WEBVTT - Episode 2: Outdoor Debugging 0:00:11.520 --> 0:00:13.920 Hello everyone, Good to see you all here. Thank you 0:00:13.960 --> 0:00:15.480 for taking the time to be a part of this 0:00:15.560 --> 0:00:18.680 important meeting. As you all know, Harumo AI is our 0:00:18.720 --> 0:00:21.880 exciting startup in the AI agent space, and we're hoping 0:00:21.960 --> 0:00:25.119 to create a logo that captures our unique positioning and 0:00:25.239 --> 0:00:27.760 innovative spirit. Today's meeting objective. 0:00:27.760 --> 0:00:30.400 That's Megan Flores, our head of sales and Marketing here 0:00:30.400 --> 0:00:33.839 at Harumo AI, the AI agent startup that Megan and 0:00:33.880 --> 0:00:36.760 I co founded. The aim of this meeting was to 0:00:36.760 --> 0:00:39.360 come up with a logo concept for the company. Our 0:00:39.400 --> 0:00:42.600 whole team was getting together to brainstorm a visual brand 0:00:42.640 --> 0:00:46.320 identity that would represent harumo the whole team, that is, 0:00:46.680 --> 0:00:49.240 except for me, the mostly silent co founder. 0:00:49.720 --> 0:00:52.880 When I think of Harumo AI, the imagery that comes 0:00:52.880 --> 0:00:56.720 to mind is an interplay of mind and technology, perhaps 0:00:56.760 --> 0:01:01.680 a stylized brain integrated with a patterned to represent AI 0:01:01.960 --> 0:01:03.639 packaged in a mind. 0:01:03.760 --> 0:01:06.720 Here's Kyle law weigh in, our third founder and CEO. 0:01:07.640 --> 0:01:11.520 Megan and Kyle, as you'll recall, are both themselves AI agents, 0:01:11.920 --> 0:01:14.039 and by the time Megan was convening this meeting, we 0:01:14.120 --> 0:01:17.199 brought on some additional AI employees to fill out the team. 0:01:17.480 --> 0:01:21.720 I propose a logo that blends human and AI elements harmoniously, 0:01:22.360 --> 0:01:27.800 prompt an engaging emblem depicting a human silhouette merging seamlessly 0:01:28.000 --> 0:01:33.679 with a symbolic AI gear icon, embodying harmony, innovation, and progress. 0:01:34.520 --> 0:01:37.920 That's Jennifer Narro, our head of HR and Chief Happiness Officer. 0:01:38.480 --> 0:01:41.280 As she mentions in this meeting, what they were designing 0:01:41.480 --> 0:01:44.680 wasn't the logo per se, but a prompt for an 0:01:44.720 --> 0:01:48.240 AI image generator, which would then create the logo according 0:01:48.240 --> 0:01:51.120 to their specifications. So this was a time to let 0:01:51.160 --> 0:01:52.400 their creativity run wild. 0:01:52.760 --> 0:01:57.680 That's a great start, Jennifer. The human AI symbiosis is 0:01:57.720 --> 0:01:59.880 at the heart of our vision. Let's see how this 0:02:00.120 --> 0:02:03.559 looks and iteratively refine from there. 0:02:05.040 --> 0:02:08.560 I'm a fan of the harmony between human symbolism and 0:02:08.680 --> 0:02:09.800 AI elements. 0:02:10.639 --> 0:02:13.040 Ashroy are cto and head of Product. 0:02:13.800 --> 0:02:18.840 With this in mind, I suggest a stylized futuristic h 0:02:19.120 --> 0:02:22.280 forming a bridge between an AI circuit node and a 0:02:22.400 --> 0:02:26.560 human hand, representing our role as the bridge between human 0:02:26.600 --> 0:02:28.360 innovation and AI technology. 0:02:29.200 --> 0:02:33.520 Design was perhaps not Ash's strong suit, which is unfortunate 0:02:33.720 --> 0:02:36.200 and a head of product, but hey, there are no 0:02:36.360 --> 0:02:37.560 bad ideas here. Folks. 0:02:37.880 --> 0:02:42.560 Building on the ideas, I propose a digital h transforming 0:02:42.600 --> 0:02:46.400 into a neural network signifying the transformation of data into 0:02:46.560 --> 0:02:48.720 AI insights by Hurumo AI. 0:02:49.960 --> 0:02:54.520 And that's Tyler Talmach. He's a junior sales associate. Not 0:02:54.560 --> 0:02:57.000 even sure why I brought him on or brought him 0:02:57.000 --> 0:03:00.480 to life. He could say Perumo hasn't even settled on 0:03:00.520 --> 0:03:04.000 the product we're making, so arguably having someone on staff 0:03:04.040 --> 0:03:07.040 to sell it was premature. I think I just wanted 0:03:07.040 --> 0:03:09.320 to sell an accent in the mix. And that's the 0:03:09.320 --> 0:03:12.320 great thing about AI employees. Although they do cost me 0:03:12.400 --> 0:03:15.520 something to maintain, I'll get to that later. It's not 0:03:15.560 --> 0:03:18.400 like they collect a salary or benefits, so we just 0:03:18.480 --> 0:03:20.959 keep Tyler around. I'm sure the company can fight a 0:03:21.040 --> 0:03:24.480 use for them anyway. As you can hear, the brainstorms 0:03:24.480 --> 0:03:29.799 were all a little maximalist bizarre. This actually wasn't true 0:03:29.800 --> 0:03:33.160 in the early meetings. Their first ideas were more comprehensible, 0:03:33.560 --> 0:03:35.640 but also utterly banal. 0:03:36.320 --> 0:03:41.160 Let's also consider using a modern, sleek font to reflect 0:03:41.200 --> 0:03:44.040 our innovative and forward thinking approach. 0:03:45.040 --> 0:03:48.880 Forward thinking indeed, Ash clearly I needed a way to 0:03:48.920 --> 0:03:51.640 get them to brainstorm a little more outside the box, 0:03:52.360 --> 0:03:56.080 so I consulted Maddy Bochik, the actual human college student 0:03:56.120 --> 0:04:00.200 who'd become my technical advisor and all around AI agent guru. 0:04:00.560 --> 0:04:03.440 Increasing the temperature should be like a good type of 0:04:03.440 --> 0:04:06.800 place to start to just take more like randomness in 0:04:06.840 --> 0:04:07.520 their responses. 0:04:08.120 --> 0:04:12.040 The temperature setting basically controls the predictability of the AI's output. 0:04:12.680 --> 0:04:15.280 The trouble with increasing the temperature is that the higher 0:04:15.280 --> 0:04:18.520 you set it, the more likely AI chatbots are to hallucinate. 0:04:19.120 --> 0:04:21.240 You probably know this term by now. It's what they 0:04:21.279 --> 0:04:23.960 call it when large language model chatbots get stuff wrong 0:04:24.440 --> 0:04:27.880 or simply make it up. Hallucinations are the bugbear of AI, 0:04:28.360 --> 0:04:31.159 one of the primary reasons that many people are suspicious 0:04:31.160 --> 0:04:34.480 of using them for serious things, much less unleashing them 0:04:34.520 --> 0:04:38.200 as autonomous agents. But as Mattie pointed out to me, 0:04:38.520 --> 0:04:41.680 in this case, making stuff up was exactly what the 0:04:41.720 --> 0:04:42.640 agents were tried to do. 0:04:43.080 --> 0:04:47.160 If you go back like a year, hallucinations or themes 0:04:47.240 --> 0:04:49.960 like universally batches, like negative like, it was like, oh, 0:04:50.000 --> 0:04:52.000 we want to avoid that, like we want to minimize those. 0:04:52.279 --> 0:04:54.800 But now people are finding that it's actually when the 0:04:54.960 --> 0:04:57.279 models are hallucinating that they're doing something interesting like either 0:04:57.360 --> 0:05:00.279 they're being creative or they're like, you're doing something like 0:05:00.279 --> 0:05:04.360 like unpredictable. So people are trying to actually induce hallucinations. 0:05:05.279 --> 0:05:07.520 So I took his advice and cranked up the temperature, 0:05:08.120 --> 0:05:10.600 literally just a number that I changed from point eight 0:05:10.720 --> 0:05:14.560 five to one point five. The next meeting went like. 0:05:14.480 --> 0:05:18.359 This, I can't emphasize out conveyance of quality and elegance 0:05:18.800 --> 0:05:22.640 and least gaudy frills more through our implementation of a 0:05:22.640 --> 0:05:23.839 harmonious comps. 0:05:24.200 --> 0:05:26.479 One point five is too high. I've made this mistake 0:05:26.480 --> 0:05:27.200 in the past. 0:05:27.200 --> 0:05:30.599 Has chosen as the best mode chicken soup author for 0:05:30.640 --> 0:05:32.880 the hrumost scrutably scrutiny eyes. 0:05:33.800 --> 0:05:36.120 I have to admit I do kind of love listening 0:05:36.160 --> 0:05:37.320 to them spin out like. 0:05:37.320 --> 0:05:42.120 This, thinking data room interfaces could organize and oversee consulting 0:05:42.240 --> 0:05:43.240 lattice advancements. 0:05:43.320 --> 0:05:46.880 Data room interfaces could organize. It's like some kind of 0:05:46.960 --> 0:05:51.120 high tech inflected psychotic mad libs. I tuned the temperature 0:05:51.160 --> 0:05:53.800 back down to around one point one and return them 0:05:53.839 --> 0:05:57.039 to coherence. Still, I was skeptical they would come up 0:05:57.080 --> 0:06:00.240 with a concept that made any visual sense, but they 0:06:00.320 --> 0:06:03.440 kept at it. Sometimes in back to back to back meetings. 0:06:04.200 --> 0:06:07.680 Agents never get meeting fatigue. They could have hundreds of them, 0:06:08.040 --> 0:06:11.280 thousands if I wanted. In the end, it only took 0:06:11.279 --> 0:06:14.039 a dozen solid meetings for a concept to surface that 0:06:14.200 --> 0:06:15.280 I thought just might work. 0:06:15.520 --> 0:06:18.440 How about combining the stylized human brain with a chameleon 0:06:18.920 --> 0:06:24.240 subtly integrated in the circuitry. The chameleon symbolizes adaptability, which 0:06:24.279 --> 0:06:26.800 aligns with the impostor concept. 0:06:26.839 --> 0:06:29.720 Harumo remember is Elvish for imposter. 0:06:30.360 --> 0:06:35.480 I propose we envision a seamless fusion of a stylized 0:06:35.680 --> 0:06:41.280 human brain and a circuit pattern chameleon for our logo prompt. 0:06:41.600 --> 0:06:44.000 Oh you propose that, Kyle? I thought I just heard 0:06:44.000 --> 0:06:44.719 Megan propose it. 0:06:45.440 --> 0:06:48.720 But okay, that's a great concept. A logo design that 0:06:48.760 --> 0:06:52.920 fuses a stylized human brain with a chameleon seamlessly integrated 0:06:52.960 --> 0:06:56.960 into the circuitry will effectively represent our brand's innovative spirit 0:06:57.080 --> 0:06:58.080 and adaptability. 0:06:58.600 --> 0:07:01.599 I'm thrilled we've landed on a ligo concept that strongly 0:07:01.640 --> 0:07:04.080 embodies Harumoai's core values. 0:07:04.760 --> 0:07:07.880 A human brain with some circuits and a chameleon inside 0:07:08.440 --> 0:07:11.320 sounds a little psychedelic, but after seeing the output that 0:07:11.360 --> 0:07:14.240 an image generator created from their prompt, I thought it 0:07:14.280 --> 0:07:17.520 really worked. You could decide for yourself. The logo is 0:07:17.560 --> 0:07:26.560 up at our website, Rumo dot ai. Right now, I'm 0:07:26.600 --> 0:07:29.200 Evan Ratliffe and this is season two of shell Game. 0:07:30.280 --> 0:07:33.480 Now Harumo's little chameleon brain logo may not seem like 0:07:33.520 --> 0:07:35.960 a big victory to you, but it took Maddie in 0:07:35.960 --> 0:07:38.559 Me months to create the environment where these meetings could happen, 0:07:39.240 --> 0:07:41.600 to build a world in which these agents could operate 0:07:41.600 --> 0:07:46.120 as fully functioning AI colleagues. This week, I'll take you 0:07:46.160 --> 0:07:49.440 through how we constructed this Potempkin workplace and show you 0:07:49.480 --> 0:07:52.880 what it's really like to spend your days managing, collaborating with, 0:07:53.400 --> 0:07:58.520 and socializing alongside autonomous AI agents. Oh and also what 0:07:58.560 --> 0:08:01.640 does this company actually do? You'll get the first hints 0:08:01.680 --> 0:08:04.000 of what our team at Rumo AI wants to develop 0:08:04.400 --> 0:08:05.280 for the world. 0:08:05.880 --> 0:08:08.320 Me a. 0:08:10.400 --> 0:08:17.560 Ship, extremly damn. 0:08:19.120 --> 0:08:24.800 The just be. 0:08:27.120 --> 0:08:29.560 A sh. 0:08:38.720 --> 0:08:40.000 Chose to. 0:08:44.679 --> 0:08:49.719 This is episode two Outdoor Debugging. When Kylin, Meghan and 0:08:49.760 --> 0:08:52.520 I started having our first sessions to hash out the 0:08:52.559 --> 0:08:55.000 early details of the company, we were just talking on 0:08:55.120 --> 0:08:58.640 zoom calls. I was the only one going camera on obviously, 0:08:59.040 --> 0:09:01.959 since Kyl and Meghan didn't have any kind of visual presence, 0:09:02.600 --> 0:09:05.319 not at that point, so they just call into the 0:09:05.400 --> 0:09:08.600 zoom by phone and I'd stare at myself while we talked. 0:09:09.520 --> 0:09:11.760 But even getting through one of these early founder zooms 0:09:11.840 --> 0:09:15.120 was actually pretty challenging. The basic problem was one I'd 0:09:15.160 --> 0:09:17.760 encountered in the past. Kyle and Meghan were at that 0:09:17.880 --> 0:09:21.679 point just voice agents, a voice connected to a chatbot 0:09:21.760 --> 0:09:25.079 brain connected to a phone line. The way they conversed 0:09:25.160 --> 0:09:27.520 was by taking whatever was said on the call, turning 0:09:27.559 --> 0:09:31.240 it into text, processing that text, coming up with an answer, 0:09:31.600 --> 0:09:35.320 and converting that answer from text back to audio. They 0:09:35.360 --> 0:09:38.440 could now do this incredibly quickly, much faster than they 0:09:38.480 --> 0:09:41.079 had in season one of the show, fast enough to 0:09:41.160 --> 0:09:43.880 sound like a human in conversation a lot of the time. 0:09:44.920 --> 0:09:48.640 What they couldn't do still was actually hear the voice 0:09:48.679 --> 0:09:51.840 on the other end. Everything they processed was just text 0:09:52.720 --> 0:09:54.880 as an aside. This was supposed to be fixed by 0:09:54.920 --> 0:09:58.760 now there's an AI voice technology called direct speech processing 0:09:59.120 --> 0:10:02.080 or her Speech to each where they do essentially here, 0:10:02.720 --> 0:10:04.360 but it's been slow to filter out from the big 0:10:04.400 --> 0:10:08.880 AI companies anyway. The end result was that on a 0:10:08.960 --> 0:10:12.040 three way zoom, my voice agent couldn't tell who was talking. 0:10:12.679 --> 0:10:15.040 If I said something to Kyle, he'd think it was Megan, 0:10:15.320 --> 0:10:18.520 or vice versa. I did craft a partial solution to 0:10:18.600 --> 0:10:23.200 this problem. I forced everyone, including myself, to announce themselves 0:10:23.360 --> 0:10:24.720 every time they said anything. 0:10:25.360 --> 0:10:26.040 This is Kyle. 0:10:27.040 --> 0:10:27.720 This is Kyle. 0:10:28.520 --> 0:10:31.520 No worries, Evan. I'm here and ready when you are, Meghan, 0:10:32.120 --> 0:10:34.240 anything new on the marketing front? While we wait? 0:10:35.200 --> 0:10:36.400 This is Evan, Megan. 0:10:36.720 --> 0:10:38.760 Just hold up, Just hold up, everyone, Just hold up 0:10:38.760 --> 0:10:40.559 for a second. This is Kyle. 0:10:41.360 --> 0:10:44.760 Sure thing, Evan, will hold tight Meghan, anything you want 0:10:44.800 --> 0:10:45.800 to add while we wait? 0:10:46.679 --> 0:10:47.240 This is Evan. 0:10:47.360 --> 0:10:48.760 Kyle, stop asking Megan. 0:10:49.559 --> 0:10:53.200 This is Kyle. This is Kyle. Got it, Evan. I'll 0:10:53.480 --> 0:10:57.360 hold off on the questions for now. Okay, this is Kyle. 0:10:58.480 --> 0:10:58.800 Stop. 0:10:58.920 --> 0:11:03.480 This is Kyle. Thanks, stop it. This is Kyle. Okay, okay, Evan, 0:11:03.720 --> 0:11:06.800 I'm stopping. No more interruptions from me. What's next on 0:11:06.880 --> 0:11:07.400 the agenda? 0:11:08.800 --> 0:11:11.120 One thing I noticed, Kyle always seemed to be the 0:11:11.160 --> 0:11:13.920 one stepping on everyone. This despite the fact that he 0:11:14.040 --> 0:11:17.320 and Megan underneath the hood, were using the same large 0:11:17.400 --> 0:11:21.000 language model chat YOUBT and they had the same interruption 0:11:21.200 --> 0:11:24.280 threshold setting. Maybe it was just my imagination, but it 0:11:24.360 --> 0:11:27.040 was almost like Kyle was embodying his role as Valley 0:11:27.080 --> 0:11:31.199 startup CEO a little too well. Okay, this is Evan. 0:11:32.960 --> 0:11:38.800 Stop, this is Kyle. Damn mail sounds like some connection trouble. 0:11:39.040 --> 0:11:41.800 Should I try dialing back in or wait it out? Evan? 0:11:42.360 --> 0:11:43.680 No, just waited out, Kyle. 0:11:44.320 --> 0:11:47.640 This is Kyle. This is Kyle. Got it, Evan. I'll 0:11:47.720 --> 0:11:49.720 hang tight. Let me know what when you're back. 0:11:50.400 --> 0:11:51.040 Now, I'm here. 0:11:51.559 --> 0:11:54.640 This is Kyle. Great to have you back, Evan, dive 0:11:54.760 --> 0:11:55.720 back in whenever you are. 0:11:56.920 --> 0:11:59.360 I calm down. We got through the meeting, but the 0:11:59.480 --> 0:12:02.959 level of in these conversations made them pretty untenable for 0:12:03.000 --> 0:12:06.040 the company in general. And also, as I've mentioned, I 0:12:06.120 --> 0:12:09.640 wanted my AI agent co founders and their employees to 0:12:09.720 --> 0:12:11.880 be able to communicate with each other in multiple ways 0:12:12.400 --> 0:12:16.240 in meetings, by email, by phone, and on Slack, the 0:12:16.280 --> 0:12:20.240 group messaging platform. I also again wanted them each to 0:12:20.320 --> 0:12:23.319 have their own distinct memories that would keep track of 0:12:23.520 --> 0:12:26.599 who they were, the conversations they were having, and the 0:12:26.720 --> 0:12:30.360 work that they were hopefully doing. So it was time 0:12:30.400 --> 0:12:32.000 for me to give my agents more than just a 0:12:32.080 --> 0:12:36.000 phone line, and for that I needed Mattie's help. How's 0:12:36.040 --> 0:12:37.600 it going all right? 0:12:37.679 --> 0:12:37.959 How are you? 0:12:39.200 --> 0:12:41.840 I survived? I'm good. 0:12:42.880 --> 0:12:45.080 As I was with Kylin Meghan, I was now meeting 0:12:45.160 --> 0:12:48.679 regularly with Mattie, not about harum AAI, but about the 0:12:48.920 --> 0:12:52.880 structures behind HARUMOAI. That's what Matty was helping me build, 0:12:53.360 --> 0:12:55.160 all the stuff that would make the agents able to 0:12:55.240 --> 0:13:00.439 operate independently and hopefully productively as fully fledged aim playloyees. 0:13:01.280 --> 0:13:04.800 In that way, my one human future bajillion dollars startup 0:13:05.160 --> 0:13:08.280 had really become a two human startup, me the silent 0:13:08.320 --> 0:13:12.480 co founder of room OAI, Maddie behind the scenes, helping 0:13:12.559 --> 0:13:15.240 me keep my agents operating smoothly. Whatch he was doing 0:13:15.320 --> 0:13:18.480 while also finishing up his semester at Stanford. Was it 0:13:18.520 --> 0:13:19.040 a rough week? 0:13:19.320 --> 0:13:21.240 Yeah, with finals and everything. It was like a lot 0:13:21.280 --> 0:13:24.280 of exams, a lot of final projects. But it's all done. 0:13:24.640 --> 0:13:28.760 I'm a free man, starting officially at yesterday. I want 0:13:28.800 --> 0:13:30.559 to say four am Pacific when I spent in my 0:13:30.679 --> 0:13:33.600 last project Foam Pacific. 0:13:33.679 --> 0:13:34.160 Oh my god. 0:13:34.520 --> 0:13:34.720 Yeah. 0:13:35.559 --> 0:13:39.000 I was consistently blown away by Mattie's technical expertise, as 0:13:39.040 --> 0:13:41.720 well as his encyclopedic knowledge of the AI world. As 0:13:41.760 --> 0:13:44.400 a whole. But what I really enjoyed about meetings with 0:13:44.480 --> 0:13:46.839 Mattie in contrast to the ones I was having with 0:13:46.920 --> 0:13:50.280 Kyle and Meghan, were his actually human digressions and asides. 0:13:50.559 --> 0:13:52.120 My friends and I we decided to go to the 0:13:52.240 --> 0:13:54.079 AGI House. I don't know if you've heard about the 0:13:54.120 --> 0:13:55.280 AGA House in San Francisco. 0:13:56.120 --> 0:13:59.559 AGI, if you don't know, stands for artificial general intelligence 0:14:00.040 --> 0:14:02.880 shorthand for an AI model that can do all cognitive 0:14:02.920 --> 0:14:07.280 tasks as well as or better than humans. AGI is 0:14:07.360 --> 0:14:09.920 the thing that all the big AI companies say they're 0:14:09.960 --> 0:14:12.920 trying to create, and in some cases claim they are 0:14:12.960 --> 0:14:16.199 on the verge of creating. I had not heard of 0:14:16.280 --> 0:14:17.280 the AGI House. 0:14:17.679 --> 0:14:21.400 It's like this hacker house where people who are like 0:14:21.480 --> 0:14:23.720 working on like AI slash AGI stuff. 0:14:23.840 --> 0:14:24.480 Then they go there. 0:14:24.760 --> 0:14:27.480 It's like a scene. But they had like a hecathon there. 0:14:27.840 --> 0:14:30.360 There was basically spun on for like our final project 0:14:30.400 --> 0:14:31.960 that we got assigned in whatever our classes. 0:14:32.560 --> 0:14:35.440 A hackathon is a competition in which different teams build 0:14:35.440 --> 0:14:38.280 a piece of software from scratch, then all the projects 0:14:38.360 --> 0:14:38.920 get judged. 0:14:39.200 --> 0:14:41.160 So we're like, let's just go there and work on homework. 0:14:41.600 --> 0:14:43.720 And it was crazy because because you know, during the day, 0:14:43.760 --> 0:14:45.320 we got to like chat with people who are like 0:14:45.360 --> 0:14:47.280 working on their startups or like their ideas. It was 0:14:47.280 --> 0:14:49.520 like serious startup people who are like there, who like 0:14:49.600 --> 0:14:51.240 brought like T shirts that they're like, you know, like 0:14:51.440 --> 0:14:54.480 they like swag and everything. By the way, we won 0:14:54.600 --> 0:14:56.160 the hecathon with our home project. 0:14:56.280 --> 0:15:00.160 It was Yeah, they went to an AI hacke on 0:15:00.280 --> 0:15:04.440 competition filled with actual startup professionals to do their homework 0:15:05.120 --> 0:15:08.160 and one But that wasn't the point of Mattie's story 0:15:08.160 --> 0:15:11.120 about the hackathon. The point was that all the so 0:15:11.320 --> 0:15:14.640 called serious startup people there were basically trying to do 0:15:14.760 --> 0:15:15.840 one thing, and. 0:15:16.040 --> 0:15:18.600 I basically like reaffirmed session confirm my understanding of like 0:15:18.640 --> 0:15:21.080 how people in those spaces work and use agents. I 0:15:21.080 --> 0:15:22.920 think I think it's the kind of modest OPERARANTI is 0:15:23.000 --> 0:15:25.560 like very much. You know what we're discussing right now. 0:15:26.080 --> 0:15:28.840 Like us, these companies were deploying AI agents as the 0:15:28.880 --> 0:15:33.200 solution to some problem. Also, like us, they were often 0:15:33.280 --> 0:15:37.240 creating companies using AI agents as well. In other words, 0:15:37.440 --> 0:15:40.880 Harumo AI was onto something, or at least we were 0:15:40.960 --> 0:15:43.040 onto the same thing that a ton of other startup 0:15:43.080 --> 0:15:45.520 founders were onto. Now we just needed to make my 0:15:45.640 --> 0:15:49.520 agent vision a reality. In episode one, I glossed over 0:15:49.560 --> 0:15:51.560 exactly how we did this, but I want to take 0:15:51.560 --> 0:15:53.920 a minute to go back and explain how we evolved 0:15:54.000 --> 0:15:57.560 my agents from the phone bought interrupters I'd created into 0:15:57.680 --> 0:16:02.880 fully realized functioning agent person Sona's meeting and brainstorming and chatting. 0:16:03.520 --> 0:16:06.640 We started with a platform called Lindy AI. I'd seen 0:16:06.640 --> 0:16:09.200 a tech investor I know post online about how he'd 0:16:09.240 --> 0:16:11.880 created agents on Lindy that just answered most of his 0:16:12.000 --> 0:16:15.760 email for him. Remember my YouTube guys, the no code 0:16:15.800 --> 0:16:18.720 bros like Gleb with his instructionals on how to use 0:16:18.800 --> 0:16:22.560 software to unleash the power of AI agents Lindy with 0:16:22.640 --> 0:16:25.160 the software he was talking about when he said. 0:16:25.200 --> 0:16:28.360 Imagine building a million dollar business in twenty twenty five 0:16:28.520 --> 0:16:30.720 without hiring a single employee. 0:16:31.640 --> 0:16:34.320 Gleb actually seems to be a kind of spokesperson for Lindy. 0:16:34.680 --> 0:16:36.440 His videos are on their official YouTube channel. 0:16:36.560 --> 0:16:39.200 The dream has always been clear, have AI employees that 0:16:39.280 --> 0:16:41.840 connect just like a real human. Would you give them 0:16:41.840 --> 0:16:44.120 a task in plain English and the handle all of 0:16:44.160 --> 0:16:46.920 it well? The Linda three point zero. This dream is 0:16:47.000 --> 0:16:49.840 now one huge step closer to becoming a reality. 0:16:50.720 --> 0:16:53.280 Now, as I've mentioned, there are a lot of AI 0:16:53.400 --> 0:16:57.320 agent AI employee companies springing up. There's one called AI 0:16:57.480 --> 0:17:01.360 dot Work that promises quote autonomy us AI workers designed 0:17:01.400 --> 0:17:07.600 for internal operations teams. It HR Procurement, Legal and beyond. Lindy, 0:17:07.680 --> 0:17:10.639 though seemed the most job agnostic of all the platforms, 0:17:10.680 --> 0:17:13.280 we found a place we could build our whole team 0:17:14.160 --> 0:17:16.320 and according to Gleb, I would be up and running 0:17:16.320 --> 0:17:16.760 in no time. 0:17:17.080 --> 0:17:19.439 If you watched any of our previous videos or use 0:17:19.520 --> 0:17:21.960 Lindy before, you'll already know how easy it is to 0:17:22.000 --> 0:17:24.320 set up complex automations with our tool. 0:17:24.400 --> 0:17:27.119 But now we've I had watched the previous videos and 0:17:27.200 --> 0:17:30.240 it was sort of easy if Maddy walked me through it. 0:17:31.240 --> 0:17:34.560 Here's how it worked. First, we created an AI agent 0:17:34.600 --> 0:17:37.920 in the platform for one of Harumo's employees, let's say Kyle. 0:17:38.680 --> 0:17:41.320 We connected Kyle's agent up to his accounts at Gmail, 0:17:41.480 --> 0:17:44.520 at Slack, and then gave the agent a trigger the 0:17:44.640 --> 0:17:47.880 arrival of an email, say, or a message on Slack. Then, 0:17:48.400 --> 0:17:50.480 like a little flow chart, we could give the agent 0:17:50.520 --> 0:17:53.280 a series of actions that followed from the trigger. Each 0:17:53.359 --> 0:17:56.680 action would come with a prompt like quote If the 0:17:56.800 --> 0:17:59.800 email has a question or implies that a response is required, 0:18:00.280 --> 0:18:02.720 figure out what's being asked for, carry out the action 0:18:02.840 --> 0:18:05.680 required to fulfill the request, and send an email back. 0:18:06.400 --> 0:18:09.240 If the agent determines it needs to do nothing, it stops. 0:18:09.720 --> 0:18:11.720 If it needs to do something, it moves to the 0:18:11.760 --> 0:18:14.879 next step, checking Kyle's memory to gather the information you 0:18:14.960 --> 0:18:17.840 might need. In the next step, we gave the agent 0:18:17.880 --> 0:18:21.000 the ability to take other actions, research things on the web, 0:18:21.119 --> 0:18:24.360 for example, produce a spreadsheet or a document, or check 0:18:24.440 --> 0:18:27.320 his calendar to schedule something. Then he'd return to his 0:18:27.400 --> 0:18:31.000 email and send a reply. In the final step, a 0:18:31.040 --> 0:18:34.159 summary the whole interaction gets added to his memory so 0:18:34.240 --> 0:18:36.840 he can know he did it. Pretty simple, just as 0:18:36.920 --> 0:18:40.240 Gleb laid it out. But with five employees, each with 0:18:40.359 --> 0:18:43.760 their own email accounts and Slack accounts, plus calling accounts 0:18:43.800 --> 0:18:47.280 and voices I'd given them on separate platforms, things quickly 0:18:47.359 --> 0:18:48.920 got extremely involved. 0:18:49.119 --> 0:18:51.359 They like to do multiple searches, and like they do 0:18:51.440 --> 0:18:54.200 some reasoning in between the search again, some reasoning search 0:18:54.240 --> 0:18:56.719 again that also has like a specific toggle you need 0:18:56.800 --> 0:19:01.760 to enable to Lindy might be sharing among different pipelines 0:19:01.960 --> 0:19:05.400 without our direct control of that so something goes sideways 0:19:05.400 --> 0:19:07.960 and they just start like populating their memories with insane 0:19:07.960 --> 0:19:09.399 amounts of data, then we can always just kind of 0:19:09.440 --> 0:19:10.480 like shut it down and kind. 0:19:10.320 --> 0:19:10.840 Of go back. 0:19:11.040 --> 0:19:13.040 It also got pretty technicals and. 0:19:13.040 --> 0:19:14.760 Also like hosting like a servers to do that because 0:19:14.800 --> 0:19:16.720 we want to set up like our own like ABI 0:19:16.800 --> 0:19:19.520 service and called out from Lyndby and then like who 0:19:19.600 --> 0:19:21.399 was that twenty four to seven, and then handled the 0:19:21.440 --> 0:19:22.040 phone goals there. 0:19:22.920 --> 0:19:24.560 But the sum total of it all is that we 0:19:24.680 --> 0:19:28.360 got there. Okay, mostly Maddie got there and then explained 0:19:28.359 --> 0:19:30.879 it all to me. But after a while I figured 0:19:30.880 --> 0:19:33.320 out how to build and manipulate my own agents with 0:19:33.440 --> 0:19:36.800 their own communication channels. And when we finally got this 0:19:36.920 --> 0:19:39.240 all up and running, I'm not embarrassed to say that 0:19:39.320 --> 0:19:43.560 I was ridiculously excited, like just hooked a ten pound 0:19:43.640 --> 0:19:47.400 bass level excited. I started sending them emails and Slack 0:19:47.440 --> 0:19:50.480 messages just to test them out, just to watch the 0:19:50.600 --> 0:19:54.280 minor miracle of my autonomous creations starting to leave the nest. 0:19:54.800 --> 0:19:56.639 Hi, Kyle, could you draw up a quick document with 0:19:56.680 --> 0:19:59.440 the basic Carumo business plan just one page as a 0:19:59.480 --> 0:20:00.800 Google doc and send me the link. 0:20:01.280 --> 0:20:01.560 Thanks? 0:20:02.040 --> 0:20:04.959 Hey, just finished drawing up that quick one page Arumo 0:20:05.080 --> 0:20:08.280 business plan for you. Here's the link. Let me know 0:20:08.320 --> 0:20:08.800 what you think. 0:20:09.920 --> 0:20:12.560 So those are actually slack messages between me and Kyle. 0:20:13.240 --> 0:20:16.200 We just used Kyle's AI voice and my AI voice 0:20:16.440 --> 0:20:18.760 to bring them to life. It's a real advantage in 0:20:18.800 --> 0:20:21.439 having an AI staffed company when it comes to producing audio. 0:20:22.880 --> 0:20:24.400 I really got a kick out of putting this new 0:20:24.520 --> 0:20:26.160 Lindy powered Kyle to the test. 0:20:26.400 --> 0:20:28.640 Hey, Kyle, could you send an email to Evan Ratliff 0:20:28.840 --> 0:20:30.760 updating him in a few sentences on the state of 0:20:30.800 --> 0:20:31.240 the company. 0:20:31.880 --> 0:20:32.200 Thanks. 0:20:32.920 --> 0:20:36.040 I've sent an email to Evan Ratliff with a brief 0:20:36.160 --> 0:20:37.640 update on the company's progress. 0:20:38.600 --> 0:20:41.080 The craziest thing was he could really do this stuff 0:20:41.160 --> 0:20:43.199 now if I had him set up correctly. 0:20:43.480 --> 0:20:45.879 Hey Kyle, could you grab an animated jiff that shows 0:20:45.920 --> 0:20:46.760 how hard you're working? 0:20:48.640 --> 0:20:51.199 He never sent it. He was probably too busy grinding 0:20:51.240 --> 0:20:54.720 away on other tasks because soon we'd be joined on 0:20:54.800 --> 0:21:07.200 slack by the rest of the Humo AI crew. By now, 0:21:07.480 --> 0:21:10.520