Ep #148: Why DIY AI Solutions Might Be Riskier Than You Think with Brad Micklea (CEO & Founder, Jozu)

[00:00:00] Adil Saleh: Hey, greetings everybody, everybody in the us Happy Independence. And we had this special episode with, with, with a special guest. And I was trying to explore a lot of a lot of products that are making a market is so hard. These are by the way, to, you know, cherry pick these platform that are, that call themselves like AI, power, Gentech frameworks like the building agents and all. So we've had more than 30 in the past, I would say 16 months that are like calling them agents co. Also, this time around we want to dig in deep, like how does agent workflows actually operate at an enterprise at scale? Because a lot of people are thinking about. You know, it's just a wrapper of chat, GPT or any NLM that they can use and they can serve as startup or an SMB or they can do the same with the, with the enterprise. But today we are going to explore that what kind of nuances that take place when you take this product at scale. Your ML models, your AI models, and you know, how, how different problems that you guys can run into and how big are. And it is for, although a lot of these folks, they have RA raised good, healthy seed round, even pre-seed round and, and they're happy spending more, but I think this episode will give them a real good insight into where to spend towards you know, in, in terms of resources, effort, direction, vision, all of that. So today we had Brad, who's the CEO and father founder of Jozu. That is there's a platform enabling teams, enterprise teams to manage their AI ML models and workflows at scale. Thank you very much Brad, for taking the time. [00:01:28] Brad Micklea: Thank you, IL. Yeah, looking forward to it. [00:01:31] Adil Saleh: Perfect. So I know that you have, you have this technical background all these years. You know, you've been graduated late nineties from McGill and you know, then you have you, you insert into corporate and you know, a lot, a lot of that you have seen you know, compared to someone like me and a lot of these listeners. So how do you see it in terms of. Not just the ai, I don't want to put in like something very general in terms of you know, these capabilities that AI is developing really, really fast into the N ml NLP into into the agent workflows you know, compared to your industry or your niche that you're targeting or otherwise you know, how do you see it like panning out for you as a, as a founder? [00:02:13] Brad Micklea: Yeah. It's interesting. I mean, I, I had been part of startups early in my career. The last startup by co-ran Code Envy was the first web IDE that used containers kind of under the hood and was fortunate to sell that to Red Hat in 2017. Then spent a bunch of time in Red Hat and AWS as an executive, but I wanted to come back. To the startup world now, because I feel like this is one of those points where a lot of things are shifting, a lot of things are changing, and historically those are always the points when startups tend to have the big, the biggest chance for growth. Now, when we looked at the market, myself and my co-founders, we really saw that we didn't, we didn't want to do, kind of like you said, a wrapper around. A public LLM we didn't, we were concerned about how kind of, defensible that technology would be competitively in the long run. It's relatively straightforward or simple for those public LMS to kind of add on capabilities. You see that now already with, you know, Claude adding code and now open AI is adding code, and so you have a, a cursor that now has to battle these huge giants as well. Instead, excuse me. Myself and my co-founders, our backgrounds are more developer tools and DevOps tools, and that's where we saw the gap. We saw that the all of the ML ops tools that people think of, weights and biases, ML flow, Jupyter Notebooks, et cetera, they're all designed for only developing the EI model. They don't really touch. Anything with traditional microservices. But we said, well, that doesn't make sense because you know, and this was a, A couple years ago. A couple years ago, kind of when it was early stages, everyone was only talking about the model. It's just the model. And then, oh yeah, we've got all these other apps, but they're totally different now. Finally, people have begun to see, okay, no, we need AI agents, and that's gonna mean traditional code and models. Kind of intermingling. And so we saw that gap early and we said, well, if that happens then organizations are not going to want one set of tools that only work with models and one set of tools that only work with things that are not models. They're gonna want tools that actually can kind of combine both. And so that was really what we focused on. We wanted to make sure that, that we could build a platform that would connect models to applications. Maybe more importantly would bring the same confidence and discipline, security, governance, reproducibility that people had with traditional applications, but bring those to the kind of more black box ai. And I think the, the last piece of the puzzle for us is we realized like that the pace of innovation in AI is so fast that in 2022 models were a little bit slow. You know, we'd ask them a question, it would be a couple seconds for them to kind of come up with the answer. But we said, and that's what everyone was focused on. How do we speed up the model? How do we speed up the model? We said, that's not gonna be a problem for very long. But what will be a problem for a long time is can I trust the model? Do I know where the model came from? Do I know who touched the model? Who trained the model? What happened? What was the data that trained the model that affected what it's gonna do? Can I prove that no one tampered with anything in that chain before it went to my customers? And so we said, let's focus on building trust, reproducibility, and discipline for enterprises that need to host their own AI and connect it with their own applications. That's, that's really where Jozu came from and I think it's mm-hmm. [00:06:03] Adil Saleh: Very interesting, very interesting. And, and, and touch us a little more on, on, on how you know, it, it becomes a challenge building something to this, this kind of DevOps really, really good for enterprise because you know, a lot has you know, enterprises are always globally divided. Even if you target like non-Americans the data side. Will be covered in the same for a lot of, a lot of these enterprises. But how do you cover the use cases of different AI models? Is there any specific story that you want to share? Because we want to really understand like you know, a lot of these companies, like I mentioned, the off record as well, like they're building agent frameworks, PIL agents, but they never know and, and they're going up market. It's like more than a year down, they're going up market. They're now trying enabling enterprises to integrate these AI models with their system. So how are you actually thinking about this enterprise segment using AI models and able to manage it, keep it secure and actually relied in longer term view? [00:07:03] Brad Micklea: Yeah. So it's interesting, I think that the answer is gonna be different for different types of businesses, which I guess is a bit of a cop out. But I'll explain kind of what, what it is for us. Obviously in building Jozu, because we're talking about building the kind of infrastructure, the tooling that people use to develop, productize, and manage these models, you know, in development through to production. We don't really need to care about what the model is. You can use it with an old school predictive ML that you might see in a financial institution. You can use it with a vision model as we have in, in some organizations that are using us already and LLMs whatever you want. The hard part for us though is, and, and you kind of said it, Adell, is that the enterprises vary. The tools that they use vary. But also what doesn't vary is they have a very high bar for what tools that they use need to do. So when we got our funding, at first, we kind of sat down myself again and the co-founders and said, how do we want to, how do we wanna split this up? How do we want to hire? And we were absolutely inundated with companies saying, Hey, we can get you, you know, 50 super cheap people in Country X, country Y, country Z. Right away. We said, for us, that's not gonna work because an enterprise wants to know that this thing was built to a very high standard, high, you know, really deep understanding of security, really deep understanding of their tool chain, their requirements. Now luckily we had that in our background from our own experience, but we knew that we wanted very senior people and a lot of organizations where a lot of companies when they first start kind of go, we need lots of people and we need to work fast. We went a little bit of a different direction and we said, no, we're gonna hire seven people max for now. And all of those people should be very, very senior. There should be extremely experienced. We're okay paying a bunch of money for each of them because of how good they are, and we are going to spend a little bit of time upfront. Before we start writing lots of code, thinking about how do we write this, how do we build this so that it can smoothly just enter the enterprise? We're not fighting battles that we don't need to fight. So for example, everything inside of of Jozu is stored in a company's existing enterprise registry, their container registry. That's critical because if we had had our own storage, which would've been easier. Much more obvious, if you will, to say, oh, we're just gonna create our own storage layer. Cool. We didn't do that because we said, well, the minute that we are storing something, then the security team at that, every enterprise is gonna say, well, show me how you're doing this. Show me how you're doing that. How are you deleting? How are you encrypting? How are you storing? What about in transit? All these questions. Who, who's gonna control authentication? Who's gonna control auth authorization? And those are hard and long conversations to have. We said no, if we put everything in the existing container registry, they've answered all those questions already. They're already happy with it. They own it. And at that point now, there is no question. They're just like, okay, well Josie works because securely, because it's using our infrastructure, which is secure, and so a very secure company, it's going to be very, very secure. In a less company, in a less secure company, it's gonna be a little less secure, but they're okay with that. So as choices like that early on that we said, let's make those, let's think about those choices, let's make those choices early, and then we can take a, a very strong team of very few people and say, okay, now build, now build something great based on that. [00:10:52] Adil Saleh: Another thing I I reckon that you did really good is you absolutely defined the ICP in the first place. You are only gonna be targeting enterprises. No. Like mid-market, no small companies, no. Like a lot of these companies, they're, they call themselves product market fit, post product but they're not just. The product market fit based companies, they're pre-product market fit and they're not using technologies, spending money on technologies that are really, really mature, as you mentioned in the beginning, like they, an enterprise. The tooling and tech stack is always, always a state of the art, and you get to work with, with that. And that is one thing that, you know, you, you pretty much secured in the first place. So now thinking about. Also I'm not that technical by the way. I'm learning from you as, as as we speak. So thinking about these agent flows that we are talking about, like applications these platforms that are totally powered. Some of these companies are like you, you call them enterprise, they couple of billion in evaluation. But let's say talking about that company that is itself and AI. On top, something built on top of ai. Mm-hmm. Sort of an egen framework. So how Jozu actually gets to work with that. What kind of workflow that you, you have a lot of that will be listening to this episode too. [00:12:01] Brad Micklea: Yeah. So I think, you know, we've got a number of folks in that, in that bucket. So for example, one of the biggest areas of growth for us are. What are called strategic integrators. So they're these big companies that enterprises trust and say, Hey, we've got a problem. We want to create our own AI using our own private data to give us a competitive advantage. We wanna put it out into the market, but we're not quite sure how to do that. Maybe we've got a couple data scientists, but it's not something we're expert at. Can you help us? And so right now, a number of those companies who are getting these requests from enterprises are saying to themselves, Hey, we should just create. Secure private cloud with all the tools that are needed and tell the enterprise, Hey, yeah, just give us your model and we'll run it through all the tooling and we'll hand you back something that is trained, that is secure, that is ready for production. And so we've kind gotten connected with a bunch of those folks and some very, very big names and obviously some, some smaller, more local folks as well. And. They're coming to us and saying, okay, well we need to be able to prove security. We need to be able to prove governance. We need to prove that this thing has not been tampered with, that the data is safe. How do we do that? And so that's where we come in and we help them to make sure that they can give those same assurances to their end customers. So some of that is our networks. We're, as I said, we're fortunate because all of us have been in this market for 25 years. So certainly there's that, but also some of it is our open source product. So we have an open source project that we started actually before the co, the commercial company, and has absolutely taken off. It's called Kit Ops. It has over a hundred thousand downloads. It's now part of the, one of the biggest software foundations for open source in the world. Okay. Can you repeat the name one more time please? Sorry. Yes, kit. Kit ops. So K-I-T-O-P-S. [00:13:54] Adil Saleh: GitHubs. Oh. Might have seen you guys on GitHub. [00:13:57] Brad Micklea: Yeah, you can find it as well. GI find a bunch of stars. You can find it@kitops.org as well. So that was just a secure packaging model for folks. And so it's a, it, it solves a very, very narrow use case, but a very important one. And so as it accelerated, people began to look at the project and say, well, this is great. It's super helpful for us. I wish it did this, this, this, and this. Then they would look and say, wait, who built this? Oh look, it's these Jozu guys. Yeah, you got the validation [00:14:30] Adil Saleh: and [00:14:30] Brad Micklea: you got, that's right. Then you come over to Jozu.com and they're like, Hey, I saw your open source. It works in my environment. It's secure. It's simple. It, you know, it saves me a lot of trouble. Now tell me about everything else you're doing and that is a great kind of door in for us as well. [00:14:46] Adil Saleh: Yeah, I mean, I, I, I'm super glad looking at, you know, you recently raised funds and now that, you know, the avenues are expanded, but you being someone really, really experienced have to make that strategic decision. You being a CEO, how, what, what makes you excited and what kind of avenues that has, has it opened up? I know that nobody says that funding is bad, but a lot of folks, they come up and say, Hey, funding. It's something that doesn't give you freedom. Gotta get the right investors, a lot of that. So first tell us like how was that process, like the funding process you wanted it or it just came up naturally or how was it and what's what's the plan post-funding as a founder, you share what kind of avenues and initiatives that you, you guys taking and doing it really, really fast? As you mentioned the beginning. [00:15:24] Brad Micklea: Yeah. Yeah, so because of what we focused on, and you're absolutely right, Adele, like we knew from the start we're gonna focus on big enterprises, governments, secure regulated industries. These are hard places to get into. And so we knew, as I said, we knew we needed to have a small number of highly paid, very experienced engineers. That's not cheap. It's very hard to bootstrap that kind of team unless you've got a lot of money or unless everybody is yeah, I don't need to get paid for a year, which most people, you know, are not in that position. So we knew we needed to get the VC money. And the other thing that we knew having experienced this from the last startup was good VCs can help you to get into some of the places where you need to get into, and they can help you make some of these tough decisions when the tough decision times come. So right from the start, we knew we're gonna go get VC money, and we knew what that meant. The last startup I was at, which like I said, we'd sold, but that was a VC-backed company. So I understood what it meant to be at a VC get backed company, but I'd actually started my career with a small startup here in Toronto, Canada that was not backed. And so I also understood. What that's and it is very different. And, and all things equal. Honestly, I prefer the non VC-backed, but to do certain types of, to solve certain types of problems, you have to have the big, big money behind you. And so yes, we went out and we looked for investors. The first investor we found is Ali Corp outta New York City. Kevin Ryan was the founder of one of the founders of Mongo of Business, insider of Gil. Like he's probably one of the most successful serial entrepreneurs in the world, but certainly in New York. And so that was a fantastic I. You know, they've been fantastic with us. And then we kind of added from there. So we were able to get half court, which is a, a smaller vc, but fantastic operators themselves, serial entrepreneurs, a lot of experience kind of similar to ours in the infrastructure world. They've been great with advice. Brightspark here in Canada has been fantastic. They're one of the most successful early stage funds in Canada and Mozilla ventures, which was really important for us because of the open source side. We wanted to make sure we had somebody kind of in our investor network in backing us who really, really understood open source, who understood. Some of the sacrifices you have to make. There are things that we can't do with the open source product project that we could do with a freemium product that would help Jozu, but we can't do them because they would, they're gonna scare away the community. They're gonna actually kill that project. Not all investors understand that. So when you don't do those things, you can end up with investors that say, Hey, you're making a mistake. You're, you're being stupid. Just Yeah. You're not [00:18:06] Adil Saleh: trying to monetize, you're doing it open source. Exactly. You should be [00:18:08] Brad Micklea: monetizing, you should be monetizing. What are you crazy? You, you know, you know what you're doing. And those distractions we wanted to avoid. So having Mozilla there, they're kind of calming influence. They can say, no, you guys are doing the right thing. It's, it's growing. You're absolutely right. Just stick with it. It's, it's awesome. You know, and so that's, but it was not easy. I don't want to, you know, even having. Had a successful startup in the past, having been in industry for a long time, I still talked to over a hundred different VC funds before we closed. Before you [00:18:41] Adil Saleh: go. [00:18:42] Brad Micklea: Yeah. It is, it is a numbers game and I, you know, wanted to make sure that I talked to as many as I could, partly to find the ones that were the best for us. And so that it's a lot of work. [00:18:55] Adil Saleh: Yeah, it's a lot of work. And it's, it's, it's, it's something that is you know, for, for a product like this, as you mentioned, like it is mandatory for you to, you know, do it right. Otherwise it's not gonna, it's not something like a B2B SaaS company that you, you, you know, you go to Y Combinator, you do three months of coaching. You get some, some of their peer group, you're slowly graduate like. Something like this, it could, you know, you're building it for enterprise. Either you do it right, or you fail. It's, it's just as simple. [00:19:22] Brad Micklea: It's a highwire act for sure. [00:19:24] Adil Saleh: Yeah. [00:19:25] Brad Micklea: You get one shot. [00:19:27] Adil Saleh: Yes. And like how did, how, what kind of leverages did it give you? You raise a among back. What kind of niches are you guys, you know, getting your heads together, getting back to the drawing board on the marketing side go to market side. How are you gonna penetrate fast development side? What kind of initiatives are you guys think about? [00:19:44] Brad Micklea: Yeah, I mean, it was, it was a game changer. It certainly, it allowed us to expand our dev team a little bit more, which was, which was great. But we already had a really solid dev team, so that was not actually where the majority of the, of the dollars are going. The majority of the dollars are actually going into our go-to-market. So we were running really a shoestring budget for go to market before we got the funding. It's really myself and one of the other co-founders who, who handle kind of all that stuff on that side. So Jesse Williams runs, runs marketing for me, and Gokin runs technology. And so Jesse, I mean, has really done amazing work on a shoestring budget for a long time. But now we were finally able to say, okay, now you have the money to do big events. So we went to Cube Con, we went to Red Hat Summit. Now we can do bigger spends on the marketing. On the marketing side and actually get our message out. More broadly. Unfortunately, another thing with dealing with enterprises is there tends to be a much higher spend to reach those enterprise decision makers. They're kind of behind a set of walls, if you will. Yes, [00:20:48] Adil Saleh: yes. You gotta get through the red tape and you gotta spend money. Time is sales cycle is always gonna be big. And if you're in North America and you're talking about the East Coast, it's always, you know, it's always very noisy. Especially with enterprises. 'cause a lot of people are chasing. At the same time. But again, as you mentioned if you're able to, you know, convince these investors that you have on the list, I think that's you've done the hard yards and you know, a lot of this will change the conversations that people see you will change. You'll, you'll see the change you know, especially in the, in, in the kind of industry you're talking about, like SF Austin, Texas, San Antonio is going to go big and next close to a year is gonna be our biggest metropolitan well, Texas and then. And New York City is already big definitely. And also, you guys also thinking of like hosting your events out in here in the United States or what, this year? [00:21:37] Brad Micklea: Yeah. Well, probably not this year. I think that's probably a little bit more into the future to do our own event. But it, it has allowed us to spend more time in Europe with the EU AI Act driving regulatory compliance in Europe. We've seen a lot of pickup. From organizations there in Europe. So being able to spend a little bit more time in Europe go to certain events in Europe, become more known in Europe, has been very helpful for us. And that would've not been really possible without, without some of that funding, for example. [00:22:07] Adil Saleh: Yeah. This, let me ask you this this is kind of an opinion for a lot of people, a lot of founders, especially from the, from the states. That Europe is a really hard market to play in, in terms of compliance and regulations and the mindset of people like spend of these mid to large scale enterprise a lot of these opinions that have been formed and they, they think that United States is the biggest market. We don't want to deal with Microsoft teams. It's a lot of that, you know, already. So how do you see it? What's your viewpoint? [00:22:35] Brad Micklea: So I think actually that's mostly true. You know, the US really is the biggest market and you generally can grow a startup to quite a good size if you're just dealing in the us. So I don't know that for most businesses, I don't think there is a need to necessarily go outside the us or, or North America in our case, because so much of what. Like our audience is in regulated industries. They are high security type or organizations. It made more sense for us. Like the things that make Europe hard for a normal business are actually tailwinds for us. You know, and so I think sometimes you have to kind of think broadly, a little bit more broader about your, it's lessing [00:23:18] Adil Saleh: in disguise for you. [00:23:19] Brad Micklea: That's right. Exactly. That's right. And with the last startup at Code Envy, for example we actually were originally picked up first by people doing embedded software for like little devices. And although we. Didn't have many connections in apac or had not done a lot of business in Asia. One of our biggest customers was Samsung. And so we kind of pivoted quickly and said, okay, well let's spend a little bit more time in Asia, because that is where a lot of that hardware embedding was happening. So you also gotta be open to it. Like you can say, we're only to go North America, but if you start getting picked up elsewhere, you, you should look at it and go. Maybe there's a reason, maybe I need to be there a little bit more. [00:23:55] Adil Saleh: Yeah. Middle Eastern market is, is growing big. Riyadh, like South Arabia, it's big Qatar that's going to go big in the next five years. And there's a huge competition. The tech Google is investing, you know that a lot, like having their data centers and all billions of dollars being spilled, you know, it's no time. So that's a, that's a big market you know, to tap in. I'm glad that you're already thinking about that. Okay, so talking about go to market, like you, you already have your enterprise, wherever you are, depending on it does not literally like whether it's in North America, European, middle East, and whatever. So now thinking about how to penetrate fast in a, in a right sweet balance with, developing the technology really fast as well. So as a, as a founder, how, what, what kind of measures are you thinking of that makes you excited this year? [00:24:42] Brad Micklea: So I think what we started with was really optimizing for feedback. So when we did reach outs when we were at shows, although we wanted sales, of course, of course we want sales. That wasn't actually the thing we would first push for. We would always push first for. We're building this thing. It's for these types of organizations, people who care about security, people who care about reproducibility, about being able to roll back safely, all these kind of things that are more production oriented, not prototyping. I mean, in the end, companies prototype with open AI and anthropic, but most of them know that to keep their own data safe and secure, to keep their differentiation, their competitiveness intact. They need to self-host. And so we'd look for those kind of companies as we were at these shows or, or whatever have you doing talks and say give us feedback. How, what would you need to see? We've built these three things. Do they make sense? Do they need more? And getting that feedback early on I think was, was super helpful. Some of those folks, of course, who provided feedback and who we were getting these conversations with have ended up being some of our best, our best users. Mm-hmm. It's also interesting because some of them, you know, you would there was, there was actually, I can think of a, there's a huge European kind of manufacturing company that we spoke to early, early on, probably a year and a half ago. And before we had a product and we just said, Hey, we're thinking of building this. And they were like, that's cool. We're building the same thing internally, so you guys should definitely do it, but we won't buy it 'cause we're building our own. And we're like, okay, that's, that's cool. You're huge, like 40,000 people. How about it? About a month ago, that person came back and said, Hey building this thing has been really hard, actually a lot harder than we thought. Let's talk, I think maybe we may want to go with you guys actually. And that's happened now twice with two different people. And I think, I think sometimes, especially as an entrepreneur, and I'm the same, you get, it's scary. You know, you don't have any, I. Any parachute, you don't have any net underneath you. And so the idea of letting somebody walk away is, it's hurts when you're like, I've spent time with you. What do you mean you're gonna build this yourself? Like I could do this better than you. But sometimes you gotta let them walk away, discover how hard it is, and then come back. 'cause once they come back like that. They've convinced themselves like, I don't need to sell them. You don't have [00:27:10] Adil Saleh: to sell. Yeah, that's right. Sale is done. [00:27:12] Brad Micklea: They've sold themselves. And that's so powerful. If I would've pushed too hard, I could have broken that relationship. And then even when they failed, they would go looking for somebody else. 'cause they'd be like, oh, that Brad guy's a pain. I don't wanna deal with him. You know, [00:27:26] Adil Saleh: very interesting story. So now thinking about you know, a lot of, a lot of this is, has to do so much with how you evolve as a product. And you, you mentioned the feedback loop from enterprises. You can, you can get like different things like scatter things. How, how, how do you make sure that all of that is. Absolutely aligned with with the product vision as, as a founder, because that's big of a pain. Like a lot of enterprise, they pay with bucks. You have dedicated account managers or you know, technical team, like implementation managers for them. But the, the contract value is big enough for you to say no a lot of times. But it, it is always important to say no when it, when it matters. [00:28:03] Brad Micklea: That is hard. And I don't think that there's a. Single formula just works all the time. But when we started this company, the reason why there are three of us as co-founders is so that we can check each other. And so Gorham Erkin, who's my, my CTO co-founder, when he and I are looking at the product. We, we kind of red team each other. So whenever somebody wants to add something, the other person will go, okay, but do 80% of the people need that Because if 80% of the people don't need it, it shouldn't be there yet. We're too small to be building things for less than 80% of our users. I think just asking that kind of question and kind of pushing each other a little bit on that. Sometimes we'll land and it, and it's, yeah, I, I think this really is needed by 80% of the people. Let's go do it. So we're, you know, we added audit logging for example. And, you know, that's a pain honestly, to add into the product. And you know, we, we looked at it and I said, I, I really think we need audit logging. And it, and we tried to kind of push back, but then eventually Gorka himself was like, yeah, I, I think we do this is I. This is something enterprises need. I think they're all gonna need this. If we make it simple enough, that's in itself a big benefit. And we'll look at other things and be like, no, 80% of people don't need this. 20% need it. So we'll kindly tell them how they could build onto our product with that, that they built themselves and connect the two things, but we ain't building it. And that is hard because yeah, there are times when somebody's well, I'm not gonna go with you 'cause you can't do this one thing. That's a tough, a tough call. Sometimes you're gonna land on, okay, I guess we're gonna build the thing that 80% of people don't need, but we really need it for this one. 'cause it's a huge deal. And other times you're not. I think Code Envy was a great example of when we won that Samsung contract, there was a very specific thing they needed. Most people didn't. At the time, that was like a, I think it was like a $750,000 contract. And so considering that that was more than all of the rest of our revenue historically combined, at the time we were like, I think we're gonna build it. Yeah, we need seven $50,000. This would be very helpful. And so we built it. And [00:30:23] Adil Saleh: that also, you know, that also takes, a longer term view of foresight and especially your CTO, like if not today, how we can replicate this for other, you know, customers in the future. Like how is gonna be this problem that that the same feature can be built for or maybe iterated slightly, or we can definitely utilize these this, this time and resources that you're spending now. For this for this enterprise in the future or maybe a lot of these features, a lot of c tell say we built this feature and we start looking for, for those customers. Because we are so early, so small, pre-product market fit, we never know what's gonna be the right customer. So we built it and then we learn that, hey, this. Let's, let's, this is really important, this feature that we built only for this customer, and now let's, let's, you know, see we, we get like lookalike customers or something like this. [00:31:07] Brad Micklea: Yeah, and you can do that. I, I, I'm a big fan of using, you know, like Figma for example. So we have a UX designer and she can build these incredibly realistic. You can kind of click through. And so in a, in a video demo, like there are features that we have not actually built, we've just built in Figma and then demoed them. And when people either respond as eh, then you know, okay, that one's gonna stay in Figma forever. And then other ones you're like, okay, let's build this. 'cause people are getting really excited. That was actually how we found the audit logging was we just locked it. Mm-hmm. People were so like, no, that's [00:31:40] Adil Saleh: the, that's the best way to do it. Yeah, that's the best way. Because Figma has this you know, you can connect all the screens, all those buttons, make it directive, like a prototype, and you can show in less than two minutes to validate it in the first way before, you know, putting all the resources and everything. That's a smart promotion. Now, thinking about taking some of the devices around this notion that people have these AI platforms, they are like, of course, not reliable at this moment. For some of the use cases, especially biggest is finance and healthcare. You know, you cannot rely on the data. You cannot rely on a lot of models that are actually storing the data and servers that are storing the data. A lot of these teams are trying to. Have AWS but of course it's so expensive to do that scale. Like even we are spending a lot like a very small early stage startup. So now what's your viewpoint along this as a founder building a tool? Just regardless of a developer tool or any, any tool in that is in the development space. We, we, we spoke to a lot of developer tools too. So companies can rely on those like enterprise that rely on Zu you know, is it more about engineering? Is it more about, you know, you, you mentioned that it's not about doubling up your resources or, you know, it's, it's about doing it really, really high standard. So what's, what's your viewpoint on it, regardless of your experience at this, but you just give us a general overview of your, [00:32:57] Brad Micklea: I mean, my opinion is that the AI models are progressing so quickly that we're already seeing that the open source models that people can self-host are good enough to do 80% of what. Of the different jobs that people want them to do. And that's gonna become a hundred percent in another year, max I think probably less than a year. I, I think that the only logical thing for a company to do long term is figure out how to self-host their own models. It is so much cheaper and ultimately you're protecting your data better. You're keeping your own competitiveness, like that's the thing that blows my mind. So more than anything is that you've got all this data that you have worked, blood, sweat, and tears, or paid large money, you know, to have that data that is yours, that is different than your competitors, and then you're gonna go and put it in an open I. Like OpenAI, [00:33:56] Adil Saleh: LLM, yeah. [00:33:57] Brad Micklea: To train that LLM the next round. And guess who's going to use that? LLM, your competitor. And now they've basically got all the goodness of their data and your data from OpenAI. How does that help you? It doesn't, it hurts you. So I think companies know that for cost reasons, for security reasons, and from competitiveness reasons, they're gonna self-host. It's the same reason why. Each bank has its own mobile app. Could they all use the same one? Of course, they could all use the same one. Yes. But then there would be no differentiation. Yeah. Yes. I get [00:34:25] Adil Saleh: your point Follow up question on this. Like how, what it takes for a, you know, small startup for or from very early days to have their models self host. What it takes at this, at a level of standard that they can sell it to enterprise y. [00:34:41] Brad Micklea: Yeah. So I think that it, when you're very small, I don't think it necessarily makes sense to self host when you're starting at, especially pre-product market fit. And by the way, I think you, you said something earlier, a di that I couldn't agree with more. Most people have much too low a bar. For product market fit. I, I don't even think now with Jozu we have real product market fit. I think we're still a little bit pre-product market fit. When you're before that, you have to be just optimizing for speed and, and capability. Speed and capability. You get way better from a, a public provider, like an anthropic or whatever have you. Use that for kind of as long as you can. And that could either be because it gets too expensive, so now you can't afford to use it. Now you've gotta self host. Or because you, you do something in the market and, and your, your, you know, your, your audience, your, your customers say, no, we're, we're no longer comfortable. You need to self-host wherever. There's gonna be some pivot point that comes and you gotta move. It's not actually that hard to self-host models you know, typically. There's platforms out there that do it for you. They tend to be more expensive if they're just for models. We tend to default to using Kubernetes. Because although Kubernetes isn't easy, it is a technology that has a very, very broad adoption. And so there's lots of technologists, lots of people that have experience with Kubernetes. And it's maturing quickly in its support for, for models. But it really just comes down to, to the individual details of the, of the company. But yeah, for me, if I were doing something where we had to use AI to inside and, and I, I did in the past public first. Then self-host when that becomes cost prohibitive or are not secure. [00:36:27] Adil Saleh: Yeah. The only thing that's important in the beginning, in the first maybe a year or two, is the capability and speed. Yeah. And you can use, you, you can leverage these l and m's open source models to, you know, increase the capability and speed with a lot of people are doing. And then at scale you can, you know, over time is it gonna be as hard of a problem to do it, engineer it to self, force it today, then tomorrow. Or just gonna be an equal amount? No, [00:36:51] Brad Micklea: it's gonna keep getting easier. Absolutely. Okay. Yeah, for sure. I mean, it's you know, one of the privileges I guess, of being old like me is that I remember when, you know, just running a website. On your own was really, really hard. Like my, my first kind of work out of university was literally building websites for companies that didn't have the infrastructure knowledge on how to do a website. Now, I mean, the idea that I. Somebody doesn't, can't build a website is insane. Like you're like, my grandmother could build a website, you know? And that's the way it's gonna be with this stuff too. Like AI models right now are a very niche technology. It is hard, harder to run them today. Next year it's gonna be easier and easier and easier in 10 years, honestly, it'll be like click a button and you've got an AI self-hosted that you're running on your own. [00:37:42] Adil Saleh: Yeah, so the biggest advice coming from Brad folks is to eventually transition into having, building your own you know, AI models, ML models, and host them yourself for scale. So that's, that's, that's the best thing that we have heard so far in a very long time because a lot of people that think about although they're rappers and they trying to build something impactful. Meaningful for a lot of these G TM tooling and all, and they try to preach, preach something that is hey, this, you are just relying on OpenAI, you're relying on Deepsea that like, we had a good amount of noise for good two, three weeks or month and then it got all got shut down. I'm not sure. What's your viewpoint on, on Deepsea part? The way [00:38:21] Brad Micklea: I, I think it does deeps seek was, I think important for two reasons. One is it showed that other countries can build powerful LLMs [00:38:32] Adil Saleh: at a cheaper cost. [00:38:33] Brad Micklea: At a cheaper cost. Exactly. It shows that, that that kind of constraint, which startups get that constraints, can actually breed innovation and success. I think a lot of times people just assume that they need massive money in order to have success. You don't, sometimes the constraints are what drive it, so that's one. And then two was that I think it was a bit of a wake up call because China has such strong technologists. [00:39:00] Adil Saleh: Researchers as well. [00:39:01] Brad Micklea: Yeah, researchers technology. Incredible. And I do think that in North America we have not taken. Science, technology and engineering seriously enough in our schooling. And I don't think we are as far ahead as people think we are. I think China's way ahead of where North America is, and I think it's a big wake up call that we need to get people focused on that. [00:39:23] Adil Saleh: Yeah. Some of the researchers that Facebook bought, I would say bot the word, there's recently, they're like half of them, more than like from these countries Chinese, originally Chinese. Great. It was really very nice meeting you, Brad. You know, it was bound to have have you on even on on the Independence Day. I never cared. And you know, it was all worth it. Thank you very much for your time, your insights, your concrete vision, and very much you know, infectious energy. Thank you very much for this. [00:39:51] Brad Micklea: Yeah, thank you. And I'll just end, if you don't mind by, by just reminding everybody that. The future of AI isn't just faster or smarter, it's about being trusted. And so mm-hmm. If you are building with AI and you care about data security, you care about IP security, you care about compliance we'd love to talk. [00:40:11] Adil Saleh: Yeah. Perfect. And one more thing. If, if, if people and companies go to these markets that are really, really smart and capable. Two years, three years later, they'll of course knock down by the security and they come back to the tools and infrastructure that are absolutely reliant, secure. And again, he mentioned that Salesforce, that they can, they can actually have, they have this reputation. So they'll, they, and two years later they'll share repetition, they'll share security and all of that. So it's, it's just about making sure that you are making a move for tomorrow. [00:40:39] Brad Micklea: That's right. Thanks very much. It was fun. [00:40:42] Adil Saleh: Yeah, absolutely. Likewise. Take care. Have a good rest of your day, Brad.