E58: Building AI Employees

Speaker A: You should just get a shure mic. It's like 150, right?

Speaker B: It's something shure, uh, SM57 or something.

Speaker C: I have a couple of them at home. I have a couple of them. It's just that I, I usually travel a lot and I do not have a setup that's mobile enough for this. So I, I haven't figured that out yet. It's very rare that I can actually take those because this, they, they help a lot when you're recording and stuff as well. Like I, I do a bit of music and it's. I have a couple of those. It's just that I never have the chance to use them properly for this, you know, so apologies ahead of time.

Speaker A: No, it's all good, it's all good.

Speaker B: Uh, on the topic of music, uh, I'm also. If I'm not working, I'm playing around with music, not making music.

Speaker A: That's a lie. Lie.

Speaker B: It is.

Speaker C: No, it's true. What do you do?

Speaker A: There's no way.

Speaker C: Is there any.

Speaker A: He just became a dad. Lie, lie, lie, lie, lie, lie.

Speaker B: Apart from that.

Speaker D: Yeah.

Speaker C: Okay, congrats, first of all. Congrats on that. Uh, there's anything that we can ever listen to. If you share them anywhere. No links or. It didn't happen. Not at all.

Speaker B: I didn't make music. No, no, no, no. I'm about to buy a Roland TD716, if you guys know what I'm saying. It's like a electric drum kit as well. Okay, very nice. Super hard to get into Peru, but, uh, yeah, ah, I found a official distributor in Peru and I'm pumped, so excited. But I'm not going to release any music. I'm just going to jam, you know? Yeah, that's nice.

Speaker C: So you played the drums. Did you play anything else?

Speaker B: Piano as well?

Speaker A: Yeah.

Speaker B: Okay. Okay.

Speaker C: Very nice. Beautiful.

Speaker A: Before he was a father, he played the drums.

Speaker C: Not anymore.

Speaker B: Drums is one that I haven't. You know. I played drums for 10 years and I got pretty good, you know, over the course of 10 years of lessons and teaching, um, myself. But then you can't travel with a drum kit. I left England when I was 18 because I started my first. Oh, no, not 18, 20, because I started my first business. So from that point on, with no more drums. Can't travel with a drum kit. So.

Speaker C: No, come on, you always have your, you always have the air drums, man.

Speaker B: Come on. Be very autistic. I'm tapping on everything. If you don't know a drummer in your life, they're incessantly tapping on things.

Speaker D: That's.

Speaker B: Yeah, that's my thing, unfortunately.

Speaker C: Yeah, you don't have that with pianists. You know, like, you don't see this around, but like with drummers it's always like that.

Speaker A: This is drummers, Drummers do this with their drummers drums with the tapping.

Speaker B: Yeah, yeah, yeah. Fortunately, I'm not, you know, air drumming. That would be like. Yeah, full blown Asperger's. Yeah. But uh,

Speaker D: Jack, I expect a, uh, I expect a recording of something with you and your baby in the next couple, uh, of weeks here.

Speaker B: But this is why I want to get the kit. I want to teach him when he's, uh, you know, old enough. Uh, I saw this video, my girlfriend sent it to me of like a one and a half year old child playing like Green Day on a drum kit. And I was like, yes, this is cool.

Speaker D: Honestly, it's probably a lot like something they could pick up a lot faster than most other instruments.

Speaker A: Yeah, percussion is the easiest for kids.

Speaker D: Um, but.

Speaker A: And then harmonica, actually, surprisingly too is.

Speaker C: It is. I, I imagine, I imagine harmonica is not the easiest for the parents at least, you know, so if they're like, if you have the kid constantly playing around.

Speaker B: Dude, I used to have like a acoustic drum kit and man, that's gotta be the hardest one for neighbors as well. We had like quite a lot of land and like, I had a shed sort of structure with my drum kit in it. And man, every neighbor, you know, they let us know whether it was a compliment or it was like a, you know, uh, you know, a suggestive compliment. It was, you know, it was very loud. They should make some time to talk about hardware, software.

Speaker D: Right?

Speaker B: This is something that, you know, I'm becoming increasingly exposed to. You know, anyone who has a Tesla is exposed to it. But I just bought an owlet. I saw that they raised over like $133 million. I was like, wow, uh, that's incredible. You know, um, there's a lot of this software, uh, out there that's um, you know, subscription based. Not subscription based. Whoop bands. Google's, um, new, you know, Fitbit bands and stuff like that. It's a whole market I've never really thought very much about, but seems to be.

Speaker D: Jack, you got it. You got to ask me for the baby recommendations. Uh, I got one. I have an outlet and another one and I like the other one better.

Speaker A: No, why do you guys do any. Wait, why are you doing any baby monitoring? You guys are. What are you doing? This is psychosis.

Speaker D: Wait, wait, you don't, you don't have like, uh, a camera with the crib or anything like that?

Speaker A: No, I use a thirty dollar security camera.

Speaker D: I mean, there's also that option. You can go. The target one is actually probably the best one that we got. We have three different ones. Each child. We got a different one.

Speaker A: Yeah, dude, I got a $30 eufy. It's like a shitty security camera.

Speaker D: That's our last one and it's great. It does exactly what we need.

Speaker A: Okay, I don't call that a baby monitor. It's a security camera that happens to double as a baby. And the reason. No, no, no. This is, this is important, Jeremiah, because the baby industry is set up to screw you. Like, everything is like 3x the price of what it should be. And it plays on like parental psychosis of fear that something will happen to their kid. And, and like, between those two things, I hate it. I hate all of it.

Speaker D: Uh, okay, but it, it is a real thing. Like the, like. Okay, Jack, did you get like the sock and stuff that out? That outlet has like. Because they've got the camera. They've got the sock.

Speaker A: Yeah, this is what I'm saying.

Speaker C: I was like, what is that? What is that?

Speaker B: It's like a baby whoop band. Yeah.

Speaker C: Baby whoop.

Speaker B: Yeah.

Speaker D: Have you tried it?

Speaker B: No, I can't set it up. Okay, so on the topic of like, baby brands want to screw, uh, you over. I think the regulation wants to screw you over. I have to buy this little dongle to spoof it into the US because if I use in any other country, they'll nerf it because of regulation.

Speaker D: Right?

Speaker B: They'll say, hey, no, you're not allowed to use the full capabilities of the device. So, uh, yeah, I'm waiting on Amazon, sending a little dongle so I can, you know, use it internationally.

Speaker A: This is the government babying you the same way you're babying your baby. Now you know how it feels. Stop it, dude.

Speaker D: I, I have, uh, no problem with people disagreeing with me on this because I think everybody's like, as a parent, everybody has a different level of risk that they're, they're open to. Right? Like, there's. Being a parent is really hard because at the end of the day, you've got like this weird competition of, uh, priorities here where you need to protect your kid and, and you also need to give them independence and the ability to learn and grow and become whoever they're going to be. And it's like those two Things are constantly in conflict with each other. Uh, and then you have this whole aspect of like the world around them might be out to get them and you don't know who, who is part of that and who is not. It's a, it's a very. It is like maybe like one of the most complex aspects of life is, is being a parent, if you end up being a parent. So the sock was one of those things. My wife and I looked at that and we were like, that is just too much information. I don't want something waking me up in the middle of the night telling me my kid is probably dying when my kid probably is not dying. Right. Like, it's just like, uh, a. Like the odds of that are so low. So you have this constant game up between of like, what is probably going to happen and what do you fear is going to happen? And those things are like, yeah, they're uh, they're at odds with each other quite a bit.

Speaker B: I think there's like distribution to everything. This for me is when he's sick so we can get the information to the doctor. I don't care about any of it. I just want the doctor to have all of the information that he needs to.

Speaker C: But is that actually medically useful information? Like.

Speaker D: Right.

Speaker B: Only if you're in the US So I have to buy this dongle, you know, so that, yeah,

Speaker A: Canada is going to say, jack, I'm not allowed to look at this because it's illegally retrieved information.

Speaker B: Right? I mean, yeah, very good point. All of the complexities. Yeah, we have, you know, fortunately we have, you know, uh, a good system there. It's just the, uh, getting the tech to work is tricky. Um, so, yeah, I mean, I say it's tricky, but I just have to buy a dongle. Like 20 bucks on it.

Speaker D: Here's my other thing with stuff like this is like the software is always terrible. Like all of these, their software is so bad.

Speaker A: This is where AI should come in. Okay.

Speaker D: Right.

Speaker A: This is like the, the craziest thing is of all of the places where we are saying like, oh, AI, like I can just vibe code it myself. Actually hard. I mean, not Teslas, but these types of things like my EUFY camera software. What instead they should do is just give me an endpoint to the effing video and say, now you do whatever the hell you want and I will build my own app to watch the videos in the way that I want to and things like this. And that is an insane opportunity, which I can't, for the Life of me understand why is not possible.

Speaker B: Yeah, Bear in mind guys, my uh, fiance is like a, um, she's incredible. So now that she's a full time mom, she has to have an outlet. So we have this app called Huckleberry. This is a couple based in San Fran. And it's. The world's like the UI is so ugly, but she tracks everything you can see. There's like every waking hour, sleeping hours, tracked sleep, uh, you know, feeds diapers.

Speaker C: But does she track them? Does she track them manually?

Speaker A: Yes.

Speaker D: Yeah, yeah.

Speaker A: Dude, this is a, this is a first baby special.

Speaker D: That's exactly. We did that with baby number one. Haven't touched it since.

Speaker B: And these guys raised, they raised 12 million I think for this app and they can't get the UI right. And I'm like, oh, I just want this in, I just want this in chat. I want it in Claude. I want to be able to just like query it and not have to, you know, um, you know, I maybe dictate as well. I just tap one button, I say, okay, this just happened. As opposed to tapping buttons that suck, you know.

Speaker D: Also, I love how, you know how much everybody's raised for all of these different things.

Speaker C: Yeah, that's true.

Speaker A: Yeah.

Speaker C: Oh man.

Speaker B: It's the first question I asked, I'm like, oh, there's an app for this. Cool. How much did they raise? Owlett was crazy, you know. 12 million for huckleberry. Yeah, I get it. 133 million for outlet. They're crushing. I don't know what you uh, know what their revenue looks like, but yeah, there are a lot of people believe in them apparently.

Speaker A: It's an insurance company, man. That's why it's so financially viable.

Speaker B: Mhm. What do you mean insurance company?

Speaker A: Why do you buy an outlet sock?

Speaker B: Yeah, good point. Yeah, yeah.

Speaker A: Ah, I mean this is, the thing is like the business models are so important to truly understand because that's how you value the company. Once you understand what the business is, then you can value it. Right. Otherwise there's nothing. I mean no one's going to value a sock.

Speaker C: Maybe, maybe.

Speaker A: Right, but.

Speaker B: Right. So Barack, Barack. Why aren't you making baby products? Why are you making baby?

Speaker C: I mean, I have, this is like you guys got me in shock. Like I'm. Apparently I'm the only one without a baby. I'm not a parent yet, so it's, I have absolutely no clue what's happening in this whole space. I haven't interacted with those software, you know, I barely have any Idea about like those kinds of problems. I haven't even thought of any of these. But I feel like at the end of the school we're going to sort of start building a baby company. Sort of like a baby products company. Software, Kind of a. Kind of a thing, all of us together, you know, um, I'm down. Don't get me wrong. Like, let's just go for it, you

Speaker B: know, it's not even just baby software. I mean, we got these cloth diapers, smart diapers or something. We got 32. It came to, like, what is a smart diaper?

Speaker C: What is it? I'm sorry, it's like freaking my ignorance.

Speaker D: What is this?

Speaker C: What is a smart diaper, man?

Speaker B: Oh, I thought I was like high tech diapers. No, it's like, um, it's like a cloth diaper, a washable diaper. Apparently, according to my, my fiance, it has no tech. There's no tech at all. But like, they cost like, uh, 30 bucks a pop. Okay,

Speaker C: then why are they called smart, though? Like, there must be something that we're

Speaker D: not getting 100 years ago.

Speaker B: Exactly.

Speaker C: So it turns out that was smart. That was smart.

Speaker A: Barack, how big is the company now? I guess maybe you should just share what it does, how big it is right now. Yeah, just do a quick lay of the land and then we just sort of like talk about what it's like to build a software company.

Speaker C: So we're building an AI data team. Uh, it's basically a bunch of AI agents that have a very deep understanding of your company data and, uh, can answer any questions through Slack, through WhatsApp, through um, Discord teams or our web UI. And uh, essentially reduces the time to insight. We're a small company where 15 people distribute team across Europe and yeah, been building this for two and a half years.

Speaker B: Incredible. Yeah, I was really excited to talk to you because we just spoke to, um, Jim Benson from Adapt, and we also spoke to Jacob Postal from hq. And these products, I'm sure they're more different than they are similar. But on face value, to somebody who might be buying, it can seem very, very similar. So like, I wonder, like, it clearly is a good space to be building in if so many serious players are building in it. But like, what makes you guys different? What is it that you guys are trying to really solve? When I go to the website, I can get an impression of the, uh, problem that you're trying to solve. But yeah, how would you stand out against competitors? And I should say that when I say competitors, I mean just like Somebody raw dogging it in Claude, setting up their own flat bot, getting Claude connected to everything and then having that in their channel. Uh, for Jim Benson at Adapt, it was like they're a supercomputer, they're an AI supercomputer in the background. But yeah, I'd love to hear what you guys are. How are you seeing yourself as less replaceable?

Speaker D: Right.

Speaker C: A lot of the products out there essentially focus on the idea of like, okay, give me a file that I'm going to analyze this and give the results back to you. It's like, I don't know, you throw me an Excel file, a PDF and you get to data analysis. Um, that's not really how actual data analysis work. When you're, uh, operating above a certain scale where your data doesn't actually fit into an Excel sheet and most of the time the data is not actually coming from a single source. It's not like just your. You just have that data on Shopify. You actually have a lot of questions that span a lot of different data sources. Like order data comes from Shopify, you have data that's coming in from, I don't know, Stripe, and then you have your marketing data across six different platforms. So the reality of answering data for real life business questions is a lot less about the AI part of things. It's a lot more about the underlying infrastructure. Otherwise, it's like you just hire this very confident intern that can just throw out answers to any question that you ask them. Um, but the underlying work, uh, is what makes this a lot more successful than just having Claude Rodock in your Shopify store. Um, so essentially we have started our life as building the whole data integration layer. We were building that before we were building anything around AI or agents. And over time it became a very natural fit. So that the reason that it works very well today is because we actually understand the full journey of that data all the way from hundreds of different sources that we connect down to all the dashboards, all the machine learning models, all the reverse ETL usage that you do, where you push this data back into different platforms. And at that point when you ask a question, AI actually has a very good understanding about all the types of data that you have available, all the metadata, all of their relationships, et cetera. This means that we're essentially talking about like hundreds, if not thousands of tables, tens if not hundreds of different sources of data for any sizable business. And, uh, you don't get to solve that problem without a proper data infrastructure. At that point, our competition becomes Not Claude. It becomes essentially While you hire 10 people and you try to build this internally versus while here's a ready made product built for this. So that's the, that's the argument.

Speaker D: Are you, are you using anthropic or OpenAI to power?

Speaker C: We're using both. We're using both, yeah.

Speaker D: How has that impacted uh, the building of this over two and a half years? Because the capabilities two and a half years ago were so different than they are today. Like I, I think one of my big challenges with AI is it's like I feel like as soon as I build something on top of the infrastructure that exists, the infrastructure changes. Um, usually in a beneficial way, uh, but maybe not always. So I'm curious how you've had to deal with it.

Speaker C: Um, so we, we struggled a lot with that initially when we were building the. So I'll take a step back around a year or so ago. Anyone that has tried this product has essentially failed because the models were not that good. We did not have as good of an understanding of the problem as we do now. And um, it just, the ROI was not really there, you know. So I think the first hurdle that we had to overcome in terms of like these models and their evolution was we had to first become believers that this will actually work. You know, because we had to, we had to overcome this, this feeling that like man, we tried this a month ago, we tried this two months ago, we tried this three months ago and it doesn't work. It's not as good, you know. And then like we had to go around this and try it again and again and again. And eventually we found that while it works beautifully, um, on the dependency side of things, we have initially been very cloth pilled I would say. I think there was a time where Anthropics, uh, models were a lot better than OpenAI. Although over the past three months that has changed very significantly for us. On production we're running an ensemble, so we're running um, a harness that has access to both OpenAI's models and in anthropic primarily we're running um, GPT 5.5 for um, the bulk of the work and then we have OPUS to help with some other parts. Um, but a lot of the dependency is now on the OpenAI side. But because we had to switch things so often and keep benchmarking things so often, we had to build a lot of layers of abstractions around the model providers and the things they supply to make this as easy for us to adapt as possible. Uh, and Right now I can confidently say we're not dependent on a particular model. We are generally dependent on the high level paradigm of large language models. And we keep experimenting on the ideas on the basis of what would we want to build if these things were 10 times better, 100 times better, and then try to build for that and hope that the models are going to catch up to that which is sort of, it's already happening. Like things we thought that wouldn't be possible 6 months ago are becoming possible already. So some of the things that we tried that sounded very stupid at the time are becoming quite obvious and uh, quite reasonable to do today.

Speaker A: There's a few things on your side. I mean the product and platform makes sense to me and you have some comparisons with like FiveTran, Error Byte, Things like this. We, um, actually recently turned off of fivetran and built something uh, on our own. Interestingly, I'm curious, do you think about your go to market from like a, uh. Hey, all we need to do is go and attack 5 Tran customers whose bills are getting exorbitant. Because like, by the way, that was like me nine months ago.

Speaker C: Hm. Right, Bill.

Speaker A: They switched their billing model and I was like, I know this is insanity.

Speaker C: It is insanity. The whole space is insanity. Like data platform, data infrastructure. This industry is mad. It's just pure madness. That has been like that for many years. But we have tried that, we have played with that. It worked to a certain extent. But that's not really a core part of our go to market simply because we already open sourced all of this. All of our connectors are open source. You could today replace everything you do with fivetran with our open source tooling without paying us a dime. You know, like you just, you already have all that ability.

Speaker A: But you and I both know that's not the problem. The reason why people don't churn off of fivetran is because it's two to three months of work of like two of your best engineers to do the changeovers. So I guess what I'm actually asking is why not have that be your go to market is like I'm going to supply you the two engineers to do the migration.

Speaker C: You could do that. But I would much rather. I don't think so. On a high level, there's companies that would try to cut costs and at their size and scale the costs would be so significant that they would actually care buying and paying someone else to reduce that cost. Uh, and on the other hand there's companies that, that have the Money that have the budget, that are more interested in the upside of new value that you can bring into them. Um, we are trying to focus on the latter because for the companies that we tried this with, where they want to reduce the costs, essentially this becomes a bidding war with regards to who can do all this work with much less. If you do the servicing part of this, you essentially become a hands on consultancy business for which I don't want to build that. You know, that's not what gets me up uh, from the bed every morning. I'm not excited about that idea. So. And when we think about like okay, could we just put out a product that does already that. Yes, but if you are not willing to migrate, either you don't care about this as much as we would think you would or the ROI is not there for you. And if that's not there for you, there's probably not much for us to earn from there as well. Um, on the other side I can actually eliminate the work of like 10, 15, 20 people team and help you avoid hiring another 1020 over the, over the course of a year. That becomes a much, much better value proposition for me to go for than trying to undercut 5tr. Like if you're paying 5tr 10k a month, you know how much am I going to make off of you? I'm going to reduce your bill to 1K and they charge you.

Speaker A: But the problem is that there's, I mean we were paying them, we, I mean we were going to have to pay them 80k a month. So I mean like I, this, my, my, my hypothesis is, is more that, and, and this is ah, a question I, not my business. Right. So I have no idea. I find it hard to believe that you're not going to end up in a world where there are a enormous number of companies with an enormous amount of data. Like look, the number one thing about AI is we now have more data, not less data. Not and not by like 20% by like a hundred X. I am sure

Speaker D: by the way this is why Fivetran changed their pricing. They felt like they had leverage in this context.

Speaker A: Exactly. And so now that I have a hundred x the data that I need to move around and deal with like, I mean like truly insane scale for a company our size. I've got to believe that the number of people for whom it is worth it. And then, and then to your point on like I don't want to build a services business. I mean right. Like every other podcast on tech in at least in the Us is like every business has to become a service business for the next two years. And so like, I separately am curious why you believe the opposite, because I just don't believe that there is such a thing as a, as a software company that cannot provide services that gets very big. I literally don't believe it.

Speaker C: Think there can be software companies that won't, um, provide services, but the decision for me is more like we can and we will provide services. What are the types of services that we want to provide and what is the upside? What I need to optimize for today is every service work that I do, I am doing it not to make money. I'm doing it to cover some of my costs. But more importantly, I need to learn something out of that work that I do. You know, So I have two paths

Speaker D: in front of me.

Speaker C: I could either do some services to help people move off of, um, 5chan and make some money on that process, or I can now help organizations adopt agents for their analytical workloads and I do some, some servicing work. I'm always going to prefer that ladder because there's so much more to explore, so much more to learn there that would be a lot more applicable to the, to the rest of the world than moving people off of 5chan. You know, eventually I imagine we will do that. You know, like, this is already happening on its own. We already have like this inbo flow with regards to like every time fivetran bumps their prices, like we get, we get inbound leads and stuff. But that's not where I want to put services, effort. You know, I don't, I don't find that interesting. Like we already take care of that. That's already a done market. In my perspective, this is a lot more exciting where if I am going to do services, I will be personally involved with all of those conversations as well. And I need to make sure me, my co founder, uh, my organization learned something out of that, you know, so it's not that I don't want to do the service. You know, we do. We get our hands very dirty in a lot of these cases. It's just that, you know, cutting the costs is, is not where I want to put that time and effort.

Speaker D: One thing that I think is really interesting about your business, I don't know much, but the, uh, the core of this is open source. I think is really interesting because I there, I mean, for a long time open source drove so much, and now with AI, it's really changing that landscape. And my, my actual, my first start in software actually was building, uh, open source software as well. Um, the WordPress ecosystem, uh, a little bit different, not as technical, but the point is like, we would put everything out for free and basically we had like, we made our money by effectively selling services, uh, in the form of a better package and better customer service. And so I'm, I'm curious how you're thinking about open source in this context. Like, what does it do for your business today and what do you think the theme nature of that is, given where you're at today?

Speaker C: So we initially started building everything open source for a couple of reasons. First of all, we are selling core data infrastructure. Like companies that we work with, um, build a lot of stuff on us, they depend on us a lot. Uh, and we wanted to make sure that them trusting a small company like ours in the early days would not put them in a very hard position. If he were to ever run out of business, if he were to ever have any trouble, they would have all of the software available for them. They could, at worst, like if all of us got in a bus and we, we died, they could pay someone a couple thousand bucks to get everything up and running in a day or two, you know.

Speaker D: So this gives a big selling point early on.

Speaker C: Absolutely. It still is. Like we're, we're still, like, we're still small. Um, we're in a much better position than like two years ago. But it's still, it gives people a lot of peace of mind that all of the core tech, everything they depend on, is part of our open source tooling. And we try to do this in a sustainable way where intentionally, all of our core product design starts from the open source and then flows through our cloud offering as well. So that means that it is incredibly rare. Other than like some enterprise governance weird features that we built into our cloud product, pretty much everything goes through our open source setup, you know, so my team contributes. If you take a look at our repos, there's tens, if not hundreds of deployments that we do every day to our, uh, open source software and then, and then take it to the cloud. I think with AI, this is becoming an even better advantage for us because part of the reason why we were building this open source was also, okay, people can contribute and we want to cover a lot of different touch points, a lot of different integration points. We're going to make mistakes, we're going to fall behind, we're going to, you know, there's going to be APIs that we don't know how they Work people can always contribute. With AI, these contributions are becoming much easier. You know, people can just jump in and submit a PR and uh, get something deployed in about a day. That both solves their problem. And now we actually have a new integration point that someone else just committed for everyone else to use and benefit from. You know, AI reduced that barrier quite a bit. It is adding a bit of noise, to be honest, on the maintenance side. Um, but, but regardless, it's definitely a net win for us. I just enjoy putting all of this out in the open. Part of that is also emotional. I feel like we're giving back to the community that we have also benefited from.

Speaker B: I want you to tell me why I'm wrong. Um, and uh, Rashad, if I mischaracterize femaat, uh, and this, please tell me why I'm wrong as well. But uh, I really like this focus on solving time to insight. Um, trying to allow operators to be informed and make decisions, move time to action a little bit faster, but more informed as well. Um, I think in a lot of the spaces that I've been in around incredible people, they're overconfident and that's incredible because they just move forward. I used to call it Raging Bull. All of the most incredible people I knew would make these decisions based on two data points and an and condition and they'd be like, yes, let's go. Right, um, and they didn't actually need, or may maybe it was this aggregate gut, ah, instinct of a lot of incredible information or it was just enough to be persuasive for them to take action. Um, but actually getting really good data into those people's hands and letting them feel even more confident that they haven't had to scribble things together in convoluted spreadsheets or trust someone else for that information. Uh, so I think it's great that you're solving that for those people, those operators. But I think there's like a, there is an end, right, to getting good information. You want it to inform good decision making, good action, right? Are there any plans for you to move from time to insight to time to action and actually empower the actions? Because if I look at Vermont, right, FAMAT is interesting. I don't know too, too much. I keep seeing FAMAT in all of these really big stores and I'm like, ah, oh, cool, what they're using Vermont. That's nice. But it's a lot of action, right? It's not actually a lot of like trying to get information into these People's hands. Maybe it is. This is why, correct me if I'm wrong, uh, Rashab, but a lot of it is actually like let's take the best information and let's act on your behalf to achieve like a business outcome.

Speaker D: Right.

Speaker B: And I imagine this is, you know, maybe no falls in the first camp, right. Time to insight, get that information that lets you act. Whereas Verma maybe falls in that second camp. Like get the information but then actually act with it.

Speaker D: Right.

Speaker B: And solve that business case. So yeah, what are the plans for the product? Are you ever going to move into time to action?

Speaker C: I mean I don't think you're wrong at all. I think um, we already moved into that in the sense that I consider us as like a decision making business. You know, in the end we're trying to make, help people make better decisions faster, um, and cheaper. You know, and ideally we're going to help them do that by eliminating some of those decisions and automating quite a lot of them. So using our agents today, we have quite a few customers that are automating their marketing campaigns, customers that are automating their CRM, customers that are operating uh, up running a B tests on their play store listings or on their landing pages. We even use it for our own internal landing pages and marketing material that constantly updates itself. So we already are there in terms of taking those actions. Um, however, this is not like a, this is not a problem that a company like ours can solve it by itself because a lot of these actions are dependent on other products, other companies, other people to be able to actually take those actions. So until that is there it's going to be very hard to um, automate all of these. But my sort of internal feeling about how we're going to talk about.

Speaker A: Yeah, I'm sorry to cut you off but um, as you're saying this, does that mean at some point you're gonna start competing with like a High Touch?

Speaker C: Basically we already are. Like it's not just like, it's not even High touch, kind of like reverse. We already do that. We've been doing that for three years.

Speaker A: Um, the way that I evolved, right, Like High Touch doesn't do reverse ETL anymore. High Touch is squarely now in like agent data pipelines to drive marketing. So they have chosen to narrow their area that they're going to operate in to marketing actions. And, and so like actually what their focus now is is like uh, like driving systems of action where uh, like they don't have the system of like they don't have the email provider, they don't have the SMS provider, they don't have the ad provider. But the marketer can log into their system and then say hey, understand this data and make act, take action in this way. So that way they can drive. In particular, when you're doing it at that scale, you can drive for example much better personalization or much better nuanced rules for when to drive a campaign and when not to. But if that's what you're saying, then yeah, like the jack is right, like the opportunity size just becomes a bunch of bigger because the set of opportunities for companies like High Touch and again High Touch happens to be focused on marketing, um, is pretty large. Right. There's like another competitor, ah, called Oxia. There's a bunch of them.

Speaker C: Yes. So we are, we're moving in that direction. Um, I don't think in the short term we're going to be, we're going to be building any industry specific sort of touch points. It's likely going to be a lot more like um, we are building that platform for people to build these sorts of automations on top of our tech. Which means that we do a lot of work around um, making our agents use both the data warehouse as well as the browser very well so that customers can actually build browser based automations on top of what they want to do based on the data that they have in their warehouse and then getting these agents to take these actions. I do not see us building tooling to make um, marketing people get better insights from um, their marketing campaigns. To be quite frank, I think we're moving towards a future where um, you don't have marketing people or you have 99% less, you need a lot less of these people because the people that you have can manage a lot more today or you don't even need them because you can manage all of these things because the number of decisions that you need to make is 90% less now. So yes, we are moving in a, there is a universe where we are competing with High Touch or kind of like marketing automation tools in one place, CRM automation platforms in another, um, data integration tools with, with another. And it might seem like where we're taking fights wars with everyone, with everywhere. I don't find that scary. I think it's a, it's where things are evolving and no, I don't think

Speaker A: it's, yeah, I don't, I don't think it's scary. I just think you need to convince a few friends to build some service businesses.

Speaker C: I guess. You know, you seem to be very into building services. Why don't you help us out? Well, let's build it together.

Speaker D: I am over subscribed.

Speaker A: I'm, I'm oversubscribed, man. I'm like, I, I, here's, here's what I'll say. Uh, Jeremiah knows this about me. I, I think, Sorry, I should have asked this much earlier. Is, is your focus today SMB or enterprise?

Speaker C: SMB. Okay. We have like quite a few enterprise customers but again like we're trying to optimize the learnings. I need people, I need companies that can move faster. So it's like companies that are smaller than 200, 250 people is sort of our uh, our target.

Speaker A: Yeah, yep, yep, I got it, I got it. Yeah, I think so. I mean, so first of all, as long as you're focused on SMB and like you're driving the platform capabilities, everything, everything that I said up until now you can like entirely ignore because obviously the business model does not support it. Just to clarify, right, if and when you decide to go to enterprise, the thing that is I am um, 100% sure underappreciated is enterprises. The way that, the way you need to like chat with the CXO of an enterprise, like these are the most sophisticated people on the planet. Like you don't get to be the CDO or CMO of a multi billion dollar company unless you're really good at what you do. And uh, these people are very sophisticated and they, when you say to them I'm going to help you move your organization forward in this tangible way. They already also have a hundred person engineering team. And so what actually becomes the problem is given the level of sophistication with which I run my business, if you don't come here and spend three months working alongside my team, the probability of you actually moving the needle for my business devolves to zero. That's actually what ends up happening. And they're correct by the way. And the consequence of that, at least in the US is there's this massive dearth of uh, what is now being called Forward Deployed Engineers. Services, right. Formerly known as services. And it's so massive that everyone is asking anyone who is competent with AI to please come and sit next to their team to enable something that will actually move the business forward. It is really, it is really hard. Like um, I can't understate how hard it is. And, and I think that underneath really hard is massive effing opportunity. Just like massive opportunity because the Amount that it's worth to them is tens, if not hundreds of millions of dollars. And therefore, the amount that it is worth to vendors like us is millions of dollars. Right. And so it's just that I think that the enterprise opportunity is 100% understated. I mean, like, it is almost comical. Did you read the SpaceX S1? I'll just, like, shift topics a little bit to make it a little fun. Did you guys read the SpaceX S1M?

Speaker C: I haven't.

Speaker A: There's this chart in the SpaceX S1 which shows the total addressable market for all of their different markets. And it's like half a trillion launch business, 2 trillion of, like, Starlink, and then it's like 23 trillion enterprise AI. Okay, so all of their previous businesses look like little, like, dots. And then it's like enterprise AI is like 90% of their future TAM or 95% of their future TAM. I mean, there we go. Look at that beauty. Look at that beauty.

Speaker B: But, I mean, this makes sense.

Speaker D: I mean, if you think about it, we're talking about this technology is disrupting the way that we work. And the way that we work was built over decades. Especially when you talk about, like, a. I don't know, a Fortune 500 company that's been around for 50 years. There's 50 years of doing things in a certain way and building on top of this foundation that now has to be undone. That is not easy. That's going to be extremely expensive, extremely disruptive, and it will take a lot of time and money to make those changes. This is not a, uh. Like, it's a very obvious reason to me why this is such a massive opportunity.

Speaker C: Is there anyone that says this is not a massive opportunity?

Speaker D: Uh, also, I'm not trying to convince you, by the way. I'm just.

Speaker A: No, no, no, no, no, no, no. I'm just wondering. I'm just wondering. Today you're learning by offering two SMBs, and we're all agreeing. We're doing the podcasting, we're saying the same thing, and we're all agreeing, which is wonderful. So when is it that you will switch to offering to the enterprise? What's the target?

Speaker C: The services part? Like us offering a product to enterprise versus us offering the services part to enterprise is. I find it very different. I would probably say, like, at least a good couple more years for us to ever get into that. Um, it's not because I don't think there is an opportunity. It's just because I. That opportunity does not excite Me, you know, there might be like massive opportunities there. There's massive opportunities everywhere. I don't care. You know, I think there is a. We're operating in a. I would much rather be stripe than being McKinsey. You know, I would much rather bet that some of the SMBs, some of the smaller companies, some of the future looking companies that we're working today with, are going to become the enterprises of tomor and then try to build for them from solid foundations. I find that idea much more exciting than going in at a Fortune 100 company and trying to like change the way completely. They work 100%. There's a huge opportunity.

Speaker B: Huge, Huge.

Speaker C: Huge. Like no disagreement at all. I just don't find it exciting. Um, so I guess, yeah, what's your

Speaker D: goal for your business, for yourself over the next five, 10 years? Because that I think, I know we get into these conversations a lot on this podcast, but that's, that definitely helps guide where you end up. Because I'm curious what, you know, what your, what your ideal outcome is personally.

Speaker C: So I have this, I have this internal sort of um, thinking that the time to insight or time to action, like Jack put it, is basically supposed to approximate to zero over time. So if we're successful, the companies that end up working with us, the companies that we've been helping along the way are going to be the ones that are going to be making the most amount of uh, decisions and they're going to be the ones that can iterate the fastest. So the goal is you take a look at the Fortune 500 five years down the line and half of them are less than 10 years old and half of those are our customers. Maybe there's a future that, that will happen and I would like to be on that part of powering those kinds of organizations where they're actually much smaller in headcount, much bigger in operations, much bigger efficiency gains and we're challenged in that regard, sort of making that a reality. I do not see us winning over business from Oracle. I do not see us going behind companies that Microsoft holds very close and very dear large enterprises. So that will be my dream scenario. That will be the goal that we're building towards. To have um, a few years down the line. That's what I would like to see. That's what we're building towards.

Speaker B: How do you guys rank order information? Right? Uh, I'm sure you've all seen it and everyone's begging for your attention. Rashab, Jeremiah, at your companies, um, how do you take information that's coming from your team or just very true information and decide what order it should come at you. What I mean is that I've seen decisions made before based on accurate data and sometimes enormous amounts of data, and they have the exact opposite conclusion about the action that should be taken.

Speaker D: Right?

Speaker B: This is true and this is true. Right? But both of them are arriving at different conclusions, right? I hate to say that maybe it comes down to opinion at the end of the day, what excites you or what is, uh, the most persuasive to the decision maker at that moment of time and links to the most amount of information that they have when they read it. But, uh, yeah, I'd love to hear how you guys think about that. I see this all the time. People making decisions based on some amount of limited data or a lot of data, right? They're taking action, but, uh, how are they rank ordering it? I guess it's the same question as how do you know which question to ask?

Speaker D: Right?

Speaker B: Because you might be asking the entirely wrong question to inform like a decision.

Speaker D: Right?

Speaker A: Yeah. I have this funny thing that I say inside the company, which is either it matters or it doesn't matter. And it's like, which sounds obvious, but, man, people don't think that way. Like, people come to me with like a, uh, oh, like, they said this, this. And I'm like, doesn't matter to them. And they're like, oh, this. And I'm like, no, no, no, no. Can you answer if it matters or if it doesn't matter? Because we only matter if it matters. Like, it only matters if it matters. And, and like, the consistency with which you have to say that is astonishing. Like, I'll give you a good example, okay? Let's just say someone inside your company, like, Jeremiah, I'm sure you deal with this. Someone is complaining about someone else is work output, okay? It's like person A complains about person B's work output. And I'm like, does it matter? Okay, this is my number one question every time I this fucking happens, okay? I'm, um, like, does it matter? And they're like, you know, it should be this, this, this. And I'm like, it can be whatever the hell you want. Let's pretend person B's work output was zero. Has your life changed? Okay, now let's pretend person B's work output was fucking exceptional. The best on the planet. Has your life changed? Your life doesn't fucking change. So stop complaining like either it matters or it doesn't matter. Like, and, and this is like a crazy thing that people don't understand. Like, people don't understand that companies have to do a small number of things to move the needle and therefore it matters or it does not matter. And just look at that list and tell me, is this action that you are taking an action that matters? And if it's not, then just please leave me alone. That is it.

Speaker B: Yeah. I think there was this warm fuzzy feeling that everyone got the first time AI told them they were stupid and that they were wrong and that maybe they should be paying attention to something else. And people do this on purpose, right? They finish their prompts with tell me why I'm wrong or whatever it happens to be. Or they, you know, um, set up their skills to constantly tear them to pieces. But, um, yeah, you know, I wonder to what extent you have to be opinionated. Right? Like, I imagine that you come into contact with this a lot. Jeremiah, with no right. If you leave it to the operator to decide which questions they want to ask, sometimes they're going to go through a sequence of different things that don't matter right. Before they arrive at a thing that you already know because you have all of the data, that is probably, probably a good place to start.

Speaker D: Right?

Speaker B: Um, in terms of a question on a post purchase survey, that's actually going to inform good decision making. But your customers, I imagine, Barack, and I imagine this applies to famaat as well. You have to use what matters and you have to be opinionated about what matters and what doesn't matter because there's swathes of data that can inform, uh, decision making. But Barack, if the operator is in charge of the question that they ask, how do we make sure that they get the, the right thing? Right? Uh, is that the model coming back to them and saying, actually, no, don't pay attention to that. You don't need to know about that. Pay attention to this.

Speaker D: Right?

Speaker C: Shab was gonna say something about this before I jump into it because he

Speaker A: got like, no, no, no, I was saying. Yes, I was saying, Jack, you fucking got it. That's all I was saying. I was like, this is a great fucking question. No, no, no, I'm in, I'm in. Brock, I want to hear it.

Speaker C: Yeah, I think, I think the way I think about this is there's types of organizations that. So there's two types of pilots that we do.

Speaker D: Okay.

Speaker C: And I was not expecting this to happen because it became much more apparent with the whole AI stuff that we're doing. And I find this funny. So, um, in one set of pilot, we set things up for the organization. We give them two weeks to use the product for free. They can explore every aspect of our agent, our integrations, pull data, push data, they can do whatever they want and they going to hammer the product. You know they're going to like bankrupt us. They, they do like they throw a lot of people on the product, a lot of questions, a lot of, they're very demanding, constantly sharing feedback, et cetera. And then there's this other group where um, they, they sometimes like they use the product moderate amount but most of the time it's for very superficial stuff that they want to ask, they want to learn and, and we've tried to make them better at that. You know like we're sort of like is this really what you should care about? Like shouldn't you ask this kind of stuff? Like you're like an E commerce company. There's like these and these and are you tracking these metrics? Do you know what these mean, et cetera? A lot of them do not know what they want to ask. You know a lot of them do not have an idea about what matters for their, what matters for their business. I think there is nothing rational behind this, you know on a grander scale I think it's a lot of emotions and gut feeling about what they think matters uh, versus what it actually matters. So at this point I'm of the opinion that if a company does not have a very clear idea about tons of stuff they would throw at if they had like unlimited amount of question answering machines, there's probably not much of a chance that we could actually save them from, from the state that they are in, you know.

Speaker D: So I think a lot of that's pain driven ultimately. So if somebody has some sort of like because I have the same exact problem in our business and it is so hard, really easy to get somebody to do one thing, it's really hard to get them to do a uh, second thing, third thing, fourth thing, unless you can address a specific pain point that they have and oftentimes like for me I only get to interact with a small portion of what somebody uh, like whatever my job is, if I'm coming in and I'm using this piece of software, it's only addressing a portion of my job. So I don't know what's going on with the other. I uh, don't know 80% of the things that that person is dealing with. And so for me to push something to them, I know that I need them to do this thing, I know it will be better for them, but to actually get them to engage with it is extremely hard because they have all of these other pains that they're dealing with. Uh, or maybe not always pains you. It could be things they're excited about. They're just not actively thinking about the problem I'm solving for them. And if they're not thinking about it, it's hard. As soon as they're thinking about it and they're engaging with it, it's extremely easy to get them to do additional things.

Speaker B: Maybe this relates back to, um. You guys see the new Codex release, like goal, slash, goal as, like a, as a command, right? And you can set a goal. Maybe this comes up to, like, decision makers and founders and operators, like, setting, you know, deciding what should be valued and then orienting around that. Uh, I imagine all of us have a friend who pays too much attention to politics and talks about it too much. And you're like, oh, stop.

Speaker D: Right.

Speaker B: I saw, uh, my girlfriend was watching some South Korean cooking show the other day and she showed it to me. She was like, oh, look at how incredible they are. And I said, how weird would it be if they just started talking about Trump and geopolitics? They started having an opinion. They're focused on something. They have a goal. Make great food or whatever it happens to be within the time constraints. So, yeah, maybe that's, maybe that's just the role of the founder, right? And the, the operator to set the direction and what the goal actually is and how people are orienting themselves.

Speaker D: Right.

Speaker A: I'm going to say something that's going to sound so bad, especially if my tier team hears this, but life should be simple, okay? There's a very small number of things that matter to a business. Increase revenue and reduce cost. You should know what your business formula is to increase revenue and to reduce cost, because it's usually pretty simple. If you're not working on those things that matter to how you're doing that or to m. That matter to how you can get there, then just stop talking. The problem is that people are good at certain things and they want the things that they are good at to matter because it makes them feel good. Now, I don't care how people feel. I care about growing my business. And therein lies the kerfuffle. Right? And, and, and that's the, like, sad reality here is, like, people want to feel progress and sometimes they're feeling progress on something that has no benefit to the end business. And, and, and that's like the really tough thing to nail and to nail it in a way in which does not demoralize your team so much that they don't even do the work. But instead you need to get them um, to recenter on the things that do matter. And doing that in a way in which is like. No, no, no, listen, like Rishabh, you're not a dummy. It's just that you're using your strengths in the wrong way. I understand that you're tempted to use it in this way, but just stop it because it doesn't actually help you. And that is the central problem. Right? Yeah. And so somehow you gotta do that.

Speaker C: Yeah, yeah.

Speaker D: I mean we had, when I took over stamps, like one of the things that was a problem there was just uh, optimizing for the wrong thing ultimately. And so like the, the technology there, the team had like the, the biggest signal of this to me was that uh, in three years the business had not grown. The actual, like actual business outcomes that we were driving hadn't changed. But the uh, we were doing 10 times as much work with the information that was happening there, uh, as it had been three years before, which is 10 times as much cost work. All these things happening to deliver the same outcome, like that means that you have optimized for the wrong thing. The optimization, the optimization was towards showing bigger numbers, not towards driving an actually different outcome. And like that's the, that's the problem ultimately in that context.

Speaker C: But how do you, how do you, okay, you guys have more experience than this than, than I do. How do you guys do this without you actually being in every room where every little decision is being made?

Speaker D: No, no, no, no.

Speaker A: Dude, this is, you want to know the craziest unlock that I had? So there are two unlocks that I had recently that are potentially interesting. And then I'm sorry, I feel like I have to hop every single time. Like at the end we gotta find a way to like, like get ourselves on time. But the, the. There are two things I did. One is I made an agent that is um, like knows what the priorities of the business are. And every time someone does something it will like chime in and say like whether it matters to the business or not. So now what's happening is I'm um, depersonalizing the feedback because now what the person feels like is that a bot is giving them feedback and they're actually much more open to getting feedback, which is a very weird phenomenon. They were much more averse to asking me for input when I would tell the person to their face, hey, this doesn't matter. But when my bot says this doesn't matter, it doesn't hurt their feelings, which is a fascinating dynamic and has been a, uh, massive unlock. So that's one thing that I did. The other thing that I did is I have standups. I don't have weekly meetings. I have standups every other day. And I tell people that their actions and their behavior is well motivated, but that it's incorrectly placed. And I think that people just need to hear that their actions are well motivated. And if you say those words out loud, then they feel validated enough that you can actually then say, but it's wrong. Like, I understand and it's wrong, right? And so just apply that same thing to this other direction and they're more, much more willing to hear it. But those are the two things that I do is, uh, because I do it every other day, it doesn't feel like I was, like, sitting on it. People feel like it's real time. And when people feel like it's real time, again, it becomes less personal, um, because it becomes about the action, not about the person.

Speaker C: But does this mean that. Does this mean that you're sort of like, with that use of, like, agent, you're essentially mimicking, like, you being in every room. Is there a way that this could become the company's part of the company culture, where there's some percentage of the company that sort of internalize all of these, uh, what actually matters for this business kind of thinking and every part of the company without you actually being the one that drives all of that? Of course, you are the strongest pillar to culture. But could it work without you being there or without agent being there?

Speaker A: 100. Because now what's happening is people give feedback back to my agent when they think it's wrong. And so what's actually happening is my agent is not a reflection of just me. It's a reflection of the best parts of everybody. But it is founded on what are the things that actually matter for the business.

Speaker B: So.

Speaker A: So, because. Because, see, the thing is about training an agent is you can write anything to the memory file. So I'll see, like, an idea from someone else and I'll say, like, hey, and then I'll like, tag my agent, command that to memory. That's a good idea. And so now it actually is not literally just me in every room. And so it is actually something that can permeate through the company. Does that, does that make sense?

Speaker C: So you rely on agent being sort of like, this collective, um, set of learnings and, like, the bearer of all these learnings into. Into the rest of the organization, which

Speaker A: is what every company is going to look like.

Speaker B: Yeah. It's an interesting, uh, version of Time to insight.

Speaker D: Right.

Speaker B: It's an interesting version of get the data. Whereas to get the data, you don't get back on goal.

Speaker D: Right.

Speaker B: Get back on target. Yeah.

Speaker D: Yeah.

Speaker B: Cool.

Speaker D: Yeah.

Speaker C: This is Jeremiah.

Speaker B: Uh, Rashab. I know you've got to go, so, uh, let's wrap this up. Barack, this was awesome. It was great to meet you.

Speaker D: Uh, thank you.

Speaker C: Thanks a lot for having me. It's really appreciated.

Speaker B: Thank you. Talk soon. Cheers.