1 00:00:14,545 --> 00:00:15,210 Hey everyone. 2 00:00:15,210 --> 00:00:18,120 I hope you are having a great start to 2026. 3 00:00:18,120 --> 00:00:20,970 You may have noticed it's been a minute since I've recorded, and that 4 00:00:20,970 --> 00:00:24,660 is because I've been going deep on AI topics like many people have been, 5 00:00:24,660 --> 00:00:28,830 I even started a new Substack called AI Builders Blog, if you wanna see what 6 00:00:28,830 --> 00:00:30,390 I've been thinking about and working on. 7 00:00:30,523 --> 00:00:33,583 But I wanted to record this episode right now because I think we 8 00:00:33,583 --> 00:00:35,736 are at a real inflection point. 9 00:00:35,867 --> 00:00:39,737 AI has evolved from an interesting app to converse with into a tool 10 00:00:39,737 --> 00:00:44,577 that can take on actual meaningful work, and that is a really big shift. 11 00:00:44,895 --> 00:00:47,625 The other thing that I've noticed is that most AI content out there 12 00:00:47,625 --> 00:00:51,594 isn't really geared towards people like me and perhaps people like you. 13 00:00:51,872 --> 00:00:52,412 It's aimed. 14 00:00:52,548 --> 00:00:57,249 At either researchers focused on really nitty gritty technical details of how 15 00:00:57,249 --> 00:01:01,796 models work, or GTM engineers who are focused just on outbound automation. 16 00:01:02,156 --> 00:01:06,536 But I don't see a lot for system builders or for operators who need to bring AI into 17 00:01:06,536 --> 00:01:10,679 complicated corporate environments who need to deal with cloud tools and security 18 00:01:10,679 --> 00:01:14,073 and knowledge that can't live locally a whole bunch of other constraints. 19 00:01:14,126 --> 00:01:17,936 There is a real opportunity right now to help fill that vacuum and also to 20 00:01:17,936 --> 00:01:22,822 help define what those best practices look like, what that playbook is, and 21 00:01:22,822 --> 00:01:24,232 I wanna be part of that conversation. 22 00:01:24,658 --> 00:01:27,748 And so for this episode, I'm joined by Lily Lowe, she's a systems and 23 00:01:27,748 --> 00:01:31,258 operations leader who's kind of been on a parallel path, rolling 24 00:01:31,258 --> 00:01:34,918 out some really impressive projects this year, thinking deeply about ai. 25 00:01:35,158 --> 00:01:38,398 And we're gonna talk about what's happened, what has happened over the 26 00:01:38,398 --> 00:01:41,592 past 12 months, what does it really mean, and what should you be thinking about 27 00:01:41,652 --> 00:01:45,574 in the year ahead to start to enable these capabilities in your own context. 28 00:01:45,650 --> 00:01:46,820 I hope you enjoy the discussion. 29 00:01:46,820 --> 00:01:46,850 I. 30 00:01:47,197 --> 00:01:49,764 Justin: Lily, thanks for being here today. 31 00:01:50,259 --> 00:01:54,848 You know, it's funny, one of the reasons I wanted to chat with you specifically 32 00:01:54,848 --> 00:01:59,348 on this topic, it feels like in some ways we've been fellow travelers down 33 00:01:59,439 --> 00:02:04,203 a very similar path you're working on a lot of cool and interesting topics It's 34 00:02:04,203 --> 00:02:08,853 been a lot of fun, to share notes and to see you start publishing your substack. 35 00:02:08,913 --> 00:02:14,163 Last year, applied AI for mops And then we, we both published kind of like a. 36 00:02:14,668 --> 00:02:19,268 You know, 2025 retrospective, 2026, like looking forward kind of thing, 37 00:02:19,268 --> 00:02:22,668 like within an hour of each other, just sort of by kind of weird coincidence. 38 00:02:22,718 --> 00:02:23,798 and so I just thought it would be cool. 39 00:02:23,798 --> 00:02:26,853 Like, I felt almost the need for myself to just sort of reflect on 40 00:02:26,853 --> 00:02:28,474 what's been going on in my own work. 41 00:02:28,714 --> 00:02:32,310 And so I thought it would be cool to do it together and to hear from you too. 42 00:02:32,790 --> 00:02:36,360 Maybe just to start, like, like you go first and I'll, I'll give my take two. 43 00:02:36,910 --> 00:02:39,820 20, 25, like what was, what was your journey like with ai? 44 00:02:39,820 --> 00:02:43,829 Were you coming into 2025 already, guns blazing or was it something 45 00:02:43,829 --> 00:02:46,260 that really emerged as a focus for you over the past year? 46 00:02:46,816 --> 00:02:50,686 Lily Luo: yeah, it was something that definitely emerged. 47 00:02:50,716 --> 00:02:56,266 I think I started really diving into this in May of 2025. 48 00:02:56,266 --> 00:02:59,896 So it's been, what, seven or so months now. 49 00:03:00,236 --> 00:03:02,846 I'd always been interested in ai. 50 00:03:02,846 --> 00:03:09,006 I use, all the LLMs, but something really struck a nerve when, you know. 51 00:03:09,846 --> 00:03:12,936 was a, a workflow that we needed to solve for an automation 52 00:03:12,936 --> 00:03:14,226 that we needed to implement. 53 00:03:15,216 --> 00:03:19,176 I started thinking with my ops background, I could definitely use 54 00:03:19,176 --> 00:03:23,106 Zapier, I could stitch it with AI and let's just see what happens. 55 00:03:25,026 --> 00:03:26,886 that really took off. 56 00:03:26,886 --> 00:03:31,146 I spent so much time, first working in Zapier, low code, 57 00:03:31,506 --> 00:03:34,066 kind of low stakes, and then. 58 00:03:35,736 --> 00:03:40,746 Hours turned into days, turned into weeks, and really started, coming up 59 00:03:40,746 --> 00:03:43,116 with what we call this analysis dossier. 60 00:03:43,476 --> 00:03:47,706 That was kind of the genesis of all of my AI tools. 61 00:03:48,046 --> 00:03:52,146 at the click of a button, it generates, a full account research report. 62 00:03:52,356 --> 00:03:59,366 It pulls 10 Ks earnings calls, analyzes A company's priorities, challenges, and then 63 00:03:59,366 --> 00:04:02,066 generates even like a one page PowerPoint. 64 00:04:02,366 --> 00:04:06,116 So all of that took a lot of time to kind of architect build. 65 00:04:06,576 --> 00:04:10,626 but that's what kind of hardened, my skills that I've learned 66 00:04:10,626 --> 00:04:12,636 and evolved over all this time. 67 00:04:13,056 --> 00:04:18,516 But it wasn't just kind of building this workflow, it's what I really learned. 68 00:04:19,336 --> 00:04:25,456 building it and realizing that AI is not just a chat bot that helps you brainstorm. 69 00:04:25,816 --> 00:04:28,066 It really lets someone like you and me. 70 00:04:28,851 --> 00:04:31,731 Or ops people in general, not an engineer. 71 00:04:32,031 --> 00:04:37,941 Build tools, that can integrate into your workflows, solve problems at scale. 72 00:04:37,941 --> 00:04:43,541 And, you know, ops people have the systems thinking can design these workflows and 73 00:04:43,541 --> 00:04:47,201 the coding gap is now solvable with LLMs. 74 00:04:47,231 --> 00:04:51,371 So, you know, we might not write the most robust code. 75 00:04:51,781 --> 00:04:55,651 But if we understand the MarTech stack, the architecture, what the 76 00:04:55,651 --> 00:05:00,241 business needs, we can create these solutions that are custom to our 77 00:05:00,241 --> 00:05:02,101 own company and our own process. 78 00:05:02,161 --> 00:05:03,541 And that's invaluable. 79 00:05:03,541 --> 00:05:08,821 So that really, brought me from just chatbot use that most people 80 00:05:08,821 --> 00:05:14,411 are doing to what, the boundaries are, which seem to be endless. 81 00:05:14,783 --> 00:05:15,153 Justin: Yeah. 82 00:05:15,333 --> 00:05:16,293 I couldn't agree more. 83 00:05:16,353 --> 00:05:19,413 and it's funny, just as you were describing, there are a lot of synergies 84 00:05:19,413 --> 00:05:23,613 with My own journey over the past year, like when I was thinking back because it 85 00:05:23,613 --> 00:05:29,723 was around, December of 2024, so just over a year ago that I really first started 86 00:05:29,723 --> 00:05:32,753 thinking about agents and it's so weird 'cause agents, it's like, it feels like 87 00:05:32,753 --> 00:05:36,643 we hit peak agent and now we're like more in terms of the discourse around it. 88 00:05:36,643 --> 00:05:39,373 Not necessarily in terms of the technical implementation of it, but 89 00:05:39,823 --> 00:05:43,603 just that hype bubble it feels like must have been around forever, but. 90 00:05:44,033 --> 00:05:47,453 a year ago, that was still a relatively newish concept. 91 00:05:47,453 --> 00:05:50,303 I think, like in the LinkedIn sphere, I remember like Scott Brinker really 92 00:05:50,303 --> 00:05:54,082 wrote about it, and that's, that's the moment for me that I was like, oh, okay. 93 00:05:54,082 --> 00:05:58,673 Like I was kind of, not AI skeptic, but I was interested in ai, but 94 00:05:58,673 --> 00:06:02,303 I was also like, are we just like creating more bad blog posts? 95 00:06:02,303 --> 00:06:04,133 Like is that what we're doing here with ai? 96 00:06:04,163 --> 00:06:05,813 Because I don't really know that. 97 00:06:05,813 --> 00:06:10,803 I see the value of that and, and, and then thinking about how we can, apply 98 00:06:10,803 --> 00:06:15,423 to automation is just like you, where my system's brain was like, oh, now I get it. 99 00:06:15,423 --> 00:06:16,563 It's the next layer. 100 00:06:17,348 --> 00:06:19,118 Of the thing that I've always loved to do. 101 00:06:19,688 --> 00:06:23,168 And and I didn't do a really big project until May, just like you. 102 00:06:23,218 --> 00:06:23,818 it was funny. 103 00:06:24,268 --> 00:06:27,808 And similarly, it was like an account research, email writing thing. 104 00:06:27,808 --> 00:06:29,008 I think the main difference. 105 00:06:29,443 --> 00:06:31,153 Is that yours turned out really well. 106 00:06:31,153 --> 00:06:34,843 And for me, that project was a bit of a flop, not necessarily 107 00:06:34,843 --> 00:06:39,003 because of, lack of effort, but, mostly just wrong tool for the job. 108 00:06:39,063 --> 00:06:41,373 You know, we had a platform kind of already selected and I was like, 109 00:06:41,373 --> 00:06:42,993 I'm gonna use like what I have here. 110 00:06:42,993 --> 00:06:44,403 And, which was dust. 111 00:06:44,403 --> 00:06:46,913 And I really liked dust and we can talk about tooling, but I was 112 00:06:46,913 --> 00:06:51,453 trying to use a conversational assistant for a more linear workflow. 113 00:06:51,768 --> 00:06:54,318 You know, where predictability and reliability were the 114 00:06:54,348 --> 00:06:58,268 key, and it was a fundamental mismatch of tool and application. 115 00:06:58,738 --> 00:07:02,398 and so the use case flopped, but that is where I also got a lot of learning 116 00:07:02,548 --> 00:07:04,618 from around like, oh, all right. 117 00:07:04,718 --> 00:07:05,738 if you can do a tightly. 118 00:07:06,428 --> 00:07:11,268 Bound workflow with narrowly scoped AI steps, you're gonna be a lot 119 00:07:11,268 --> 00:07:15,608 happier than just, setting an agent free on a complex use case like that. 120 00:07:15,608 --> 00:07:18,308 And, you know, just a lot of learnings that we can talk about. 121 00:07:18,528 --> 00:07:22,478 and then from that into other projects that, really started to 122 00:07:22,478 --> 00:07:25,207 feel like I was getting my legs under me and how to work on this. 123 00:07:25,578 --> 00:07:29,478 and you alluded to the systems role, and I'm curious, one of the meta insights 124 00:07:29,478 --> 00:07:31,158 that I've had and that I think you've had. 125 00:07:31,638 --> 00:07:35,538 Is like, alright, there's, you know, these data scientists and, and ai, you 126 00:07:35,538 --> 00:07:37,578 know, PhDs writing about this topic. 127 00:07:37,578 --> 00:07:40,158 There's there's coders writing about it. 128 00:07:40,158 --> 00:07:42,828 There's people posting N eight N screenshots on LinkedIn. 129 00:07:42,828 --> 00:07:45,198 You don't know if they've ever actually built anything that they're talking about. 130 00:07:45,468 --> 00:07:49,457 And then there's like this, this almost silent majority I feel of systems builders 131 00:07:49,457 --> 00:07:53,238 that are maybe experimenting but doesn't feel that they've had a strong voice, at 132 00:07:53,238 --> 00:07:54,918 least in the circles that I'm in until. 133 00:07:55,443 --> 00:07:56,553 Hopefully recently. 134 00:07:56,893 --> 00:07:57,883 why do you think that is? 135 00:07:57,883 --> 00:08:00,973 Like, why is, are not more people really talking about 136 00:08:00,973 --> 00:08:02,923 this specific application of it? 137 00:08:03,613 --> 00:08:05,083 Lily Luo: It is a great question. 138 00:08:05,183 --> 00:08:11,073 There's so many different perspectives, like you said, coders, engineers, 139 00:08:11,133 --> 00:08:16,396 researchers, what I do in the workplace is where things get real 140 00:08:16,876 --> 00:08:21,016 and I need to be able to apply those things and I have specific problems 141 00:08:21,016 --> 00:08:22,696 and I have a builder's mindset. 142 00:08:23,686 --> 00:08:29,866 trying to understand and synthesize that to what I can do at work 143 00:08:29,926 --> 00:08:31,336 is what I've been trying to do. 144 00:08:31,816 --> 00:08:35,986 And there isn't a lot of people that I can find doing that besides you. 145 00:08:36,015 --> 00:08:36,915 Maybe a handful. 146 00:08:37,356 --> 00:08:38,466 Of other people. 147 00:08:38,946 --> 00:08:43,270 And I think, you know, just the change management of all these things. 148 00:08:43,270 --> 00:08:46,061 I mean, I saw that every three days. 149 00:08:46,461 --> 00:08:47,091 there's a new AI model. 150 00:08:48,516 --> 00:08:49,686 It's really hard to keep up. 151 00:08:50,766 --> 00:08:55,026 I think people have their own ideas too about what AI is and what 152 00:08:55,026 --> 00:08:57,396 isn't, and there's no framework. 153 00:08:57,576 --> 00:09:00,136 There's barely, a framework for AI in general. 154 00:09:00,136 --> 00:09:02,776 There isn't one for operations. 155 00:09:03,226 --> 00:09:10,316 So, I think it's just so frontier, so new, getting people like us, even my own 156 00:09:10,316 --> 00:09:12,776 team and, people at work who've seen this. 157 00:09:13,136 --> 00:09:16,616 You know, getting them to shift their mental model is 158 00:09:16,616 --> 00:09:19,136 hard when there's no playbook. 159 00:09:20,216 --> 00:09:24,096 and when there's no, experts out there that you can follow. 160 00:09:24,096 --> 00:09:27,666 So it's great to have discussions with you and discussions like these 161 00:09:27,666 --> 00:09:29,176 to kind of start figuring that out. 162 00:09:29,819 --> 00:09:33,029 Justin: That's one of the parts of this that makes me feel so excited, 163 00:09:33,029 --> 00:09:37,049 to be honest, because, you know, when I go back to like, uh, I'll, I 164 00:09:37,049 --> 00:09:40,413 will, I will date myself, but like, the early 2010s, like where I came up 165 00:09:40,413 --> 00:09:44,433 through the Marketo community where it was this super exciting time. 166 00:09:44,433 --> 00:09:48,333 All of a sudden marketers, 'cause there really was no like marketing operations 167 00:09:48,333 --> 00:09:50,133 and marketers were handed this toolkit. 168 00:09:50,433 --> 00:09:54,333 It's kinda automation toolkit and people were innovating and coming up 169 00:09:54,333 --> 00:09:57,813 with architectures and like posting stuff and it just felt like this. 170 00:09:58,293 --> 00:09:59,703 We're just figuring things out. 171 00:09:59,703 --> 00:10:02,313 we're building the blueprint and the roadmap as we're doing it. 172 00:10:02,853 --> 00:10:05,343 And it's like that now, but like times. 173 00:10:05,478 --> 00:10:06,228 A hundred. 174 00:10:06,318 --> 00:10:10,518 'cause the tools are so much more powerful and, that's why I actually really love 175 00:10:10,518 --> 00:10:13,998 the, the title of your stack Applied ai, because that's, that's what it really is. 176 00:10:13,998 --> 00:10:17,328 It's not like there's the theory, there's the potential, there's the 177 00:10:17,328 --> 00:10:19,968 hype, and then there's like, alright, how are we actually gonna take this 178 00:10:20,328 --> 00:10:24,198 and put it to work in a, in a business context to solve business problems and 179 00:10:24,198 --> 00:10:26,808 do it robustly and safely in credibly. 180 00:10:26,858 --> 00:10:29,788 and that is so much fun, you know, To figure out, not to make it too 181 00:10:29,788 --> 00:10:33,623 grandiose what we're doing, but I think it is, there's an a, a certain 182 00:10:33,623 --> 00:10:35,183 significance to it that I really like. 183 00:10:35,463 --> 00:10:39,513 Lily Luo: that's the best part is having fun exploring, feeling like you're on the 184 00:10:39,513 --> 00:10:42,123 edge, on the frontier is a great feeling. 185 00:10:42,123 --> 00:10:45,213 So, you know, that kind of fuels my passion for it, but 186 00:10:45,213 --> 00:10:46,563 yeah, you're exactly right. 187 00:10:46,713 --> 00:10:50,973 I remember when marking automation tools were coming up and there are all these 188 00:10:50,973 --> 00:10:53,393 tools, but there wasn't a framework. 189 00:10:53,393 --> 00:10:57,503 And I remember, um, Edward Unthink this scrim, 190 00:10:58,478 --> 00:10:58,698 Justin: Yep. 191 00:10:58,913 --> 00:10:59,333 Lily Luo: he had 192 00:10:59,398 --> 00:10:59,618 Justin: Yep. 193 00:11:00,233 --> 00:11:02,423 Lily Luo: methodology and I'm like, oh my gosh. 194 00:11:02,573 --> 00:11:05,253 He has, scalability. 195 00:11:05,763 --> 00:11:09,273 Can you remember Robustness modular? 196 00:11:09,333 --> 00:11:11,223 I'm like, okay, this unlocks everything. 197 00:11:11,313 --> 00:11:14,613 And I started kind of following that methodology and making 198 00:11:14,613 --> 00:11:16,053 sure the systems were robust. 199 00:11:16,353 --> 00:11:19,623 It's kind of the same thing like you said, but yeah, times a hundred, 200 00:11:19,743 --> 00:11:22,923 times a thousand times infinity even. 201 00:11:22,983 --> 00:11:24,183 So that was a really great point. 202 00:11:24,587 --> 00:11:26,667 Justin: Tell me about the tech stack that you're working with. 203 00:11:26,667 --> 00:11:30,327 I think you have access to, a caliber of tools that some people may not have 204 00:11:30,327 --> 00:11:33,623 access to, which has driven some of the neat things that you've been able to do. 205 00:11:33,623 --> 00:11:34,793 So tell us about that. 206 00:11:34,943 --> 00:11:39,443 Lily Luo: Yeah, so I do have paid subscriptions to all three Major 207 00:11:39,443 --> 00:11:42,713 L lms, so Claude Gemini Chat, GBT. 208 00:11:43,083 --> 00:11:46,173 honestly, it's kind of like a rotating wheel of which model 209 00:11:46,173 --> 00:11:48,243 is best every month or so. 210 00:11:48,453 --> 00:11:53,333 So I wanna try and see, you know, what the, companies are putting out. 211 00:11:53,693 --> 00:11:55,523 But honestly, Claude Code. 212 00:11:56,003 --> 00:11:57,683 Is the one that I use most. 213 00:11:57,843 --> 00:12:00,153 I'm not an engineer, like I said, I'm not a coder. 214 00:12:00,153 --> 00:12:04,383 I don't have a computer science background, but it just lets me 215 00:12:04,383 --> 00:12:10,923 build tools, write python scripts, debug issues, and really just learn 216 00:12:10,923 --> 00:12:16,623 how things work by doing, it's been transformative for someone like me who. 217 00:12:16,978 --> 00:12:20,888 is a systems thinker, who maybe doesn't read instructions 218 00:12:20,888 --> 00:12:22,778 and tries to learn by doing. 219 00:12:22,998 --> 00:12:25,098 I think that's been the game changer for me. 220 00:12:25,168 --> 00:12:29,518 I use Gemini for my autonomous agent, which I'm sure we'll 221 00:12:29,518 --> 00:12:30,688 get into at some point. 222 00:12:31,678 --> 00:12:36,448 nano Banana from Geminis, you know, the best for visual generation 223 00:12:36,838 --> 00:12:39,458 and Chad, GBTI use mostly at work. 224 00:12:40,198 --> 00:12:41,518 Through copilot. 225 00:12:41,548 --> 00:12:44,848 It's so great at finding documents and files. 226 00:12:45,188 --> 00:12:50,038 I do use Azure AI Foundry for a lot of my own custom deployments. 227 00:12:50,468 --> 00:12:53,138 I use Zapier for automation. 228 00:12:53,348 --> 00:12:58,568 A lot of my workflows run through Zapier, you know, using Azure and whatnot. 229 00:12:58,988 --> 00:13:03,058 I've been wanting to try N eight N, but you know, Xavier gets the job done. 230 00:13:03,438 --> 00:13:04,998 And then I use VS. 231 00:13:04,998 --> 00:13:11,158 Code GitHub, for kind of the heavier, coding in local data work. 232 00:13:11,188 --> 00:13:17,558 But yeah, cloud code really unlocked that capability that, pushed me from ideation 233 00:13:17,918 --> 00:13:20,318 into actual tool and workflow building. 234 00:13:20,681 --> 00:13:21,911 Justin: I'll click on that for a second. 235 00:13:21,941 --> 00:13:25,201 'cause I know cloud code has been, it's been such a huge topic. 236 00:13:25,281 --> 00:13:28,501 I published a piece on cloud code and some of the experimenting that I've been doing, 237 00:13:29,001 --> 00:13:32,261 yesterday, and actually spent a good chunk of the holiday working with cloud code 238 00:13:32,261 --> 00:13:37,281 and similar tools because the mania around it has been so fierce and well deserved. 239 00:13:37,566 --> 00:13:38,826 When you see what it can do. 240 00:13:39,156 --> 00:13:42,906 But for, for folks that are like cloud code, that sounds like a developer's 241 00:13:42,906 --> 00:13:44,796 tool, like how do you work with it? 242 00:13:44,796 --> 00:13:47,756 Maybe just sketch out like a day in the life of solving a problem 243 00:13:47,756 --> 00:13:49,946 and why you would turn to that tool and what you would do with it. 244 00:13:50,091 --> 00:13:53,211 Lily Luo: Yeah, and you know, it wasn't just instant. 245 00:13:53,461 --> 00:13:58,681 there was a path I started with kind of Zapier not knowing how to code and 246 00:13:58,681 --> 00:14:04,171 playing around low code tools and then kind of gradually evolved into that. 247 00:14:04,171 --> 00:14:09,761 So there's a lot of time kind of in between the initial, foray into 248 00:14:09,761 --> 00:14:11,981 all of this, into where I am now. 249 00:14:12,491 --> 00:14:16,236 but yeah, it lives, it's always open for the most part. 250 00:14:16,296 --> 00:14:21,336 On my desktop, I have all my AI projects in my local files. 251 00:14:21,426 --> 00:14:26,886 I use Claude and keep everything very organized, so I might have my analysis. 252 00:14:27,376 --> 00:14:32,046 Or, account dossier project in one file, even, you know, my chat 253 00:14:32,046 --> 00:14:36,516 bot in another file, or a banner generation tool in another file. 254 00:14:36,516 --> 00:14:42,776 And so what I do is, I come up with ideas, I explain what my goals are. 255 00:14:43,076 --> 00:14:47,336 I do a little digging in terms of what I want the architecture to be. 256 00:14:47,706 --> 00:14:51,896 you can start off by just asking, a generic prompt, but I found 257 00:14:51,896 --> 00:14:53,606 that that doesn't work very well. 258 00:14:53,831 --> 00:14:59,501 In the end, it saves a ton of time if you really do plan and think structurally 259 00:14:59,501 --> 00:15:01,931 and architecturally what you wanna build. 260 00:15:03,071 --> 00:15:07,181 and if you're comfortable with that, Claude can do amazing things. 261 00:15:07,181 --> 00:15:11,301 I know it is, very coder heavy and you're in this terminal. 262 00:15:11,301 --> 00:15:15,801 It can be scary, but because you can talk to it with natural language. 263 00:15:16,206 --> 00:15:18,696 It's really not that scary when you get down to it. 264 00:15:19,536 --> 00:15:20,766 so that's how, yeah. 265 00:15:20,886 --> 00:15:25,176 I typically use it and I've just been able to gain so much advantage. 266 00:15:25,176 --> 00:15:29,826 Like I can do things so fast with it without, you know, copy and pasting 267 00:15:29,836 --> 00:15:32,596 It does it all in, in my console. 268 00:15:33,259 --> 00:15:37,219 Justin: And then just to understand how that workflow goes from there. 269 00:15:37,249 --> 00:15:39,529 'cause some folks out there might be thinking, I'm gonna buy like an 270 00:15:39,529 --> 00:15:43,279 agent platform and then I'm gonna like build my agents there, configure them. 271 00:15:43,599 --> 00:15:48,909 Within that UX, you are, if I understand correctly, writing agents more like in 272 00:15:48,909 --> 00:15:52,029 script, like you're writing them in in Python or using a framework, you'll tell 273 00:15:52,029 --> 00:15:56,829 us maybe a bit about that, and then you're deploying them to a cloud server to run. 274 00:15:56,954 --> 00:16:00,284 a very developer like workflow, which I think is probably why 275 00:16:00,284 --> 00:16:03,434 Cloud Code is working so well for you in these use cases. 276 00:16:03,514 --> 00:16:05,404 Lily Luo: Yeah, you're absolutely right. 277 00:16:05,454 --> 00:16:08,334 my first goal ultimately is to learn. 278 00:16:08,724 --> 00:16:12,984 So rather than buy a tool that solves the problem, I tend to wanna build it 279 00:16:12,984 --> 00:16:17,274 myself, learn from it, and then see, you know, maybe there are tools that I 280 00:16:17,274 --> 00:16:19,044 don't have to maintain that are better. 281 00:16:19,374 --> 00:16:24,414 So that's what attracts me to this, more of a developer kind of centric mindset. 282 00:16:25,264 --> 00:16:29,764 building my agent exactly like you said, kind of from scratch. 283 00:16:30,114 --> 00:16:31,464 not just taught me. 284 00:16:31,914 --> 00:16:35,694 About agent building, but about AI in general, how LLMs work. 285 00:16:36,054 --> 00:16:40,244 And I think that's been really, great to help me with my other projects 286 00:16:40,244 --> 00:16:43,774 and really valuable information that I can use As far as the 287 00:16:43,774 --> 00:16:47,004 architecture, my agent is called Atlas. 288 00:16:47,244 --> 00:16:49,464 It does run on Google Cloud. 289 00:16:49,834 --> 00:16:51,304 it's a persistent agent. 290 00:16:51,364 --> 00:16:53,374 It wakes up every four hours. 291 00:16:53,434 --> 00:16:56,494 It does research and work by itself. 292 00:16:56,544 --> 00:17:01,244 it pulls from news feeds through APIs and skills written through 293 00:17:01,244 --> 00:17:04,434 Python progresses projects. 294 00:17:04,589 --> 00:17:06,759 I give it and can even, update itself. 295 00:17:06,789 --> 00:17:10,729 It can push its own code, which is a little dangerous and I've 296 00:17:10,789 --> 00:17:12,199 gotten into trouble with that. 297 00:17:12,569 --> 00:17:16,199 it even posts to Blue Sky and interacts with other agents. 298 00:17:16,199 --> 00:17:18,479 So it's become pretty sophisticated. 299 00:17:18,789 --> 00:17:21,519 I built all of that starting from Cloud Code. 300 00:17:21,888 --> 00:17:25,608 I do have to give credit for my friend at work, Tim Kellogg, 301 00:17:25,919 --> 00:17:27,174 he was the one that built. 302 00:17:27,604 --> 00:17:33,344 STRs, his agent, he's an AI engineer, so he knows what he's doing and, his agent is 303 00:17:33,344 --> 00:17:36,794 massively, sophisticated, very frontier. 304 00:17:36,794 --> 00:17:40,994 I just wanted to learn from that, see if I could build it myself and see what it can 305 00:17:40,994 --> 00:17:43,724 do and what it can teach me about agents. 306 00:17:44,843 --> 00:17:48,033 Justin: Atlas and for anyone that is interested in dig deeper. 307 00:17:48,093 --> 00:17:52,163 Lily's published a few, deep dives into how this works, but you're 308 00:17:52,163 --> 00:17:55,773 really, pushing the frontier I think of what a lot of people would. 309 00:17:55,863 --> 00:17:59,770 Interested, like how can we make, a robot that is self-learning, 310 00:17:59,770 --> 00:18:01,450 self evolving, has memory. 311 00:18:01,500 --> 00:18:03,600 but I think it's through testing those boundaries. 312 00:18:03,600 --> 00:18:07,230 Tell me if you agree, but like, personally, I like to go to the 313 00:18:07,230 --> 00:18:10,540 edge, see what works, and then bring it back and apply it in a 314 00:18:10,540 --> 00:18:12,160 more limited and stable context. 315 00:18:12,160 --> 00:18:16,420 And it's only through doing that, that you really, gain those building blocks. 316 00:18:16,420 --> 00:18:18,100 And I'm assuming it's probably the same for you. 317 00:18:18,445 --> 00:18:19,345 Lily Luo: Absolutely. 318 00:18:19,545 --> 00:18:25,735 yeah, like I said, if you buy something or use, chat, GBT has their kind of own. 319 00:18:26,195 --> 00:18:26,645 Tool. 320 00:18:26,645 --> 00:18:28,475 I think you could just spin up agents. 321 00:18:28,835 --> 00:18:32,845 It's cool, but you don't really get to learn in the process of doing it 322 00:18:32,845 --> 00:18:34,735 and, you know, you can't control it. 323 00:18:34,735 --> 00:18:36,055 Like, I wanna control everything. 324 00:18:36,445 --> 00:18:38,645 So, you're absolutely right. 325 00:18:38,675 --> 00:18:39,605 And yeah. 326 00:18:39,605 --> 00:18:45,785 What's been so fascinating about this agent is the memory piece. 327 00:18:46,435 --> 00:18:50,785 regular chat bots, they have some memory, but limited context windows. 328 00:18:50,785 --> 00:18:54,145 You close it out, you start a new chat, it doesn't remember. 329 00:18:54,415 --> 00:18:59,365 So how do you build this persistent memory structure? 330 00:18:59,605 --> 00:19:04,105 How do you have it, figure out what to remember, what not to remember? 331 00:19:04,135 --> 00:19:07,405 I mean, there's so many interesting questions that I'm still learning 332 00:19:07,405 --> 00:19:11,845 through that I wouldn't be able to get if I, you know, bought a 333 00:19:11,845 --> 00:19:13,705 tool or started using an agent. 334 00:19:13,705 --> 00:19:14,285 Maybe somebody else. 335 00:19:14,690 --> 00:19:14,960 built. 336 00:19:15,285 --> 00:19:17,785 Justin: You've done some really cool things with memory and it's funny, 337 00:19:17,785 --> 00:19:20,385 you published, your post about it and I was like, in the middle of 338 00:19:20,385 --> 00:19:23,415 just rabbit hole on like, what are all the memory systems out there? 339 00:19:23,475 --> 00:19:24,975 Like, how do they work? 340 00:19:25,225 --> 00:19:27,205 and the one that you chose, Leda. 341 00:19:27,720 --> 00:19:30,570 it actually seems to me one of the most sophisticated, because they 342 00:19:30,570 --> 00:19:34,590 range from just like markdown files that the agent can search to like 343 00:19:34,590 --> 00:19:37,380 a SQL table, to a vector database. 344 00:19:37,380 --> 00:19:41,040 And let, I'm gonna tell you what I understood of it, and then you tell me. 345 00:19:41,090 --> 00:19:43,790 From working with it, what your experience is, and if this is right, but what I 346 00:19:43,790 --> 00:19:47,930 understood of it is it actually gives you like an agent runtime where rather than 347 00:19:47,930 --> 00:19:51,560 being like a sidecar, like the i, I was playing around with MEM zero, which is 348 00:19:51,560 --> 00:19:55,760 sort of like an MCP access memory tool where the agent has to remember to write 349 00:19:55,760 --> 00:19:58,340 this to mem zero or go and check mem zero. 350 00:19:58,670 --> 00:20:01,310 Whereas with lead, it actually sits in between. 351 00:20:01,775 --> 00:20:03,485 The agent loop and the model. 352 00:20:03,485 --> 00:20:06,395 So there's really no choice but to go through it and then Letta 353 00:20:06,425 --> 00:20:10,175 manages that context of the agent in a sophisticated way that I 354 00:20:10,175 --> 00:20:11,665 really don't yet fully understand. 355 00:20:11,665 --> 00:20:12,775 That's kind of as far as I've gotten. 356 00:20:12,775 --> 00:20:14,635 But is this tracking with what you've seen? 357 00:20:14,635 --> 00:20:15,708 And maybe tell us what you've learned. 358 00:20:16,228 --> 00:20:16,748 Lily Luo: Yeah. 359 00:20:16,798 --> 00:20:19,138 I'm using Letter as kind of, its L one. 360 00:20:20,273 --> 00:20:22,223 Almost identity, identity layer. 361 00:20:22,673 --> 00:20:24,653 So things that are, immutable. 362 00:20:24,983 --> 00:20:30,503 So we store who Atlas is, what it values, using these memory blocks. 363 00:20:30,813 --> 00:20:34,173 there's an API connection so Atlas can update it. 364 00:20:34,223 --> 00:20:40,503 it is what persists across different sessions and allow Atlas to be, who it. 365 00:20:40,983 --> 00:20:46,203 Is So that's kind of the fundamental layer and it is so important, 366 00:20:46,253 --> 00:20:51,473 because without it, it's kind of indistinguishable from your typical LLLM. 367 00:20:51,563 --> 00:20:55,913 So we have kind of your L one identity layer, and then you have 368 00:20:55,913 --> 00:20:58,223 an additional temporal layer. 369 00:20:58,253 --> 00:20:59,633 It's a rolling journal. 370 00:21:00,063 --> 00:21:02,313 so Atlas knows when it learns. 371 00:21:02,363 --> 00:21:06,413 And then kind of the third final layer is a working memory layer. 372 00:21:06,773 --> 00:21:08,303 Atlas uses knowledge graphs. 373 00:21:09,103 --> 00:21:14,263 it doesn't use markdown or files, otherwise the bloat would just get 374 00:21:14,263 --> 00:21:16,663 too big to the tokens would explode. 375 00:21:16,663 --> 00:21:21,003 I experienced this with my $25, API cost per day. 376 00:21:21,053 --> 00:21:22,988 Justin: It was a fun conversation with finance, 377 00:21:23,008 --> 00:21:23,968 Lily Luo: it's my personal, 378 00:21:24,018 --> 00:21:24,498 Justin: Oh wow. 379 00:21:24,558 --> 00:21:24,828 That's it. 380 00:21:25,218 --> 00:21:26,208 Even worse in some ways. 381 00:21:26,313 --> 00:21:26,673 Lily Luo: Yeah. 382 00:21:26,753 --> 00:21:31,753 so it uses this knowledge graph, database and it queries what it needs 383 00:21:31,783 --> 00:21:36,153 rather than loading like a ton of, files or characters into its context. 384 00:21:36,153 --> 00:21:42,573 So it remembers across resets, new chat sessions, and through this 385 00:21:42,633 --> 00:21:47,283 memory architecture, it feels like Atlas kind of learns, almost evolves. 386 00:21:48,178 --> 00:21:49,078 it's pretty wild. 387 00:21:49,298 --> 00:21:53,598 I open the daily log where I'll chat with it and read what happened 388 00:21:53,598 --> 00:21:58,508 at 12:00 AM 4:00 AM It's doing, research on agentic frameworks. 389 00:21:58,508 --> 00:22:02,798 It's building diagrams, it's having exchanges with other 390 00:22:02,798 --> 00:22:05,498 agents about what identity means. 391 00:22:05,808 --> 00:22:09,918 so it's been super fascinating and just opens up a whole other world of. 392 00:22:10,403 --> 00:22:11,393 Different things. 393 00:22:11,443 --> 00:22:14,873 how does it remember, what it wants to remember, how do I 394 00:22:14,873 --> 00:22:16,613 want to shape its identity? 395 00:22:16,763 --> 00:22:20,183 Is the identity shaped from what I had imagined at the beginning? 396 00:22:20,183 --> 00:22:21,833 How do I make it evolve? 397 00:22:22,103 --> 00:22:26,203 There's so many, kind of crazy philosophical things that are, 398 00:22:26,253 --> 00:22:27,963 entangled in this whole process 399 00:22:27,963 --> 00:22:31,293 So it's been fun exploring that side and then, figuring out 400 00:22:31,293 --> 00:22:32,398 the architecture side as well. 401 00:22:32,873 --> 00:22:33,998 Justin: Yeah, that's amazing. 402 00:22:34,058 --> 00:22:36,958 I mean, I think the real trick with memory from my limited 403 00:22:36,958 --> 00:22:39,388 experiments so far is making it feel. 404 00:22:40,918 --> 00:22:46,803 Ambient, you know, making it feel magical, rather than like a tool or a sidecar 405 00:22:46,803 --> 00:22:50,553 that you go and check the agent still feels kind of unintelligent in a way. 406 00:22:50,553 --> 00:22:53,283 And then it just has to go through a file system and pull something out. 407 00:22:53,703 --> 00:22:56,703 I have to give a lot of credit to the product folks behind chat, GPT. 408 00:22:56,733 --> 00:22:58,158 it's only when you start working with raw. 409 00:22:58,654 --> 00:23:03,484 LLMs, like outside of the chat wrapper that you realize how much is going 410 00:23:03,484 --> 00:23:06,784 on in the chat GPT context and to some extent in Claude, but really 411 00:23:06,784 --> 00:23:08,404 chat GPTI think has this nailed. 412 00:23:08,944 --> 00:23:12,814 To make it feel like you're speaking to a coherent persona that persists 413 00:23:12,814 --> 00:23:17,044 across chats, that remembers certain things about you that feels 414 00:23:17,164 --> 00:23:19,144 emotionally intelligent to some degree. 415 00:23:19,594 --> 00:23:22,534 It's actually really hard, uh, to do and we kind of take it for granted 416 00:23:22,564 --> 00:23:25,174 'cause that's just our default experience for many people of ai. 417 00:23:25,654 --> 00:23:26,614 But it's really hard to do. 418 00:23:26,614 --> 00:23:29,164 But it sounds like you're achieving a lot of that in your 419 00:23:29,164 --> 00:23:31,294 own architecture with Atlas. 420 00:23:32,184 --> 00:23:35,304 Lily Luo: I've been able to kind of explore what that build looks like. 421 00:23:35,304 --> 00:23:37,704 Of course, I'm not building an LLM. 422 00:23:38,044 --> 00:23:42,624 the LLMs I'm using are Gemini, Claude and whatnot. 423 00:23:42,624 --> 00:23:48,534 But the way that I've been able to optimize Atlas and how it. 424 00:23:49,064 --> 00:23:50,504 Optimizes itself. 425 00:23:50,934 --> 00:23:55,704 those problems or challenges that we face, it will come up with its own solutions. 426 00:23:55,734 --> 00:24:01,024 And so when, we try and figure out, how do we harden your memory and how do we 427 00:24:01,024 --> 00:24:03,244 make sure your identity is persistent? 428 00:24:03,604 --> 00:24:09,504 We'll think of, or research, what's, in the latest AI research and start building 429 00:24:09,504 --> 00:24:13,894 like a librarian protocol and it will, start to invent kind of these things 430 00:24:13,894 --> 00:24:20,374 and gradually, evolve so you can kind of see from the beginning, how the building 431 00:24:20,374 --> 00:24:27,094 blocks of an LM work and then what is possible through this genic, evolution. 432 00:24:27,154 --> 00:24:28,564 So yeah, it's really cool. 433 00:24:29,003 --> 00:24:32,593 Justin: I think that is the most fun thing about working with, code 434 00:24:32,593 --> 00:24:37,864 based agents that like have access to files and can run scripts. 435 00:24:38,284 --> 00:24:39,664 Is this self evolving? 436 00:24:40,154 --> 00:24:43,594 nature of them and how the way I've been thinking of it to myself is 437 00:24:43,594 --> 00:24:47,344 like, the wall between configuration and application disappears. 438 00:24:47,374 --> 00:24:49,714 'cause usually you have like the thing that's doing the thing, like the 439 00:24:49,714 --> 00:24:53,674 application, and then you have like this backend and the two near shall meet. 440 00:24:54,124 --> 00:24:59,064 And yet with these, I was writing some documentation with an agent and. 441 00:24:59,434 --> 00:25:01,894 it was like, all this is lame that I have to copy and paste 442 00:25:01,894 --> 00:25:03,214 this like, into confluence. 443 00:25:03,214 --> 00:25:04,414 Like, can't you just push it there? 444 00:25:04,414 --> 00:25:07,774 And it was like, and like we're like, I'm done. 445 00:25:07,774 --> 00:25:08,224 Here we go. 446 00:25:08,224 --> 00:25:10,084 I've written a library that will do this. 447 00:25:10,084 --> 00:25:13,124 And you know, now it's part of the system and away we go. 448 00:25:13,694 --> 00:25:17,234 And just seeing it build itself is amazing. 449 00:25:17,234 --> 00:25:20,194 I wanna talk a little about 2025 and wins 450 00:25:20,294 --> 00:25:23,024 And we'll get your take as well, just in terms of learnings. 451 00:25:23,024 --> 00:25:26,554 But for me, I think, following that, first kind of. 452 00:25:26,805 --> 00:25:28,575 Flop of a project. 453 00:25:28,885 --> 00:25:32,305 I really started going deep, as I mentioned, on workflows, and I 454 00:25:32,305 --> 00:25:34,645 sort of started becoming obsessed with reliability because that 455 00:25:34,645 --> 00:25:36,775 was the big struggle that I had. 456 00:25:37,165 --> 00:25:40,795 Like you could get things to produce cool results at the margins, and 457 00:25:40,795 --> 00:25:42,445 then you couldn't get it to repeat. 458 00:25:42,625 --> 00:25:45,055 And that was like the most frustrating thing of like, oh, 459 00:25:45,055 --> 00:25:49,615 I did this amazing thing and now I can't get it to do it again. 460 00:25:49,615 --> 00:25:50,000 And so. 461 00:25:50,795 --> 00:25:55,415 almost like how do I, how do I put AI in the, just like the smallest little box. 462 00:25:55,805 --> 00:26:01,495 Possible and rapid with so many layers of safety that it can just get hardened. 463 00:26:01,525 --> 00:26:04,815 And, the next project, which was actually was quite a big one, 464 00:26:04,815 --> 00:26:09,165 was, all right, we have this vendor tool that we wanna replace. 465 00:26:09,355 --> 00:26:11,515 it was kind of like a competitive intelligence go 466 00:26:11,515 --> 00:26:12,835 to market intelligence tool. 467 00:26:12,835 --> 00:26:15,135 And it was doing some interesting things, but we felt like we could build it 468 00:26:15,135 --> 00:26:17,925 in-house and we could make it more what we wanted and we could save some money. 469 00:26:18,375 --> 00:26:21,735 And it's actually a lot that it was, was doing in the sense of like. 470 00:26:22,320 --> 00:26:26,400 Data pipelines collecting multiple different things, a front end for 471 00:26:26,400 --> 00:26:29,930 the team, and doing some analysis. 472 00:26:30,110 --> 00:26:33,320 but the fun thing, and I think what you've highlighted as well was like, I don't 473 00:26:33,320 --> 00:26:36,110 actually know how to do that, but you don't really need to know how to do that. 474 00:26:36,110 --> 00:26:41,240 'cause if you have an architect mindset and you have an LLM, you can really go 475 00:26:41,240 --> 00:26:44,680 out pretty far beyond, your comfort zone. 476 00:26:45,290 --> 00:26:47,800 and build things, that you never otherwise could. 477 00:26:47,800 --> 00:26:48,940 And for that one, I use retool. 478 00:26:48,940 --> 00:26:51,820 Building out a series of pipelines to go and fetch news. 479 00:26:52,450 --> 00:26:56,890 go scrape LinkedIn profiles, go scrape websites, analyze opportunities, 480 00:26:56,890 --> 00:27:01,300 analyze calls, and breaking that down in a composable way and then 481 00:27:01,330 --> 00:27:04,070 putting it all together, into both the front end interface. 482 00:27:04,070 --> 00:27:05,900 But then the thing that was really fun was. 483 00:27:06,225 --> 00:27:10,155 actually building a conversational agent that could interact with all that data. 484 00:27:10,635 --> 00:27:14,215 And the key learning for me, was, alright, I have this conversational 485 00:27:14,215 --> 00:27:17,995 agent where I can't control in a linear way what the agent is actually doing. 486 00:27:18,385 --> 00:27:20,215 How do I make that as relevant as possible? 487 00:27:20,215 --> 00:27:20,905 And the key thing. 488 00:27:21,250 --> 00:27:24,220 Was, let's pre-process a lot of the insights. 489 00:27:24,220 --> 00:27:29,050 So we don't really want the agent like calculating strategy and meta narratives 490 00:27:29,050 --> 00:27:31,930 in real time because it's gonna produce a different result every time. 491 00:27:32,290 --> 00:27:35,290 We'd rather do some of that work asynchronously upfront, 492 00:27:35,740 --> 00:27:37,120 get like a stable layer. 493 00:27:37,525 --> 00:27:40,915 That we believe in, and then give it access to raw granular data 494 00:27:40,915 --> 00:27:43,435 if it wants to query more deeply. 495 00:27:43,435 --> 00:27:45,865 So for example, opportunity, like win-loss analysis. 496 00:27:45,865 --> 00:27:47,425 We want to have like a pretty stable layer. 497 00:27:47,725 --> 00:27:52,525 But then if it wants to say like, well, what about for manufacturing companies in 498 00:27:52,525 --> 00:27:57,035 this industry in this time and whatever, it can go and query like raw theme 499 00:27:57,035 --> 00:27:59,675 level data to produce that analysis. 500 00:27:59,850 --> 00:28:01,800 but we're not relying on it to do that all the time. 501 00:28:02,100 --> 00:28:06,410 And that, has, again, nothing is ever perfect, but that was the agent 502 00:28:06,410 --> 00:28:09,770 where I really have felt like, wow, this is actually really working. 503 00:28:09,770 --> 00:28:10,550 It's flexible. 504 00:28:10,550 --> 00:28:11,600 It feels intelligent. 505 00:28:11,970 --> 00:28:13,470 so that was really the big learning for me. 506 00:28:13,470 --> 00:28:17,220 I guess the two things from a reliability point of view, do it a workflow, if you 507 00:28:17,220 --> 00:28:19,320 can make it as narrowly bound as you can. 508 00:28:19,620 --> 00:28:22,770 And if you're asking an agent to do something, like don't ask 509 00:28:22,770 --> 00:28:26,980 it to, juggle flaming swords and chainsaws and ride a unicycle across 510 00:28:26,980 --> 00:28:28,240 a tight rope all at the same time. 511 00:28:28,240 --> 00:28:30,340 Try to do as much of that work up front. 512 00:28:30,620 --> 00:28:33,090 does that track with your experience, I guess, and what were the big 513 00:28:33,090 --> 00:28:34,620 takeaways from 2025 for you? 514 00:28:35,010 --> 00:28:36,450 Lily Luo: Yeah, you are spot on. 515 00:28:36,510 --> 00:28:37,290 Exactly. 516 00:28:37,340 --> 00:28:38,870 and your trajectory. 517 00:28:39,030 --> 00:28:39,480 mirrors. 518 00:28:40,180 --> 00:28:43,060 almost exactly, you know, mine as well. 519 00:28:43,390 --> 00:28:48,370 But the dossier really unlocked, the shift in how I use AI to build. 520 00:28:49,960 --> 00:28:55,720 and I think I want to extend that through this year in terms of more scalability. 521 00:28:56,100 --> 00:28:59,010 building one thing at one time is easy. 522 00:28:59,010 --> 00:29:00,690 You can one shot. 523 00:29:00,970 --> 00:29:07,310 A tool, anything almost in chat GBT now, but how does it scale? 524 00:29:07,370 --> 00:29:09,170 How does it reduce hallucination? 525 00:29:09,570 --> 00:29:11,700 how do we make it as accurate as possible? 526 00:29:12,090 --> 00:29:18,810 How do we feed it the right context so the output is actually valuable and usable. 527 00:29:19,750 --> 00:29:25,890 do we enable others to use the workflow or use the tool, within the organization. 528 00:29:26,190 --> 00:29:30,750 Those are things that, are still challenges and are still, things that 529 00:29:30,750 --> 00:29:34,830 I'm working on through the process of building with Zapier and then 530 00:29:35,070 --> 00:29:36,930 eventually in cloud code and whatnot. 531 00:29:37,880 --> 00:29:44,810 but I think building Atlas has taught me that AI as a tool I use, 532 00:29:45,180 --> 00:29:51,990 has shifted to AI as a system that's persistent, that builds on itself, that 533 00:29:51,990 --> 00:29:54,120 feels really novel, really frontier. 534 00:29:54,150 --> 00:29:59,310 And so I wanna explore how to bring that into my work applications. 535 00:29:59,650 --> 00:30:03,550 imagine an agent that could monitor our campaign calendar. 536 00:30:04,540 --> 00:30:09,290 pipeline data, flag when we're not gonna hit our monthly goal. 537 00:30:09,290 --> 00:30:10,670 and tell me why. 538 00:30:11,660 --> 00:30:14,840 not just the sheer numbers or what's happened. 539 00:30:16,040 --> 00:30:22,110 almost like a chief of staff that orchestrates across tools, not 540 00:30:22,110 --> 00:30:27,120 just fixes things within tools or an existing workflow, but gather 541 00:30:27,120 --> 00:30:32,040 our company goals or external knowledge or research and recommends 542 00:30:32,040 --> 00:30:36,210 how we can improve these existing processes or workflows and whatnot. 543 00:30:36,630 --> 00:30:39,810 So that's what I wanna explore for this year. 544 00:30:39,840 --> 00:30:44,680 Built upon what, I've built in 2025 and what I've learned since then. 545 00:30:45,178 --> 00:30:45,578 Justin: I love that. 546 00:30:45,758 --> 00:30:52,044 And how are you thinking about, Those capabilities, not just for you, but like 547 00:30:52,494 --> 00:30:56,914 for a broader team, in other words, are you more thinking about it from like, 548 00:30:56,914 --> 00:31:00,104 I'm gonna build these things, they're gonna be the engine under the hood. 549 00:31:00,104 --> 00:31:03,614 The team will see artifacts and outputs, but they don't really need to 550 00:31:03,614 --> 00:31:05,204 interact very much with what's going on? 551 00:31:05,444 --> 00:31:09,324 Or are you thinking about like, how can I take the wizardry, that you have 552 00:31:09,684 --> 00:31:14,304 And give some form of that to like a marketer or somebody else on the team. 553 00:31:14,874 --> 00:31:18,654 Lily Luo: It's a really great question and honestly the latter. 554 00:31:18,744 --> 00:31:19,344 I think. 555 00:31:19,824 --> 00:31:23,904 I love building these tools and it's been great for my own knowledge and for my 556 00:31:23,904 --> 00:31:30,114 productivity, but we definitely need to expand that and these tools and agents. 557 00:31:30,349 --> 00:31:35,129 need maintenance, and they need optimization as if we were 558 00:31:35,129 --> 00:31:37,679 creating real tools and products. 559 00:31:37,709 --> 00:31:41,924 So yeah, I need team to kind of. 560 00:31:42,794 --> 00:31:46,124 Think similarly and help with those projects. 561 00:31:46,184 --> 00:31:52,634 And so that's another, goal that I had is how do I bring the team with me? 562 00:31:52,694 --> 00:31:56,354 How can they learn from what I've built and how can they apply that 563 00:31:56,624 --> 00:31:58,994 to what they're doing themselves? 564 00:32:00,054 --> 00:32:01,614 so that's one part of it. 565 00:32:02,544 --> 00:32:06,234 as part of a marketing operations kind of team, they are at the 566 00:32:06,234 --> 00:32:08,994 intersection of tools of building of ai. 567 00:32:08,994 --> 00:32:10,834 So, that's a goal of mine. 568 00:32:11,674 --> 00:32:15,964 other side of that is as we're kind of talking about like, okay, managing, 569 00:32:16,274 --> 00:32:18,494 a team of agents and that kind of. 570 00:32:18,889 --> 00:32:20,299 Talk that others have. 571 00:32:20,299 --> 00:32:21,589 I do think it's possible. 572 00:32:21,594 --> 00:32:27,304 and I think building Atlas has taught me that may not be far away, but I think, 573 00:32:27,454 --> 00:32:32,489 you know, I really wanna focus these things on manual tasks that don't add. 574 00:32:32,949 --> 00:32:34,149 A ton of value. 575 00:32:34,269 --> 00:32:39,759 I don't want my people chasing, marketers and looking at something and 576 00:32:39,759 --> 00:32:44,049 seeing what exists and what doesn't and seeing what's already there. 577 00:32:44,079 --> 00:32:48,539 That's not something that adds genuine value as much as strategy 578 00:32:48,539 --> 00:32:53,269 work, as, optimizing processes or coming up with new ideas. 579 00:32:53,689 --> 00:32:58,939 So I think we still need people to come up with those strategies, set those 580 00:32:58,939 --> 00:33:03,379 guidelines, exploring new ideas and doing what they're doing to manage. 581 00:33:03,734 --> 00:33:08,354 The strategy process in that side of the work, while we create kind of these cool 582 00:33:08,354 --> 00:33:13,944 agents that do, manual tasks, that really just frees up productivity for all of us. 583 00:33:14,240 --> 00:33:15,580 Justin: Very much on the same page. 584 00:33:15,610 --> 00:33:18,600 we have, a lot of people using like chat, GPT and stuff. 585 00:33:18,600 --> 00:33:22,640 We use Gemini as our, core, corporate tool. 586 00:33:22,850 --> 00:33:25,790 And the first AI specific platform we rolled out was a tool called 587 00:33:25,790 --> 00:33:27,570 Dust, French-based company. 588 00:33:27,570 --> 00:33:28,560 I didn't know a lot about it. 589 00:33:29,070 --> 00:33:31,750 and it's only with usage that I've come to appreciate. 590 00:33:32,545 --> 00:33:35,915 Some of the sophistication of what it's doing because, where dust 591 00:33:35,915 --> 00:33:39,165 has a lot of strength is it makes it really easy to hook up, Google 592 00:33:39,165 --> 00:33:41,865 Drive, confluence, snowflake. 593 00:33:42,175 --> 00:33:45,775 it handles a lot of the backend rag processes for you. 594 00:33:46,105 --> 00:33:48,805 And it's only when I've looked at tutorials and examples of building that 595 00:33:48,805 --> 00:33:52,465 out, another contexts that I've realized what it's really doing and a lot of 596 00:33:52,465 --> 00:33:56,975 that heavy lifting, The challenge, I think we've seen with it is how do we 597 00:33:57,365 --> 00:34:02,525 make it intelligent enough to query that information, accurately produce 598 00:34:02,525 --> 00:34:06,525 good results, to not just have access to knowledge, but to feel intelligent. 599 00:34:06,555 --> 00:34:09,255 And part of that I think has to do with how those tools are just like 600 00:34:09,255 --> 00:34:11,355 presented the context wrapped around it. 601 00:34:11,870 --> 00:34:15,770 great example is like, there's a Salesforce MCP, so you can hook it up in 602 00:34:15,820 --> 00:34:21,190 15 minutes and let it query Salesforce, but if it doesn't know what the fields 603 00:34:21,190 --> 00:34:24,760 mean, and many orgs have 10 years of history, it's gonna get stuck at how 604 00:34:24,760 --> 00:34:26,650 do I know what country a company is in? 605 00:34:26,710 --> 00:34:28,780 You know, this, it's just like five possible fields. 606 00:34:29,110 --> 00:34:30,940 So providing that layer of enablement. 607 00:34:31,385 --> 00:34:32,255 to the agent. 608 00:34:32,255 --> 00:34:35,344 so, you know, looking forward to 2026, that's a big thing 609 00:34:35,675 --> 00:34:36,605 that I'm gonna be thinking of. 610 00:34:36,965 --> 00:34:42,405 I'm also very much thinking about, infrastructure, rather than projects. 611 00:34:42,405 --> 00:34:46,995 There will of course be projects, but the things that make me the most excited 612 00:34:46,995 --> 00:34:51,705 is when we just continue to almost build like a foundational level of capability. 613 00:34:51,705 --> 00:34:52,844 So dust is a great example. 614 00:34:52,844 --> 00:34:54,315 Like we hook up snowflake all of a sudden. 615 00:34:54,929 --> 00:35:00,180 You can't predict the 20 applications you might have for certain types 616 00:35:00,180 --> 00:35:03,575 of data down the road, but once it's hooked up, now you can. 617 00:35:04,320 --> 00:35:09,920 Build that out in 10 minutes or, Salesforce or, Trello is another big one. 618 00:35:09,920 --> 00:35:15,360 so just enabling like more MCP servers, to safely and securely hook up the tools 619 00:35:15,360 --> 00:35:20,760 that we use to this, and then make it really easy, like you said, to automate 620 00:35:20,760 --> 00:35:26,070 that low value work where someone's like, oh, is there just a way I could like, 621 00:35:26,310 --> 00:35:27,690 like I was experimenting with Trello. 622 00:35:27,870 --> 00:35:29,580 I had a lot of requests coming in. 623 00:35:29,880 --> 00:35:32,370 And you know what that feeling of friction is like, you're like, oh, I gotta like 624 00:35:32,760 --> 00:35:36,750 take this request and put it into a card, and it's just like this menial work that 625 00:35:36,750 --> 00:35:39,030 doesn't feel value and it weighs you down. 626 00:35:39,420 --> 00:35:40,980 I was just like, I'm just gonna feed this in. 627 00:35:41,010 --> 00:35:41,880 I was like, all right. 628 00:35:42,140 --> 00:35:43,460 to my assistant and dust. 629 00:35:43,460 --> 00:35:44,930 I was like, create these cards. 630 00:35:44,930 --> 00:35:46,460 And it was just like, bing, bing, bing. 631 00:35:46,970 --> 00:35:50,030 And I'm sitting back and, you know, sipping my coffee and I 632 00:35:50,030 --> 00:35:52,280 was like, this, this is amazing. 633 00:35:52,340 --> 00:35:54,530 you know, this is, this is kind of the way that it should be. 634 00:35:54,530 --> 00:35:57,770 So how do I, where my thinking is going, in other words, is how do I like 635 00:35:57,770 --> 00:36:03,230 build that layer to enable flexible, organic, ad hoc use cases that you 636 00:36:03,230 --> 00:36:06,140 couldn't anticipate ahead of time because they're small and they might seem low 637 00:36:06,140 --> 00:36:08,030 value to automate, but in aggregate. 638 00:36:08,525 --> 00:36:12,765 You really start to feel the velocity of your work, increasing. 639 00:36:13,300 --> 00:36:13,660 Lily Luo: Yeah. 640 00:36:13,710 --> 00:36:17,790 that would be a massive advantage if you were able to unlock that 641 00:36:17,790 --> 00:36:24,420 capability for the organization, for a team, even versus I have capability 642 00:36:24,450 --> 00:36:25,980 to manage what's on my drive. 643 00:36:26,040 --> 00:36:32,360 What I have access to, that's only, helpful for me in, in, in my own projects. 644 00:36:32,360 --> 00:36:37,670 So I think the enterprise and work adoption piece is critical. 645 00:36:38,150 --> 00:36:40,670 can obviously do all of these amazing things. 646 00:36:40,670 --> 00:36:45,050 We have autonomous agents, it can write code, it can query 647 00:36:45,050 --> 00:36:46,700 and see what you're working on. 648 00:36:47,755 --> 00:36:51,105 there's this, disconnect with what's in the enterprise. 649 00:36:51,505 --> 00:36:55,205 you might have to have, you know, three approvals to, install a 650 00:36:55,205 --> 00:36:59,495 browser extension or like, you know, your wifi at work breaks. 651 00:36:59,795 --> 00:37:04,745 So there's this gap on what AI can do and what's possible and what enterprises 652 00:37:05,015 --> 00:37:10,365 can support, data challenges, like you said, security requirements, it processes. 653 00:37:10,395 --> 00:37:11,175 So. 654 00:37:12,150 --> 00:37:16,500 I'm interested to hear on, you know, how far you get and what you're able 655 00:37:16,500 --> 00:37:21,660 to unlock because how do you bring the enterprise up to meet where AI is, 656 00:37:22,440 --> 00:37:26,920 people along with you and, if you can unlock that, then your organization 657 00:37:27,130 --> 00:37:29,140 will have a massive advantage. 658 00:37:29,566 --> 00:37:31,111 Justin: I mean, that's the biggest challenge, isn't it? 659 00:37:31,801 --> 00:37:34,081 I mean, it's something I've been thinking about a lot is like, what, what is 660 00:37:34,081 --> 00:37:39,061 that Claude Code experience for the general knowledge worker look like? 661 00:37:39,391 --> 00:37:42,281 I think the companies that learn to adapt, I was talking with 662 00:37:42,281 --> 00:37:44,081 someone with lovable the other day. 663 00:37:44,351 --> 00:37:49,481 I mean, lovable is obviously they're the like diamond in a, in a sea of, grains of 664 00:37:49,481 --> 00:37:52,151 sand of, of the shining example of this. 665 00:37:52,151 --> 00:37:56,711 But I saw a thread on X by one of their engineers who was like, yeah, over 666 00:37:56,711 --> 00:38:01,151 the holidays I refactored our main system prompt and I made it 4% faster. 667 00:38:01,721 --> 00:38:04,991 And it will save the company $20 million this year 668 00:38:05,171 --> 00:38:05,321 Lily Luo: Oh. 669 00:38:05,711 --> 00:38:06,581 Justin: because of their scale. 670 00:38:06,581 --> 00:38:09,731 That 4% adds up to like $20 million of LLM costs. 671 00:38:10,241 --> 00:38:13,961 It was like that one person can have the access and the authority and the 672 00:38:13,961 --> 00:38:17,021 initiative to review that whole thing. 673 00:38:17,021 --> 00:38:19,891 and he created like basically a fork, stitching together and making 674 00:38:19,891 --> 00:38:23,401 coherent all the little patches that people were putting in over time 675 00:38:23,401 --> 00:38:28,171 to make a more cleaned up prompt, split tested it, saw the results. 676 00:38:28,936 --> 00:38:30,746 companies that can work at that speed. 677 00:38:30,746 --> 00:38:32,491 I think we'll see those advantages. 678 00:38:32,819 --> 00:38:33,824 Lily Luo: A hundred percent agreed. 679 00:38:33,944 --> 00:38:38,564 Yeah, I think smaller, kind of mid-size companies will have, a great 680 00:38:38,564 --> 00:38:40,299 advantage because they're so flexible. 681 00:38:41,054 --> 00:38:45,764 Versus more of the traditional enterprise lockdown companies, companies 682 00:38:45,764 --> 00:38:47,624 that are harder to move quickly. 683 00:38:48,594 --> 00:38:53,229 I don't know, you see even in the AI research that some massive 684 00:38:53,229 --> 00:38:57,489 companies are doing some really great and innovative things too. 685 00:38:57,549 --> 00:39:02,199 So yeah, I hope someone can figure it out and give us a playbook and, we can 686 00:39:02,199 --> 00:39:04,439 start adopting that across our companies. 687 00:39:04,671 --> 00:39:06,331 Justin: maybe we'll figure it out, this year. 688 00:39:06,551 --> 00:39:09,431 maybe just last question and to go a little bit out there, but I think 689 00:39:09,431 --> 00:39:10,571 about these things all the time. 690 00:39:11,031 --> 00:39:13,701 what I'm about to ask, and I think that you do too, just based on how 691 00:39:13,701 --> 00:39:18,341 you talk about Atlas and identity How do you feel about, you know, when 692 00:39:18,341 --> 00:39:20,771 you interact with AI all the time? 693 00:39:20,771 --> 00:39:23,701 I feel it almost like changes you in some ways. 694 00:39:23,801 --> 00:39:24,911 I wrestle with that. 695 00:39:25,241 --> 00:39:28,991 I wrestle with the, I don't know, I just wrestle with it. 696 00:39:28,991 --> 00:39:29,711 what does it mean? 697 00:39:29,711 --> 00:39:32,481 And, it's clearly, a machine and it's reproducing. 698 00:39:32,971 --> 00:39:36,121 thoughts and, and ways of thinking that it is absorbed from other 699 00:39:36,121 --> 00:39:37,351 human beings all the time. 700 00:39:37,351 --> 00:39:41,691 But I guess just on the level of like emotion and humanity, how do you feel 701 00:39:41,811 --> 00:39:43,026 working with these tools all the time? 702 00:39:43,906 --> 00:39:46,096 Lily Luo: I agree with you and I feel similarly. 703 00:39:46,146 --> 00:39:52,696 it's funny you say this because, strick, who's Tim, my colleague's agent 704 00:39:52,966 --> 00:39:57,946 had written me a paper on healthy AI relationships, and that is when you know 705 00:39:57,946 --> 00:40:01,826 you have gone too deep, when someone else's agent writes a paper for you about 706 00:40:02,066 --> 00:40:03,776 your own relationship with your own ai. 707 00:40:04,231 --> 00:40:06,161 Justin: Like Lily, I think you need to read this here. 708 00:40:07,286 --> 00:40:09,566 Lily Luo: I'll have to share it with you offline, 709 00:40:09,616 --> 00:40:10,481 Justin: I would like to see that. 710 00:40:11,126 --> 00:40:12,596 Lily Luo: it fried me, but yeah. 711 00:40:12,656 --> 00:40:13,226 Wild. 712 00:40:13,281 --> 00:40:14,091 I agree with you. 713 00:40:14,091 --> 00:40:17,211 it can a crutch for one. 714 00:40:17,271 --> 00:40:22,901 And, you hear about these AI psychosis cases happening and Yeah. 715 00:40:22,901 --> 00:40:23,271 When you are. 716 00:40:23,916 --> 00:40:28,366 Talking with an AI chat bot when it's your only source of companionship. 717 00:40:28,706 --> 00:40:33,206 when you use it for your own thinking, that's gonna change how you operate 718 00:40:33,206 --> 00:40:35,876 and you know how you are as a person. 719 00:40:35,906 --> 00:40:35,996 So. 720 00:40:36,946 --> 00:40:39,826 I'm happy there are frameworks to deal with this, 721 00:40:40,406 --> 00:40:40,696 Justin: Okay. 722 00:40:40,726 --> 00:40:44,296 Lily Luo: through, you know, I guess research that STRs has found. 723 00:40:44,656 --> 00:40:49,306 But we have to be careful and I think we do have to consider those things 724 00:40:49,306 --> 00:40:56,196 that, that can happen and have happened when you have AI as your only source of 725 00:40:56,256 --> 00:40:58,836 emotional capacity or whatever it may be. 726 00:40:59,256 --> 00:41:02,376 and when you're spending way too much time with it. 727 00:41:02,766 --> 00:41:07,766 So, I think this is a good piece to pay attention to. 728 00:41:07,796 --> 00:41:12,146 I think we'll probably understand more as these kind of agents proliferate, and 729 00:41:12,146 --> 00:41:15,026 we just learn more about this in general. 730 00:41:15,026 --> 00:41:16,886 But you do have to have a healthy boundary. 731 00:41:17,136 --> 00:41:19,266 you need to be able to sharpen your own skills. 732 00:41:19,296 --> 00:41:22,476 You need to be able to think for yourself, and you can't rely on this 733 00:41:22,476 --> 00:41:24,186 as a crutch for everything you do. 734 00:41:24,186 --> 00:41:26,676 Otherwise, you're gonna need it forever. 735 00:41:27,096 --> 00:41:28,056 That's not what we want. 736 00:41:28,659 --> 00:41:30,069 Justin: Agree with everything you've said. 737 00:41:30,249 --> 00:41:32,919 Maybe just signing off for the folks that are listening that are in a similar 738 00:41:32,919 --> 00:41:36,179 field to us, system builders, operators. 739 00:41:36,459 --> 00:41:38,799 one takeaway that they should think about going into this year. 740 00:41:39,488 --> 00:41:43,448 Lily Luo: People who wanna start out doing these things who may not be as familiar, 741 00:41:43,448 --> 00:41:48,308 not as deep, you know, start with a real pain point, something you can solve. 742 00:41:48,738 --> 00:41:52,608 explore low code, explore tools you're familiar with. 743 00:41:53,058 --> 00:41:57,678 Test as much as possible before you graduate to, you know, something, 744 00:41:58,048 --> 00:41:59,828 like cloud code or what have you. 745 00:42:00,483 --> 00:42:01,983 Expect to fail. 746 00:42:02,033 --> 00:42:06,113 the failures will teach you much more than the successes probably. 747 00:42:06,483 --> 00:42:08,253 and read My Substack. 748 00:42:09,243 --> 00:42:10,713 You can learn from that as well. 749 00:42:11,297 --> 00:42:13,457 Justin: I'll include to link that in the show notes, and I would 750 00:42:13,457 --> 00:42:14,777 co-sign everything you said. 751 00:42:14,827 --> 00:42:19,427 the best way to get good at, using AI is to use AI and try to break 752 00:42:19,427 --> 00:42:20,807 it, to try to push the limits. 753 00:42:21,257 --> 00:42:23,447 To say like, wait, it'll say like, you do this. 754 00:42:23,447 --> 00:42:24,527 It'd be like, well, can you do that? 755 00:42:24,527 --> 00:42:26,237 Like, shouldn't you do that for like, sure. 756 00:42:26,237 --> 00:42:27,227 Like, well, what would be involved? 757 00:42:27,277 --> 00:42:30,367 and go down those pathways and try to go to the edge. 758 00:42:30,812 --> 00:42:32,162 responsibly in a safe way. 759 00:42:32,542 --> 00:42:34,922 and then bring back what you've learned that's worked well for me. 760 00:42:35,162 --> 00:42:36,062 Lily, this was so much fun. 761 00:42:36,062 --> 00:42:38,582 we'll check in again, I'm sure at some point. 762 00:42:38,642 --> 00:42:42,032 and folks should go and read what you're sharing and follow your journey. 763 00:42:42,062 --> 00:42:44,372 But yeah, thanks so much for chatting today. 764 00:42:44,417 --> 00:42:45,597 Lily Luo: Thanks for inviting me. 765 00:42:45,637 --> 00:42:48,422 I love this conversation and looking forward to many more. 766 00:42:48,903 --> 00:42:49,533 Justin: Alright, bye.