AI Company Brains and Personal Agents

Speaker 0: You're choosing an animation to generate Speaker 1: more each field build out the next field. Speaker 0: AI, everyone. I I saw the huge hands that were this is your first AI, team for a group. How many of you have ever been to this building before? Okay. We got half a dozen, 8, 10 people. Speaker 0: For all of you who did not raise your hand, this is not the main campus of Chatham as when you walk into this parking lot and think, where am I? The main campus is actually about 3 blocks out and not just for alumni. And we do that because we received a grant a few years ago from the California office of small business advocates, just a record. It was 1000000 bucks over 4 years, and it allowed us to open our doors to the public. So the whole community is able to take advantage of the programs that we have here other than the classes that are taught in Oshkosh. Speaker 0: But everything else that we do here is available to the community at no cost. It's available to you. So I urge you to do that. It's it's something that's very special. You know, I talked to a couple folks earlier. Speaker 0: They had no idea that we did these types of things and that they were available to people like yourself. So, I urge you to take advantage of that. We are looking at starting this summer a workshop series, maybe a half a dozen series of how to use AI in a way that's been meaningful to you at the end of those 6 weeks. So if you have a start up and you're trying to build up to an agent that's gonna be useful that actually you can deploy, that's gonna be what we're trying to do. If you're advanced already and you're trying to make your AI work better for you, we're gonna have, you know, a section for that as well. Speaker 0: We're still developing a new we're gonna have, you know, a section for that as well. We're still developing the program itself, but I just wanna give you that kind of heads up that if you guys are interested in something like that, I would love to be able to connect with you. We may be. Yes. Let's talk. Speaker 0: Yeah. So, you know, again, that's part of what this this program is gonna be is, like, finding some of the best people in that particular area, whether it's, you know, at the very beginning, understanding what custom painting is. It's not always just about showing straight into technology. Learning how to interview somebody so you can understand what, you know, they're really saying. How to listen. Speaker 0: You know, all those things that, you know, a lot of more technical Anyway, so just giving you kind of a a preview that we're gonna be doing on this Tell, sometimes I think we've done this so many times that I forget to maybe talk about the point of the demos and speakers and stuff like that. So let me back up just a tiny bit and just to say that the whole idea behind AI Tinkerers is to shares request directory services, which was crazy. I got out, went to Boeing. I did high performance computing to get a complete flip of the script. I went to Linux sysadmin, eventually a systems engineer. Speaker 0: And after there, I went to we're continuing. I'm a escalation engineer, for Code infrastructure and platform. That's my day job. My true passion is what I'm about to show you, this is what I've been working on and basically spending Jun providing a whole operating system runtime for not just 1 agent, but many agents and people to work in the same environment together. Now the key thing is here that this is all running on local hardware, including this website. Speaker 0: Model to orchestrate and call bigger models, tools, everything else to get the job done. So, so now on the, DX Spark, I'm actually running plan, 3 6, a 27 milligrameter model, and I've used that for a reason. Now the way this works is the orchestrator model, and the agent, the main agent, and it is, going to route basically everything through that orchestrator model. And when, it has a basic well, it has an intent engine to evaluate if it's going to feed deeper reasoning for the task at hand. And if it does, then it reaches out to that reasoning model on the other hardware. Speaker 0: It uses it as a tool. It sends its inputs, it gets read link, and it comes back to the orchestrator and the orchestrator figures out what it has to do next. Oh, do I need to reason on that more? Do I need to search the web, get more context? And so all of this to say, it it's an operating system and I think the industry is moving closer towards, calling them company brands. Speaker 0: That's a big thing. I recently flew home to Ohio, and I have an aunt who runs a small business. She's a photographer, and she's trying to grow her business. But you find that a lot of people running these small businesses have to wear a lot of different hats. So most of the time is if they wanna grow their business, they have to tap into skills, learn new scale, like how you market, how do you find new customers, how do you do stuff like that. Speaker 0: So that really got me thinking, and I changed my entire website actually over the past couple of weeks around this idea of building that can function as that company's brain and put in a package that, anybody with a little bit of hard work can run and has a useful agent. Because you'll find that this little orchestrator agent, has been benchmarked. You can go look, look at and it might not compete with, like, 4 6 and 4 8 and all that stuff now, but it was, competing with, like, 4 5, I think, Opus 4 5, competing with the Frontier models. With the right setup, it's you have to have the right harness, I spent a lot of time architect architecting, how things are gonna communicate and what data I can take from the system operating so I can fine tune the orchestrator model itself so it's just smarter in general. So I have less time, I can less less harness engineering, and more of the fun stuff. Speaker 0: And there's model level. So, since it's running at home, it's connect to the Internet and it'll redeploy everything for me. But I I rebuilt that project with AI about 30 times until I figured out how to develop with AI and how I could, what I needed to what guardrails I needed, how I could make it not completely destroy the code base. That was part of the reason I had to rebuild 30 times. Luckily, now code the agents are really, really good at coding, so you don't have to do that as much. Speaker 0: But, you have to be careful. Final caveat. Believe it or not, all of this is a 100% AI generated. I haven't written a line of code with myself in, like, yeah, at least probably a year almost. But here's a lot. Speaker 0: Okay. So I've talked a lot. Let me just actually show you some stuff. It actually works. I know I don't have a lot of time. Speaker 0: I wanted to show you my back end infrastructure. So I've set up I had a tunnel that runs. I set I built my own identity service that runs, so I don't pay for any services. I just pay for electricity in my AI that I run a small mobile. But let me go to the bot that I made for my aunt after I talk to her. Speaker 0: So there's a front end on my orchestrator. And so I designed this around, her being a small business and, had a short conversation about, like, what her workflow's like. And 1 of the tools she could use is she's had it forever and now she's locked in. CRM tool, Sprout Studio. And it works mostly, but there's these things that don't work about it, and she can't move off the platform. Speaker 0: It they got her in with a lifetime, purchase at this point. But, but here it is. So I just built this simple app, and, this is all based around, a bot. I read this 1, I actually met with somebody who's looking to implement AI in their business. They're a construction firm, about a 150 person company, and they wanna accelerate their business with AI. Speaker 0: And they already don't want to move to cloud. They want to keep it on prem. They wanna keep their data local. 2nd they're every they're already aware of, how much token Roast are increasing, like, for these Orange models. And if you're building your businesses on Frontier models, well, it's just gonna get more expensive. Speaker 0: So I hope you're bringing enough revenue to cover whatever you're spending on it. Yeah. Am I at 0? Well, $200 a month in ClozMax is a little bit of that stuff. Anyway, let me go back here. Speaker 0: Okay. I don't know. Let me try again. I swear it was working before I left, but I had backups. Yeah. Speaker 0: Yeah. Yeah. Yeah. Yeah. Sorry. Speaker 0: We can hand the, take the mic round. Anybody have questions for David? Hi, David. Really appreciate this. Are you, open to using more of the, lecture models instead of hosting those locally? Speaker 0: Sorry. I'm blind. Sorry. Oh, sorry. AI I more interested in well, I mean, no. Speaker 0: Because, I built this whole thing so that I can configure my agent harness to run this entire thing for MAI, and I wanted to do it autonomously. And sure, I could use flagship models for Matt, but for the work workloads loads I'm intending to expand this to, Matt would be Jun exorbitant amount of money. Right? Like, MAI, right now, have, like, over a 100 autonomous routines that run-in the background on the System. And most of what they're doing right now is just doing things that, AI, collecting data, doing stuff AI building my knowledge graphs, my code Make. Speaker 0: I was AI, I'm running I have agents. I have a system that uses GitHub actions and a local GitHub actions runner. So I here's what I so here, let me show you this real quick. This the back end for this, I built in about 14 minutes using a small on GitHub actions. I had Vite Jun, like, 12 agents using Get, chat c p t 4 l. Speaker 0: And then it produced the entire back end. I had to do a lot of work for the front end, but it did the back end fine. And City did that in 12 minutes 2nd product City, and then I AI my local agents review City, and it review, put a comment on the PRs that, hey. This is what we found that's wrong for the spec, and then we got it fixed. 2nd show yeah. Speaker 0: I don't know. I don't see the need. I think open models are just gonna get better and better, and I just won't need to. How many hours does it take to use to estimate this whole thing up to boot into a Linux VM, and I actually did this just a little bit ago. Show part of this is because I'm doing it all from scratch, it also means I get really good at figuring out how to deploy it. Speaker 0: So I actually build a custom Linux AI though, that put it's actually for this Claude bot that I'm gonna deliver to a customer. Ethics is actually the end. And I just recorded this a little bit ago. Get you see CEO starting DartBot, and so they boot Startup, and then you get City, and you run your networking, and then you have a bot deployed on your network. And then I have a whole thing to, like, register and AI multiple, and also run them on just regular endpoints because the model is small enough if you have it can it can Jun, obviously, not that fast, but then you can have an agent fleet running on any PowerPoint you have depending on what model you're using. Speaker 0: So Awesome. David, thank you so much. Thank you. Alright. And, it's John. Speaker 0: You're already in place. Also, by the way, you'll see if you go to the AI Tinkerers site for Orange County, you can see all the information about the speakers, their, project, if they have, like, a GitHub Stories real you can CEO, some of the things that they've shown. So there's a lot of resources 2nd then connecting as well. If folks have filled out their profiles, you can connect with them on LinkedIn and all that good stuff. So, John, yeah, go for it. Speaker 0: Yeah. So I wanted to share some of the work that I've been doing, and it's all around marketing as well as creating AI digital Media Startup I mean by that is pictures, Stories, or or Ideas, Jun, of course, even songs. Built, specifically, I wanna do this is what I'm hoping to get out of it, or I wanna see you guys make a smile a little bit, chuckle, maybe even laugh because I think it's a fun story. Rethink author, Jun story. Rethink about how if you're pitching something so I'm not a startup founder. Speaker 0: I work at Microsoft, and I've been there for a very long time, like, over 20 years. But I've been in the pitching part of it because I'm asking for a very long time. I'm Can we log in this as 1? So credit card over here. I think if you yeah. Speaker 0: H t my x from that's right. Yeah. And then can use Show, we won. So it's pretty happening. And here's my takeaways. Speaker 0: Make it real quick. The music initially, I did it without the music, and I was like, hey. Can you give me some music and sequence it with the motions that you're doing? I thought I did a pretty good job. The other lesson as I reflected was, you know, this is very personal for me. Speaker 0: And big tech and even tech and general, there's a lot of things to be kinda disappointed a APEX Accelerator. So Apex stands for agentic practitioner excellence. And while I wanted to pitch was getting, a bunch of our architects together, they've lived all across The US and even my wife, AI world is evolving facts and so must we. It's time for a different kind of CSA. AI it's it's a quick story that talks about her story about how she became, when you are able to fine tune, like, a person's personality, how fast it is to be, what they say, all of that is great. Speaker 0: But what's the trade off? The trade off is driving, you don't look at them, and you throw all of my content in there, it's able to create Moh. John, with the content that you're creating, are you having are you writing that or are you having the AI write that? Yeah. Good question. Speaker 0: So what I do is think about the theme Yep. And the outcome that I want. Yep. And I just think, you know, check sheet, the TV, blah. This is what I wanna do. Speaker 0: Help me create the prompt. And then I go back and forth. So the question that I have because I've been running into this myself in the content creation, written content, but similar. The challenge that I have is that I'm starting to sound like everybody else. Live you figured out yet how to I mean, I I'm getting there, but have you figured out ways to actually make it sound AI, well, to your point, I and not AI? Speaker 0: Yeah. No. You know, I can't say I figured it out, but it depends on the content that you It's interesting with the syncopancy of AI that it like, it'll say something and you're like, oh, damn. I sound good. And you're like, oh, wait. Speaker 0: I sound like like that person and that person and that person. So, yeah, if you find other ways to find somebody to be willing to do so. So 1 more question? Yeah. Just a second. Speaker 0: Hey. So, John, thank you for much for that. I love the litig thing. That's fantastic. So 1 of the challenges AI had recently, and I actually put together a music video with a 118 different 8 second slides, is trying to take the vocal from the song and sync it with the actual visuals AI I found was, like, literally impop I just gave up 2nd said, okay. Speaker 0: We're never gonna see engines actually singing. We're Jun gonna see them holding a microphone coding that. Have you seen any ways or to try to get around that synchronization piece of it? Because I can't for the life of me. Have Have you tried Omni Flash? Speaker 0: No. I've not. So So, again, that was it just came out just a couple days ago. 2 months ago, I was gonna be getting the full stuff. I wish I had that because I was Yeah. Speaker 0: That would be great. So it is That's awesome. So fast. Check out Oh, I didn't move on. Awesome. Speaker 0: Thank you so much. Alright. AI. Thank you so much. Alright. Speaker 0: Excellent. AI. Our next, person to present is Travis Johnson, and he's going to using Claude for like, used Claude ever? Used Claude? Okay. Speaker 0: In the terminal? In the terminal? Have you used Claude in the terminal? Okay. Who's your big who's used Claude over a year? Speaker 0: I've been using it for a while. Okay. Who's recently tried codex? Because everyone on Twitter is talking about codex even though you've been on cloud phone now. Okay. Speaker 0: Yeah. So that's why I'm presenting today because I, been really busy working. I've got a AI consulting company and we do train implementation and billing. And, I'm, like, trying to keep up with everything just like everyone else. And everyone's been talking about codecs and how great it is. Speaker 0: And I'm like, you know what? And also how long it runs before it quits. And I'm like, I wanna get some time and actually put this to the test and see if I give a really build, heavy task to both coded and clog with kind of the right engagement Get up, what will be Center? Right? How will they work? Speaker 0: What makes them different? As well, for those who are really dirty into Claude, they ship like Craig, but as a result, their product, surface areas are rough. Right? Like, the CLI is different than the terminal, which or sorry, which is different than the development app. And so it's just, like, not product packaged very well. Speaker 0: And, and Code has done a really good job lately of making their, app on the desktop, super well productized and easy to use. And so that was my goal here, was to try to say, hey. I'll kinda get a claw to run next to Code. You get a feel for, how good or not they are or a big task. And also trying to give them a little bit more, a long task to compare how it does it over time. Speaker 0: So I'm gonna try to present I don't know why I'm presenting a notion, but it's just, that was an easy way for me to capture my thoughts AI. 1 quick question Orange Travis while we get the best Personal up and everything. Yeah. No. I'd agree with you. Speaker 0: I feel like Claude shipped like crazy for, like, a month and a half there, but their product teams aren't communicating right. But they, like, they innovated fast. Coded got a chance to kinda, like, fast follow and productize their offering a little bit better. And we'll see. AI do I do agree. Speaker 0: I feel like there's better polish right now on the Codex Spooner experience for sure. And I AI, that was part of my experience as well. Yeah. So Awesome. Thank you, I miss. Speaker 0: Appreciate it. I think we're doing okay on time. I'd like to welcome Sebastian. Now Sebastian with a company called PostHog, and some of you all may have used this, tool of your development and everything. He's gonna share more about it. Speaker 0: 1 of the things that, we're doing with PostHog as far as AI teams members globally, the founder of the organization is called named Joe AI, and he's got, again, 220 something chapters around the world. He's been doing, some global sponsorships, I guess you could Jun. And PostHog is 1 of those. So they've been doing you're not use yourself, but your your counterparts in the company have been traveling around the world driving presentations like this. And so we'd like to, have you share what we've got for us today. Speaker 0: And I think about 15 minutes or so you're good. Alright. Thank you. Okay. But, so before we get started by show of hands, how many of you guys have used agents on a regular basis Tom vibe enable to have well defined endpoints and build a tool that's bespoke to our needs. Speaker 0: So we built it. And so ARC has been very popular. And if you go to software, we'll be in currently. And our goal is to completely replace your CRM by the end of this year. All of this data is being piped into the CRM and the format that's beautiful for us that we can act out for our customers, there should be a version of this part of our agents. Speaker 0: Right? Everyone of us at VisColl basically, will be have some kind of each that we're using in our data workflow. And Show in very simple terms, this is just a really well architected skill. That's all it is. But it plays to all the things that make an agent do its best work. Speaker 0: It's called CS debug, and I'll basically give it a customer aug project that we talked to. And the question is, do we have any sale feature flags? Which ones? And also, this is the real demo, so I And part of our task today is to find a way to capture that expertise in a digestible format for agents to pull from efficiently and then go and produce your work, do your work for you at a larger scale. Right? Speaker 0: So for us, there are 3 context layers that we know through lived expertise that we always check for customer issues. We have mechanics, So we're finding a ton of data for our customers of course, and we know internally what data matters for specific issues. So these context pillars, these context pillars serve as like places to check for an agent to think, okay. I'm gonna go check the mechanics. But that is basically is. Speaker 0: Engagement is just looking at how customers, interact with post docs. We can look at all that data as well. So it understands what specific users are impacted. And then lastly, instant state is just, it is trying to accomplish a cadence of excellence that's predefined, and it will eventually ask for a promotion, much like we do as humans when we work. And CEO, so that's what I wanted to share with you today. Speaker 0: Thank you. Good stuff. Questions? Speaker 1: What harnesses have you played your skill with, and what LLM models have you been playing with, like, when you Get at your desk? Speaker 0: Just Claude right now. Just Claude? And I'm now thinking about FedEx because of, Speaker 1: Yeah. And what about harnesses? Like, when you have multiple Agents, I mean, you have Gist using Claude UI? Speaker 0: Just Cloud UI. Speaker 1: Okay. Speaker 0: I actually Speaker 1: don't have to watch about harnesses. Speaker 0: So if anyone wants to teach me about that, I'll give you a pair of those soft socks to exchange. Speaker 1: Have you played with linear at all? Speaker 0: Do you Speaker 1: know what Fine AI haven't used it. Because it'd be a great integration for your post hoc. Speaker 0: Very cool. RHA will use it, but, yeah, that sounds great. Speaker 1: It's vital. You guys gotta have that. Speaker 0: Right now, it's generating outreach. I have a very specific tone that I try to work for. And when I'm in a race with customers, I always use emojis, and I use lowercase, very casual. I I wanna make it accessible. And I found that cloud is Speaker 1: Gist Moh on your GitHub to download, or shares this scale available? Speaker 0: Right now, it is internal. My goal is to make it something that is balanced Thank you, Sebastian. Alright. We've got our last but not least speaker. We've got Luke Freiler, who's the CEO and founder of CenterCode, here locally in Orange County. Speaker 0: And, Luke, you're gonna share with us about UserVolley, the AI agent, and driving real validation. So Yeah. All good? AI what I might do is is, I'll do a quick announcement while Luke's getting connected. So I was gonna say it was for the very end of the webinar now in case anybody has to go. Speaker 0: We are super excited to say that we've been trying as a goal of the AI here is to get the next meeting planned in or by the time we have the meeting that we're in. So meaning that, we've got July 21, AI, at the Center code. So thanks to Luke and the team and everything for coding us on July 21. It's another Tuesday AI, and we're super excited to to be there. And it's their brand new office space that they've moved into. Speaker 0: I think that it is still maybe a little, wet. And by then, it'll be dry so that we coded do the the meeting. So, anyway, we're gonna shift over there for July, and then we're we've got a few other venues sorted for the the rest of the year. But, we just wanna let everybody know that, yeah, the next meeting is planned, and we'll have that ready to go. And we'll get the meeting invite launched, and everybody can start signing up. Speaker 0: So 6 weeks from now, we'll see you at the center code HQ. So that take just long enough that you needed to get alright. Is a platform that has over 900 API endpoints. There's over 400 streams to it. There's 12,000 streams right now for language, and it is fully and multilingual. Speaker 0: It is everything you could expect out of an enterprise application and I went to my CTO who's also my cofounder, and I basically said, look, I'm gonna make this thing with the help of the team as I need them, and your job is actually just to clear a path. I need you to do everything you can to reduce any amount of friction at every point, and I will tell you when to start adding that friction back when we need to get more serious about things like security or performance or whatever it happens to be. So we've gone through all that, and I'm happy to answer any questions about what it's been like to to build a large application in as a really interesting thing of, like, yeah, it doesn't matter if it kinda stuck along the way as long as it works. And the next question becomes, and and this is kinda where it's a little interesting to me, is what does that mean? What does doesn't work mean in your context? Speaker 0: So we sat down and we workshop for a couple days, and, you know, workshopping is very different now. It used to be we actually had them reach shut their laptops off because otherwise, they'd just be on Slack. Now we put our laptops on because if you aren't consulting Claude for feedback as you go, you're gonna behind the earth. So we sat for 2 days with product team engineering leadership, and then we got together. And we decided that to us, does it work with 6 things? Speaker 0: It was 3 things that are meaningful to the product side that they would be ultimately accountable for and 3 things that would be, on accountable for. And on the product side, we decided that we were responsible for value. Is it valuable? Does it deliver something to the customer that they need that solves their problem and achieves our vision? Is it usable? Speaker 0: Are they going to be able to use it in order to extract that value? If it's not usable, it doesn't matter. And then is it delightful? And this is something, again, I'm just gonna keep picking on post doc as well as it gets. Nobody has mastered delightful like post doc. Speaker 0: If you guys haven't been to their website, you will go down and grab it. It's amazing. For product, it was those things. Is it valuable? Is it usable? Speaker 0: Is it delightful? And everything that we do, everything that goes out has to check those gates in order to work. Now the engineering side, again, we serve, an enterprise audience. We have to deal with scale. We have to deal with PII and all these things. Speaker 0: AI? Or soft 2? We have to do all these things. Right? The next 1 is, is it best? Speaker 0: Is it performing at the scale we need it to be? It doesn't matter if an AI or human really needs to perform at scale for our customers to be happy. Again, that's what we said linear as our front front star. And then the last 1 is reliable. The very obvious, you know, is it actually kind of performing and doing what's expected? Speaker 0: Are these new types of user experiences and very different things, going to be what, you know, we need them to be? And I think that was weird. Again, this isn't something we put out on GitHub. I'm happy to share it with you the road. We did it for us. Speaker 0: We did it because we felt it fit the context of our business very well. So, another challenge to this, which I'm happy to talk about in the questions, So with that, I'll give you guys a quick tour here. And again, there are 400 screens, so it's not gonna be that extensive a tour that Abby showed everyone people were interested in. But we wanted to start with some language that that we thought hit our audience very well, which is we're actually looking for 2 different types of people. Our our job is to connect companies and their customers to have them experience their products early and help evolve and put out the best product possible. Speaker 0: So sort of a much more modern interpretation of what a traditional beta test would be. And we looked at that as having 2 sort of primary channels in. The first was anybody that wants to shape the products that enrich their lives, we want them as testers. We want people that understand that the purpose of technology is to to enrich our lives, and and they wanna be part of shaping that and have a seat at the table. And then the audience that we monetize, of course, is is is the companies, the companies that are building these products and need these products to be amazing. Speaker 0: And with each of these, and as you might imagine, you know, this is of course built with Foglight with everything else here. This is just a part of the application. This is all 1 big model that is effective with the entire company at this point. And what this goes through is this explains that that we as product people are facing 4 new forces, 4 things that we built this product to solve for. And the the first, and I'll I'll paraphrase these and actually go into the products, you don't have to read it. Speaker 0: The first is that we are seeing team compression everywhere. You know, we work with hundreds of large tech companies. Layouts are, you know, pretty much the the name that came at all of them, and they're expecting their teams to do more with less using AI. So that's the first sort of challenge we we wanna have to deal with. The second was what we're sort of referring to as role collapse. Speaker 0: It's not just that I have fewer people, it's that I actually have fewer distinct roles, and I now have more people operating as generalists as opposed to specialists, and therefore they have to fill skill gaps that they didn't have before. And again, they're expecting to use agents to do this. The next 2, which again, everyone in this room will experience different parts of is this AI engineering thing is new. It moves very quickly. Things change very rapidly, and there are no proven frameworks grounded to the right. Speaker 0: So there's a lot of risk in that in putting out any sort of new product, feature, or release. And then finally, this amazing AI thing is allowing us, ironically, to create entirely new user experiences. And those are untested. We don't know how people are gonna interact. And I think we generally assume that the future is not just a prompt because most people aren't sort of creative enough and and driven enough to to use that prompt effectively. Speaker 0: But it almost certainly looks different than what we had before. And there's been a bunch of talk or anywhere in the world. They can hyper target people. They can run different surveys that turn into whatever sort of insights they want and integrate with virtually every tool you you can imagine. This thing again is, somewhere around 800 I'm sorry. Speaker 0: 400 different, screens wide. Huge. And it is not something I I look forward to demoing in just a few minutes. I'd rather spend some time answering questions and whatnot, and I can drive it as VP. But it is a big piece of software developed from scratch using this new methodology to replace something that we have 25 of big business of doing. Speaker 0: And, again, this is a platform used by, you know, most of the biggest companies in in tech, To do this, we took all of that and did build our entire own harnesses, our own skills. I love seeing the most on skills. We have the same things behind the scenes here, running between virtually everything. And it's just been a phenomenal fun journey. And, again, I I don't know where to go with this exactly because usually when I speak about this, it's a couple hours longer. Speaker 0: But if you haven't answered any questions, talk about any part of the journey, any of the tools that I have approached to actually anything from the team to hiring to to anything and everything. We've been pretty deep in the AI since day 1, and and we're not selling that. But can you actually clarify something that I may have misheard? Sure. You said that you'd let your CTO just kinda stand back and clear the way. Speaker 0: And I don't know if I've read that right. Was your first experiment, like, a lockdown sandbox? Like, you weren't worrying about security and this and that? You're using any Show the the very first version, was literally I I had this sort of epiphany. I was on the flight, back from Japan, family trip, and I prompted out everything I could think of to to build a prototype. Speaker 0: And at this time, I was brand new to all of this. My CTO introduced me to Cursor a couple months earlier and and given, you know, my day to day, I didn't have a lot of time kind of front end stack. I took that to the team, to engineering, and said, okay. This is the direction. This is where we're going. Speaker 0: And then we did a lot of research and a lot of work into, okay, what is the real stack gonna be? Where do we wanna go? And then we spent another 2 weeks building, came up with something that was kind of been where it was so much in that few weeks that we had to start over. That was mid to late December. And then the beginning of January is when we said, okay, we're comfortable enough and don't feel the need to rebuild it from scratch, which is the thing you're gonna wanna do a 100 times initially, that we just started going forward. Speaker 0: And his job, what I know by that, is our traditional development processes are as rigid as any other. We have many tools, but I could do it at such a pace that that by the time, you know, I approved something in our audit, I could move on very quickly. And he has been phenomenal about this. The the difference between T and I, we're both cofounders, is he is very much the with the pure engineering path. He's the guy that spends his time in security reviews and soft stuff, which would drive me crazy. Speaker 0: I went more the product side and and kind of took on the the CTO moniker in the company. And out of the 2, I actually felt far more empowered, as a product person with this new set of tools versus engineers with this new set of tools. And, you know, it's it's a whole other story that 1 of the first things we did at CenterCode when we fell in love with this and kind of wanted to onboard all of the product and all of engineering is we created this thing called Center for Glass and and you can go to labs.centercode.com. The very short version of the very long story is, through AI Aviso, that we would use in a marketing model to attract leads. And we sort of created this big company wide hackathon. Speaker 0: And what we very quickly realized is our engineers were generally very bad at like, what we were doing even 6 months ago when we were kind of planning this out was, yeah, we're gonna need engineering really early and they're gonna have to be very deep. But in the back of my mind, I was hoping, and it it basically happened, that we were really skating to where the buck was going, and we knew that the models were gonna get better. We knew and and we just sort of got another plan. If they didn't get better, it was engineering was gonna come in, and then I was gonna have to have my whole team, you know, jump in. That didn't happen. Speaker 0: The models got so good that we and we got somebody using the haven't had to do as much and therefore our engineering team is actually still primarily on our legacy products, our traditional products, and they have taken on the refactoring project, almost just to see if they can do it at this point exclusively with AI, which when we first started that refactoring project, AI couldn't you're gonna be able to write a few lines on what's gonna happen for you. And we're getting closer to that, Matt. Yes. I think, you said Speaker 1: teams were gonna be utilizing your software here. AI? And I have a feeling Matt gonna be more super agents of the teams larger gonna be going into your deal. So there's not gonna be a human touching this. Matt what I Make, super agents that use all Frontier models. Speaker 1: And I can see a team building the super agent and then utilizing this. Speaker 0: I don't think so, actually. I I I temporarily didn't think so. So so 1 of the 3 conversations that I had that led me on this path was actually about that. There was a a CEO of a a very large company in the testing space that that basically told me over dinner that he expected within 2 years he wouldn't have any users. And that all of his users would be agents, basically. Speaker 0: And the next time we spoke, he had shifted his perspective to, you know, we're still gonna make the software. We're just gonna make the agents to use the software. And that's actually what this is. What, you know, I obviously have a shot or never even show anything, is when you go to to do anything material in the project, it is all just tell me what you're trying to achieve and I'm gonna go do it all for you and you get to choose where you wanna be. Because 1 of the things I believe that and I I think everyone in this room should probably take this to heart is there's not gonna be an overnight transition to everybody just trust AI. Speaker 0: So I believe we're gonna have a trust path that we're gonna have to start with them picking and choosing and seeing what's coming in and improve things, and then eventually they're gonna take their their hands off the wheel and let EXcellence. Cool. Thank you. MAI AI we I I know we could have done a lot of work with all the folks who demoed tonight, but I had this idea that I'm gonna be paying Mario, like, a certain amount of minutes if it were not out. So I think what might be awesome next time is because we have so much we have to be CEO