just pick up. If you think about our space, it's quite different. For better or worse, we have a very involved post-sale motion where we talked about our conversation designers, but they're part of a much broader team that we use to actually work with our clients and get their agents stood up and help write these agent operating procedures because we have the benefit of having the expertise. and because you have that post-sale motion it's i think it's pretty unlikely that there's going to be an app that replaces it um but over time i think the application layer is like so thick that there's a lot a lot of stuff to build so i actually i think there's going to be a pretty bright future for most application companies what just as somebody's sort of so close to the space and how things are developing like how do you think the world looks differently in in five years uh specific to our space or just no no just broadly like what do you think like i don't know yeah for the regular person just watching this trying to understand how ai is going to change their lives like what do you think feels the most different in five years because of what people are building today uh so a couple themes so the first one in our space we kind of touched on this but i think the way that consumers are going to interact with any brand is going to be fundamentally different and you know hopefully we are again a major player in that but if you think about the previous shifts, right?

1:20:59Like last one, let's say from internet to mobile, right? That basically created like entirely new UI for people to interact with, you know, the brands that they need to interact with, whether to buy things or to get something done, et cetera. And AI is, you can almost think of it as like a new UI. And this UI is conversational. And so you can talk to it on the phone, you can chat with it. But that is something fundamentally that's going to change. and you're going to have agents people like to push back on that some people like shopping you know like some some of these cases it's like where people talk about like travel and it's like people like spending time booking travel in some of these cases like do you really think it's going to be like hey like text in go figure out my travel go figure out well i don't think it's necessarily like a clean replacement right like going from web to mobile some people still if i'm on my computer i'll just use the website people go to stores right but it's more like a creation of a new medium.

1:21:56And yeah, there will be situations where I would rather talk to the agent. There might be other situations where I'd rather use the mobile app. And so you have this like new UI that's created. And this UI is a lot more friendly for consumer agents because consumer agents can also go to your website and try to click around. But that's like a very crude approximation. But agents, again, are very good at conversation. And so the two agents are talking. You just, like back to my original point, you're basically just massively increasing the number of healthy interactions between consumers and brands whereas right now a lot of them i would describe as like unhealthy or there's not even happening because i can't be bothered to do them right so that's one big thing that will change i think this is an argument that's like oh if you hate talking to ai customer support like don't worry any i know you're trying to be better than the loops that everybody got and so in a lot of cases they are happy to talk to you but there is also the further pitch which is soon enough you can have somebody talk to us and then get the readout of like what transpired and you don't actually have to do it yourself, which I think people will be excited about.

1:22:54Okay. So that's one. Yeah, that's one. That's specific to our space. So we're really excited about that. And I think that's why the market is so large and why there's so much interesting work to be done in our space. More broadly, I think something that will happen a lot more is right now people think of AI as it's like a tool there. Kind of like Google, right? You go to Google and you like search something and like, okay, I go to my, I go to AI and like, get something done. But I think in three to five years, it'll be much more normal to just have like AI running continuously. And it's just like always running on whatever.

1:23:29And it's kind of happening in the background, right? Because right now it's kind of like you ping it and it does something. But the sort of the standard practice that we're moving towards is just like long running, autonomous things. And that's cool for a bunch of reasons. One, you can just accomplish like way greater tasks and two you can really increase leverage because like you don't have to be involved like it's just doing things and so this is a this is a big trend in the coding space if you're following those like uh a lot of what you start to sleep wake up and see what's happened overnight exactly yeah or what started with like it's it's kind of tab auto completing things now it's uh you know it moved to okay you give it a some prompt it can like write whole things then you review it and then now it's moving towards like okay it's just like hey let me give you some instructions and then like you just kick it off and it's like a colleague doing work right that's the same concept again behind duet where we've kind of created this this new concept of like a long-running agent in the background and i think that will happen with consumer agents as well where you just give it like long tasks or it's constantly listening to you and it's just like a always there like helper great jesse thank you so much for coming on the newcomer podcast cool eric thanks for having me and congratulations for being on the list thank you all right sweet Thank you.

1:24:43Thanks for sticking around to the end. Please, if you've made it this far, you've got to like, comment, subscribe. Excited to grow the channel. And of course, you can find our writing and reporting on startups and venture capital at newcomer.co. We also host events. You can check out what we're doing at newcomer.events. Share the podcast, help us get distribution, support the channel, comment, support, like, comment, subscribe. You know, you know the deal. Thanks. Thanks for being on the journey with us. All right. See you next video 2.