The Exchanges

Q You were doing how much in revenue, more or less?

A We were doing just under three, two, 300 K of revenue. But, like, solopreneur, like, you know, ACVs are five, 10 K, right? Um, and so we were just, like, servicing that archetype of customer, generally speaking. And we're, like, at a few hundred K, we had raised our seed round. Um, and we're like, okay, cool, you know, let's go get to two, three, and raise an A, and then we'll try to get to 10, and so on and so forth. But again, going back to, like, if my co-founder and I lose the conviction that this is a 10, fifty, hundred billion dollar outcome, um, we need to instantly change direction. Even if there's still a lot of room to go, right? Even if, like, we're at the beginning of a hill, it's still a hill. As soon as you see that end state being 10 to fifteen million of revenue, not five hundred million or a billion of run rate, then now it's opportunity cost. Now we're wasting everybody's time. We're wasting our time. We're selling a story to the world. We're sharing a story to the world that we no longer believe in. Right. And so that obviously was like a very big moment. The other thing we started realizing also was we had just built a more enterprise grade platform that a lot of our earlier customers were only taking advantage of 10 to 20% of the actual machinery we had built. It almost felt like at some point we started selling a tank to somebody who wanted a bicycle. And …

AI assessment note: “We were doing just under three, two, 300 K of revenue.”

Q When did you decide to leave Amazon and go all in on this?

A So July, 20, 24 is when I quit Amazon and had probably started this fact-finding journey maybe in February or March. So it was three or four months of making sure and looking at what's happening in the market. You know, when I looked, there was a lot on the freight side, but there wasn't a lot on the warehousing side. I thought it was an underserved market. Um, and I also thought that having some domain expertise was helpful, and we'll talk about this, but we still use that today. Again, just making sure that when we're going and talking, I'm using specific examples to cite, like, are you having this exact problem, and is it causing this exact pay point for you so that they could relate? And so after July, 20, 24, left Raise Venture within weeks, actually, of quitting, and then we were kind of off to the races.

AI assessment note: “July, 20, 24 is when I quit Amazon and had probably started this fact-finding journey”

Q But from the customer's perspective, because we've all, we've all had this, we're doing a task and we're like, this is an annoying task. I wish I didn't have to do it. I wish I could do it. That's probably the moment where they're like, oh, Record, and then they just do the task they were going to do anyway, and you pull that into an SOP?

A Yeah, exactly. So they were already going to do it, they can hit record, and walk through, and as they're walking through, of course, they can verbally articulate anything that they want. From there, once we have it, we'll process mine it, so we'll actually make sure that it's the fastest route to get from A to Z. What we see is that historically, maybe they're pulling down an Excel spreadsheet, doing some sort of manipulation, and uploading it somewhere else, where obviously They don't need to do that anymore, so we'll make sure that it's efficient. After it's built, it'll go to staging, and then they can actually push it to production. And really where that pulls ahead is, let's say that there's two different things. Number one, let's say the use case is in flight, but then there's an edge use case, and they're like, oh, I totally forgot to tell backups that when customer A, B, C has this problem, we do DEF. They can just record, you know, a two-minute of, this is how you handle this specific, right? And we can very quickly adapt the agent. And then the second part is that on that enterprise-grade side, Now, if it's that claim and finance wants to track it, they can see what the current process is. So there's no more silos and they can say, Hey, well, every time that happens, we have to do our process. So can we just attach our process on? And the answer is yes, they can just…

AI assessment note: “Yeah, exactly. So they were already going to do it, they can hit record”

Q And what about like in person, anything in person you're doing?

A We're doing a lot in person, you know, surprisingly, the conferences have been really, really successful for us. We were just at manifest, uh, in February and actually closed two deals for manifest already. And so that was great. And I think it's a good way to get in front of either folks that are in our industry or maybe even that new. And then for our actual customers, yes, we're flying. I think this month I'm in six or seven different cities meeting with prospective customers. I think one of the big things that we're also trying to do is these like onsite activations with our customers. So if they're just becoming a customer, when we need to get all that context information that we talked about, getting that in person, and even if the contract's not signed, we'll fly out. And meet with them. And it just helps to conceptualize like what this would actually be like, because we're workshopping it together. We're answering their questions live. They're bringing it people into the room that are potentially going to be the blockers down the line of kind of the gatekeepers. And we're able just to work everything out in person. And this is before a contract is even signed. And so it's an investment on our side to do, but it's paid off multiple times. And so I think that that's also been a huge unlock because a lot of This cell is educating the customer on AI as well. And so you have…

AI assessment note: “We're doing a lot in person, you know, surprisingly, the conferences have been really, really successful”

Q I've found is just The challenge, like the failure mode is you just get stuck somewhere or it just starts taking too long and, you know, time kills all deals, priorities change, whatever, because you're already in at that point. Clearly they like what you have. What have you found works in order to shorten the time between interest and some closing of some sort, even if it's a pilot?

A I think the pilot thing for AI startups can be very dangerous and we've learned with lessons too. I think, you know, when we look at Those pilots, and what I say by learn by lesson, you don't want to, after the pilot, be in this purgatory of like, where is this going to go? Yes, we're going to go to commercial contract, but that's got to go right back through legal, and you know, you might be still having your solution deployed, and so I think it's a dangerous area. The way that we do, we always try to push for a one-year contract, but if we have to do a pilot with, and with many of these large, large enterprises, they require some sort of observation phase, we'll build the pilot into the one-year contract. So it's a one-year contract, But it will include a 30 day pilot, 60 day max, but 30 day, hopefully, pilot that will auto convert to a one year. And that way, all of the legal documentation and infrastructure and all of that's pre-approved. It's all part of the normal agreement and everybody's on the same page. Obviously, you need to make sure that you deliver in that 30 or 60 day window so that you can get that auto conversion, but that's hands down. The best way. And to answer your question, when you talk about like, how do you get it? That's a really good way because it still is a light enough ask to the customer. Their feet aren't necessarily held to the fire if they don'…

AI assessment note: “we'll build the pilot into the one-year contract”

Q been working on for the last few years. So We'll definitely talk about the new company. We'll talk a lot about AppDynamics because that's obviously a well, well known name. So maybe let's go back to that time, right? As a starting point, like if we're talking, you know, mid late 2000, what was going on in your world? How do you end up kind of becoming part of AppDynamics?

A Yeah, absolutely. So AppD was started in 2008 by, you know, Jody Bansal, who everybody knows sort of like a multi unicorn founder. Uh, so Jody and I were working together at a company called Wiley. Wiley Technology, this was in South San Francisco, Brisbane, and they were, at that time, the leaders in application performance monitoring. If you go back to about, 2005, 2006, what's really happening is more of software is basically running business, right? And Java was one of the predominant sort of ways you would build an application. And what they would do is instrument Java application, and when I say instrument, really add probes into the, into the code that is actually executing in Java. And use that to sort of really monitor the application to tell you, hey, is it healthy? How is it sort of with respect to how it has been doing? Set up dashboards around it and things like that. So, so Jyoti and I were both at Wiley. And in 2008, when Jyoti started AppDynamics, I was the first one he brought on. We were friends. So in a way, we really disrupted the category that was application performance monitoring, you know, which Wiley was really good at. And the way we were thinking about it is applications Just around 2007, 2008 were getting super distributed. Initially you would have a few servers running the application. And then as you started going into 2008, 2009, applications were…

AI assessment note: “in 2008, when Jyoti started AppDynamics, I was the first one he brought on.”

Q And then jumping kind of fast forwarding a lot, like from 20 10 to, I guess, like 2019, tell me about the acquisition. Like, you know, how, how did that come about? And then what was it like when that moment really happens and, and the company exits for, you know, many billions?

A Yeah, absolutely. So I think the whole thing started off with Cisco being a customer, right? So in our selling cycle, Cisco was a lead. Um, one of our reps was basically really Working with the internal IT team there to sell AppDynamics to them. And, uh, in fact, I did a bunch of like, you know, sessions with them to get to the point where they made the buying decision on, hey, we're going to get AppD into our internal IT. That's how they started seeing the tool. That's how they started seeing how powerful it was. And as they started thinking about, hey, this is something that we want to have, you know, as part of our suite, as part of like our company. That's when the conversation started. But by then we were Pretty much on our way to, because, I mean, in startups, as you know, like, you'd run things parallely. You never stop one or the other, and you keep going in multiple directions. We had filed the S-One. We were pretty much ready to go. In fact, like, some of the team was in New York. Like, they had their suits and everything else, and it was that close. Like, we're a couple days away. This was going on in the background. You know, we were talking to them in the background. Not, obviously, everybody knew. The fact that, like, we were in Soma. That's where our office was. But, uh, the fact that Cisco was just in San Jose, super easy drive, et cetera, made it pretty easy. Y…

AI assessment note: “So I think the whole thing started off with Cisco being a customer, right?”

Q Mainly on features, like on capabilities, that's, that's what you're winning on?

A No, I would say on fit. So customers just, I think, felt more familiar in our platform. It felt like this is a lot of just ease of use. It was easier to now to get around. Salesforce has quite a learning curve. A lot of the features that our competitors had were, they were integrations really. So, you know, they would say, well, we have X feature that you need. For instance, uh, you know, FileVine built its own e-signature tool, Vinesign. It is a 98% attach rate for our customers. Pretty much every customer drops DocuSign or Adobe Sign or whatever it is. They drop that and they buy our e-signature tool for a bunch of good reasons. Like it's embedded into FileVine. It's a completely seamless experience. It's really amazing. And You know, a customer would say to Salesforce, okay, hey, we want an e-signature feature too, and they would go, great, here's 10 integrations you can choose from. None of them really do it quite like FileVine, it's not seamless, it's sort of janky, but it's right there. And by the time the Salesforce competitors would line up the five or six integrations you needed, that didn't quite work right, and FileVines all were sort of embedded in this bespoke environment to law firms, first of all, they were more expensive than us, and all of our features just worked. And you didn't have to go buy them and maintain all these outside integrations.

AI assessment note: “No, I would say on fit. So customers just, I think, felt more familiar”

Q But do you compete with the Harveys and Legoras?

A That's a hundred, a hundred percent. That, that, that is our competitive set today. So it's no longer the case management systems. It is a hundred percent. These other AI tools. That is, but that's basically all we compete with. That is where I would say near a hundred percent of my mental attention is. The company Litify that I mentioned, those are irrelevant to me on a day-to-day basis. The only companies we really think about are legal AI companies, Harvey Lagora, who we think we stack up much better than, you know, from our perspective, Harvey and Lagora were GPT wrappers when they started, and I think they're basically still GPT wrappers today. I think they know that that's a problem, by the way, that that doesn't provide a very significant moat, and they're trying to become more like systems of record and operating systems for legal that you hear them both talk about this in their podcasts and kind of publicly. We already have that. Filevine already built the operating system for legal. So now it's reimagining in an AI native way How lawyers want to interact with an operating system. And that's the journey file lines on today.

AI assessment note: “That's a hundred, a hundred percent. That, that, that is our competitive set today.”

Q How did you raise from A-sixteen? Like, what was that process like for the first round?

A So my other co-founder, Dan, who runs most of our work with investors and also like our commercial business, he left Palantir in 20, 22 and had been warming up certain relationships like a year plus in advance. And so he did a lot of the groundwork that allowed us to go like very quickly. The Palantir alumni network and mafia network also was very advantageous because You know, folks want to support ex-Palantir founders. There's a lot of former Palantirians that are at some of these investment firms, like Michelle Volz was an example. She was at A-sixteen-Z, former Palantirian, and was kind of our first connection there. He was building a relationship with A-sixteen-Z prior to that, like, three-day process, and that allowed, like, Dan to get to know and trust some of the GPs there that would later on go to support us, so we weren't going into it cold. And we knew obviously for what we're doing, which is at the intersection of AI, security, defense, infrastructure, there's really no better first partner there because they just have such a network and system across those different fronts. So they were at our top of our list. When it came together, it just happened quickly. But we put in, Dan especially put in a lot of like work beforehand to build relationships to make that happen.

AI assessment note: “Michelle Volz was an example. She was at A-sixteen-Z, former Palantirian, and was kind of our first connection”

Q it was then. And certainly there hadn't been a time like that. I mean, at that point, since, you know, people would talk about comparing that to like the .com era, right? So these are very unique moments in time. Maybe as a first question, because it's not, your product is not the simplest product. Like maybe just tell us a little bit about what it is that you do.

A Yeah. So basically AirBite is a data infrastructure product, and the value we provide as a product is wherever you have data, you can bring it into a place where you will be able to act and make decisions out of it. So our mission as a company is to make data available and actionable. So a very simple example is like for an analytics use case, you might have data about your customer across 20 different systems. And you want to be able to monitor everything that, uh, your customers are doing, like the health of your customer, what type of attachment you've had with them. And in general, what you're going to do is you're going to be using a warehouse for that. So it can be Snowflake, BigQuery, Redshift, you name it. And the data needs to be going from this operational system like Zendesk, Salesforce, and other into these warehouses. And normally your executives are looking at this dashboard on a daily basis. So you need that thing to be Reliable. And that is what Airbyte does is we are basically the pipe between every single type of operational system into a warehouse for the case of analytics.

AI assessment note: “we are basically the pipe between every single type of operational system into a warehouse”

Q And when you get to YC, what, like, is it just pure idea stage, or do you have something built? Like, what stage are you at?

A Yeah, I would say, getting into YC, we were in an idea phase, which is, we knew what problem space we wanted to address, but we did not really have a solution yet. So, it was very clear, like, we want to make access to data. As easy and as reliable as possible. But how do we do it? That was the, the question. And yeah, the, the first months we actually, like Jean and I have a, we're pretty good at just getting to talk to people, uh, reaching out to people. And that's always been part of the, the air bite story. And we spent the first month just talking to as many potential users or like even batch people, batch members about What kind of data issues they were saying, try to understand what could be a good, a good solution. And funny enough, at the end of January, we started on a marketing type of use case. Marketing was very, very hot, completely transformed by data. And that was the, the initial idea we started with was providing more relevant data to, uh, to marketing teams. And then COVID happened.

AI assessment note: “getting into YC, we were in an idea phase”

Q Tell me a little bit about monetization, like through 20, 21, you know, usage is obviously growing like crazy as your community is growing. Is it all free or do you start to charge something somewhere for it?

A No, we, we, we started to, uh, to charge, but it was more early support packages for, like, specific users. So we're not trying to monetize at the time, but we knew that there were people that were building critical pipeline on us, and for having a, a financial relationship with them was just making them feel better about adopting Airbytes. So we did that, and we were just supporting them. We didn't have that many at the time, but the community, though, was growing And the usage was just growing massively. We're just adding thousands of new users of Airbyte every single week, every single month. And so that led to the, to the Series B. And over twenty-twenty-two, we started to really have that new internal effort at Airbyte on, okay, we have a great brand awareness. We have great usage of Airbyte. Let's start thinking about, like, the, the next steps for, for monetization because, you know, Project market fit and product market fit are two different things. And that's when we started to build the cloud product that we released at the beginning of 2023. So basically it was around like mid, mid 2022 that we started to work on the, on the cloud product. And we released cloud in the, in 2023.

AI assessment note: “we started to, uh, to charge, but it was more early support packages”

Q We have tens of thousands of people who have followed the show. Are you one of those people? You want to be part of the group. You want to be a part of those tens of thousands of followers. So hit the follow button. And what were some of the use cases that, you know, X or Spotify were, were using it for?

A So for Spotify, what they were exploring, and they still use for this today, but like they were exploring building kind of like live concerts, uh, using live kits. So being able to have like an artist, you know, go up there and like do kind of half talk show, half live concert or play some pre-release stuff that they're working on, like for their Closest fans. That was like the Spotify use case. Oracle's was crazy. That was like a cyber truck police drone. Um, they use us today too for a bunch of different initiatives, but that was like the first one that they had an idea for. It was like, I don't even know if the cyber truck was out at that time, but they were trying to build these like police drones using Tesla vehicles and crazy. It's pretty wild. They do a lot of stuff with the government and military as well. X was looking at Twitter spaces. And, uh, and then there was Reddit who was trying to use us for Reddit talk. Um, Reddit uses for some other stuff now today, not Reddit talk. Reddit talk ended up getting shuttered and shut down post kind of coming out of the pandemic a couple of years later, but Reddit talk was what they were exploring to use us for initially. So we were talking to them and like, you know, when I say these use cases like Reddit talk, Spotify stuff, Twitter spaces, like these are all like really massive scale kind of like use cases. The communities on …

AI assessment note: “So for Spotify, what they were exploring... building kind of like live concerts”

Q And do you compete with like a VAPI, for example, or they sit on top of you?

A So VAPI and like land and retail and some of these, most of them actually are built on top of LiveKit. And so they use our infrastructure for all of the transport. Then they have their own infrastructure for orchestration and like they do their own deployment thing and And all of that. But I think the key differentiator there, they are using our underlying transport, but like the key differentiation is that like they kind of come in at a layer above on the abstraction level. So like their lower code and where they have an advantage or where they deliver a better experience to some cohort of people is that the time to market is very quick. You click some buttons and you put in a system prompt and you click go and you're You know, you're off to the races. LiveKit is more of a high code solution where you need finer grain control and like, how is this agent going to act? And I think of it personally as like Webflow versus like Next.js and Vercel. They both are solid businesses and they address a different segment of customer. But my personal opinion is that eventually you need to go to code. Eventually, if you have like 10 people that all come into a low code solution and through clicking some buttons, And getting a system prompt together, have like a viable product. Eventually those folks are all going to compete with each other. You know, the barrier to entry for competitors is …

AI assessment note: “most of them actually are built on top of LiveKit”

Q And do you compete with like a VAPI, for example, or they sit on top of you?

A So VAPI and like land and retail and some of these, most of them actually are built on top of LiveKit. And so they use our infrastructure for all of the transport. Then they have their own infrastructure for orchestration and like they do their own deployment thing and And all of that. But I think the key differentiator there, they are using our underlying transport, but like the key differentiation is that like they kind of come in at a layer above on the abstraction level. So like their lower code and where they have an advantage or where they deliver a better experience to some cohort of people is that the time to market is very quick. You click some buttons and you put in a system prompt and you click go and you're You know, you're off to the races. LiveKit is more of a high code solution where you need finer grain control and like, how is this agent going to act? And I think of it personally as like Webflow versus like Next.js and Vercel. They both are solid businesses and they address a different segment of customer. But my personal opinion is that eventually you need to go to code. Eventually, if you have like 10 people that all come into a low code solution and through clicking some buttons, And getting a system prompt together, have like a viable product. Eventually those folks are all going to compete with each other. You know, the barrier to entry for competitors is …

AI assessment note: “most of them actually are built on top of LiveKit.”

Q We have tens of thousands of people who have followed the show. Are you one of those people? You want to be part of the group. You want to be a part of those tens of thousands of followers. So hit the follow button. And what were some of the use cases that, you know, X or Spotify were, were using it for?

A So for Spotify, what they were exploring, and they still use for this today, but like they were exploring building kind of like live concerts, uh, using live kits. So being able to have like an artist, you know, go up there and like do kind of half talk show, half live concert or play some pre-release stuff that they're working on, like for their Closest fans. That was like the Spotify use case. Oracle's was crazy. That was like a cyber truck police drone. Um, they use us today too for a bunch of different initiatives, but that was like the first one that they had an idea for. It was like, I don't even know if the cyber truck was out at that time, but they were trying to build these like police drones using Tesla vehicles and crazy. It's pretty wild. They do a lot of stuff with the government and military as well. X was looking at Twitter spaces. And, uh, and then there was Reddit who was trying to use us for Reddit talk. Um, Reddit uses for some other stuff now today, not Reddit talk. Reddit talk ended up getting shuttered and shut down post kind of coming out of the pandemic a couple of years later, but Reddit talk was what they were exploring to use us for initially. So we were talking to them and like, you know, when I say these use cases like Reddit talk, Spotify stuff, Twitter spaces, like these are all like really massive scale kind of like use cases. The communities on …

AI assessment note: “X was looking at Twitter spaces.”

Q Do you remember any like mini ChatGPT moments? Like obviously ChatGPT is where everybody saw the possibilities. It sounds like you're describing well before ChatGPT, you seeing that things were on the come. I'm wondering if there were demos or products where you're like, oh, wow, I can't. Interesting that this is possible. Do you know what I mean? That, that we're kind of showing you a certain trend.

A To me, it was something very, very simple that blew my mind. Funnily enough, it wasn't even ChatGPT like, but Early on, you know, one of the issues with computer vision machine learning is you would have to deal with, like, let's say, large volume of legal contracts or bank statements. And the typical way of extracting information from that used to be computer vision, right? So you train models, you get them to try to recognize pixels and boxes, basically, and try to extract that out. And with BERT, entity extraction is becoming a thing. So a model Could know if that's a name or not a name. Is it a number or not a number? Is it what have you, right? Like a dog or not a dog. And that alone made a dramatic improvement to extraction rates. So, right, so you went from needing to train thousands about thousands of documents to 10 documents, maybe, to extract things like dollar amounts from invoice totals, uh, simple things like that. But it's happening in really, really large volume. And the amount of effort required To train a model was just so cost prohibitive before it got released that 10 tens of samples rate that it just never made sense even for really high volume repetitive grunt work. And that's actually what got me to take a big leap of faith and kind of bet on that trend line and basically make it possible to help anyone make machine learning models easier. That's kind of …

AI assessment note: “with BERT, entity extraction is becoming a thing.”

Q What was, by the way, the business that you were running before Runway?

A So I was running a company called Sandbox VR in early 2020. So I joined the company as chief product officer, and then around when we were raising our series B, I was appointed CEO. Um, the founder and the board thought that was the right move because I raised the round for our series A and I was going to do the series B and I was doing most of the job of the CEO anyway. So that happened, and then basically six months into the job, COVID hit. So we're 400 and some employees, and overnight we went down to 15 because we had to hibernate the company. But the process of getting there required us to make these models, because nobody knew how long COVID was going to last, right? Elon said in March of 2020 that it was going to be over by June.

AI assessment note: “I was running a company called Sandbox VR in early 2020.”

Q Um, and do investors doubt you, stakeholders, employees? Like that's the other thing is you got to keep everyone on this. We're going to get their employees.

A Definitely. Investors less so if you have the right investors. Yeah, it's very hard. And what made it easier are a few things. Oh, one is you, you try to talk to customers as much as you can, and you are constantly reinforced that this is a real pain. It's empirically true that there's a real pain because the market already exists. There's multiple examples of billion-dollar companies in, in the space. Um, so you know that this is a valuable thing. You also know that the status quo is bad. One of the things that helps is having other examples. So I had actually an expectation coming in that this would be a three and a half to five year build. From day before I started a company, like, that was my basic expectation. And that is because when I look at the most valuable companies that, products that I admire, Figma, Notion, Airtable, Coda, all of these were three and a half to five years to launch. It was six years for Figma to get to one million dollars in AR. Six years. And I remember talking to Dylan, Dylan's one of our investors, when I think I was like two years in, I asked him what was the most surprising thing about taking Figma from zero to where it is today, and he said, what's surprising the most is how few people made it all the way. And the more I thought about the answer, the more I think that's remarkable. Because Figma, now we know, is one of the most successful pri…

AI assessment note: “Definitely. Investors less so if you have the right investors.”

Q Cause what was the value of that document? Is it for it to train new employees or disseminate knowledge, like within the employees? Like how was it mainly used?

A Yeah. So, uh, we call those documents scribes. The primitive is a step-by-step guide with screenshots and instructions on how to do something. It's used anytime you have to explain to somebody how to do something, which shows up in a lot of different places, like onboarding people for sure, onboarding people. Brenda's retiring. She's been here for 20 years. We don't know what Brenda does. A lot of like new tools. We're rolling out Workday. Like, we gotta enable everybody on how to use Workday. New processes, like, hey, I'm in sales ops. Look at what a lot of sales ops people do. They spend a lot of time just trying to get sales people to do things correctly in the CRM. And they send them emails with, like, exclamation points, underlined, bold, like, please, please listen to me and, like, do this thing differently. And so we solved any of those use cases. Like, we're now making that automatic.

AI assessment note: “It's used anytime you have to explain to somebody how to do something”

Q What does a killer, 32nd demo look like for cybersecurity for Tenex?

A Cybersecurity really is the example, right? What we're doing, what we're talking about is detecting and responding to threats, and the way that this works, you might get a bunch of different alerts. My firewall, my endpoint, so like CrowdStrike says I've got this problem on this machine, and the way that this normally happens, somebody gets that alert, you know, an analyst working in your company or working for a managed security company, they get an alert, they get to it, they pull it up, They read it. They try to understand what it means. They then go, is there anything else that's going on in my company that looks like this, that might be relevant to this? I need to go try to pull all this together. I need to go run a bunch of searches and queries to try to figure out what's going on. It's a process that everybody in cybersecurity understands. It might take you. 15 minutes in a very nice. Everything went perfect. Maybe that was 15 minutes more often. It's like an hour. Two hours of work that I'm doing where I'm running all these queries, waiting on the data to come back, everything else. I come and pull up that same alert, except the AI, and we've got, you know, seven or eight different clusters of agents that are constantly working on our behalf. The AI has already clustered 30 different alerts that are all the same thing that you all want to look at. It's already produced …

AI assessment note: “the AI in 1:02 has done an hour's worth of human work.”

Q So eight years in, man, did you ever imagine it would get to this size? You ever imagined you would be in it eight years later?

A No, you know, it's one of those things where you found a company, you have a vision, the vision takes turns and you hit curves and you hit walls and, you know, different permutations. When we founded the company, myself and my co-founders, I think we all knew that we wanted this to be our last job and our last job, not in the context that we were going to make phenomenal wealth and then retire, but in the sense that we wanted to build products and culture that we would never want to leave behind. That this would be our home. This would be our final employment. So I think it's played out in that exact way, and we're still building as a founder-led organization, but no, I could not have imagined that we'd have 23,000 employees, that we didn't operate in multiple countries, that we'd have raised this much money. No, I couldn't have fathomed in my wildest dreams.

AI assessment note: “No, I could not have imagined that we'd have 23,000 employees”

Q So those are kind of the trends that are maybe in the background, right? Fueling you, but what was kind of the exact product or service that you wanted to build at the outset?

A So the original product was really tied to how we could leverage artificial intelligence. So neural networks and computer vision to create seamless checkout free experiences in the real world. So we wanted to start with mobility and within mobility, we started with parking, but we immediately realized that our vision or our product could extend across the mobility landscape, whether it was gas stations, carwash, quick serve retail, Tolling or parking. We could leverage computer vision to create personalization and seamless checkout free experiences in the real world. So the perfect example was parking. You could pull into any metropolis enabled facility anywhere in the United States, get a text message when you arrive and get seamlessly charged when you leave. And the way we did that was through the deployment of neural networks. We could create a fingerprint of your vehicle leveraging computer vision. Sending you this text message, identifying you when you arrived and then seamlessly charging you when you leave.

AI assessment note: “within mobility, we started with parking... get seamlessly charged when you leave.”

Q And first of all, you're creating 330 accounts, you're just doing that manually? Like, you're just going in, open, new account, new name, new whatever, boom, boom, boom, boom, boom.

A Initially, what we did is I just paid people overseas to warm up new accounts, have VPNs in the US, and then post. Echo does this automatically, which is the second product. We warm up new accounts using agents to automatically build, follow, and like a bunch of different accounts, and then post it. Because the other half of the battle is that, like, let's say you open up a brand new account, you post it, that's also useless, because it's not target audience posting. And so the ideal play, let's say you had a fitness app, is you want to create a new account, watch all the fitness YouTubers, follow all the fitness YouTubers, have your feed be curated by fitness-related content, and then by the time you post something, it will go to that related content. But then the secondary problems is around VPNs and proxies, which is a whole separate issue about being able to make sure that even if it is fitness related, it's fitness related towards the country you're supposed to be in. And so with my original operation, we had like a bunch of people be able to do that at scale. Uh, and I realized that like you can then automate this with software. And so when I bought Clover, I'm like, that is going to be the most immediate thing we're going to do. And then figure out the outliers of videos and things that are working at scale.

AI assessment note: “Initially, what we did is I just paid people overseas to warm up new accounts”

Q many keywords. Let's say again, back to the startup, but like there aren't, 10,000. Like there's just like, you know, you're going to run out, but I guess with blog, you can always, I mean, I'm doing new episodes. You can make a new blog post about every episode and maybe that drives traffic and it's incremental and whatever. When did you launch this video product and what is it?

A Yeah, we launched it about two months ago. It's called Echoes. Uh, and so what we did is that one of the biggest demands we had, which is an obvious demand is that like, once you solve SEO organic hacking, the next best big thing is like video hacking. And so right now Echoes only works for Instagram and like half of TikTok is like a little bit buggy, but It's going to solve for TikTok very soon, where we then mass post and mass produce videos at scale, all promoting your product. And so this is like AIGCs, AI slideshows, uh, AI posts in general that are just photo generated. And then we do that at a very low CPM. So average CPM, it varies depending on the industry, but average CPM is about a dollar CPM, which is like obviously a lot cheaper than if you were to do the same a thousand views on Facebook. So average Facebook CPM is about eight dollars in comparison.

AI assessment note: “we launched it about two months ago. It's called Echoes.”

Q I'm going to ask it differently. Like the typical question is like, when's the first time you, you felt like you had true product market fit, but let me ask a different thing. Like you have witnessed product market fit now across at least three, if not four products, maybe more. When do you know you have true product market fit? Like what is true product market fit to you?

A I would argue the first two companies I built were actually not the definition of how a market true product market fit. What I have seen in the last three to 45 days is that true product market fit is like purely based off referrals. If people love your product to the degree they say they do, they will refer other customers to that product. And if people can do that to say that not only you love your product and the market thinks the product is a good idea, but also the customers themselves think is a brilliant idea. And again, right now, 30% of our new top line revenue per month is coming from referrals. It's just people that are like, this is great. This works. I'm getting results. This is cheaper than Facebook ads. This is cheaper than any other mechanism that I've ever used to get a ton of customers. Let me give this to my best friend who also has an e-commerce business. Let me give this like another SaaS friend that I have.

AI assessment note: “true product market fit is like purely based off referrals.”

Q Okay. And is it true that during that time, like, Amazon is such a behemoth today, it seems like inevitable, but as I understand it, like, especially after the dot-com bust, like, it wasn't clear that Amazon would, would even survive. Is that true? Like, what was it like inside?

A Uh, I mean, I think the outside word was all about, I remember Time publishing an article saying Amazon.bomb, which was like, hey, it's all gonna blow up, and it's a bubble. But inside, I mean, I have to give credit to You know, the leadership there, Jeff Bezos, Jeff Wilkie, you know, they continue to see the massive opportunity that the internet will afford for online commerce and almost double down on innovation when everyone else was pulling back. You know, like in 2000 when, you know, the stock market had crashed and the internet stocks and the darlings had all, for one time, like really retracted back. Jeff Bezos and Jeff Wilkie, like they just said, hey, we're going to double down because we believe in this long-term This is a better way to shop. There's more convenient. You can have more selection, and you can have better prices, ah, long term.

AI assessment note: “But inside, I mean, I have to give credit to You know, the leadership”

Q 20 13 was pretty small. Like how many people, they're pre IPO. They're what, like series C company or so?

A That's right. They were around a 150 people. When I joined, maybe around 200 between when I interviewed and when I joined. Spent almost the 10 years there building infrastructure, um, on the infrastructure team. It was a very small team at the time and grew quite a bit as Shopify grew. We were doing a few hundred requests per second when I joined, and we went into soft last sales in the, in the millions of requests per second by the time I left. The majority of what you spend your time on if you're scaling infrastructure are the databases and the database layer. And that's where I spent almost all of my time caching databases and everything. The database that brought me the most trouble, both in terms of the product that we could ship and in terms of the operational experience was the search engine at Shopify. And I never thought that I was ever going to work on that again. But after Shopify, I encountered the problem again of working for a friend's company, an actual bootstrap company. I worked there for a few months, just helping them in infrastructure challenges. I'm doing a short consulting stint. And they asked me to build a recommendation engine, and so I built a small recommendation engine and used some vectors, because, I mean, vectors are great, right? They're a very good representation of what content is.

AI assessment note: “That's right. They were around a 150 people.”

Q know, it's a growing customer. We've got a bunch of little customers. Now you signed Notion. I would argue, 99% of founders in that situation, especially these days, they're going around, they're raising like a thirty million dollar Series A. I mean, you've got all the, like everything you need to have to go out and do that. You, you don't do that, and you haven't done that. Why not?

A There's six reasons to fundraise. Reason number one was the reason that we fundraise in January of 2024. Exactly how much money that we needed to prove to ourselves that we could find product market fit. If Justin and I could not find product market fit by the end of 2024, we were just going to close up shop. Justin and I were very clear about this. We even told the investors Who were invited for that round in 2024 that we were going to close up shop by the end of 24 if we'd not found PMF. That was terrifying to everyone but Lockheed. We raised enough money that we knew that we could fund R&D for 24. The second reason to raise is to fund growth. It's to fund marketing and other ways that you think that you could grow the business once you have a proven way to turn dollars into more dollars and mindshare. The third reason to raise And this is probably the most popular reason, and it's kind of what you're getting at here. It's for ego. It's also known as raising because you can, or because there is momentum, or because you got preempted. These are not first principle reasons to raise capital, which has real downsides, right? It dilutes employees. It ups the strike price. It has all kinds of ramifications that you now have to live with that might put your business at risk. The fourth reason To raise is to fund liquidity for the employees, right, that have believed and that have be…

AI assessment note: “The third reason to raise... It's for ego. It's also known as raising because you can”