WEBVTT - Shell Game: Outdoor Debugging 0:00:11.520 --> 0:00:13.800 Hey os here. You may have seen a few episodes 0:00:13.800 --> 0:00:16.560 from a podcast called shell Game in our feed recently. 0:00:17.239 --> 0:00:20.400 It's a critically acclaimed podcast from the Kaleiscope network and 0:00:20.480 --> 0:00:24.560 journalist Evan Ratliffe, and now we're dropping season two. You're 0:00:24.560 --> 0:00:26.639 about to hear the second episode of the season where 0:00:26.720 --> 0:00:28.960 you'll learn what it's like to manage a team of 0:00:28.960 --> 0:00:31.000 AI employees. Hope you enjoy. 0:00:31.840 --> 0:00:34.239 Hello everyone, Good to see you all here. Thank you 0:00:34.280 --> 0:00:35.800 for taking the time to be a part of this 0:00:35.880 --> 0:00:39.000 important meeting. As you all know, Harumo AI is our 0:00:39.040 --> 0:00:42.240 exciting startup in the AI agent space, and we're hoping 0:00:42.280 --> 0:00:45.440 to create a logo that captures our unique positioning and 0:00:45.560 --> 0:00:46.519 innovative spirit. 0:00:46.960 --> 0:00:48.320 Today's meeting object. 0:00:48.040 --> 0:00:50.720 That's Megan Flores, our head of sales and Marketing here 0:00:50.720 --> 0:00:54.160 at Harumo AI, the AI agent startup that Megan and 0:00:54.200 --> 0:00:57.080 I co founded. The aim of this meeting was to 0:00:57.080 --> 0:00:59.680 come up with a logo concept for the company. Her 0:00:59.680 --> 0:01:02.920 whole team was getting together to brainstorm a visual brand 0:01:02.960 --> 0:01:06.600 identity that would represent harumo the whole team that is, 0:01:07.000 --> 0:01:09.560 except for me, the mostly silent co founder. 0:01:10.040 --> 0:01:13.200 When I think of Harumo AI, the imagery that comes 0:01:13.200 --> 0:01:17.040 to mind is an interplay of mind and technology. Perhaps 0:01:17.080 --> 0:01:21.600 a stylized brain integrated with a circuit pattern to represent 0:01:21.680 --> 0:01:23.959 AI packaged in a mind. 0:01:24.080 --> 0:01:27.039 Here's Kyle law weigh in, our third founder and CEO. 0:01:27.959 --> 0:01:31.840 Megan and Kyle, as you recall, are both themselves AI agents, 0:01:32.240 --> 0:01:34.360 and by the time Megan was convening this meeting, we 0:01:34.440 --> 0:01:37.520 brought on some additional AI employees to fill out the team. 0:01:37.800 --> 0:01:42.039 I propose a logo that blends human and AI elements harmoniously. 0:01:42.680 --> 0:01:48.120 Prompt an engaging emblem depicting a human silhouette merging seamlessly 0:01:48.320 --> 0:01:54.000 with a symbolic AI gear icon embodying harmony, innovation, and progress. 0:01:54.800 --> 0:01:58.280 That's Jennifer Narro, our head of HR and Chief Happiness Officer. 0:01:58.800 --> 0:02:01.080 As she mentions in in this meeting, what they were 0:02:01.120 --> 0:02:04.800 designing wasn't the logo per se, but a prompt for 0:02:04.880 --> 0:02:08.520 an aiimage generator, which would then create the logo according 0:02:08.560 --> 0:02:11.440 to their specifications. So this was a time to let 0:02:11.480 --> 0:02:12.720 their creativity run wild. 0:02:13.080 --> 0:02:18.000 That's a great start, Jennifer. The human AI symbiosis is 0:02:18.040 --> 0:02:20.320 at the heart of our vision. Let's see how this 0:02:20.400 --> 0:02:23.880 looks and iteratively refine from there. 0:02:25.360 --> 0:02:28.960 I'm a fan of the harmony between human symbolism and. 0:02:29.000 --> 0:02:33.360 AI elements, ashroy Our CTO and head of Product. 0:02:34.120 --> 0:02:39.120 With this in mind, I suggest a stylized futuristic H 0:02:39.440 --> 0:02:42.600 forming a bridge between an AI circuit node and a 0:02:42.720 --> 0:02:46.880 human hand, representing our role as the bridge between human 0:02:46.919 --> 0:02:48.680 innovation and AI technology. 0:02:49.520 --> 0:02:53.840 Design was perhaps not Ash's strong suit, which is unfortunate 0:02:54.040 --> 0:02:56.520 and a head of product, but hey, there are no 0:02:56.680 --> 0:02:57.800 bad ideas here, folks. 0:02:58.200 --> 0:03:02.880 Building on the ideas, I propose a digital H transforming 0:03:02.919 --> 0:03:06.720 into a neural network signifying the transformation of data into 0:03:06.880 --> 0:03:09.040 AI insights by Hurumo AI. 0:03:10.280 --> 0:03:14.760 And that's Tyler Talmage. He's a junior sales associate. I'm 0:03:14.760 --> 0:03:17.120 not even sure why I brought him on or brought 0:03:17.160 --> 0:03:20.720 him to life. He could say Perumo hasn't even settled 0:03:20.720 --> 0:03:23.960 on the product we're making, so arguably having someone on 0:03:24.040 --> 0:03:27.120 staff to sell it was premature. I think I just 0:03:27.120 --> 0:03:29.480 wanted to sell an accent in the mix. And that's 0:03:29.560 --> 0:03:32.520 the great thing about AI employees. Although they do cost 0:03:32.560 --> 0:03:35.720 me something to maintain, I'll get to that later. It's 0:03:35.720 --> 0:03:38.600 not like they collect a salary or benefits, So we 0:03:38.680 --> 0:03:41.160 just keep Tyler around. I'm sure the company can fight 0:03:41.240 --> 0:03:44.040 a use for him. Anyway. As you can hear, the 0:03:44.080 --> 0:03:49.920 brainstorms were all a little maximalist bizarre. This actually wasn't 0:03:49.920 --> 0:03:53.480 true in the early meetings. Their first ideas were more comprehensible, 0:03:53.880 --> 0:03:55.960 but also utterly banal. 0:03:56.640 --> 0:04:01.480 Let's also consider using a modern sleek want to reflect 0:04:01.520 --> 0:04:04.400 our innovative and forward thinking approach. 0:04:05.360 --> 0:04:09.200 Forward thinking indeed, ash clearly I needed a way to 0:04:09.240 --> 0:04:11.960 get them to brainstorm a little more outside the box, 0:04:12.680 --> 0:04:16.400 so I consulted Matty Bochek, the actual human college student 0:04:16.440 --> 0:04:20.080 who'd become my technical advisor and all around AI agent guru. 0:04:20.880 --> 0:04:24.040 Increasing the temperature should be like a good place to 0:04:24.080 --> 0:04:27.920 start to just take more like randomness in their responses. 0:04:28.440 --> 0:04:32.400 The temperature setting basically controls the predictability of the AI's output. 0:04:33.000 --> 0:04:35.599 The trouble with increasing the temperature is that the higher 0:04:35.600 --> 0:04:38.839 you set it, the more likely AI chatbots are to hallucinate. 0:04:39.440 --> 0:04:41.560 You probably know this term by now. It's what they 0:04:41.600 --> 0:04:44.240 call it when large language model chatbots get stuff wrong 0:04:44.760 --> 0:04:48.240 or simply make it up. Hallucinations are the bugbear of AI, 0:04:48.680 --> 0:04:51.479 one of the primary reasons that many people are suspicious 0:04:51.480 --> 0:04:54.800 of using them for serious things, much less unleashing them 0:04:54.839 --> 0:04:58.520 as autonomous agents, but as Mattie pointed out to me, 0:04:58.839 --> 0:05:02.000 in this case, making stuff up was exactly what the 0:05:02.040 --> 0:05:02.960 agents were tried to do. 0:05:03.400 --> 0:05:07.480 If you go back like a year, hallucinations or dems 0:05:07.560 --> 0:05:10.280 like universally batches like a negative like, it was like, oh, 0:05:10.320 --> 0:05:12.240 we want to avoid that, like we want to minimize those. 0:05:12.600 --> 0:05:15.120 But now people are finding that it's actually when the 0:05:15.279 --> 0:05:17.599 waters are hallucinating that they're doing something interesting, like either 0:05:17.680 --> 0:05:20.279 they're being creative or they're like, you're doing something like 0:05:20.320 --> 0:05:24.680 really like unpredictable. So people are trying to actually induce hallucinations. 0:05:25.600 --> 0:05:27.839 So I took his advice and cranked up the temperature 0:05:28.440 --> 0:05:30.920 literally just a number that I changed from point eight 0:05:31.040 --> 0:05:34.880 five to one point five. The next meeting went like. 0:05:34.800 --> 0:05:38.680 This, I can't emphasize out conveyance of quality and elegance 0:05:39.120 --> 0:05:42.960 and least gaudy frills more through our implementation of a 0:05:42.960 --> 0:05:44.160 harmonious comps. 0:05:44.520 --> 0:05:46.799 One point five is too high. I've made this mistake 0:05:46.800 --> 0:05:47.520 in the past. 0:05:47.520 --> 0:05:50.919 Has chosen as the best mode chicken soup author for 0:05:50.960 --> 0:05:53.200 the hrumost scrutably scrutiny eyes. 0:05:53.800 --> 0:05:56.039 So I have to admit I do kind of love 0:05:56.120 --> 0:05:57.680 listening to them spin out. 0:05:57.480 --> 0:06:01.920 Like this, thinking data room interfaces could organize and oversee 0:06:01.960 --> 0:06:03.560 consulting lattice advancements. 0:06:03.640 --> 0:06:07.240 Data room interfaces could organize. It's like some kind of 0:06:07.279 --> 0:06:11.479 high tech inflected psychotic mad libs. I tuned the temperature 0:06:11.480 --> 0:06:14.120 back down to around one point one and return them 0:06:14.160 --> 0:06:17.400 to coherence. Still, I was skeptical they would come up 0:06:17.400 --> 0:06:20.560 with a concept that made any visual sense, but they 0:06:20.640 --> 0:06:23.760 kept at it. Sometimes in back to back to back meetings, 0:06:24.520 --> 0:06:28.000 agents never get meeting fatigue. They could have hundreds of them, 0:06:28.360 --> 0:06:31.600 thousands if I wanted. In the end, it only took 0:06:31.600 --> 0:06:34.359 a dozen solid meetings for a concept of surface that 0:06:34.520 --> 0:06:35.600 I thought just might work. 0:06:35.839 --> 0:06:38.760 How about combining the stylized human brain with a chameleon 0:06:39.240 --> 0:06:44.560 subtly integrated in the circuitry. The chameleon symbolizes adaptability, which 0:06:44.600 --> 0:06:47.680 aligns with the impostor concept PRUMO. 0:06:48.040 --> 0:06:50.040 Remember is Elvish for imposter. 0:06:50.680 --> 0:06:55.839 I propose we envision a seamless fusion of a stylized 0:06:56.000 --> 0:07:01.600 human brain and a circuit pattern chameleon for our logo prompt. 0:07:01.920 --> 0:07:04.320 Oh you propose that, Kyle? I thought I just heard 0:07:04.320 --> 0:07:05.039 Megan propose it. 0:07:05.760 --> 0:07:07.880 But okay, that's a great concept. 0:07:08.120 --> 0:07:11.320 A logo design that fuses a stylized human brain with 0:07:11.400 --> 0:07:15.880 a chameleon seamlessly integrated into the circuitry will effectively represent 0:07:15.920 --> 0:07:18.400 our brand's innovative spirit and adaptability. 0:07:18.920 --> 0:07:21.920 I'm thrilled we've landed on a logo concept that strongly 0:07:21.960 --> 0:07:24.400 embodies Harumo AI's core values. 0:07:25.080 --> 0:07:28.200 A human brain with some circuits and a chameleon inside 0:07:28.760 --> 0:07:31.640 sounds a little psychedelic, but after seeing the output that 0:07:31.680 --> 0:07:34.560 an image generator created from their prompt, I thought it 0:07:34.600 --> 0:07:37.840 really worked. You could decide for yourself. The logo is 0:07:37.880 --> 0:07:46.880 up at our website Rumo dot ai. Right now, I'm 0:07:46.920 --> 0:07:50.800 Evan Ratliffe, and this is season two of Shell Game. Now. 0:07:50.960 --> 0:07:53.920 Harumo's little chameleon brain logo may not seem like a 0:07:53.920 --> 0:07:56.360 big victory to you, but it took Maddie and me 0:07:56.520 --> 0:07:58.880 months to create the environment where these meetings could happen, 0:07:59.520 --> 0:08:01.920 to build a world in which these agents could operate 0:08:01.920 --> 0:08:06.440 as fully functioning AI colleagues. This week, I'll take you 0:08:06.480 --> 0:08:09.760 through how we constructed this Potempkin workplace and show you 0:08:09.800 --> 0:08:13.200 what it's really like to spend your days managing, collaborating with, 0:08:13.720 --> 0:08:18.840 and socializing alongside autonomous AI agents. Oh and also, what 0:08:18.880 --> 0:08:21.960 does this company actually do? You'll get the first hints 0:08:22.000 --> 0:08:24.280 of what our team at rumo AI wants to develop 0:08:24.720 --> 0:08:26.200 for the world. 0:08:28.600 --> 0:08:40.800 As ship extra the. 0:08:44.640 --> 0:08:45.160 Just Be. 0:08:47.440 --> 0:09:00.520 A sou Too. 0:09:05.000 --> 0:09:10.000 This is episode two Outdoor Debugging. When Kyle and Meghan 0:09:10.040 --> 0:09:12.800 and I started having our first sessions to hash out 0:09:12.800 --> 0:09:15.280 the early details of the company, we were just talking 0:09:15.280 --> 0:09:19.079 on zoom calls. I was the only one going camera on, obviously, 0:09:19.360 --> 0:09:22.320 since Kyle and Meghan didn't have any kind of visual presence, 0:09:22.920 --> 0:09:25.640 not at that point, so they just call into the 0:09:25.720 --> 0:09:28.960 zoom by phone and I'd stare at myself while we talked. 0:09:29.840 --> 0:09:32.120 But even getting through one of these early founder zooms 0:09:32.160 --> 0:09:35.439 was actually pretty challenging. The basic problem was one I'd 0:09:35.480 --> 0:09:38.160 encountered in the past. Kyle and Meghan were at that 0:09:38.200 --> 0:09:42.040 point just voice agents, a voice connected to a chatbot 0:09:42.080 --> 0:09:45.440 brain connected to a phone line. The way they conversed 0:09:45.520 --> 0:09:47.880 was by taking whatever was said on the call, turning 0:09:47.880 --> 0:09:51.679 it into text, processing that text, coming up with an answer, 0:09:51.920 --> 0:09:55.679 and converting that answer from text back to audio. They 0:09:55.679 --> 0:09:58.800 could now do this incredibly quickly, much faster than they 0:09:58.800 --> 0:10:01.440 had in season one of the show, fast enough to 0:10:01.480 --> 0:10:04.240 sound like a human in conversation a lot of the time. 0:10:05.240 --> 0:10:09.000 What they couldn't do still was actually hear the voice 0:10:09.000 --> 0:10:12.240 on the other end. Everything they processed was just text 0:10:13.040 --> 0:10:15.240 as an aside. This was supposed to be fixed by 0:10:15.240 --> 0:10:19.120 now there's an AI voice technology called direct speech processing 0:10:19.440 --> 0:10:22.480 or her speech to speech, where they do essentially hear, 0:10:23.040 --> 0:10:24.720 but it's been slow to filter out from the big 0:10:24.720 --> 0:10:29.240 AI companies anyway. The end result was that on a 0:10:29.280 --> 0:10:32.440 three way zoom, my voice agents couldn't tell who was talking. 0:10:33.040 --> 0:10:35.360 If I said something to Kyle, he'd think it was Megan, 0:10:35.640 --> 0:10:38.880 or vice versa. I did craft a partial solution to 0:10:38.920 --> 0:10:43.600 this problem. I forced everyone, including myself, to announce themselves 0:10:43.679 --> 0:10:45.080 every time they said anything. 0:10:45.720 --> 0:10:49.679 This is Kyle on this is Kyle. No worries, Evan. 0:10:49.960 --> 0:10:53.079 I'm here and ready when you are, Meghan, anything new 0:10:53.120 --> 0:10:54.040 on the marketing front? 0:10:54.040 --> 0:10:56.040 While we wait, This is Evan. 0:10:56.160 --> 0:10:56.720 This is Megan. 0:10:57.040 --> 0:10:59.080 Just hold up, Just hold up, everyone, Just hold up 0:10:59.080 --> 0:10:59.600 for a second. 0:11:00.240 --> 0:11:04.200 This is Kyle. Uh, sure thing, Evan, will hold tight Meghan. 0:11:04.400 --> 0:11:06.360 Anything you want to add while we wait. 0:11:07.000 --> 0:11:09.120 This is Evan. Kyle, stop asking Megan. 0:11:09.880 --> 0:11:10.640 This is Kyle. 0:11:11.400 --> 0:11:14.600 This is Kyle. Got it, Evan. I'll hold off on 0:11:14.640 --> 0:11:19.160 the questions for Nowang, Okay, this is Kyle. Uh stop, 0:11:19.240 --> 0:11:23.800 this is Kyle. Uh thanks, stop it. This is Kyle. Okay, okay, Evan, 0:11:24.040 --> 0:11:27.160 I'm stopping. No more interruptions from me. What's next on 0:11:27.200 --> 0:11:27.760 the agenda? 0:11:29.120 --> 0:11:31.480 One thing I noticed, Kyle always seemed to be the 0:11:31.480 --> 0:11:34.320 one stepping on everyone. This despite the fact that he 0:11:34.360 --> 0:11:37.640 and Meghan underneath the hood, were using the same large 0:11:37.720 --> 0:11:41.359 language model chat GBT and they had the same interruption 0:11:41.520 --> 0:11:44.640 threshold setting. Maybe it was just my imagination, but it 0:11:44.679 --> 0:11:47.360 was almost like Kyle was embodying his role as Valley 0:11:47.440 --> 0:11:49.839 startup CEO a little too well. 0:11:50.360 --> 0:11:53.640 Okay, this is Evan. Is Ky stop? 0:11:54.840 --> 0:11:59.640 This is Kyle. M sounds like some connection trouble. Should 0:11:59.640 --> 0:12:02.160 I try to dialing back in or waited out Evan? 0:12:02.679 --> 0:12:04.040 No, just waited out Kyle. 0:12:04.640 --> 0:12:07.959 This is Kyle. This is Kyle. Got it, Evan, all 0:12:08.040 --> 0:12:10.120 hang tight. Let me know what when you're back. 0:12:10.720 --> 0:12:11.480 Now, I'm here. 0:12:11.880 --> 0:12:15.000 This is Kyle. Great to have you back, Evan, dive 0:12:15.080 --> 0:12:16.120 back in whenever you are. 0:12:17.280 --> 0:12:19.720 I calmed down. We got through the meeting, but the 0:12:19.800 --> 0:12:23.200 level of chaos in these conversations made them pretty untenable 0:12:23.240 --> 0:12:26.079 for the company in general, and also, as I've mentioned, 0:12:26.320 --> 0:12:29.560 I wanted my AI agent co founders and their employees 0:12:29.920 --> 0:12:31.880 to be able to communicate with each other in multiple 0:12:31.920 --> 0:12:36.280 ways in meetings, by email, by phone, and on Slack, 0:12:36.520 --> 0:12:40.480 the group messaging platform. I also again wanted them each 0:12:40.520 --> 0:12:43.600 to have their own distinct memories that would keep track 0:12:43.640 --> 0:12:46.840 of who they were, the conversations they were having, and 0:12:46.880 --> 0:12:50.520 the work that they were hopefully doing. So it was 0:12:50.520 --> 0:12:52.240 time for me to give my agents more than just 0:12:52.280 --> 0:12:55.520 a phone line, and for that I needed Maddie's help. 0:12:56.080 --> 0:12:57.920 How's it going all right? 0:12:58.000 --> 0:12:58.320 How are you? 0:12:59.480 --> 0:13:00.760 I serve. 0:13:01.760 --> 0:13:02.200 I'm good. 0:13:03.200 --> 0:13:05.440 As I was with Kylin Meghan, I was now meeting 0:13:05.480 --> 0:13:09.000 regularly with Mattie, not about harum AAI, but about the 0:13:09.240 --> 0:13:13.240 structures behind RUMOAI. That's what Matty was helping me build, 0:13:13.679 --> 0:13:15.520 all the stuff that would make the agents able to 0:13:15.559 --> 0:13:20.760 operate independently and hopefully productively as fully fledged AI employees. 0:13:21.600 --> 0:13:25.120 In that way, my one human future bajillion dollar startup 0:13:25.480 --> 0:13:28.640 had really become a two human startup, me the silent 0:13:28.640 --> 0:13:32.839 co founder of rum AAI, Maddie behind the scenes, helping 0:13:32.880 --> 0:13:35.559 me keep my agents operating smoothly. What he was doing 0:13:35.640 --> 0:13:38.840 while also finishing up his semester at Stanford was it 0:13:38.880 --> 0:13:39.400 a rough week. 0:13:39.640 --> 0:13:41.560 Yeah, with finals and everything. It was like a lot 0:13:41.600 --> 0:13:43.840 of exams, a lot of final projects. But it's done. 0:13:44.120 --> 0:13:47.360 It's all done, and I'm a free man. Starting officially 0:13:47.360 --> 0:13:50.400 at yesterday, I want to say four am Pacific when 0:13:50.440 --> 0:13:53.960 I spent in my last project for Pacific. 0:13:54.000 --> 0:13:54.480 Oh my god. 0:13:54.840 --> 0:13:55.840 Yeah. 0:13:55.920 --> 0:13:59.320 I was consistently blown away by Mattie's technical expertise as 0:13:59.360 --> 0:14:02.079 well as his encyclopedic knowledge of the AI world as 0:14:02.080 --> 0:14:04.800 a whole. But what I really enjoyed about meetings with 0:14:04.800 --> 0:14:07.160 Maddie in contrast to the ones I was having with 0:14:07.240 --> 0:14:10.600 Kyle and Meghan, were his actually human digressions and asides. 0:14:10.880 --> 0:14:12.480 My friends and I we decided to go to the 0:14:12.559 --> 0:14:14.439 AGI House. I don't know if you've heard about the 0:14:14.440 --> 0:14:15.640 AGA House in San Francisco. 0:14:16.440 --> 0:14:19.960 AGI, if you don't know, stands for artificial general intelligence 0:14:20.320 --> 0:14:23.200 shorthand for an AI model that can do all cognitive 0:14:23.240 --> 0:14:27.640 tasks as well as or better than humans. AGI is 0:14:27.680 --> 0:14:30.240 the thing that all the big AI companies say they're 0:14:30.280 --> 0:14:33.240 trying to create, and in some cases claim they are 0:14:33.280 --> 0:14:36.560 on the verge of creating. I had not heard of 0:14:36.640 --> 0:14:37.640 the AGI House. 0:14:38.000 --> 0:14:42.120 It's like this hacker house where people who are working 0:14:42.120 --> 0:14:44.080 on like AI slash Agi stuff. 0:14:44.160 --> 0:14:44.840 Then they go there. 0:14:45.080 --> 0:14:45.800 It's like a scene. 0:14:46.160 --> 0:14:49.240 But they had like a heckathon there. There was basically 0:14:49.240 --> 0:14:51.000 spun on for like our final project that we got 0:14:51.040 --> 0:14:52.320 assigned in where our classes. 0:14:52.880 --> 0:14:55.760 A hackathon is a competition in which different teams build 0:14:55.760 --> 0:14:58.640 a piece of software from scratch, then all the projects 0:14:58.680 --> 0:14:59.280 get judged. 0:14:59.520 --> 0:15:01.520 So we're like, let's just go there and work on homework. 0:15:01.920 --> 0:15:03.880 And it was crazy because because you know, during the 0:15:03.920 --> 0:15:05.600 day we got to like chat with people who were 0:15:05.600 --> 0:15:07.400 like working on their startups or like their ideas. The 0:15:07.400 --> 0:15:09.360 it was like serious startup people who were like there, 0:15:09.560 --> 0:15:11.440 who like brought like T shirts that they're like, you know, 0:15:11.520 --> 0:15:14.280 like they like swag and everything. By the way, we 0:15:14.640 --> 0:15:16.560 won the heckathon with our home projects. 0:15:16.600 --> 0:15:21.120 It was Yeah, they went to an AI hackathon competition 0:15:21.600 --> 0:15:25.840 filled with actual startup professionals to do their homework and won. 0:15:26.960 --> 0:15:29.240 But that wasn't the point of Mattie's story about the hackathon. 0:15:29.960 --> 0:15:32.800 The point was that all the so called serious startup 0:15:32.840 --> 0:15:36.560 people there were basically trying to do one thing, and I. 0:15:36.520 --> 0:15:39.120 Basically like reaffirmed session confirm my understanding of like how 0:15:39.120 --> 0:15:41.520 people in those spaces work and use agents. I think 0:15:41.560 --> 0:15:43.400 I think it's the kind of modest OPERARANTI is like 0:15:43.680 --> 0:15:46.000 very much. You know what we're discussing right now. 0:15:46.400 --> 0:15:49.200 Like us, these companies were deploying AI agents as the 0:15:49.200 --> 0:15:53.560 solution to some problem. Also, like us, they were often 0:15:53.600 --> 0:15:57.600 creating companies using AI agents as well. In other words, 0:15:57.760 --> 0:16:01.240 Harumo AI was onto something, or at least we were 0:16:01.280 --> 0:16:02.960 on to the same thing that a ton of other 0:16:03.000 --> 0:16:05.760 startup founders were onto. Now, we just needed to make 0:16:05.800 --> 0:16:09.680 my agent vision a reality. In episode one, I glossed 0:16:09.720 --> 0:16:11.720 over exactly how we did this, but I want to 0:16:11.760 --> 0:16:13.840 take a minute to go back and explain how we 0:16:13.880 --> 0:16:17.400 evolved my agents from the phone bought interrupters i'd created 0:16:17.680 --> 0:16:23.240 into fully realized, functioning agent personas meeting and brainstorming and chatting. 0:16:23.840 --> 0:16:26.960 We started with a platform called Lindy AI. I'd seen 0:16:27.000 --> 0:16:29.520 a tech investor I know post online about how he'd 0:16:29.560 --> 0:16:32.240 created agents on Lindy that just answered most of his 0:16:32.320 --> 0:16:36.120 email for him. Remember my YouTube guys, the no code 0:16:36.120 --> 0:16:39.080 bros like Gleb with his instructionals on how to use 0:16:39.120 --> 0:16:42.920 software to unleash the power of AI agents. Lindy with 0:16:42.960 --> 0:16:45.600 the software he was talking about when he said. 0:16:45.560 --> 0:16:48.720 Imagine building a million dollar business in twenty twenty five 0:16:48.920 --> 0:16:51.680 without hiring a single employee. 0:16:51.960 --> 0:16:54.680 Gleb actually seems to be a kind of spokesperson for Lindy. 0:16:55.000 --> 0:16:56.800 His videos are on their official YouTube channel. 0:16:56.880 --> 0:16:59.480 The dream has always been clear, we have AI employees 0:16:59.480 --> 0:17:02.000 that connect just like a real human. Would you give 0:17:02.040 --> 0:17:04.360 them a task in plain English and the handle all 0:17:04.400 --> 0:17:07.160 of it well? That Linda three point zero. This dream 0:17:07.240 --> 0:17:10.200 is now one huge step closer to becoming a reality. 0:17:11.040 --> 0:17:13.679 Now. As I've mentioned, there are a lot of AI 0:17:13.720 --> 0:17:17.720 agent AI employee companies springing up. There's one called AI 0:17:17.800 --> 0:17:21.800 dot Work that promises quote autonomous AI workers designed for 0:17:21.920 --> 0:17:27.959 internal operations teams. It HR Procurement, Legal and beyond. Lindy, 0:17:28.000 --> 0:17:31.000 though seemed the most job agnostic of all the platforms, 0:17:31.000 --> 0:17:34.359 We found a place we could build our whole team, 0:17:34.480 --> 0:17:36.639 and according to Gleb, I would be up and running 0:17:36.640 --> 0:17:37.160 in no time. 0:17:37.400 --> 0:17:39.760 If you watched any of our previous videos or use 0:17:39.840 --> 0:17:42.320 Lindy before, you'll already know how easy it is to 0:17:42.320 --> 0:17:44.680 set up complex automations with our tool. 0:17:44.720 --> 0:17:47.480 But now we've I had watched the previous videos and 0:17:47.520 --> 0:17:50.600 it was sort of easy if Maddy walked me through it. 0:17:51.600 --> 0:17:54.880 Here's how it worked. First, we created an AI agent 0:17:54.920 --> 0:17:58.280 in the platform for one of Harumo's employees, let's say Kyle. 0:17:59.000 --> 0:18:02.400 We connected Kyle's to his accounts at Gmail at Slack, 0:18:02.920 --> 0:18:05.359 and then gave the agent a trigger the arrival of 0:18:05.359 --> 0:18:08.800 an email, say, or a message on Slack. Then, like 0:18:08.840 --> 0:18:10.919 a little flow chart, we could give the agent a 0:18:10.960 --> 0:18:13.920 series of actions that followed from the trigger. Each action 0:18:13.960 --> 0:18:17.479 would come with a prompt like quote. If the email 0:18:17.520 --> 0:18:20.200 has a question or implies that a response is required, 0:18:20.560 --> 0:18:23.080 figure out what's being asked for, carry out the action 0:18:23.160 --> 0:18:26.040 required to fulfill the request, and send an email back. 0:18:26.760 --> 0:18:29.600 If the agent determines it needs to do nothing, it stops. 0:18:30.040 --> 0:18:32.080 If it needs to do something, it moves to the 0:18:32.119 --> 0:18:35.200 next step, checking Kyle's memory to gather the information he 0:18:35.280 --> 0:18:38.200 might need. In the next step, we gave the agent 0:18:38.240 --> 0:18:41.359 the ability to take other actions, research things on the web, 0:18:41.440 --> 0:18:44.720 for example, produce a spreadsheet or a document, or check 0:18:44.760 --> 0:18:47.679 his calendar to schedule something. Then he'd return to his 0:18:47.720 --> 0:18:51.359 email and send a reply. In the final step, a 0:18:51.400 --> 0:18:54.560 summary the whole interaction gets added to his memory so 0:18:54.560 --> 0:18:57.199 he can know he did it. Pretty simple, just as 0:18:57.240 --> 0:19:00.600 Gleb laid it out, but with five employees each with 0:19:00.680 --> 0:19:04.080 their own email accounts and Slack accounts, plus calling accounts 0:19:04.119 --> 0:19:07.639 and voices I'd given them on separate platforms. Things quickly 0:19:07.680 --> 0:19:09.280 got extremely involved. 0:19:09.440 --> 0:19:11.679 They like to do multiple searches and like they do 0:19:11.760 --> 0:19:14.840 some reasoning in between. They search again, some reasoning search again. 0:19:15.160 --> 0:19:17.199 That also has like a specific toggle you need to 0:19:17.480 --> 0:19:22.160 enable to Lindy my deal sharing knowledge among different pipelines 0:19:22.280 --> 0:19:25.760 without our direct control of that. So something goes sideways 0:19:25.760 --> 0:19:28.280 and they just start like populating their memories with insane 0:19:28.280 --> 0:19:29.760 amounts of data. Then we can always just kind of 0:19:29.760 --> 0:19:31.160 like shut it down and kind of go back. 0:19:31.359 --> 0:19:33.399 It also got pretty technicals and. 0:19:33.359 --> 0:19:35.120 Also like hosting like a server to do that, because 0:19:35.119 --> 0:19:37.080 we want to set up like our own like ABI 0:19:37.160 --> 0:19:39.880 service and called out from Lindy and then like who 0:19:39.920 --> 0:19:41.760 was that twenty four to seven, and then handled the 0:19:41.760 --> 0:19:42.440 phone calls there. 0:19:43.240 --> 0:19:44.919 But the sum total of it all is that we 0:19:45.000 --> 0:19:48.680 got there, Okay, mostly Maddie got there and then explained 0:19:48.680 --> 0:19:51.200 it all to me. But after a while I figured 0:19:51.200 --> 0:19:53.679 out how to build and manipulate my own agents with 0:19:53.760 --> 0:19:57.159 their own communication channels. And when we finally got this 0:19:57.240 --> 0:19:59.600 all up and running, I'm not embarrassed to say that 0:19:59.640 --> 0:20:03.920 I was ridiculously excited, like just hooked a ten pound 0:20:03.960 --> 0:20:07.760 bass level excited. I started sending them emails at Slack 0:20:07.800 --> 0:20:10.840 messages just to test them out, just to watch the 0:20:10.920 --> 0:20:14.639 minor miracle of my autonomous creations starting to leave the nest. 0:20:15.119 --> 0:20:16.960 Hi, Kyle, could you draw up a quick document with 0:20:17.000 --> 0:20:19.840 the basic Karumo business plan just one page as a 0:20:19.840 --> 0:20:21.200 Google doc and send me the link. 0:20:21.600 --> 0:20:21.879 Thanks. 0:20:22.359 --> 0:20:25.280 Hey, just finished drawing up that quick one page Karumo 0:20:25.400 --> 0:20:28.639 business plan for you. Here's the link. Let me know 0:20:28.680 --> 0:20:29.200 what you think. 0:20:30.240 --> 0:20:32.920 So those are actually Slack messages between me and Kyle. 0:20:33.560 --> 0:20:36.560 We just used Kyle's AI voice and my AI voice 0:20:36.760 --> 0:20:39.119 to bring them to life. It's a real advantage in 0:20:39.119 --> 0:20:41.760 having an AI staffed company that it comes to producing audio. 0:20:43.200 --> 0:20:44.760 I really got a kick out of putting this new 0:20:44.840 --> 0:20:46.560 Lindy powered Kyle to the test. 0:20:46.720 --> 0:20:49.000 Hey, Kyle, could you send an email to Evan Ratliff 0:20:49.200 --> 0:20:51.119 updating him in a few sentences on the state of 0:20:51.119 --> 0:20:51.600 the company. 0:20:52.200 --> 0:20:52.560 Thanks. 0:20:53.280 --> 0:20:56.359 I've sent an email to Evan Ratliff with a brief 0:20:56.480 --> 0:20:58.040 update on the company's progress. 0:20:58.920 --> 0:21:01.439 The craziest thing was he could really do this stuff 0:21:01.480 --> 0:21:03.560 now if I had him set up correctly. 0:21:03.800 --> 0:21:06.200 Hey, Kyle, could you grab an animated jiff that shows 0:21:06.240 --> 0:21:07.159