The Voice AI Platform Powering a Billion Calls a Year

Author: YC Root Access · Watch on YouTube

Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.

Original language: English

Original video: The Voice AI Platform Powering a Billion Calls a Year · Channel: YC Root Access · 31:53 · Watch on YouTube

Topics: AI agents, Enterprise AI applications, Markets and economy, Multimodal models

Summary

Vapi’s founders describe building a platform for deploying voice agents, now serving about a billion calls annually for companies including Amazon and Uber. They recount years of pivots—from productivity software and meeting notes to an AI therapist—before turning their voice-agent technology into a platform after early customers struggled with latency and reliability. The conversation explores why deploying voice AI in production is difficult, how the company learned to serve enterprise customers, and why the founders believe this market will take years to develop. They also discuss scaling the company and using coaching and communication practices to strengthen their co-founder relationship.

Key points

Vapi provides a platform for deploying voice agents and reports serving about a billion calls annually.
Voice agents are difficult to bring into production: businesses need low latency, reliable performance, coherent customer experiences, and appropriate guardrails.
The founders’ early products included a friendship CRM, an investing tool, a meeting-joining utility, meeting notes, and an AI therapist.
They saw a broader opportunity when businesses building voice products struggled with latency, reliability, and configuration.
Early customers came largely through word of mouth; the founders supported many users closely and shipped requested features quickly.
Serving enterprise customers required learning procurement, contracts, and sales, and eventually moving one founder from coding into sales.
The founders argue that foundation models do not remove the substantial work needed to turn them into production-ready enterprise systems.
They describe a market of roughly three trillion annual business-to-consumer minutes and say Vapi handles a small fraction of it.
As the company grew to about 100 people, hiring, management, and strategic decision-making became major responsibilities.
The founders invested in coaching and nonviolent communication, including pausing during disagreements to check whether they understand each other.

Chapters

Start	Chapter	Summary
00:00	What Vapi builds and why voice AI is hard	The founders introduce Vapi and explain the production challenges of making voice agents reliable, fast, and safe for real customer interactions.
02:03	The conviction behind voice AI	They explain how a long-term belief in improving model performance led them to pursue voice despite its early limitations.
05:08	Early startups and the productivity tool	The founders recount their initial pivots and how a simple meeting-joining product found early users and paying customers.
09:16	From meeting notes to an AI therapist	They describe experimenting with meeting summaries, questioning the ambition of that business, and building a voice-based AI therapist.
15:24	Turning voice experiments into a platform	After encountering customers with similar voice-agent problems, they shaped their technology into a platform and won an early customer.
17:58	Finding customers and learning enterprise sales	The company grew through developer demand and customer support, then learned to handle enterprise procurement, contracts, and sales.
22:06	The market opportunity and the role of the platform	The founders discuss the scale of business phone calls and why the gap between AI models and production systems remains significant.
25:13	Scaling the company and defining its mission	They describe growing to about 100 people and their goal of helping businesses build more personal human interfaces.
27:46	Co-founder communication and lessons learned	The founders share how coaching and practical communication habits helped them work through conflict and keep their partnership healthy.
30:22	Therapy use cases and closing reflections	They briefly discuss voice AI therapy and reflect on the contrast between the years of experimentation and the growth after finding product-market fit.

Quotes

Vapy is a platform for deploying voice agents. — 00:00

the skill of voice agent building is just hard — 00:32

I just built this AI therapist to go on walks with every day. — 14:23

Making that easy for people is just a really herculean hard problem. — 24:10

Can you understand me? — 28:47

Concepts and entities

Vapi, voice agents, Superpowered, Hyperbound, GPT-4, Michael Cable, nonviolent communication, enterprise sales

Full transcript (original language)

00:00 [music] Today I’m here with Nikquila and Jordan from Vapy. Um, thanks for coming. Uh, why don’t we start with you guys telling us about Vapy? What are you guys building? >> So, Vapy is a platform for deploying voice agents. Uh, today we have about a billion calls served uh, annually and we work with folks like Amazon and Uber and into it helping to uh, automate their phone calls. >> What would you say is the biggest misconception about voice AI right now? probably the fact that um people don’t

00:32 realize how hard it is >> uh to get it to production to a great like all these businesses out there um they want to resolve 100% of their calls and it just takes so much effort um what’s hard most people don’t realize how to take their business processes and make them into agents uh making into agent that’s a cohesive experience for customers that’s low latency again at the cost that they need it’s following all the guardrails that are relevant to Uh so the skill of voice agent building is just hard and and that’s why we exist

01:02 to make the make it easier for them. >> Yeah. And and I would say like voice as opposed to chat is a very unconstrained interface. Yes. >> You can say or do anything and customers expect that it’ll say or do anything back once once they get how it works. So there’s a whole fundamental different set of challenges under extremely low latency >> low latency conditions. >> Takes a human 10 years to learn how to speak properly and you’re supposed to do it out of the box. And then on top of that, you’re putting it in front of real customers. >> Yeah. >> And so, uh, you you you better make sure it works. It’s not an internal tool.

01:32 It’s not whatever. It’s like if we are down, like your customers don’t get help that day. >> These appointments do not get booked. So, it is pretty mission critical. >> You guys have Amazon into it. Uber has customers. You process over a billion voice AI phone calls. It seems like we’re on a path where basically every phone call between a human and a business will be done by voice AI. Is this is that not inevitable? >> I don’t think we’ll ever reach every like if you think about uh banking like that’s still some still run on cobalt

02:03 etc. Like there’s always an asmtoic you know relationship for these things but inevitably we’re on that track. We’re more and more and more and it is accelerating from what we’ve seen. This seems like this is very recent. Like two years ago, >> it was hard to make a voice AI phone call actually work well. And two years later, it seems now we’re on the track for every single phone call that a business will answer >> be done by AI. What happened? >> I think it’s kind of funny looking back and we made this like in this depression we made this like AI therapist. Um and

02:35 that was fine. Uh but I remember after a while after making that we were still souping in it like is voice AI the right bet to make. >> Why were you doubting? >> Why are we doubting? Yeah, great question. >> We pivoted like 12 times up until that [laughter] point and I think like everything we tried just like kind of just like wilted like a flower. And >> I don’t know. I feel like when you’re a founder who hasn’t hit product market fit for so long, it becomes so elusive that it’s not even a real thing. >> Possibility. Yeah. Yeah. Like you think of it like like heaven, like you don’t

03:06 actually know. >> Yeah. And we were trying to then like boil it down like to something irrefutable. >> Yeah. >> Uh and like I think Jordan now has a good framework around it. Like what’s something that like in 10 years just feels like inevitable. >> Yeah. >> Like what is something we don’t know? We didn’t know it was going to be like next year or like this fast. to add. But like if we kept on this path in 10 years, can we claim that models are going to get cheaper, faster, and better and reach human performance?

03:36 >> Like will they be fast enough and the intelligence is fast enough? It was like well yeah what the hell? Like it’s like it’s just so obvious when you look at the graph that that’s going to happen. So we didn’t know the exact time stamp uh but it was just a macro trend we could really like lean on. So that was that’s what it became a irreducible fact. >> It’s almost like we on the outside got like some spot check. So like sometimes we heard a cloned voice model be like of my voice and be like wow it kind of works like they can you can say

04:06 something with my voice >> and then in the background you were building this big company that basically would >> be able to answer the phone with low latency >> everywhere. Yes. Exactly. It just wasn’t I think very obvious to people that people would be talking to models like this was before GPD4 real time and all that stuff but we had chained together transcription model our time model text to speech model together and when you talk to it it would like take 10 seconds to reply and it was really shitty and it would like sound like garbage and it would interrupt you and all that. At the

04:37 time we thought it was just reducing latency. If you could just reduce the latency then we could do it. And so if if you draw that line out of how latency is trending per dollar per, you know, per compute, um, it looked good. And so we were like, okay, well, let’s set a vector that’s ambitious. Let’s assume that everybody’s going to be talking to agents and that’ll be the default interface for everything. >> You guys have been on a long journey to PMF and now you have it. Let’s go back to sort of like the very beginning. You applied to YC with a very different company than you even did YC with with and then there was some time in between

05:08 that and what now is BAP. Take us back to the beginning. I was working an internship somewhere and then my boss told me to stay late to get this PR review like done and I was like this what the hell and I’d always wanted to be be a founder or whatever. So I I I called Nquille that night and I would like lock myself in my closet. I didn’t want anyone to hear me because I was embarrassed. I don’t know. And I made the phone call I’m like dude I think we should just do a startup. Let’s just try this. Let’s drop out of school. We’ll figure something out. And then we worked on a bunch of different ideas. is we were like we we had a a friendship CRM

05:40 uh that we called better friend. We were really good at naming at the time. Uh and then pivoted a few more times. Then we applied to YC with uh I think it was >> invest better. >> Invest better. >> Yes. [laughter] >> Uh you can see the naming convention better brand invest better. Yeah. And it was like a thing for investing in baskets of stocks. It kind of didn’t really like uh I think by the time like we got accepted we were like please go off give us a better idea. I already knew you gonna get on a better idea when I interviewed you and and you said you have the interview so we can maybe

06:10 listen to that. >> Yeah. >> I was like these guys gonna pivot to something else >> and you were a pretty big team and this they were very early on this idea. >> And this is co right? >> Yes. >> Yeah. So you applied in end of 2020. >> Yes. >> Yeah. And then did the winter 21 batch. >> Yeah. That’s right. >> What happened during the batch? >> Pivoted another 12 times as predicted. So then we ended up building a productivity tool which is a button. I remember that time being >> kind of like not super good because it wasn’t a lot of new ideas and the the the one accelerant was co and zoom >> and sort of like we had this new like

06:42 remote office experience that was new to most people >> and and that’s how this the idea that they didn’t end up working on kind of came out of right >> yeah exactly we were late to a lot of meetings because we were booking a lot of customer interviews and we realized this was a problem it was hard to join meetings so we just solved this >> because you had to go to Google calendar >> uh find the day Then open the link, then find the link, then click on the right link >> and then >> which opens another web page. Then you click to launch Zoom which opens finally the Zoom app and then you click join into the meeting. >> 10 clicks like 12 times

07:12 >> and then and then you like fiddle with your microphone all >> and not to mention about 30% of people uh were late >> to meetings >> at least a minute or two late to their meetings. And so we found very very you know strong product market fit with one user which was you uh who paid >> because I have a lot of meetings today. >> Yes. Yeah. With your office hours and stuff. So so it was about I think you paid 20 bucks a month. Uh I think we said it was 10 bucks and you were like no it’s going to be 20 bucks a month and then then you paid us 20 bucks a month and then you got your wife on it and

07:43 then our MR was 40. >> Mhm. Um, and then we did an internal Bookface launch and then our MR was like $100 and then it >> well when actually we launched on Bookface without integrating Stripe at all into the app. >> So because we just wanted to get it out, >> we were like, oh, we don’t have anything integrated. We we were like, oh, we’re just going to do it after and which was the right decision, but it’s just like so I reminisce those times where it’s like, you just cut corners. Let’s get it out as fast as possible. We have like we have six weeks before demo day.

08:14 [laughter] Every day matters. >> Yeah. And we were like, “Okay, well, how long is it going to take to build a pay wall?” It’s like two weeks. It’s like, “Okay, two week free trial.” [laughter] >> I would say one week probably. >> Yeah. [laughter] >> Then at some point you launched public, right? Like there was like And there was feedback. >> Yeah. >> Oh, I love what happened. >> Uh the feedback was like pretty positive. We did an internal bookface launch first. Uh that got us to our first 80 80 users each paying I think I think we reduced the price to 10 bucks a month because they weren’t as generous as you >> and it was called superpowered. >> Superpowered. Yeah.

08:44 >> Yeah. >> And it was what was it specifically? >> It was literally just like a thing in your menu bar that told you when meetings were coming up and then gave you a button that said join Zoom. It just worked for us at the time. We had 90% retention. Uh we had I think 20% conversion uh to paid when we actually did shift the pay wall. And so we knew we had something. And then we basically spent the next two years uh iterating on on that. >> And then how did this button evolve into something else because it became something even more interesting like later on. >> Yeah, we experimented a lot just trying to find more ways to provide value to

09:16 customers and and honestly like we went through a time where we’re just growth hacking all over the place because we didn’t know what else to do. We just couldn’t find like a path from this local minima to pro provide and and get more value from customers. >> Yeah. The way I phrase that time is like it’s like a training ground because we got really good at like looking at data >> uh and like mix panel every day was like our god just every day you look at it try to see what experiments were run to bump it up >> like five 12 simultaneous uh feature flag experiments in one moment. >> Um so that was really good as a time I

09:47 think and then yeah like one thing that really hit people was we basically built like a botree meeting note ticker >> that was well that was when GP4 came out came out. Yes. >> Yeah. And that increased like then all of a sudden the problem space opened up and we were like whoa there’s like >> this is interesting but at that point it was expensive right. So I remember asking you how many of these can you do? >> Yes. >> How many meetings can you actually like summarize basically and I I believe like you can only afford to do like a small number of meetings like basically you would >> the business model would break if you did all my meetings.

10:17 >> Yes. >> A good point. Yeah. >> And so we just 10 extra prices. [laughter] >> Exactly. Like we used to charge like people like hundred bucks a month or something and people were happy to pay it. Like we were probably the first like subscription app I feel before like Tajibility made it $200 a month like okay to do that. We’re like oh we’ll just charge people a lot of money. >> So the button turned into basically a noteaker for meetings. it turned into granola like an on device exact same >> and it wasn’t uh was it anything that was like not as good with it or like why didn’t it

10:48 >> why didn’t you stick to this idea like granola there’s a bunch of companies now that basically do this that you >> really well I mean there’s like I think we knew >> I don’t know if you would add something here but like to me it always felt like we knew how to build a billion dollar company here Michael like would really push us like Michael Cable on um hey talk to your users like where are they actually using your notes because most people don’t use notes And we talked to people and they realized that um there was wealth managers, do like nurses, doctors like they were getting a ton of value. They take notes anyway.

11:18 >> They took notes after every meeting and they they look at notes. So in services industry, you just care a lot about like notes. >> And we knew we could build businesses there, but I think at this point we were just burnt out. >> Yeah. >> Um mostly burnt out. >> And at this point, like it’s not like a bad business. Like we were half a million in revenue at the time. Uh we we had 10,000 paying like users. Like it was it was great. But I think >> you’re living in Toronto still because basically we’re here for the batch and then move back to Toronto and stay there. >> Yeah. And it basically became honestly like a lifestyle kind of setup and we were kind of like taking it easy and and

11:49 not we we didn’t treat it as like a oh man this is a generational opportunity. This is our chance to build something huge. We just kept like throwing spaghetti at the wall and I don’t think we just had the ambition. >> That was basically it. I think it just felt like like is this going to be the most important thing one of the most important things in the next 10 years? Like we were like man calendar app like after four years of count like like >> this is basically end of 22 early 23. >> Yes. >> And they were like the AI moment had not yet happened. >> No. >> If it really would fully hadn’t

12:19 happened. I think this was still after Chad GBT. Like so had come in um at least after before the noteaker. >> Yep. Well, I think one way I like how you phrase it, which I hadn’t thought about, but like our ambition just kept rising [clears throat] because we kept getting more unsettled by what we were doing. It’s just like didn’t feel right. You know, in my perspective, we had a recipe to make a billion dollar company um into healthcare or granola. But here’s what Cybel said. He’s like are

12:50 this is going to be the sixth most important productivity app in the world in the because in the last five five decade there’s only been like >> word sheet calendar email like >> PowerPoint like everything that Microsoft owns >> and like notion tries to be the sixth one but like are they really significant maybe we don’t know yet >> and it’s like are you unless you’re doing productivity at that scale like what the are you doing or what so that really pushed us to be like that ambition we should um it doesn’t feel ambitious enough and that’s where we

13:20 pivoted out completely and >> uh that fateful >> and this is probably for people watching like probably a moment that they might see themselves in. It’s like okay a bunch of ideas and now I’m like kind of running out of steam. >> What what did you guys do? >> We had a late night conversation with Michael Cable and I’m sure we we’ll send you the clip after. Okay. But he was like look it’s clear you don’t give a about your users. It’s it’s clear that you guys like something about this team is just not working. Um something about your environment is not working. You just need to hit reset. move to SF. Just move, but if you’re gonna move somewhere, move to SF and figure out

13:52 what you’re going to do and just like cut everything else out and just like heads down on this for a while. And so within two weeks, we did that. >> Just gave up like basically gave up on on Toronto and just left. >> Yeah. And then we set up a autorespond. So anytime someone sent us a support email, we refunded them their full subscription. [laughter] >> Yeah. >> Smart. [clears throat] And then when did you build a therapy app? You build a voice AI therapy app that you would talk to. >> Yeah. Yeah. >> Yeah. So we were pivoting yeah for roughly about three months I think. And then while Nikia was away uh and and I had just recently come out of a a

14:23 breakup a weekend and I was just like super sad. So I just built this AI therapist uh to to go on walks with every day. Yeah. >> And I spent two hours a day walking with it and then the rest of the day just like grinding on it to try to improve the latency and make it feel more natural. And I kind of just channeled that uh depression into obsession over like just trying to have this like human like why doesn’t it sound and feel like a person like >> was it good at giving you advice or listen to you? >> Yeah. Yeah. I remember one time I was talking to GPD4 and I had this long dialogue about my feelings and stuff

14:53 like that and it and it was like you you you notice like you always talk about your feelings in the third person. Are you trying to distance yourself from them? And I was like >> what the Did a model just like diagnose me or something? That’s pretty good for being that early. >> And I was like, okay, this isn’t like this is like a unique moment that could not have happened over over chat that has now happened over voice. >> And so the longer that we kept toying with it, we found one guy, our friend and also angel investor, his name’s Mo. Uh he I gave it to him and he would talk to it like literally every day or like

15:24 12 times a day >> just like spitballing ideas and stuff. So basically we we proved like product market fit for like chatbt voice mode that would later come in like two years. What was the state of the models at the time when you were building this? >> Really slow. >> Super slow. Super expensive. >> So the problem was it expensive and it’s like the latency was not >> latency was a thing that everybody was obsessed with. >> Yeah. >> And so there were some other companies that were also obsessed with latency. Uh Hyperbound being one of them. They were building this like AI sales roleplay training tool. Uh and I believe like you

15:54 connected us on a >> they were in the winter 23 batch maybe. >> Yeah. Yeah. Yeah. And they were pivoting too and looking for an idea and they were trying to build this like thing where you could interact with a voice agent to you know practice your sales skills >> and same thing they were struggling with latency. They were struggling with like reliability of the pipeline and like configuration. Uh and we were like I guess we could just turn this thing that this therapy thing into into a platform instead. It seems useful to these two guys. So yeah sure. And then they were our first customer >> and they are actually pretty big company now. >> Yeah. >> Yeah. Yeah. It it it worked out for

16:25 them. Was it hyperbound or was it other conversations that led you to like basically the VP platform or >> I think it was hyperbound for me. >> Yeah, it was hyperbound. We had like a couple random stragglers coming in and out, but it was really hyperbound that we like >> engage with them. So, they were like, “Do they know what they wanted? Did you know what you wanted to?” >> Yeah. >> Yeah. I I kind of I remember I was like outside of a a mall in a in a car like trying to basically pitch him. Yeah. Uh and I and I was like uh well yeah we have the super configurable platform and it’s like an API and like we reduce your latency and uh and he’s like really like

16:57 do you think this will work? And we’re like yeah we’ve done this like four or five times before of course it’ll work and so we kind of sold it without it even being we were just confident in our own ability to deliver. >> But they didn’t did not need real time or did they >> they did it was like a voice like they were trying to build like sales roleplay agents like one of the first voice a companies that Yeah. Great use case. >> So it pushed you to do real time. >> Yeah. Yeah. It was full. >> How did How did you solve that? >> Well, I think Jordan has solved a lot of the problem for the AI therapist. >> Yeah.

17:27 >> But now it was like about being building it to production scale. Yeah. >> So, which was I remember this um yeah full holiday sprint where we were both of us were couched in like Jordan’s like mom’s um place and just she was feeding us and we were while we’re just like coding all day uh trying to get them get them successful um and that was it. Yeah. They just it just worked and then they signed their first customer got successful with them and we couldn’t believe it. >> Mhm. >> And then how did it go from there? Like what happened the next six months?

17:58 >> People just came to us. Yeah. >> Well, I remember also you like I think this part is also really cool. Uh because you have the first customer, but it’s not quite PMF yet. I mean, no one knows about you, right? Uh you have like a powder, right? Powder. So Jordan like I remember scraped every YC company. Like I don’t know how he did that, but like he had he was in contact with everyone in YC who was building a voice. >> Yeah. There wasn’t that many. >> There wasn’t that many. There’s like five 10. And then I so we started working with like a few more of them who had the same problem and then three of them like died along the way just

18:28 because like that’s the nature of of it was mind-blowing. Just two years ago it was mind-blowing. People were like you could that just responded like is that a person? As we started working with more and more like small startups and honestly it was like all over the place. We had like agencies and like random people out in like middle America and like you know random tech companies and so it was just a hodgepodge of like no focus. And so at at one point it just grew and grew and grew to the point where we had about 300 Slack channels for support. >> Wow. >> And not discriminated by like how much they were spending either.

19:00 >> And and we would within five minutes of someone giving us a request, we would ship the feature to production. And and that’s roughly we we we did not do any marketing. We did not do any sales. We didn’t do anything. We just built like uh a need that existed in the market. And and again, this was aligned with our original conviction, right? Like okay, voice models are going to be there. They’re probably going to be really hard to do stuff with. So, let’s just build something that makes it easier to do stuff with them. Um, and so we just rode that. >> I’m assuming at some point these startups there was like basically self-s serve bottom up like developer marketing. At some point a big company showed up probably. >> Yes. Yeah. We had our first big company

19:31 uh show up. They were kind of a bigger version of one of like the smaller startups and I remember they would like email us like these strange emails that were like from head of procurement and I didn’t even know what that word was. and he was like, “Yeah, we want to know if you’re sock 2 compliant and stuff and and and like we want to sign like a contract.” And me and Nquille were like literally so annoyed. We were like, “Dude, just read the docs. Like, I don’t understand like what this is.” And then they were like, “Fine, okay, we’ll get on a call with you. Sure.” And like 20 people show up to the call like director of whatever of me and the killer, what

20:03 [laughter] are you talking about, dude? But eventually we start to understand like, oh, they want to deploy like millions of minutes like on the platform. they’re like an enterprise and we’re like okay like yeah you know what we do enterprise by the way here’s our contract like and and here’s our everything and I remember uh I put together uh a 40page master services agreement for them uh and then it took about like two months of red lines to get across I didn’t know what I was doing at the time and the pricing was super complicated too but luckily we got across the line and and

20:35 over time like they did take a lot longer than other customers they were way more demanding than our other [clears throat] customers um but it did push us taught you how to actually serve enterprise customers which is now most of the customer base or totally most of what the company is. >> Yes. >> Yeah. And a big consequence consequential decision here was to uh get Jordan out of coding completely >> which is really hard for us. I think we’re like what’s something we have never done before >> and here we should try to do something different was like George should just not be like one of us one of us should

21:05 try completely to be on sales and then that also really helped and accelerate but it was really hard because like up until this point like the way that we determined the value that we were providing was always like what we’re reshipping >> and to change my mindset from okay I’m building all the time to like I’m just like sending a bunch of cold emails and making zero progress. Like that was a really tough time, just being really at something for so long and showing no progress. >> But then eventually over time, we started to master that like, oh, somebody comes in and they’re starting

21:35 to spend money like I should email them and get on a call with them and present some options and send them a contract and get that signed. So we started to build this like kind of motion of uh uh building contracts. >> Um and then eventually I couldn’t ever crack like the outbound sale and like going out and like finding new customers. They always came to us. Mhm. >> Uh, and so I ended up getting a sales adviser and that was like instrumental in like just teaching me like proper sales process and like prospecting and like how to make sure that you >> qualify for pain and don’t waste time on people who are going to waste your time. Uh, and how you present commercials in a

22:06 nice way and and all that. And what I’ll add is like we didn’t know this at the time. Yeah. >> Now we do. Uh, this market is like the Milky Way. It is enormous. It’s like there’s about three trillion minutes annually that are business to consumer. >> How many how many of them do you serve? Uh today we serve uh 0.0134%. >> Wow. >> And what I’ll say is like >> and you’re still a pretty large company like by YC standards. >> And what I will say is like some of the largest telecom sort of companies in the world have like sub 1% of that volume as well.

22:37 >> Um like names that you would think of and so it’s like an impossibly difficult diffusion problem >> and and an impossibly large market. So that’s why there’s so many winners and so many verticals. At some point you there was like one of the feedback I got from other investors were like oh this is just the lab is going to do this. >> Yeah. >> And I would slightly worry but I like ask you guys and then you probably I probably offloaded some additional worry on you that was unnecessary. [laughter] >> Yeah. >> And that was probably something you did worry about. >> Yeah. >> And they kind of kept coming up. >> Yes. >> What happened? >> Yeah. I mean I think we almost gave up.

23:07 So open launched their speech model like the original chip voice and at this point we were doing well but not that well. I think we were less than a millionR. This is before our A and we were still like four or five people maybe like sixish people and that came in and that was a moment of crisis. But for us I think what came down to is our customers still needed us >> and >> what was not obvious back to us to us back then and intuitively was obvious but we couldn’t articulate it and what we can articulate now is the gap from

23:38 these models to production is so high. It takes so much effort. Like most people in this world do not know how to build a voice agent. Like you don’t know how to build a voice agent. And that’s okay. Like no one is learn how to build a voice agent. That’s true of every customer out there. >> Yeah. >> Like three trillion minutes talking about no one. And this for every business listening, every founder listening here, everything you’re selling. Most people in the world do not know how to do that. Especially in this moment where there’s a complete paradigm shift. >> Like what’s AI? You know what is prompting? Like all these things that we just know and we internalize. no one in

24:10 the world knows and making that easy for people. Making that like super like a like you can deploy in production, it works for you, it has all the requirements met is just a really herculean hard problem. >> And then we just internalized that at some point and since then I don’t think we’ve ever been worried >> and you never lost customers to then I mean they always came back. >> Yeah. >> They were like I can do this myself and then >> and we’re like okay cool we’ll we’ll set a reminder six months later they come back. Um and and I think Michael Cyel, we had a chat with him about this and

24:41 because this came up like I was yeah like this this like dilemma like oh the labs and then we go out to fund raise and they’ll be like oh the labs and like it was just like the constant FUD we’re trying to hire people same thing. Um and and the way Michael friend was super helpful he was like there’s two generational problems in AI. One is AGI and the other is just like deploying AGI in the enterprise. Both are equally difficult and all of the labs have their A players on that first problem and their B players on the second problem. You have your A players on the second problem. >> So that’s the advantage. >> It’s a very good framing and people

25:13 underestimate how large this enterprise portion is and how how long it’ll take. >> Like there is no way this will happen in 5 years or maybe even 10 years. This will happen over like our generation basically. >> Yes. Yes. Yes. It’ll take a long time. >> Yeah. Tell them about the company right now. like give me an update of where you guys are and and sort of like where’s the site set on like where you going to go next? >> Yeah, we’re about 100 people now. So, so growing a lot. We got like a full team. We’re like learning like leadership and management and I don’t know organization

25:43 >> and now competing with like the biggest in the industry like the biggest deals. Um I think I remember right after YC there’s this like presentation that um Michael delivered to us. It’s like don’t get into company building, you know, until you have PMF. And I’m so glad. >> Yeah. >> Because oh my god, company building is so hard and it [laughter] takes it takes so long. >> There’s so many random skills you have to learn. >> Amount of time I’ve heard the word alignment. [laughter] >> Yeah, it’s really hard. >> It’s necessary, but yeah.

26:13 >> And it’s fun. Um because you you have all these smart people and you don’t realize also it’s like cuz you’re just founders and you’re just doing your own thing. There’s a lot of good people in the world. like there’s a lot of great talent who knows how to scale these things uh and you need to get them in the board boat and like inspire them. >> So all those things are very necessary but >> it’s so different than just a raw like exercise of finding PMF. Um and I’m so glad we didn’t start that until like really we had a strong foundation. >> Um but that’s what takes up more of our time now hiring hiring I think making

26:45 like >> trying to make good strategic decisions. Yeah. What’s the holy grail of of voice like? Where will this and is this sort of like you can argue one way is like Twilio but bigger >> but maybe it’s something else. >> So our mission is to help every business build their human interface >> and we took a lot of time to flush out that that statement and and it kind of drives everything that we do. Like we believe that one day once we can help diffuse this technology to these enterprises, we can make these enterprises feel like people,

27:16 >> people that know you, that talk to you and remember you and and feel like as personal as like this feels. >> And so that’s ultimately what we’re trying to drive towards and we believe that voice is really like the medium of interaction that will be default. this might be the most important upgrade customer service that ever happened of all kind because basically like every single person you will interact with in the company is going to be a great representative of that company and know everything about you. >> We’d love to get to a place where we can kind of build like that that interface layer that orchestration layer that makes it possible. >> Co-founders relationships are hard and

27:46 you guys seem to build a strong really strong foundation from knowing you guys for a couple years. Like can you just like practically how do you do it? Like how do you interact with each other? People watching this might not know, but co-ound breakups is like top reason why the companies actually fall apart or die because co-founders aren’t getting along. What have you learned? Are there specific practical advice to give to people that are watching us running their own companies? >> Uh, one like that is the most important thing period >> like full stop. Um, if that is not in well functioning order, none of this

28:16 other stuff matters. Uh, so me and Nquille uh pretty early on we invested in getting uh coaches for this kind of stuff. So, we got uh interpersonal dynamics coach that helped us like actually learn how to communicate with each other uh and work through the principles of uh what’s called non-violent communication. So, there’s a great little book on that. It’s like our bible we give to every employee when they show up. >> I have it. Yeah. >> Yeah. It’s incredible. It’s and it’s it’s so simple. Um and so through that we kind of built up the skills to be able to work really well together. >> What are some examples on how you

28:47 practically do it? >> So, one sentence I think we both say often is like, “Can you understand me?” because this often happens in discussions at least I noticed with like other people who are not uh pills on NVC where you’ll be arguing about something but it’s not quite um you don’t know what you’re arguing about even sometimes or you’re arguing about like a completely different things you keep moving along >> it’s like and then taking that moment to like if you’re if it feels like we’re circling like can you just like pause and understand me >> is a such a common thing in the company for us uh I don’t know what other

29:19 examples come to you >> I guess like a bit a bit more viscerally when we’re talking and disagreeing, you can sens. And then he goes, “No.” And I’m like, “What are you saying?” And then he’s like, “Yes, that is what I’m saying.” And then the tension comes back down, >> right? That handshake of understanding uh is is super important. >> Was there a specific argument that led to you investing in this or you kind of knew that this would be a risk that you had to invest in? >> Oh, yeah. I mean [laughter] there was

29:50 like there were exfounders crying in the shower moments. [laughter] Uh there was it was a lot. >> It was like lots of fighting. It was extremely unhealthy, extremely mentally just not >> just I think the the stress of pivoting like it was a lot of skill building [laughter] on all sides to bring the tension down and really align what we’re saying. >> Um it’s hard to like it’s hard to be mad at someone once they understand you. I don’t know what it is. It’s just a human thing. >> Yeah. Yeah. And so like probably the best investment we ever made was just getting a a coach and like coach

30:22 therapist kind of person. Honestly, I think everybody could use a bit of both. >> Um someone that can help you work like cohesively together as a team and someone who can help you make sure your mental state is in a good place. Um >> do you think that the app that you built the AI therapist is actually going to be like? Yeah, the really really big is like number one one of the top use cases is therapy. I think >> yeah just like people talking to it like well at least for voice. >> Yeah. >> Yes. Yeah >> for sure. >> So it seems like you had actually this is what I love about your story. You you invented multiple products

30:53 >> way too >> all became all became big companies. So like >> you invented sort of like the sumbar that everybody knows about. You invented like circle back granola before that was a thing and they are therapists and then you sell on >> building >> building. Yeah. I think there’s a lot of stories there of where we felt like we had strength we where we could build a business too. >> Yeah. >> Yeah. >> Fun. >> All right. It’s great to see you guys. Thanks for coming and talking about this. Um, I am honestly so so excited to have you guys here and talk about Vappy

31:23 because >> there’s a you’re not it’s not a unique story of companies that come that basically like were >> trying things for a while and then AI came around and then the founders came to SF and then you >> kind of jumped on the wave that was that was the models and the AI and built something like this and you guys did it like perfectly and probably the best uh of the ones I’m working with. It’s so funny that you say the word perfectly. Like >> it’s never Oh, it’s perfect. [laughter] Hindsight. In the moment, it’s not. But it reinforces this idea that once you

31:53 find PMF, it just it just works. And then you don’t have time with other stuff because it just works. And that’s sort of how your experience was cuz >> the amount of growth you’ve had in the last two years compared to the first two and a half years. It’s just like you can’t even compare. >> Yeah. 100%. >> Thanks for coming and telling the story to everyone watching. >> Cool. >> Yeah. I appreciate it. [music] >> [music]