Want to join the conversation? Connect with us on Tribe.
Find “Founders in Arms” on Apple Podcasts, Spotify, and YouTube.
Transcript of our conversation with Jesse Zhang:
If you're doing a good job, right, if you're if you're doing a good job as AI agent, you should be kind of resolving 70, 80, 90% of, of the incoming volume.
00:00:07:22 - 00:00:24:03
Unknown
And so that doesn't have to reach the CRM that just lives in your system. So that's great, right? So then you kind of become the system of record for that. But I think more importantly, you kind of become the system record for like the business logic. And that's very important. And so that's like, I think someone coined the term like system of intelligence for this.
00:00:24:03 - 00:00:33:14
Unknown
It's like, not only just storing the tickets because that that's one thing, but you're storing like, how should I got an agent at this company behave.
00:00:37:10 - 00:00:43:00
Unknown
Hi, everyone. Welcome to Founders in Arms podcast with me in Immad Akhund co-founder and CEO of Mercury.
00:00:43:10 - 00:00:57:00
Unknown
And I'm Raj Suri, Co-Founder of Lima and Tribe. Today we have Jesse Zhang. Jesse, welcome to, the Founders in Arms podcast. You are the co-founder CEO of Decagon.ai, is that correct?
00:00:57:05 - 00:01:17:04
Unknown
Yeah. That's right. It's great to be here. Really excited to chat with you both. Quick intro on my end. Yeah. I'm one of the co-founders and CEO here at Decagon. What we do specifically is, we build a AI customer service agent. So these are systems that actually go and talk to end users for for large businesses over chat or email or voice.
00:01:17:04 - 00:01:30:12
Unknown
And the goal is to give their customers a great experience and, you know, drive a ton of efficiency for them. Before this, originally studied AI at Harvard, started a company afterwards that was acquired by Niantic. And so this is the second one.
00:01:31:10 - 00:01:54:19
Unknown
People have been talking about kind of AI agents, kind of replacing customer support. And I think Kleiner did something about, like, firing 80% of the customer support. It's hard to tell from the outside. Like what's real and what's not real. Like. Yeah, and you don't have to, like, tell us, like, the specific companies. But, you know, when you get deployed, like, like, are you replacing human labor or are you making it more efficient?
00:01:54:19 - 00:02:02:00
Unknown
What percentage of human labor do you think you can replace like You know, where are we in like, this kind of X agent game? I guess.
00:02:03:04 - 00:02:22:20
Unknown
Yeah. I think my view on this is that it's a it's a really hot narrative to kind of say that AI is like replacing humans. I think in reality, it's it's not really that. It's not really. I think most people don't think about it that way, and it's really customer dependent. So we have customers that, you know, maybe they're later stage and they do just care about efficiency.
00:02:22:22 - 00:02:43:00
Unknown
And so their goal with the deploying us is like, hey, we want to, you know, either reassign a chunk of our team or be able to, you know, downsize the, you know, PPO operations that we're using. Or it could be like, hey, our business is growing super quickly, and we just don't want to hire, you know, new people linearly with the amount of growth we're having.
00:02:43:00 - 00:03:00:04
Unknown
So I think that's one category, but then still another category. And I would say this category is bigger, is that they're kind of coming to us because, hey, we think Jenny AI is so powerful. We just want to make our customer experience better. If there's efficiencies, that's great. But those are kind of like table stakes and this lower priority.
00:03:00:06 - 00:03:07:00
Unknown
Like we want our NPS to go up. We want our revenue to go up. We want our retention to go up. And I would say that's probably like more
00:03:07:00 - 00:03:09:00
Unknown
Where. Why would the customer experience improve?
00:03:09:00 - 00:03:17:00
Unknown
so yeah, we've all used kind of like traditional chat bots, right. It's like super frustrating all the time. Or you get on the phone and you're like, all right, press the, you know, press your
00:03:18:12 - 00:03:32:08
Unknown
there's a reason for that because, without generative systems, this system is pretty like you can do your best. But like we all know, it's not a great experience, right? And the reason is because the system is quite brittle. It's pretty hard to, like, be able to iterate, like, a lot of logic.
00:03:32:10 - 00:03:51:09
Unknown
And then, you get stuck down a path and you're like, oh, that's not really what I was asking. Like, let me watch. I go back and then you're stuck. So I think a lot of people just care about that. First and foremost. And we've seen some great results of, like, you know, someone's NPS was here and then after, like, we come in with the tenure system, like the NPS, like, you know, goes up A3X or something like that.
00:03:51:11 - 00:03:55:03
Unknown
And, that that's really what we see more in the markets. And
00:03:55:15 - 00:04:18:05
Unknown
So what you're saying is, like, it's not quite there where it's replacing, like, your in-house customer support, but your current kind of like automation and, like, this kind of lame chat bot slash, like, press one plus five that you can replace, and maybe you can replace your kind of tier one BPO. Customer support. So it's like taking it's improving the experience.
00:04:18:05 - 00:04:23:19
Unknown
It's replacing some stuff, but it's not like replacing kind of your. Yeah, it's a difficult problem and you shouldn’t
00:04:23:19 - 00:04:38:08
Unknown
be thinking about it as like, hey, I'm going to fire all my in-house support agents. Like, you'll be able to downsize if you want to. So we have we have customers that literally reduce their teams by maybe up to 80%, but it really depends on, what their priorities are and, like, what they want to get out of it.
00:04:38:10 - 00:04:41:12
Unknown
And also, of course, it depends on the distribution of their tickets and stuff like that.
00:04:41:12 - 00:05:09:19
Unknown
Yeah. Probably depends on the type of company as well. We're actually looking at Decade Gone, and we are very much looking at it in terms of like this year, either we go to get a BPO, to, you know, do, kind of tier one, customer support, inquiries, all we do some sort of AI solution because, yeah, it's like I think the number was it's like 30 ish percent of the questions are like, very I mean, not very easy, but quite easy to answer.
00:05:09:21 - 00:05:22:00
Unknown
And they don't actually take the majority of the time, but it's just not like useful time. I think it's like five ish percent of the time. I mean, and like, those are like the rough orders of magnitude or like, the percentages, if that makes sense.
00:05:22:03 - 00:05:41:02
Unknown
Yeah. I would say if if the if the comparison is between, like, adopting a good agent and like a BPO, that should be a very obvious, position to be, because with Bpos, because they outsourced, by definition, none of the test scores are going to be that complicated or that involves. And then you have this process of new to train them like, oh, churn.
00:05:41:03 - 00:05:50:12
Unknown
There's like all this costs about like maintaining the, the sort of like consistency, of level across, across all the agents. And so that's yeah, that's a good point.
00:05:51:16 - 00:05:59:00
Unknown
Immad, why are you thinking of, Like, what's, benefits of the BPO? That you're thinking of Jesse saying it's an obvious decision. I'm curious where you think.
00:05:59:00 - 00:06:16:14
Unknown
Well, the reason to do a BPO is, you know, we want to. We have aggressive growth goals this year. And, like, we have, I think, six, six agents. Right? Like, we don't want to go, like, if we were going to double our users, do we want to double like, we have to go high as 60 people and train them, right?
00:06:16:18 - 00:06:41:12
Unknown
Like, that's like a very, it takes a lot to, like, double any operations team, especially when you got, like a reasonable scale. So, you know, if instead of doubling, maybe we can grow the operations like the customer support team by 50% and like, we can use AI to, like, fill to fill the gap. And why do it instead of BPO is, you know, when you're using Bpos, it's actually I mean, there's still like humans, you have to train them.
00:06:41:12 - 00:07:01:16
Unknown
You have to onboard them. They still make mistakes. So there's QA, all that stuff. Right. So it is a little easier to hire 60 people. VI BPO like the, like, more scalable, talent and and a bit cheaper. But, like, at least the dream is that, you know, something like that Decagon, does it? And this is turning into Ad for you, Jesse.
00:07:01:18 - 00:07:19:23
Unknown
But, yeah, the dream is like, you just train the AI once and you can monitor it and you can keep improving it. And yeah, eventually you can do more complicated things and it can integrate into your systems and like, you know, do all this stuff and know when to escalate to a human, etc.. And it can answer questions quicker and like in the middle of the night.
00:07:19:23 - 00:07:38:12
Unknown
Right. Like we, we actually just hired like five people in Australia and five people in Ireland, literally to answer my questions in the middle of the night. So yeah, it's there's just a bunch of things that like if an AI can do good enough and I give instant answers to a bunch of queries, it would be just as you, as Jesse said, also a better customer experience.
00:07:38:12 - 00:07:57:17
Unknown
really well right now? I think Jesse mentioned, like, a really good point. Like to, like, you know, it is annoying for all of us. And actually, this sort of, like, space of attempted. I. We're just not, like, quite perfect, but it's like, okay, is, is that it is a pretty annoying kind of middle zone right now where it's like you're someone is trying to use AI, but they're not really doing a good job.
00:07:57:19 - 00:08:13:12
Unknown
And you're like, well, you know, maybe you ask one question to get you a little bit further. And then the second question, it fails. And then it has to do human. And the human doesn't have context. Like I feel like most consumer service because at least what I use doesn't seem to have nailed this one.
00:08:14:12 - 00:08:29:03
Unknown
Yeah. I mean, so. And to your point, there's been a long history of automation in this space. Which makes sense because it's it's a very kind of, you know, obvious space for automation. Like, you originally had people building the CRM and then like, people building IVR systems, and then after that they were like the chat bots, which were a little bit smarter.
00:08:29:03 - 00:08:45:08
Unknown
They use like basic NLP stuff. You can do classification and intent routing and stuff like that. But you know, that's still good science in the current position where, you know, anytime we're on something like this, we're trying to get to humans as quickly as possible. So the step function change has been like this generation. And obviously the space is still early.
00:08:45:08 - 00:09:05:22
Unknown
So I wouldn't claim that anyone has like, including us, like, solved it. I think there's just a lot to like to build still. But obviously, I think the reason why, you know, we're growing quickly and, you know, other folks building Genii, solutions are kind of getting adoption is because people are seeing this step function change. And, I mean, we have our own unique take on it.
00:09:05:22 - 00:09:31:09
Unknown
I think other people probably have their own view, but I would not claim that anyone solved it. I think the reason why that this generation is better and like why I guess I can speak for us, like why things have gone fairly smoothly so far. You know, like knock on wood is that, if you look at the approaches that everyone has taken, including a lot of the AI players, it's kind of using this system of, hey, we have all this logic that we want the AI to do.
00:09:31:11 - 00:09:49:21
Unknown
Let me like try to program it into like some framework. Right. And like the framework in the past used to be some like decision tree or, or whatever. And like these days it could still be some version decision tree plus some code or something like that. And the issue with that approach is that it's it's very like unlike how you would teach a human naturally.
00:09:49:21 - 00:10:05:02
Unknown
Like it's really unnatural for us to teach systems that way because you have to like, figure out like, oh, okay, how do I like, jam into this protocol and you end up getting systems that are more brittle because, you know, you're stuck in one area, is not flexible, and it's just a lot slower to iterate because it takes a lot of time to build these systems.
00:10:05:02 - 00:10:26:13
Unknown
And you kind of need like a specialist group of people to maintain them. And so our approach so far has been radically different. We, we believe that you should be able to teach these AI agents in the same way to teach your human and the way you teach a human. And, you know, like you mentioned, at 60 people, it's like you have recipes that are like, okay, I've defined it like written down instructions for new agents to come in and read.
00:10:26:14 - 00:10:41:21
Unknown
And, so we've created what we call a ops agent operating procedures. And that's, that's kind of like the, the main innovation that we've made, which is allows you to kind of use natural language to define these procedures and in a very robust way. And, then you can iterate a lot quicker and you get to higher quality a lot faster.
00:10:41:21 - 00:10:51:12
Unknown
And, it's cheaper to maintain as well. So, that's been our approach. And like that's why I think this generation has been better. But yeah, I would not say it's like solved yet.
00:10:52:00 - 00:11:07:12
Unknown
Jesse, you, We had dinner recently, and you were giving some insane stats. I don't know what's Publix. Like, what is like your, What is your public kind of, out there about, like, how fast you're growing and, Yeah. What metrics do you talk about?
00:11:08:01 - 00:11:28:12
Unknown
Oh yeah. No, I appreciate that. I mean, the only thing we're public about is our funding. So we we basically started, late 2023. So this is a post GPT four and we kind of did our process of discovery figuring out like what use cases could make sense. So we decided, you know, customer support and customer service or just customer experiences broadly is is like the the best space.
00:11:28:14 - 00:11:54:00
Unknown
So we start building we raised a seed A and B, with within about 12 months, 100 mil total raised. And yeah, I mean, I think that's, that's what we're generally public about. And I think there's still a lot of room to grow, obviously. Like the space is gigantic. Like we don't know what the reason we're not that public is because I don't think we want to brag about what we've done so far because I still think, like we
00:11:54:06 - 00:12:12:10
Unknown
These numbers that you see on, like, on X or Twitter or whatever, real like when people are like, oh, they say our company got 100 million in revenue in like two years. And, you know, you see a bunch of this stuff like, I'm a little skeptical when I see these numbers, like how how much of it is real from your perspective.
00:12:12:10 - 00:12:21:00
Unknown
And like, I guess, like, not to, not to put you on the spot here. Yeah. What do you think of the other companies that do profess these, like, big numbers?
00:12:22:12 - 00:12:30:00
Unknown
So I think there's three modes here. The only the only company I'm aware of that's, like, remotely close to that is, cursor.
00:12:32:09 - 00:12:34:00
Unknown
Yeah, yeah.
00:12:34:00 - 00:12:40:09
Unknown
that's kind of like if you look at the spectrum of, like, you know, prosumer, like consumer, that's like, yeah, I think those things can be pretty explosive.
00:12:40:11 - 00:13:00:02
Unknown
I think on the enterprise East side, I think that's a lot of folks that are growing quickly. And like selling to SMB or other startups. I think that's a little shaky because the churn is actually quite high. And, I mean, I actually don't know much about, like, the AI for sale space, but that's what I've heard.
00:13:00:02 - 00:13:10:00
Unknown
Like, I think a lot of those tools, it's just like you're getting a ton of customers and so you get a lot of growth, but they're all small customers. And then it's like, oh, I don't really see the value. And they churn.
00:13:11:12 - 00:13:15:00
Unknown
space, generally our contracts are like 6 or 7 figure contracts.
00:13:15:00 - 00:13:32:13
Unknown
And so there's a lot more of like people like really quantifying ROI and and kind of measuring that way. And so I think, I guess to summarize my answer your question, I think the ends of the spectrum are more real. So like if you're selling to enterprise or like you're something like cursor like that feels very real.
00:13:32:14 - 00:13:47:00
Unknown
I think the intermediate feels a bit more shaky. Not to say it's like not real, but I would be more skeptical about and I don't I don't think there's that many companies on on the ends. Like I don't think there's that many companies that are growing quickly. So enterprise and I think that there's that many cursors either.
00:13:47:00 - 00:14:10:16
Unknown
I also think every business I mean I assume enterprises do like medium sized businesses telling every team and saying like hey what AI tools are using like how you driving efficiency or growing sales using AI. So like there's a lot of like top down pressure for everyone to go experiment with this stuff. So it creates a, it's actually kind of like exciting to be one of these companies because you have like a sales opportunity.
00:14:10:18 - 00:14:22:12
Unknown
But yeah, it's like you can you can kind of be a head of product market fit. Right? Like you you can get like sales quickly. And actually your product doesn't work that well. So it turns I can definitely say that.
00:14:22:15 - 00:14:40:07
Unknown
You know, Jesse, what you're doing actually reminds me a lot of what I used to do at Presto. Where, you know, we did, AI for drive thrus, which is another type of customer, service. And, we were we were quite early. I mean, we were doing this before ChatGPT, but, you know, it obviously got better afterwards.
00:14:40:09 - 00:14:59:02
Unknown
And, that type of narrow use case was actually really good for AI because it's like, very clear where you're going when you go into a drive through that you're there to order food, and there's only limited amount of things that you could order. So a lot of it was just trading on the menu. And the escalations were pretty, you know, also self-evident.
00:14:59:04 - 00:15:16:19
Unknown
I think even in that space though, like, has it really reached full penetration or anywhere close to, like, even 20% penetration yet? You know, it's such an obvious one, like, yeah, I go to drive through, like, you know, the humans in the restaurant are super busy making the food and doing other things. And the AI can upsell much better than human.
00:15:16:20 - 00:15:38:12
Unknown
Like it actually remembers to, like, you know, ask you if you wanted to shake with your burger, but still actually hasn't reached, you know, significant penetration. So what do you think the timeline is of, like, you know, like, what do you think we'll get, you know, 10% or 20% of, like, our customer service is like, AI or do you have any predictions on that?
00:15:38:22 - 00:16:00:05
Unknown
Yeah. For your question, I'm also curious to hear your take on why, at least in that space, the penetration hasn't been that high. So I think the, the reason why, like, there's, like, so much hype around AI is not so much that, like, people are. I think that the tech is cool, but I think like, especially in the enterprise, there's hype because like, I think a similar to what I might say, like senior leaders have a ton of pressure to adopt AI.
00:16:00:07 - 00:16:24:03
Unknown
So there's like so much tailwind for them. So just like push their team to like adopt something and just like be able to tell their board or tell their investors like, hey, like we're using AI now there's like efficiency here. And that's great for our space because, that means that, you know, some sales cycles that may have taken like a year and a half can be shortened to like, you know, half that time because, like, there's just a lot more movement and, you know, you have to get through like security, compliance or whatever.
00:16:24:03 - 00:16:44:19
Unknown
But like if the senior leaders really care about it, like that process becomes a lot quicker. So, I do think the timelines will be accelerating, 10 to 20% penetration. I actually think we'll be there in like a year or two. And it's kind of like the, the crossing, the chasm curve where, like, I already think that, the, like the early adopters.
00:16:44:19 - 00:17:00:09
Unknown
I think that was like last year, I think this year is already like, that, you see, like the early majority actually adopting and the, you know, there's there's like large contracts and stuff like that. Again, the space is still early, like who knows what's going to happen with these contracts and stuff like that. But yeah, I do think that adoption will be pretty quick.
00:17:00:09 - 00:17:05:12
Unknown
So yeah, here's why you think the adoption has been slow for for the drive through space.
00:17:05:14 - 00:17:26:19
Unknown
You know, I think, the value is is very clear, but, you know, I will say one thing that it's dependent on is also the cost of labor and availability of labor. So, you know, that's a major determinant, right? Like, you know, you're comparing an AI solution against, you know, labor and solution. So the cost of labor and availability makes a big difference.
00:17:26:19 - 00:17:45:19
Unknown
When there was, a big labor shortage a couple of years ago, you know, especially coming out of the pandemic, you know, the adoption tend to be much faster since that there was, an easing of those conditions. So then the value prop switched to more of the upsells and like the, you know, driving more sales, which is, which is a good thing.
00:17:45:19 - 00:18:02:10
Unknown
But the, you know, you want the labor part as well. Are you typically competing with, like, with folks who are using, like, offshore, support, you know, is that the typical competition or legacy solution?
00:18:02:10 - 00:18:19:01
Unknown
For like enterprises when you're talking to enterprises like everyone has, like, the idea, it's very rare for enterprise to have, like, a full onshore support team because the volumes are so large. And so you need that, like, scale and you need the flexibility. So, yeah, I think that's that's almost
00:18:20:20 - 00:18:48:14
Unknown
Seems like the difference for you. Difference for you. Rogers. Like every fast food place has a person there already. So there's, like, like a very direct trade off. It's like, oh, you could hire 5% more people. Whereas, like, with customer support, you might have a thousand people. And like, if you can, like, you know, have a 20% gain, it's like it's a lot more, scalable and easier to apply and do the reasoning off probably.
00:18:48:16 - 00:18:50:12
Unknown
Would you say that story?
00:18:51:15 - 00:19:11:01
Unknown
Yeah. I mean, I would say like, it's also because you needed a person physically at the restaurant. It's not easy to find that person. Right? Like, it's, And that person tends to be busy, so there's a reason for that as well. I mean, yeah, if you have a massive team of people in a call center somewhere, then, you know, you could get more efficiency, I guess, from that.
00:19:11:01 - 00:19:18:00
Unknown
But, there's also there's friction in, in also distributing the calls internally. To some degree.
00:19:18:00 - 00:19:42:00
Unknown
Jesse, the, you know, as a as an investor, actually. And I could go on. But in this space in general, like the two things that always come up, as, like, you know, when you look at a startup is, you know, number one, especially as these models get better, is it's super easy for another startup to come around and say, hey, you know, we also using open AI, we're also going to have, x agent and that I mean, that has happened in this space, right?
00:19:42:00 - 00:20:06:00
Unknown
There's like a hundred fricking CSS agents. And then number two, all of these platforms out there, whether it's like Zendesk or which I don't know. Well, all the other ones are the that do CSS, stuff like that will obviously also going like, hey, we need oxygen. So, I think your space actually like very much personifies both those problems.
00:20:06:02 - 00:20:26:00
Unknown
I would love to hear and not just for decade gone, but in, in general, like, you know, what are things that, like on both sides are like, moats that startups can have, like, you know, how can they not be just, GPT wrapper and how they can compete against, like, these incumbents that already have distribution.
00:20:26:17 - 00:20:42:11
Unknown
Yeah. Good question. So it's like the classic. Oh, if you go super quickly, that's awesome. But the the counter to that is that means that it was actually quite easy to go quickly, which means other people can actually come in and the barriers are low. And like they could possibly go quickly to And I think that's true.
00:20:42:11 - 00:20:59:23
Unknown
I think that's true of like, I don't think that's even true of like just agents. I'm sure there's like many SAS spaces in the past that were like this. And so typically what happens is, you know, there's like a lot of players and then eventually like probably two, maybe three, if the space is large, like, people kind of consolidate and like become the winners.
00:21:00:01 - 00:21:19:00
Unknown
And if you look at like, why they become the winners, I think in reality, in my opinion, it's just mostly like they just have the best teams, they move the quickest. They're able to like, build the best product, like faster. And so that's more or less what we are focused on right now. It's like, okay, we it's like a very heavy like execution thing.
00:21:19:00 - 00:21:37:13
Unknown
And again, like most of the spaces are like this. So like you just have to make sure like every day your whole team's locked in. There's like a ton of execution, excellence there. But then of course, structurally, you want there to be some, like, lasting advantages from the products. And historically, this is like the classic all your either your system record or your system of, you know, behavior or whatever.
00:21:37:15 - 00:21:56:10
Unknown
And, I think every AI space that's kind look a little bit different. I can kind of going could kind of explain what it looks like in our space, which is that originally you had, you know, CRMs. Right? So like, those are kind of like the system of records for when you have, you know, a million support tickets and they all live in the CRM and like, you have information, everything.
00:21:56:12 - 00:22:15:11
Unknown
All the agents are living in their. So if you have a good agent, you want to be thoughtful about how you can kind of take some of that advantage and get people like in your system as much as possible. If you're doing a good job, right, if you're if you're doing a good job as AI agent, you should be kind of resolving 70, 80, 90% of, of the incoming volume.
00:22:15:13 - 00:22:17:19
Unknown
And so that doesn't have to reach the CRM that just lives in your
00:22:20:12 - 00:22:31:18
Unknown
record for that. But I think more importantly, you kind of become the system record for like the business logic. And that's very important. And so that's like, I think someone coined the term like system of intelligence for this.
00:22:31:18 - 00:22:48:16
Unknown
It's like, not only just storing the tickets because that that's one thing, but you're storing like, how should I got an agent at this company behave. Right. And so I kind of consolidate all the knowledge around, you know, how do we handle this workflow? How do we handle this for, like, what? What should the tone be like?
00:22:48:16 - 00:22:51:12
Unknown
What words can you say? What? What can you do? And if you're
00:22:51:12 - 00:23:15:12
Unknown
say system of intelligence, are you talking about actually integrating into the backend, like database and all these things and actually executing things on behalf of, like, the agent? Okay, so it goes way beyond just like, like a read-only kind of thing that just, like, answers things that actually goes and like, does the thing in the back end and like, goes and like, resolve something or ship something or something like that.
00:23:15:15 - 00:23:33:15
Unknown
Yeah. Yeah. Exactly. Right. So let's say, like, you know, working with a financial services company, it's like I have a credit card, and I lost my card, and I need to reorder a new one. There's, like, a very specific way that that company wants to handle that process. And it's usually not simple. Usually you have to, like, hit many data points and like take an action at the end and maybe have to lock their old card or whatever.
00:23:33:15 - 00:23:55:00
Unknown
Right. And so that is like the intelligence that you need to learn. And it takes time to learn this. And like the, the, the company will continually like iterate on this over time. And so if you're here in ideally you've built up so much intelligence there that it is actually like hard to like replicate that. And there's just like a lot of well, I don't like
00:23:55:00 - 00:23:57:12
Unknown
like the system of intelligence. I've never had that before. Have you
00:23:57:12 - 00:23:58:12
Unknown
had that before? Right.
00:23:58:12 - 00:24:00:00
Unknown
Well you
00:24:00:00 - 00:24:02:12
Unknown
should you should make it a thing, Jesse.
00:24:02:12 - 00:24:22:00
Unknown
it's kind of like the. So yeah, system of records are strong because there's just so much there is like, oh, like, I have I got a new thing I'm going to have to, like, somehow explore all this stuff over and then like, all my metadata will be messed up and like, you know, so that's, that's like the same idea and, I think they'll probably be more and more common for AI agents.
00:24:22:00 - 00:24:40:12
Unknown
Yeah. And if it's your space. Very well. Like a, you know, sales agent doesn't necessarily have, system of intelligence as much. I mean, they can answer questions, but, you know, there's not like that. There's not that. Go do this thing in some internal system that, like, is going to be complicated and have lots of, like other systems to integrate with the logic associated with it.
00:24:41:09 - 00:24:55:00
Unknown
Yeah. It's interesting. I actually, I hadn't really thought deeply about this for other types of agents because if you look at like coding agents, which is like the other big space in addition to support right now, it's like, does this exist?
00:24:55:04 - 00:25:10:12
Unknown
Every company has like different coding style. Like they have different rules around like what kind of testing frameworks you have to use and. Yes, but but a lot of that stuff does get standardized because of like, open source and like developers want to use like the same systems between companies. Maybe.
00:25:10:12 - 00:25:28:00
Unknown
Yeah, yeah. I think, the other way to put it, I think the level of customization for our space is very high from customer to customer. Like, each company has their own way of doing support. It might be quite different. Whereas for coding, I imagine it's a little bit more similar from company to company. So yeah, interesting stuff to think about.
00:25:28:11 - 00:25:49:09
Unknown
Jesse. It sounds like you're also doing like, a kind of a hardcore enterprise sales company. You know, similar to what I did as well. I know that's a pretty, steep learning curve for founders. I used to always, kind of enjoy competing with other startups because I knew how hard it was, to be, like, really good at enterprise sales.
00:25:49:11 - 00:26:04:00
Unknown
And I found most founders, technical founders coming into enterprise sales really struggled with, like, the sales motion and the skills that you needed to win, which are very different than just building the product. Like, you know, it's, you have to, you have to build a good product. But that's only step one. Like, you have
00:26:04:04 - 00:26:07:12
Unknown
What was, Raj? What was the hardest thing to learn?
00:26:08:19 - 00:26:28:05
Unknown
Just how much of that was relationship driven to some degree? And, and also how to hack those relationships. I found ways to, like, hack, like, even though I didn't have the relationship with found ways to hack them and also how much of it was also, based on, you know, like hacking the incentives that enterprises have, right?
00:26:28:05 - 00:26:47:06
Unknown
Like they have incentives to, like, not be fired, obviously. Right. So, like, the buyers are very risk averse. So how do you position yourself in a way that you're the you're the safer option in every scenario, whether you're competing against a big company or a small company. So like and then also how do you like, you know, make sure you win the process every time.
00:26:47:08 - 00:26:51:00
Unknown
Like there's all sorts of different tactics there which, you know, probably took maybe
00:26:51:00 - 00:26:54:00
Unknown
What's it? What's your best relationship hack?
00:26:54:00 - 00:27:10:00
Unknown
Find people who worked with the buyer directly, like closely for many years and find ways like, like to incentivize them, you know, to be part of your company so that they'll, they'll win the deal for you. And,
00:27:10:00 - 00:27:15:00
Unknown
you mean, like, consultants or stuff like that? Like consultants or GSEs? Like people like that? Or.
00:27:15:03 - 00:27:32:04
Unknown
Ideally, it's like a, like a colleague of that person who worked with them for many years and like, like someone who's like, work hand in hand or work for them or with their boss and like, find ways to get that person on your team in some way. Like you could make them an advisory board member. You can make them, you know, a consultant for you.
00:27:32:04 - 00:27:45:00
Unknown
You could, you know, you could just go out and wine and dine them, you know? But the key is to have that person have like, be able to call the buyer as a backchannel and get deals done for you. And like that by itself is way
00:27:45:00 - 00:27:50:00
Unknown
That sounds like way too much work. Raj.
00:27:50:00 - 00:27:59:00
Unknown
Jesse, tell me if I'm right or wrong. I don't know if you want. You're in a head to head fight on a bake off and stuff. You know, you'll realize that this, this, this is the thing that puts you over the top.
00:27:59:00 - 00:28:17:13
Unknown
I think that's true. I mean, I do agree it's like it's been a linear learning curve, but, my last company was consumer. I would argue that I was, like, way harder, at least for me. I think it's like, maybe person dependent, like consumer is just like you just you can be bashing your head against the wall for like, just like a long time and grinding and like, there's there's a lot of luck involved.
00:28:17:13 - 00:28:31:04
Unknown
There's a lot of intuition involved. Whereas at least for enterprise sales, it's like more systematically. You have a lot of signals that are real. It's like like at the big signals, like, here's a buyer and like they're willing to pay money, like, can he get the money? So I think that that part has made it better. But yeah, I do think there's definitely a learning curve.
00:28:31:04 - 00:28:49:11
Unknown
Like, both my co-founder I like relatively young. And so there's, there's I'm sure there's advantage of like being older, having more of a network and stuff like that. But, we've enjoyed it so far. I think it's like, we're very competitive people. We like the feeling of like, hey, there's like there's a concrete deal's concrete, like goals to go after.
00:28:49:13 - 00:29:14:12
Unknown
And that gives us a lot of energy. So yeah, but I do think the learning curve like this, like you mentioned Rogers, like stuff like that that, you know, we've had to adapt to. There's, like, how do you get in front of these people? Right. Like right now. It is it is almost always a very network based thing because, like, yeah, when you're at that level of enterprise, like it's just like this, people have strong networks and there's a lot of relationship building.
00:29:14:14 - 00:29:25:00
Unknown
So, you know, we've relied on of our investors. We've relied on a lot of like, you know, successful founders that are helping us. And so that is that's been helpful for us to get off the ground, at
00:29:25:05 - 00:29:42:00
Unknown
Well, what, Like, what has been the most useful in terms of networking with enterprises is, like, you just ask, like, 100 people that are backers and, kind of investors and friends of yours or or. Have you found, like, one particular channel is, like, particularly good in, like, helping you kind of get to these enterprises?
00:29:42:22 - 00:30:02:04
Unknown
Yeah. I mean, it is mostly our backers. So you were pretty, like, open with them about like, hey, here's the accounts. Like, can you get us in front of them in, like, we have amazing backers, and so they're really helpful. Also like, yeah, big shout outs, like the major VCs that we've raised from A16z XL. Bain.
00:30:02:04 - 00:30:22:00
Unknown
Like, if they like, I think there's a common trope for like VCs on the operating side are not that helpful, but at least I think is obviously situation dependent because we had a bit of momentum because we have a lot of interest for enterprises like they've really helped us accelerate, like getting in front of people. And that's yeah, that's been quite helpful so far.
00:30:22:02 - 00:30:42:17
Unknown
Yeah. I actually think VCs are more helpful in enterprise sales than, like, in a lot of other things. I yeah, we we mostly do kind of SMB, startup sales. And there, you know, it hasn't been that helpful to go like that deep to break that into, like, one single enterprise. But, but yeah, a lot of people, like other portfolio companies.
00:30:42:20 - 00:30:51:08
Unknown
I mean, we also backed by Andreessen Horowitz and yeah, they're like, wow, they put me in front of this CIO and I closed the sales because and I'm like, oh, that's actually like super value add.
00:30:51:17 - 00:30:53:11
Unknown
Yeah. Yeah it is.
00:30:53:11 - 00:31:13:08
Unknown
One thing that tends to help is like, You know, for any target company, they generally have a financial component or like a capital markets facing component. So like any board, like, you know, you either target customer. Right? They may have someone who's like, quite, you know, maybe working at a public company.
00:31:13:13 - 00:31:36:22
Unknown
And those people tend to be easier to get Ahold of. And, and VCs can kind of triangulate their way into that. That board member, and, and the board member get you access to the CEO. So, you know, I think, like going through the capital markets route into a, into a, a customer, even if the VC doesn't have a direct connection, they can go through that route tends to get you more access.
00:31:37:00 - 00:31:39:11
Unknown
That way, we found that was very successful for us.
00:31:39:11 - 00:31:59:14
Unknown
Yeah. And that even, like, a lot of the enterprises have innovation teams, right? Or they have, like, a VC arm. And generally the advice is like, it's not usually that useful because, I mean, like, they're incentivized just to talk to a bunch of people and so on. But yeah, at least in our experience, we've met some great innovation teams and they've actually like, you know, made meaningful progress for us.
00:31:59:14 - 00:32:05:23
Unknown
And so I think it's just you got to try a bit of everything. And, you know, hopefully you can get the relationship strong enough.
00:32:05:23 - 00:32:35:07
Unknown
think it helps because you're. You know, the innovations team mandate right now is probably. Go find AI companies that can make us more efficient. So, like, all the all the incentives kind of, line up, what's your kind of general take on, like, this kind of model war, right. There's like every, every week there's a new state of the art model, whether it's like DPC or the new new cloud one.
00:32:35:09 - 00:32:43:23
Unknown
I guess, like, do you think there's one that's better than the rest? Which ones do you use? Like what is your general thoughts on like are the different algorithms?
00:32:43:23 - 00:33:15:06
Unknown
Yeah. I think it's great that there's a model where, it benefits applications like us a lot, both in terms of obviously models getting better as good. But, you know, price goes down and and so on. I would say that's the best ones are still very clearly, OpenAI anthropic. So I think like those are. Yeah I think any perfect like us we will have like a lot of Yuval set up internally for us to evaluate models based on like what we care about, for example, whether things we care about the most is like, how good is the model at following instructions?
00:33:15:06 - 00:33:39:06
Unknown
Because that's what really matters for our customers. I think Gemini is like it sounds like is making a bit of a resurgence, at least from my like, friends of mine that have talked about it. So maybe those three. I would say that, I don't think there's like a clear winner between OpenAI AI anthropic. I think their models are good for different things.
00:33:39:08 - 00:33:39:17
Unknown
And
00:33:39:17 - 00:33:47:17
Unknown
How quickly can you test a new model? Right. Like when DPC came out. Was it like, hey, let's go test it in, like, within a week, you can run your evals. There's a really quick.
00:33:47:17 - 00:33:49:17
Unknown
less than a week. Like within a day. Yeah.
00:33:49:17 - 00:33:57:15
Unknown
a day. Okay. So it's super easy for you to swap these out. Like, if, if. Yeah. Tomorrow. Like, the new llama comes out from meta.
00:33:57:17 - 00:34:01:17
Unknown
You can put it in. If it's better. You can like start sending some traffic. That way.
00:34:02:00 - 00:34:03:17
Unknown
Yeah, exactly.
00:34:03:17 - 00:34:13:05
Unknown
That's interesting. Do you think like, these companies are going to have, like, long, long running moats, given how easy it is for applications to kind of switch them out?
00:34:14:15 - 00:34:24:05
Unknown
It's hard to say, but, like, the other thing is, Oh, I was gonna make the cloud provider analogy, but it actually is not easy to move cloud, so I
00:34:25:17 - 00:34:35:06
Unknown
makes sense, but I do think it's like the advantage they have is that there's already like a determined set number of companies that could win as a model.
00:34:35:08 - 00:34:53:09
Unknown
So like they basically already won because even if there's like I mean there's like West Sub five, right? So there's like so five models that people will like, because he will use that scale and like it's very difficult for other people to catch up. I mean, maybe DPC like distilling models has like shown that that's, that's easier.
00:34:53:09 - 00:35:06:17
Unknown
But, I do think there's going to be like a oligopoly. And so it doesn't really matter that much if it's easier to switch between them, because as long as they're all kind of on on par, there, they're all going to be like gigantic businesses.
00:35:06:17 - 00:35:19:17
Unknown
And maybe there'll be some that are better than others at different things like anthropic seems to be like focusing on coding a lot more and winning there. So maybe they'll be like each model will have like a slightly different flavor to it of specialization.
00:35:20:00 - 00:35:21:06
Unknown
Yeah. That's fair.
00:35:21:21 - 00:35:40:06
Unknown
Jesse. How do you. What do you think of the space in general? Like there's a lot of venture dollars going into AI companies. You know, there's some some eye popping valuations. Like, what do you see is real? What do you think is like, you know, kind of just, hype at this point.
00:35:42:07 - 00:36:00:04
Unknown
I think yeah. Obviously necessary I think the some of the like prosumer companies are real and some of the enterprise companies are real. The the middle ones I, I'm like I'll reserve judgment. I'm like more skeptical of those I think is just more like. And I feel like this doesn't get talked about enough, like the incentives for VCs and founders are actually, like, not that aligned.
00:36:00:06 - 00:36:17:15
Unknown
Because if you think about what a VCs incentivize, right, they raise the fund, they want to, you know, look good for their LPs and they want returns. And they know the AI is like super hot right now. And so they're just trying to deploy like AI dollars into like, as much as possible. As many companies as possible.
00:36:17:15 - 00:36:33:07
Unknown
They're okay at the valuations are inflated. These they think that there can be 1 or 2 huge winners as long as there's 1 or 2 huge winners like you. Good. But like, let's say the our ten companies, just one winner, the other nine that they gave huge valuations to are kind of screwed because for those founders you're like all right I got a great valuation.
00:36:33:09 - 00:36:47:06
Unknown
But like my company successful it's not like it's not like vast successful right. It's not like a it's not like a, like a 1000 x return on top of AGI evaluation. And so for a VC that's fine because it's like, yeah, it's like whatever. Right? I have like
00:36:47:08 - 00:36:48:18
Unknown
The.
00:36:48:18 - 00:36:54:00
Unknown
But for the founders AI really sucks because now your press stack is ruined.
00:36:54:00 - 00:37:25:13
Unknown
So, like any, any sort of like exit that you can have is going to be limited. But, also, I just kind of like it sets up all the employees for not like, a good result because now they're, you know, options are at a very high price. And, yeah, I feel like this is. I feel like it's just like it's important for founders to be really responsible on this because, VCs will be very aggressive about kind of giving you money, and you might think like, oh, that means like, the fair bet.
00:37:25:14 - 00:37:39:18
Unknown
Fair market price for me is like this. So, like, I should get my fair market price, but their settings are different from yours. And so you should just kind of be careful around, picking the valuation that like you're comfortable with and like that's should be based on like your revenue growth ideally.
00:37:39:18 - 00:37:52:06
Unknown
I think it's really hard for a founder not to take the money. I mean, yeah. You're saying Jesse this thing. But at the same time, you did three rounds in a year. So, like,
00:37:52:06 - 00:37:58:06
Unknown
I don't think our valuations that high. Like, I think, with our revenue growth, it's like a reasonable valuation.
00:37:58:06 - 00:37:59:18
Unknown
We have already been fancy many
00:37:59:18 - 00:38:05:18
Unknown
times for like, subsequent rounds and we decide to hold off like for this reason because, like, I just
00:38:05:18 - 00:38:20:14
Unknown
advice normally to founders, and I think, like, it's very hard to say no to, like, a big valuation. And it's like, I think that would require, like, almost superhuman strength, especially because, like, you know, you're saying like, hey, our revenue growth is good. Like, all of these founders also believe that, like they're like, hey, we're growing fast.
00:38:20:14 - 00:38:46:02
Unknown
This is a huge market, right? Like everyone feels these things. So my no more advice is like, don't spend the money, and raise enough money. I think actually like a somewhat worst case scenario has here is like someone like raises at $1 billion valuation, but only raises like 50 million and then spends it very aggressively. Now they just don't have like, they didn't get like, you know, disciple die down at some point, especially for their space.
00:38:46:04 - 00:39:00:13
Unknown
And when the hype dies down, if you still. Yeah, if you actually in the same situation, if you'd raised 150 million and you only spent 20 million, sure, the hype dies down, but now you have 140 million in the bank. And you can, like, try to grow into the market. Right. So that's one piece of advice.
00:39:00:13 - 00:39:21:09
Unknown
And then the second one is always take the secondary that way at least. I mean, it doesn't help the employees by at least like you've, you've taken part in this like, upside in the hype and you've set up, I mean, yeah, it doesn't, like, necessarily set you up for the future. But yeah, if you really believe you can build a $10 billion company, it might still be worth taking that billion dollar rounded.
00:39:21:09 - 00:39:39:17
Unknown
Like, now you have the time to do it right. Like, I think a lot of the time it just takes time to actually manifest that like huge size. Like it's just like, yeah, in these hype cycles, everyone's like, it's going to happen in three years or whatever. But like the reality is most of the time to build a real like even to build $1 billion company takes like ten years.
00:39:39:17 - 00:39:46:06
Unknown
And to build a $10 billion copy, it only takes like 15 plus years. So, you have to, like, survive long enough to make it.
00:39:46:06 - 00:39:55:05
Unknown
what's your view on? If the valuation is jacked up to, like, if the multiples too high, that means that, like, employees, their options could be underwater for a while.
00:39:55:16 - 00:40:16:19
Unknown
It's true. It's tricky. I mean, it screws up your 49A, There's a few ways around that. I mean, in general, don't go too far. Like, if the market is like, you know, valuing you, like, if most VCs are saying, like, hey, you're worth 1 billion, like, let's put like 3 or 4 people said, and someone else comes in with a 2 billion valuation, like, don't go push it that far.
00:40:16:19 - 00:40:40:00
Unknown
Right. Like there's the like, you can go 80th percentile but don't get 95 fifth percentile. So don't go over the top of what the market's saying. Be yeah. You can reprice, options if you're like when the hype dies down, like a ton of companies did this during, post like the 2021 hype. Everyone had these crazy valuations for nine days.
00:40:40:00 - 00:41:08:04
Unknown
Were crazy. You know, 2023 came along. A ton of companies revalued the four nine, a, based on like the new situation and then like reissued options. So there are ways to like make employees whole if you know you don't live up to it afterwards. It is much more complicated. The, ideally you just grow fast enough that it's fine, but like, that's that's not the way and like, I mean, the other thing you can do and you probably have to be a more mature company to get, there's, you know, alongside a big round.
00:41:08:04 - 00:41:14:06
Unknown
Also do an employee tend to office, you know, like, ideally the founders are not the only ones getting a secondary. They're.
00:41:15:18 - 00:41:25:06
Unknown
Immad, how early should, founders or employees do a secondary, like, nowadays, a company, they're doing series A, $1 billion valuation, so. Yeah.
00:41:25:12 - 00:41:49:20
Unknown
I mean, my take on it is, it's very personal dependent. Right. Like I think, yeah, I let's take off my investor hat. I'm talking about the founder. Right? Like, you know, $1 million a young person. I don't know what Jesse's situation is, but, yeah, it's a lot of money. Right? And, if you are worth a billion in theory, you as a founder might have 100 million of, like, this paper worth, right?
00:41:49:20 - 00:42:11:02
Unknown
So, like, I think it would be very rational for you not to take some money out. Right. Just. But, I mean, you're still, like, mostly like if you if you sold 10% of your position or 5% of your position, right. That can be extremely meaningful to you personally. Investors don't really care, like, especially if it's a hot round, like they want to get in, and actually, like.
00:42:11:02 - 00:42:32:06
Unknown
Yeah, that actually, to Jesse's previous point, it actually aligns you with investors closer, right. Like, now you're thinking like, hey, I'm like successful. I've got money at this company. Now. I just want to make it the biggest thing possible. That is. Yeah, that's that's the mindset that the investor wants and I have. And yeah, because it is rational to try to build a $10 billion company.
00:42:32:06 - 00:42:50:04
Unknown
Like no sane person should do that. Right. Like if you just want money, you should just sell relatively early and become like a wealthy person, right? Like it's so much headache to build a huge company. So, like to get, like, actual alignment with the investors, I think, like, taking a secondary is reasonable. Yeah. I think if you ever get above, like, 100 million valuation, you should.
00:42:50:08 - 00:43:09:22
Unknown
And like, assuming you're not already like, wealthy, I think you should do the rational thing as a, as a founder, at least, for employees, it probably doesn't make sense at, like a 100 million valuation. Just like that, the stake is not as significant. Normally not that many people have invested that much. So maybe you do an employee secondary a little later.
00:43:10:00 - 00:43:16:06
Unknown
But yeah, I think actually like both investors and founders, are like relatively aligned and doing it relatively early.
00:43:16:17 - 00:43:18:18
Unknown
God. Yeah, that makes sense.
00:43:19:04 - 00:43:27:06
Unknown
Jesse is like just is like, oh my God, I should have done more in the last round. It's all right. Jesse would take one of those preamps.
00:43:27:06 - 00:44:00:00
Share this post