BUILDING THE FRAMEWORK: With Steven Walchek, Co-Founder of Liminal

From ChatGPT Moment to Liminal's Founding Vision Maybe everyone, welcome back to another episode of our Building the Framework podcast series. Today, I'm joined by Stephen Wolchek, CEO and cofounder of Luminal. Thanks for joining us today, Steven. Speaker 2 Yeah, thrilled to be here. Thanks for having me. Speaker 1 Absolutely. So maybe jumping right into things, can you share a bit about your background and what led you to start Liminal? Speaker 2 Yeah, absolutely. Without spending too much time on my background, this is the fifth company that I've either been the founder of or been a part of the founding team of and have had a few good wins along the way. I've done big company as well and LED global technology partners in native US silent go to market for their emerging technology and launched them services around AI and and and video analytics and went to other very large fintech company actually to become their chief innovation officer in an EVP there where we built a corporate venture studio and launched a couple new companies out of that as well and saw some acquisitions, which was cool. And then the kind of ChatGPT 3 put my jaw on the floor. I think that was the the moment that the AI wave really rolled over the the world and I was certainly caught U and not as well coming out of FS words heavily regulated and I frequently partnered with our chief information security officer there and a number of people and compliance to bring any new technology into the business, realizing that there was going to be a pretty significant amount of burden when it came to regulatory compliance for any organization that subject to that, you know, think banks, financial services, insurance, life sciences, healthcare companies, state, local government and education. They all are subject to some degree of regulatory compliance that they need to meet in order to be able to leverage modern tools. And now we have the most modern tools available to them through AI, and we wanted to be able to bring that to them. So we built a a really neat way for organizations to deploy general purpose generative AI to their end users that's relevant to the vast majority of their organization, but that contains all of the security tooling they would need to meet regulatory compliance and to feel comfortable and confident in sharing private data with these downstream click models, as well as deploying 42500 in different supported slums, looms and agents all through the same interface. So we look at two sides of the same coin. Look at the administrator view, we look at the the end users view. And Liminal has packaged up a really neat experience for both of them. The companies get what they need in order to securely and safely deploy generative by end users, get multi modal, flexible generative AI and everybody wins because it's all based in a very cost effective package. So that's how we got here. The Two Broken Approaches to Enterprise AI Awesome, appreciate the background and maybe go and bit deeper in terms of, you know, that the pain point maybe that you're, you're seeing in the market. So it feels like right now they're still pretty big gap between no experimenting with AI and actually scaling it inside an enterprise. What do you see as as the kind of the biggest challenges that companies are trying to to solve women when moving from pilot to production? And you know how does that shape Luminos approach? Speaker 2 I think we're in the kind of just entering the meaty part of the adoption curve. There were earlier in the last couple years, we've seen a lot of tire kicking and a lot of exploration and, and people really genuinely trying to make sense find signal limits. The cacophony of noise that exists out in the market today. If you're doing anything in AI, I do not envy your position as a buyer. It's got to be immensely difficult to sort through what is signal and what is noise here. The encouragement I always tell people and the people that have been most successful, particularly the customers that we have, have always just decided, let's get started. We've got to get something deployed and, and if we think about the, the problems that we're solving, there are two different types of personas that sit on either side of the fence in the markets that we serve. You either sit on one side of the fence, which is I've got a policy in place that doesn't let any AI into the organization whatsoever, which paradoxically, rather than improving security, it actually makes security more difficult to enforce because people are typically moving off network. We've got over 900 million weekly active users of ChatGPT alone, let alone Gemini, let alone Copilot, let alone name any of the the modern AI tools that you're familiar with. That brings out well over a billion. There's a good chance these people are using it in their personal lives to some degree, whether it's as expensive as they would like or not yet certain. But here we are faced with the challenge of trying to maintain an understanding of what these tools are and how they'll be using them in a workplace. While again, that first persona says absolutely not. We're not allowing any of that in here. But people are going to be using this on their own. So they're probably pulling data off network. There was a static read recently. In fact, it said 71% of enterprise data leaks are now coming through ChatGPT. If you're an international company now you've got a little bit of concern with the agreement that was signed with the Department of War and Open AI. And although they've come to a little bit back step on their policies there and their terms of service, there's a notion that things can change on a dime. And you're completely powerless as an end user or as an as an enterprise to be deterministic on those changes. So that's a that's a difficult position. So that's kind of side one. As I as a buyer have decided that I'm absolutely not allowing Jenny in here, I'm going to get paradoxically, that actually breaks the security standards more so than people might think given the amount of usage we have outside the organization. The other side of the coin is I allow an open tool policy. We had one customer of ours that that had this this particular stance and we encourage them to go and take a look at their network traffic and say, yeah, see how many actual requests are moving out to Jenny I tools. And they found that over a period of three months, they ran their net scope report and they found that there were 300,000 incidences of corporate data exiting and Excel trading outside of the company. Just in that period of three months is a company of 2500 employees. And these are unsanctioned. There's no licensing, there's no enterprise agreements with these tools. These are just people bringing their own licenses in. It's also not gonna work because you have a variety of different terms of service there. We all know that the modern tool providers and their free accounts and even in some of the paid accounts are retaining data internally. They're using it to train their models. That's an opt out, not an opt in. And most people don't know where to find the opt out. There's a bunch of different things buried inside of a fifth page of the 5th part of the contract that you know, you never read. There is some clause buried in there that says they can use your data if it even Microsoft, who's got so much depth and penetration into the enterprises of all shapes and sizes with Copilot and their privacy policy, they said if you, unless you specifically fill out this particular form, they will be using your data to train their content moderation models, which nobody wants their data to be used to train anything. It's just hard in this world to trust anybody at this juncture, legal or not. And you can see that the Android user group got, you know, they saw sued Google successfully. And when Google explicitly said I'm not using your Android usage data to train our models or train anything else or to sell it. And then we found out that they've been for 20 years using it to do that. And they paid a paltry whatever, something like a $350 million plus fine, which for them is, you know, nothing to end users. That's their most discarded secrets in the public so. Speaker 3 Without spending too much time, you've. Speaker 2 Either got this open to a policy, allow anything in which is not effective, or I'm going to close everything off which is not effective. The the way that we see companies make that leap from pilot is to give people the tools they need to just get started. It comes later. 98% of end users, you're not even thinking about agentic. They're still trying to figure out how do I best prompt this thing to help my help improve my current workflow, my current job. They are not thinking about how do I automate various parts of my job away that they lack the technical capability. Even if they have the technical capability to translate that to actual requirements that make sense, it costs them the vast majority of their teammates doesn't make sense. We're in a position now, we're just getting started is actually the best move right now. Give people the tools they need in order to accomplish the job that they have better, faster, more efficiently, and that's what Jenny I can do it promises to do that, but you have to make those tool decisions wisely. And the moment you lock yourself into a single model vendor like ChatGPT, you you decide to that you want to write a policy that doesn't allow any tools in there that you introduce problems because people want multimodal flexibility. Obviously you need the security, data privacy, the governance, the administration, the observability around all of this. You need to do it in a way that is cost effective pre organization and fortunate part of Liminal is that we've made. Speaker 3 All three of. Speaker 2 Those and so we enable that path really neatly of hey, you need to get started. Here's something that's ultra secure, contains unlimited access to the most popular models, but also allows you to deploy your own models into the same delightful and user experience. And then we do this in a way that is much more cost effective than any even single model provider subscription can be, let alone having access in an unlimited capacity to multimodal providers. So super exciting, but it really is like the the leap is get started with something basic that applies to the vast majority of your end users that meets your security criteria, meets the flexibility and multimodal capability, and is cost effective for your organization. Speaker 1 Awesome. Yeah, I appreciate that overview. I'm sure it's very challenging right now for corporate leaders to to look through all the, you know, hundreds and thousands of AI providers today and which ones are doing what they say they are. Speaker 2 Being the buyer right now? Absolutely. Speaker 1 Exactly. Redefining Automation with BAP for End Users And for the corporate leaders out there who are, you know, trying to look a bit more into the future, I know you and I, we talked about the idea of this behavioral agent automation platform. And I think yeah, to a white paper on it. Can you breakdown what what Barb actually is and how it differs from maybe the traditional workflow automation? Speaker 3 Yeah, there are two different types of when you hear. Speaker 2 Agente typically today you think of two different types of ways to deploy that. The first is some off the shelf agent that goes and does something for you. It's the call centre automation bot or an agent that does something unique for the call centre. That's an off the shelf agent, typically bespoke for one particular use case for one particular group of people. Speaker 3 The other thing is kind of the other. Speaker 2 Side of that. Speaker 3 Which is a DIY framework you can build all of your own agentic frameworks that you can go and you kind of have carte blanche and what you want to try to build however. Speaker 2 But both of those are flawed and in different ways. Now, both of them are also useful in different ways. There will always be really comprehensive, complex workflows that the DIY works really well for, but you're now requiring an IT team to be the owner of that. They typically some understanding of the tool itself. So there's a technical understanding that's needed there, and then they need to be able to deploy that. But they're typically solving one thing and another burning problem inside the organization, not the mundane everyday things that take up actually quite a large swath of time for all of us. Um, if you Add all that together. Speaker 3 And the off the shelf one, again, is kind of making people fit the machine versus the machine fitting the people. It's basically saying, here's your workflow. That's what you have to do now and people are going way, but you're ignoring the nuance and styling this. So there's, there's some gaps in each one of these. You know, we, we kind of got names for all of these. So like the first one, and I like didn't discuss yet, which is this notion of prediction over proof. It's this idea that if I'm going to use a diary framework, I have to predict what end users want versus having proof of what the output they're looking for is. And typically I'm not the end user. Reminder that there's still a significant amount of AI literacy challenges that we need to overcome right now. People need to understand how to use the tool in its first. Automation is way off in the distant future, so typically some IT team member coming in and saying. Speaker 2 Buying I know your job as an ex better than you know your job here. Or I have to observe you doing your job. I have to find challenges in the workflow and then again I'm building a blanket. Speaker 3 Thing that everybody in your role has to conform to, which if I just observed you, I've got to pay attention to the nuance there, which is if I had proof of what you're trying to do, if you were to be able to tell me somehow this is what I'm trying to do, or these are the 15 things that I do 20 times every week, that that's a very different scenario and the value equation is much more. Speaker 2 Favorable for having proof that it is predicting. I think we all know that. So we want to see that the next thing is the technical. Speaker 3 Translation gap, which is this notion of OK, like let's say buying. You have the ability to say I know my problem and I have proof. Speaker 2 That I do this thing 15 * a week. Speaker 3 Well, then the next thing you have to do is boring. You're not a technical person in this hypothetical scenario, which is the vast majority of people. So do you don't have the ability necessarily to be able to translate here's what I need you to do into actual technical requirements, here's how I need you to go and build this thing. Huge gap in between understanding that there's a problem and then being able to translate that into a method that allows me as a technical person to actually go and solve that problem for you. And then the next piece we tend to face is this notion of the integration complexity challenge here. So this is the OK I've solved for the. Speaker 2 Technical translation gap we're playing. You're now able to effectively communicate technically what you need me to go and do to go and build for you. You have proof that this is the outcome that you want to generate. Speaker 3 OK, we've sold the first table. Now I'm the poor IT guy that's absorbed your technical translation. I don't do your job and I now have to go and somehow. Speaker 2 Figure out how to integrate all of these new systems together to automate that particular problem that you've been able to express to me, both technically and with proof that there's value. It's solving that on the other side. Speaker 3 Right away reported like these are three big problems to have to overcome. But let's pretend that you're able to solve all that. Now we have to think about the fact that there's the the non adaptive workflow trap, which is cool. I've solved this for Boyang based on what I've been unable to understand, Stop Lying and how he works on the technical translation that I've developed for him on the integration complexity to boss the systems that Boeing uses. But Boeing works differently than the 15 other people that are in his role. OI just ignore the nuance of all of the other people's workflows and justice work. We're going to base it on blanks workflow alone, which well, requires people to fit the machine versus the machine fitting people we want to see as I adapt to the end user, not the other way around. And and that's the challenge with these off the shelf and DIY frameworks is you are basing it on an individual's particular workflow and ignoring the nuance of everybody else. Honestly, traditionally genetic solutions just require you to build agents. Speaker 2 Before you truly understand what they're actually trying to. Speaker 3 Do O what does that, what does that mean for back? We talked a lot about Bapple BAP stands for behavioral agentic automation platform. There are kind of three things that are behavioral agentic. Speaker 2 Automation platform does really well. First, it does observation really well, so. Speaker 3 As you use generative AI, it is developing memory as you interact with your data that's connected into generative AI systems. It understands how you're doing it and it's it's looking for patterns of behavior. It's saying, hey, I want to see how you work, how you use generative AI to interface with your data, how you interface with your data, what questions you're asking, the frequency with which you're asking. And I can unpack and infer based on those types of things, what problem you're actually trying to solve. And if I get an eye, the generative AI here, you can infer what problem you're trying to solve. Well, then I can break it down into its component pieces. If you were say, doing preparation before a call as a salesperson and every time you went to your CRM or you asked me, I'm connected to your CRM as as an agent, OK, I as part of me is an agent automation platform. I'm connected to your CRM, I'm connected to your calendar. If I notice that 15 minutes before, I recall you asked me to pull data from your CRM about this particular customer that weirdly matches up to that calendar meeting. And I see you do that 20 times a week. Well, wouldn't it, wouldn't it be an easy leap for me to? Speaker 2 Make that. Speaker 3 I know that you do this thing 15 * a week. I know you do it before, 15 minutes before recall. Couldn't I just say, hey, I see you trying to do this thing 15 minutes before every call. Would you like me to do that for you? That is this notion of I don't require, I don't have to predict that. There's value there. There is value there. You do that every single day. Now comes the next piece here. So let's say I can unpack that. There's an opportunity here to drive efficiency for you. I can say, hey, would you like me to do this for you? That's the efficiency game. Um, the answer is yes. Well, I'm not asking how to go snap together a bunch of pieces. I'm not asking you to go and be an IT person or some other developer or have to go lean into code or cause no end user is going to do that. It just doesn't happen. Or if there is that they're very technical individual and and they're the amongst the few in the organization, not the many, OK. Speaker 2 I just actually kind of like. Speaker 3 Blitzed out that issue that we had earlier where we know that there's a problem, there's proof here that there's a there's a need. We know know that if I'm doing the work for you and automating or orchestrating the creation of an agent that solves for this particular workflow challenge for you. I don't have the technical translation gap anymore. I'm not worried about integration complexity. Speaker 4 I don't even. Speaker 3 Have to worry about non interactive workflow because machine saying hey I see you do this 15 minutes before it recall it might say to me Steve, who's in the same role as you boying Hey Steve I see you do this two hours before every call or at the beginning every day. So now we don't fall into the non adaptive workload trap either where I'm not paying attention nuance. And the system is, is automatically creating agents for you. It's self assembling things that improve your workflow based on your workflow. And now you can kind of say, OK, that's on the individual level. So, so behavioral genetic automation platforms are observing how you operate. And then they self assemble things, we'll call them agents, we call them capabilities, whatever you like to call them that improve that workflow for you. They drive efficiency. They say, hey, I notice you do this thing 15 * a week, 15 minutes before every call. I will look at your calendar. I will understand when that call is and I will deliver this stuff to you so you don't have to do that. And while you consider like how long that takes you, it might be 5 minutes, but you do that five times a week or 20 times a week, right? Well, that's all of a sudden, you know, anywhere from 25 minute save per week to end number of hours saved a week, that's a big deal. And that's on a new onset specific to you. These are small repetitive tasks that everybody has throughout the day. And if we watched what happened with the open cloud revolution and we notice that people are hungry for simple automations that improve their lives. But with open claw, you face all of the technical issues. It's a it's difficult to deploy if you're not in it. So behavioral genetic automation platforms, what they do, they observe how end users work. They find the inefficiencies and other workflows and then they solve for those efficiencies by self assembling automations, capabilities, agents on behalf of those individuals and deploy them. Now that's at the individual level. What's even more cool and and what, what Liminal is able to do is not just an individual level, but then look at corporately what's happening across the use of generative AI. The platform observes, hey, we've noticed that 15 different people have asked for this particular contract review and we've noticed how people have found different pieces of data that have helped them accomplish this. We've assembled agent that becomes your contract review agent that has approved elements in there for doing contract review and we've deployed that. And now the administrators have the ability to control how and who that's now deployed to and and when. But the ability to again, witness across the organization what behaviors are consistent and then be able to enforce automations and not just enforce, be able to have the platform self assemble automations on behalf of that constituency of end users based on that organizational memory is what behavioral agentic automation platforms can do. And that's what's so neat is it's taking this whole notion of magenta Ki and workflow efficiency and it's removing the technical challenge, It's removing the burden of prediction. And you just rely entirely on proof and, and and it's adapting to the end user versus the end user having to adapt to the machine. And that is? Speaker 4 Like the true. Speaker 3 Promise of AI in my compromise standpoint is like I, I get so excited because there are three judge off moments that I I think I've seen over the course of my career. There was to get under the you know, I'm old. So like with the release of the first iPhone is like a holy, like unbelievable moment when I was like, oh, smooth scroll up and zoom. That's wild. I think when I first put on a VR headset was like a mind blowing experience for me. And and when I had my first conversation with PT was that it was a mind blowing experience for me. I think what we get so excited about here is we get to deliver that mind blowing experience where you just get into a platform, you start to work and it goes, hey, I've noticed 15 things I can go ahead. Speaker 2 And just automate for you want me to do that? You don't have to be technical. You don't have to do anything. You just have to say, yeah, that's exactly what I was looking for. That's really cool. Speaker 1 Yeah, I mean, that's, that's amazing. I think that, you know, two big takeaways for me for again a non-technical person, it's comparing, you know, against the two other approaches which is off the shelf and customization is like the time value, right. The time of value something like BAP provides compared to those two solutions, it's just magnitudes faster and scalability, right. You're talking about something that really is able to grow and scale more with organization. The more you interact, the more you use it. How Liminal Discovered Behavioral Agentic Automation Going back to the the start here, what made you really start be thinking about how agentic AI should be deployed? Was there a specific specific customer or situation that really sparked the colour aha moment? Speaker 3 Yeah, I think we were getting. Speaker 4 More and more questions from customers about Agentic, but Agentic lack any real form or definition to it. So I think our customers were as confused as we were at the time. It would be 2025 for anybody listening right now where I think Agentic really started to capture imagination and hearts and minds in terms of its proliferation nationally, but certainly not in the actual use. But, but very few people actually had a, a reasonable definition around what agentic meant. And I think we started to get these questions inbound naturally, if we were considering what does this mean for a road map internally? How do we drive the next generation of agentic software forward? And because of the, the degree of confusion around the market around how typical agentic frameworks look to me, like the disruption risk for, you know, non complex agentic workflows, we were, we kind of forced us to think a little bit differently. And it was about a week before buying, we came up to the Toronto offices, if you'll recall, where the idea really started to take place, where we, we had an aha moment. And I'll never forget this. Actually, I was sitting in my office in this chair right here. And I was conversing with with Claude at the time and I was going through and saying, you know, hey, I don't think agentic workflows are doing it correctly to the current of the current frameworks have all this broken about them. What what's the right path forward? And it said, you know, and I said, here's again, has all this information from about about Liminal and what we do and said, have you ever considered that you sit on top of all this log data and there's some really great data on how people operate? And it was like, Oh my God, like, yeah, we do. We every tenant has all of the information about how people are interacting with their data through enterprise search. MCP connectors are starting to come to the rise. So we could drive in all this MCP connection connectivity and bilateral actions and read from those particular data sources connected to MCP, like, Oh my God. And then we can see if how people are interacting with their data. Oh my God, we know how people are interacting with Oh my God, we've already built the insights engine, which taps into all of that log data and just gives companies understanding of hey, what's happening across the organization, where problems are existing, where risk risks exist, where opportunities exist. We already built like the component tree here necessary for us to be able to make that leap and start to self assemble agents based on patterns of behavior that exist in how individuals are engaging with generative and their data. We have most of the componentry built and what I was able to do in that moment was break it down their lyrics. Three things you need, you need effective log data, you need observability data. You need to be able to connect to downstream data sources through a single platform, but you need multiple systems of record tapping into that real time data source like a calendar and e-mail and MCP unlock that for everyone. And then you need the third piece, which is the delivery platform, which you need to actually have a platform which end users can engage with. We had all three of those things and it was like, Oh my goodness, we we can build behavioral agentic automation. We can have our platform self assemble automations based on how individuals are working, things that are attentive to their nuanced behavior, things that accomplish the mundane for them and allow them to focus on the parts of their job that they enjoy. And we can do this not based on speculation, but based on real proof that these are the things they're trying to solve for. It was such a moment. Now you remember when I came to that the following week when we met for our our our team meet up in Toronto and I sat with you and went we've we've got like we, we got the future. We're literally we're building it right now and we were still really fleshing out the how and the architecture around it. I'm more thrilled than ever. I mean, it's like I, I told you, I want to deliver that jaw-dropping moment for people and we have the ability to do that again, that spark that ignited a revolution. I feel like we get to deliver that in agentic as well. So it's going to be really fun. I'm already seeing like little inklings of it from what we're building today. And I can tell you it's it's wow. So it's super fun. Speaker 1 Yeah, I, I definitely still remember, you know, that day when when you kind of called it, called me over in, in our Toronto office and that walked me through it and I was like, wow, this is, this is the future. Like this is how it's, it should be. When, when enterprises think about adopting AI, the the current common workflows and approaches are frankly, I wouldn't say broken, but they, they don't meet where where the customers need. That's it. They launched their solutions and kind of speaking on that, you know, from the buyers perspective, right at the end day, you want to also like your, your buyer looked good in front of their bosses and the people who are managing and monitoring their performance and they need to deliver on certain KPIs where maybe some kind of quantitative metrics, right. Do you have a? Speaker 4 Sense in something that's really interesting too though it's this notion of wow factor. I think everybody who touched Jenny I for the first time had that. And then there was all the question marks around what does this mean for my business? And then they sought the traditional ROI metrics when you know, the people who took off the fastest in this for the people who just got started and they started using it and people started to realize there's all sorts of different things that can shortchange research and help me get my job done faster. And whatever role that is, hundreds of use cases that exist across different roles that are micro in their own right. But the macro picture adds up to a lot of efficiency gains. And there's been plenty of studies that can showcase that. There's this problem that Jenny I faces for the vast majority of people in the workforce. And it's this notion that you open the door and you're like, what am I doing here? And people land on this today. What am I supposed to do with Jenny I The first thing that we teach people is how to prompt it. Here's different ways that within the context of your roles, you can show. And then, well, that's interesting. And then that kind of starts to jog the cycle here and people going, but what if we could right away? And this is kind of how we've been thinking about this is like, not only do we have this understanding of behavior across organizations individually, but we understand how roles are behaving specifically. And we can start to a lot for the nuance in every role, but also just out of the gates start to say, hey, Boyang, what are some things that you do on a day-to-day basis? Oh, well, I researched a lot of companies. I talked to a lot of founders and it goes, well, what if I were to automate the corporate research for you based on your meeting schedule? Well, you just ride away, solved the empty house problem. You get there and it says, I know why you're here. I know a little bit about you, not a lot, but enough about you to assume that I can build something for you and for you. You go, that sounds cool, go ahead and do that. And then immediately shows you that proof. OK, you've got a meeting upcoming here in 15 minutes. I've already done the research on the company for you. Here's some data about the founders and it's showing you where all that information came from and what connected sources from your internal systems of record. It's told from external that it's pulled from and you go, that's real, that's cool. Like I'll keep coming back for this and then all of a sudden you keep using it. It's still general purpose generative AI. You still could use it the way you want to, and then all of a sudden it's done 10 other things for you. And for the first time ever, an AI is engaging you versus the other way around and saying, hey, I I know a little bit about you. I've got your back here. Do you want me to go and do this thing for you? Yes, That's sounds great. Totally. It's like a robot coming to your house. You want a full your laundry for you? Do you want me to clean your dishes for you? Yeah, that sounds awesome. Totally. I didn't even know you could do that stuff right here. I mean, I get really excited about that because I just think that's what people are craving. Everyone's wanting that moment now. It's like, cool. AI is really impressive. I use it, I think at this point probably over 100 times a day in my personal life and my business life as well. And I've got agents operating on a separate machine. I even have my dad. I think this is so cool. Call me in the other day and he had like he's like setting up a new lighting system, an override for his existing like old school Lutron system that he put in his house and he bought a Raspberry pie. And he's like, I'm using cloud code and I'm actually going out. Is not an engineer, super intelligent, but he's not an engineer. He was able to build like using your Raspberry pie and cloud code, a system that overrode his like ancient Lutron system. The admins be able to control his lights. That's wild. Let me get to bring those experiences to people, which I do have, but also we do it without ignoring the core needs of a company. They need security, observability, data privacy. They need multi modal flexibility and cost efficiency and something that does the heavy lifting for them. Just like my dad needed cloud code to do the heavy lifting for him on engineering that everyone needs. That saved the really bright engineers that don't need that, but even they use it as an augment. So that's what we all want and we get to deliver that screen. So like I want to, I want people's jaws to drop. And I'm I get excited because the very thought of this drug, like drop my jaw just like you did for you when we told you we were all just sitting there going Oh my God, this is going to be amazing in the early indicators are that, yeah, it's going to be amazing. So it's pretty exciting. Navigating the "Meaty Part" of Enterprise AI Yeah, I mean, it's definitely crazy how how things have quickly changed and called in the last 12 months, last six months, last last few weeks. I mean at the end of a you know, where, where we're focused on right now, at least for luminol is, you know, servicing the regulated space, really flying and servicing that pain point. We talked a bit earlier about, you know, the adoption curve and what we're we're folks are at. What do you think the regulated industries are right now in terms of AI adoption like and where do you think you know when we hit that, you know, when we when we open the floodgates in terms of everyone in the regulated industries adopting AI? Speaker 4 Yeah, look, I think we're broadly speaking at the kind of early phase of of, you know, post early adopter adoptions. We're at the meaty, chunky part of the market where the vast majority of companies live, not the late guys, they're not their only guys. They're right in that middle market and that and I think that's exactly, we're at the beginning of that. And I can tell you that because bicycles are accelerating now, there's still apprehensive. I don't want to Ding up Microsoft too much here, but Copilot was everybody's anchor point on AI inside of enterprise. What we've heard from customers is that it's been a subpar product and it did kind of a service and a disservice probably to the market. The disservice it did to the market was everybody anchored on that as AI. That's my experience with AI and everyone went I and not for me. I can say this empirically because we we deal with these conversations literally every day. The service they did was they released a product that was subpar. Like if they crushed it with Copilot, there would not be a whole lot of breathing room for anybody else in the enterprise. But they didn't. And because of that it created opportunity for other products to have entry point, particularly in those mid market enterprises who want something turn key for them. They want to operate another network layer security platform. They don't want to operate multiple Gennai models and licenses. They don't have the ability, they don't have the time and frankly they don't have the patience to go and deploy and manage all of that. They want something that's turning. We offer that solution for them. So I think we're right at the cusp. Bicycles are accelerating. People are getting to know, hey, I want something different. I don't know how to articulate that. We find that our value proposition starts to really resonate and customers now more so than ever, not because the value positions change because people's notions of what they need to have changed. And I think that just comes with the maturity of the life cycle. Now, we're still at the beginning of that. There was that big dot graph that came out that showed a adoption and was all over LinkedIn and everything else were like what, 3 to 5 million paid accounts or something like that. It was something absurdly low. And the rest are free and then the rest are not using. And it was like, you know, 6 billion not using or something like 6.5 billion not using. And there's a pretty broad swath of folks that are still trying to figure this out. And I can tell you from the training calls, it's not because little luck desire, it's not because they like the intellect. It's just because it's new. Remember our two earlier archetypes. There are people who allow open license policy, which then only the Super user is going to be the ones demanding access to it. And then to the people who close everything off, which means no one gives access to it. And so no one's getting articulate in it. And then you have some CEO that sits on top of that. So everyone needs to be illiterate. Everyone's like, you don't give me the tools to be illiterate. And Copilot is not AI. We have a chance to to continue to kind of get that. But it's fun because we're right at the meaty part of the curve. This is like the beginning of the adoption curve where I think, um, there are a ton of early adopters. Those are the ones we hear a lot about, but really like the vast majority of the market is like, OK, let's get started. Let's go do something with this now. Speaker 1 Yeah, it's all the the graph that you spoke about 6 billion, seven billion people who still haven't adopted. I think we need more than just two approaches, right? Allow everybody there needs to be a bit more choice for the other 6 billion people out there. A Founder's Bet on Vision and Team Resilience Maybe you kind of going back and kind of last few questions back to kind of your experience as a as a founder and and more kind of, you know, servicing some of the founders and entrepreneurs that are listening to this podcast right now. Maybe going back to experience. What was it difficult decision that you made early on whether I don't know on product or go to market or team, that was really difficult at the time, but gosh, it proved to be the right decision now looking. Speaker 4 Back That's a great question buying. There's been so many as someone who's experienced in this too, and I as a personality trait, I tend to be really aggressively behind myself. Like I bet on myself all day long and feel confident. And you will get to moments where I would consider myself unflappable in my decision making the vast majority of time, but you'll get the moments in there. It's like God because it's the right thing to do. It's always the hard ones are around people. But we've had multiple moments too where we started the company as an application layer security platform. We did not start with end user tooling in mind and we also were targeting large enterprise with our platform. We realized about one year into building our platform and taking it to market that we were to try and wait out the sales cycles of these large enterprises. You would die as a company. We would not have the money to survive. Their balance sheet wouldn't allow us to float these 24 month sales cycles and Even so, the pot of gold at the end of that rainbow was not worth the the timeline it took to walk it. This was in December of 24. We had to completely shift our go to market, completely had to rebuild our product to incorporate end user tooling to focus on an end user experience without removing the security piece. We had to redo the math behind how we were going to offer models inside of our platform. We will build a bunch of new tech that didn't exist at the time. We didn't have all the tools, coding tools as well here. So engineering was still full of difficulty. Not that it's not today, but the mundane is not as much stricter as it is today. So our engineering team had to had to undo some tech debt, had to kind of rebuild and they did it quickly, like four months. We made this big shift from going and targeting large enterprises with an application layer security platform to Liminal as we know it today, which is, you know, a secure, multi modal, flexible, cost effective way to deploy generative eye to the masses with, you know, admin controls around security, data privacy, governance, administration and observability, but also the great wonderful end user tools that people have come to use in their personal lives with again, all the multimodal flexibility built in there. That was the gamble. If we if we didn't hit gold on that, we were done. The hard part isn't the leader. You hear like these wonderful tropes from people who are really bright. Like Bezos is a trip that I've always followed, which is like be relentless on vision and flexible on detail. That's his like a big line that he's always promoted events like God, you like have a vision when you start the company and you go this isn't working, but I'm supposed to be relentless on vision and flexible on detail and I'm like, I wanna be my vision as I envisioned it isn't going so well. And if I keep being relentless on this, I may not make it so you can pivot in those moments, um, or you can't. And I wouldn't call ourselves the runaway success by any means, but we've certainly been able to grow extremely steadily for the last year. We've breached some core revenue milestone. We're continuing. We're not bursting at the seams. You have to hire actually because we have no capacity left to take on our inbound now and, and our partner lead flow and some of the bits we made early on are starting to pay off. But man, like they are really bets and you as a leader have to stand behind those. And it's the hardest part isn't necessarily making the decision, standing behind that decision. And that's like that be relentless on vision and flexible and detail. It's that I made the decision. I've made my bed. Now I will sleep. And then but then also going, when is the right time to to look at the data and go, is this actually the right thing that I'm supposed to be doing? If you're a visionary CEO, which a lot of folks are, you're constantly going, well, what else can I go do? What else can I go do? What else can I go do? We got pretty fortunate in that one moment where we go, what else can we do here to have made a bet on going into the building kind of a secure perplexity, overall intensive purposes with better pricing and better security obviously for enterprise and midmarket and not being the right call. And we could have made any number of bets. There were lots of ideas on the table, but that's the one that's stuck. And you're going to have to make decisions with 20% of the data or you want to have 80%. And and that's what we have. And when I think about hard decisions that comes along with how do we staff this appropriately, how we got our go to market team that's set up for this appropriate. We had to let people go and these weren't bad. These are amazing people just hired at the wrong time and those are tough because people are everything. That's all you wanna do is do the right thing by your team and you're just sitting here banging your head against the wall. I'll leave you with this because I just feel like this is such an appropriate way to describe our journey here on this question. At least. You know, Sam Altman said doing a startup is like continuously banging your head against the wall and then you finally find breakthrough. I would add to that quote where it's like continuously banging up your head against the wall. You finally break through the wall and need to find another wall. You got to start banging your head against. He also said, and I agree, the number one, all you want to find an entrepreneurs resilience and my God, I thought it was resilient. And then you get battle tested through these big moments of do or die. And then you find out what resilience really is and you find out what yourself, what you really made of and and your team, what they're made of to. And it's just not easy. None of this is easy, but it's also, I told you earlier, I don't know what else to do. If I'm not flying around with my hair on fire going 100 miles an hour, I don't know how I'm supposed to do. I'm not having fun, that's for sure. Speaker 1 Thankfully you have people like Aaron and Michelle and. Speaker 4 Mark. Speaker 1 Right, that are just the same as you, right? Just feels like they're the people who don't know what they do if it wasn't. Speaker 4 There I agree more. They're the the wind beneath our collective wings for sure, as one on unbelievable team can do for you personally to alleviate the anxiety and the the difficulty in making hard decisions. And I feel so fortunate because I have good founding, resilient leaders around me who understand that I can, I can be completely open with and tell them exactly about the state of the company exactly filling. And they're not going, I'm afraid, because our CEO is stealing anxiety about the current scenario that we're in right now and hasn't actually got a clear path forward yet because they don't have enough data to do so. And so to have a team that you can go and say, hey, I don't have the data right now, but I need to, I've got ideas. Where do you think we should go? What do we think? And then battle test that with them and collectively make a decision to move forward. It's like rowing a boat on your own versus rowing the boat together with poor people. It gives a lot faster and it's a lot easier and you could fight all you want rowing the boat on by yourself and sometimes you'll win. But my God is it a lot easier when you have a team like I do around me. Cannot overvalue that enough. Speaker 1 I was there and probably makes it a lot more fun too. Makes a lot more worth it so. Speaker 4 They're super good people, so it's really fun too. Speaker 1 Awesome. See, even, you know, really appreciate you taking the time to speak with all of us and everyone here listening. Speaker 4 Thanks everyone for listening to me ramble and for letting me spend some time with the guys, and thanks for being great partners.