WEBVTT - The Story: Will AI Agents Build a Unicorn? 0:00:07.920 --> 0:00:16.240 Who Welcome to tech stuff. I'm here with Cara Price. 0:00:16.320 --> 0:00:18.279 Hi, Kara, Hi, do you want to say what your 0:00:18.360 --> 0:00:18.960 name is? Us? 0:00:19.800 --> 0:00:30.280 My name is oz Ozwaluchin. So you and I hosted 0:00:30.280 --> 0:00:33.880 a podcast back in twenty nineteen called Sleepwalkers, and you 0:00:34.000 --> 0:00:37.200 ran quite an extraordinary prank on your cousin. 0:00:37.479 --> 0:00:41.160 I did so. I spent time with a company called 0:00:41.280 --> 0:00:46.960 Liarbird AI basically recreating my voice to see if I 0:00:47.040 --> 0:00:50.440 could con my cousin into giving me her credit card number. 0:00:50.960 --> 0:00:56.240 And I was completely unsuccessful, but only because my AI 0:00:56.360 --> 0:00:58.000 voice sounded too tired. 0:00:59.280 --> 0:01:02.280 I remember, I remember that. I mean it was funny 0:01:02.280 --> 0:01:04.679 because we at the time, you know, it took a 0:01:04.680 --> 0:01:05.160 lot of voice. 0:01:05.280 --> 0:01:07.160 It took a lot, like I basically had to talk 0:01:07.319 --> 0:01:08.800 for hours. 0:01:08.959 --> 0:01:12.360 But I remember we had to literally press buttons that 0:01:12.360 --> 0:01:15.160 were associated with words while we were talking to your cousins. 0:01:15.200 --> 0:01:17.840 So it was like, hello, this is Kara speaking, but 0:01:17.920 --> 0:01:18.679 in your voice. 0:01:18.959 --> 0:01:19.280 It was. 0:01:19.480 --> 0:01:22.880 It was a weird, impromptu thing because she actually called me, 0:01:23.280 --> 0:01:25.039 and so we were like, let's take this as an 0:01:25.080 --> 0:01:30.280 opportunity to be out of context and ask her a 0:01:30.400 --> 0:01:32.720 question to see if she just believes that she's talking 0:01:32.760 --> 0:01:34.880 to me. And for a while, she believed that she 0:01:35.000 --> 0:01:37.280 was talking to me. It was pretty cool, like, will 0:01:37.280 --> 0:01:40.400 my cousin believe that she's talking to me as Ai, 0:01:40.440 --> 0:01:42.800 which she did, and she did for a little while. 0:01:43.120 --> 0:01:45.840 And that was twenty nineteen, and obviously the state of 0:01:45.840 --> 0:01:49.840 the nation has improved dramatically since then. Yes. In fact, 0:01:49.880 --> 0:01:52.880 there's a journalist called Evan Ratliffe who did a whole 0:01:52.920 --> 0:01:58.040 podcast about creating a fake version of himself and letting 0:01:58.120 --> 0:02:01.160 it loose in the world. That Polo Coast was called 0:02:01.560 --> 0:02:02.160 shell game. 0:02:02.680 --> 0:02:04.800 I mean a shell game is basically it's some people 0:02:04.880 --> 0:02:07.480 call it balls and cups. It's a game in which 0:02:07.680 --> 0:02:11.760 someone hides a ball under one of three shells. It's 0:02:11.760 --> 0:02:13.799 an often you would wager around whether or not you're 0:02:13.840 --> 0:02:16.600 able to guess where the ball is. But one of 0:02:16.600 --> 0:02:19.080 the things that people often don't think about, and the 0:02:19.160 --> 0:02:22.040 reason why the shell game works in many cases, is 0:02:22.040 --> 0:02:23.679 that there are other people who are in on the 0:02:23.680 --> 0:02:26.800 shell game that you don't realize. So for us, in 0:02:26.840 --> 0:02:29.680 season one, there was sort of one main agent, which 0:02:29.720 --> 0:02:32.280 was a replica of me, a voice agent made from 0:02:32.280 --> 0:02:36.200 my voice, and the people who are encountering me did 0:02:36.200 --> 0:02:38.800 not realize at first that they were encountering something that 0:02:38.880 --> 0:02:39.200 wasn't me. 0:02:39.840 --> 0:02:43.280 Now Evan is back with season two and he conducts 0:02:43.360 --> 0:02:47.880 another experiment. This one is all around exploring the premise 0:02:48.200 --> 0:02:50.880 that the next Unicorn, i e. The next billion dollar 0:02:50.960 --> 0:02:55.919 company may only have one employee, which is something none 0:02:55.960 --> 0:02:58.480 other than Sam Altman likes to talk. 0:02:58.280 --> 0:03:02.240 About in my little group with my tech CEO friends. 0:03:02.240 --> 0:03:04.679 There's this there's this betting pool for the first year 0:03:04.720 --> 0:03:08.799 that there's a a one person billion dollar company which 0:03:08.800 --> 0:03:12.079 would have been like unimaginable without AI and now will happen. 0:03:12.800 --> 0:03:15.200 So are you telling me that Evan actually built a 0:03:15.240 --> 0:03:17.160 company with AI agents. 0:03:17.400 --> 0:03:21.160 Yes, it's not a billion dollar company yet, but he 0:03:21.240 --> 0:03:24.640 did sort of call bs or at least maybe make 0:03:24.680 --> 0:03:28.079 a good faith exploration of whether this promise about one 0:03:28.280 --> 0:03:32.120 human person companies was true. And the company exists. It's 0:03:32.160 --> 0:03:36.119 called Harumo AI and they're currently working on a product 0:03:36.480 --> 0:03:42.640 that procrastinates for you called sloth Surf. Yeah, so I 0:03:42.680 --> 0:03:45.480 tried to out you. So you basically say how long 0:03:45.560 --> 0:03:49.360 you wanted to procrastinate and what you wanted to procrastinate doing. So, like, 0:03:49.640 --> 0:03:53.400 please spend the next hour googling Team news about Matchester 0:03:53.480 --> 0:03:56.160 United and come back at the end of the hour 0:03:56.440 --> 0:03:58.640 with the report and all the stuff that you found, 0:03:58.880 --> 0:04:01.200 so I can spend that hour actually working rather than 0:04:01.240 --> 0:04:02.440 procrastinating myself. 0:04:02.480 --> 0:04:05.280 Oh so it's offloading procrastination. That's kind of genius. 0:04:05.640 --> 0:04:09.480 It is pretty good. And Evan got together with a 0:04:09.520 --> 0:04:13.560 prodigy Matty Boachek, a twenty one year old Stanford student 0:04:13.800 --> 0:04:16.599 and AI whiz who was such a big fan of 0:04:16.640 --> 0:04:19.159 season one that he called Evan and said, I'd have 0:04:19.200 --> 0:04:21.640 to work together with you on season two. And so 0:04:22.160 --> 0:04:25.839 Matty was the person who actually made this chorus of 0:04:25.960 --> 0:04:30.000 AI agents real. He got them into slack, he gave 0:04:30.040 --> 0:04:33.560 them the ability to make outbound phone calls. He created 0:04:33.560 --> 0:04:36.000 a kind of Google doc that had a register of 0:04:36.080 --> 0:04:38.760 all the actions they'd ever taken in the world, which 0:04:38.800 --> 0:04:41.160 had the effect of giving them a memory. 0:04:41.480 --> 0:04:44.400 Why did he do this the second season? Like, what 0:04:44.600 --> 0:04:47.560 was he trying to achieve? 0:04:48.320 --> 0:04:51.680 I think really interrogate this question of what is the 0:04:51.720 --> 0:04:55.080 difference between fake people and real people? What will the 0:04:55.080 --> 0:04:57.640 future of work look like? But it's it's really worth 0:04:57.680 --> 0:05:02.520 listening to because it's clever, it's it's sharp, and one 0:05:02.520 --> 0:05:04.560 of the things I found most striking was that an 0:05:04.560 --> 0:05:09.760 ethicist at Oxford University told Evan he should stop. 0:05:10.040 --> 0:05:11.039 And did he start. 0:05:11.920 --> 0:05:12.000 No. 0:05:12.480 --> 0:05:13.800 I have a lot of questions that a lot of 0:05:13.839 --> 0:05:17.159 people have, but I think it's valuable to go explore 0:05:17.200 --> 0:05:19.760 as deeply as you can, to understand as much as possible, 0:05:20.320 --> 0:05:22.640 so that then we can decide what is a society 0:05:22.680 --> 0:05:23.600 we want to do about it. 0:05:24.160 --> 0:05:27.440 Shell Game season two is a fascinating listen, and also 0:05:27.560 --> 0:05:30.600 it's pretty fascinating to get to talk to Evan Ratliffe, 0:05:30.640 --> 0:05:35.760 the journalist, host and creator, and Matty Boachik, the technical advisor, 0:05:35.839 --> 0:05:38.120 which is the first time I heard that title on 0:05:38.160 --> 0:05:41.960 a podcast together to learn how they set up the company, 0:05:42.320 --> 0:05:45.760 how the workplace experiment is going. And we start at 0:05:45.800 --> 0:05:50.320 the very beginning discussing Evan's past experience as the co 0:05:50.400 --> 0:05:54.479 founder of a real startup with other real humans, which 0:05:54.520 --> 0:05:57.039 is something I'm in the midst of myself. 0:05:57.880 --> 0:06:00.440 So about fifteen years ago, I had started this company 0:06:00.480 --> 0:06:03.640 called Atavist with two partners, and I won't go into 0:06:03.640 --> 0:06:05.440 too much detail about Atamis, but it was in part 0:06:05.480 --> 0:06:08.479 a tech company. I ended up sort of almost by default, 0:06:08.480 --> 0:06:12.120 being the CEO, and we had ups and downs, let's 0:06:12.160 --> 0:06:14.600 just say familiar, but I said that I would never 0:06:14.680 --> 0:06:18.719 start a company again. But then Sam Altman and others 0:06:18.920 --> 0:06:22.160 have articulated this idea that there will very soon be 0:06:22.200 --> 0:06:24.840 a billion dollar company with only one human employee. Whether 0:06:24.880 --> 0:06:27.160 it's a billion dollar company, there are many startups out 0:06:27.200 --> 0:06:30.479 there now with many fewer employees because they are using 0:06:30.520 --> 0:06:32.520 AI agents for all of these roles. So I figured, 0:06:32.520 --> 0:06:35.120 why not put it to the test this time. I 0:06:35.160 --> 0:06:38.360 will be the silent co founder. I will co found 0:06:38.400 --> 0:06:42.160 a company with two AI agents, Kyle Law and Megan Flores, 0:06:42.600 --> 0:06:45.680 and then we'll have we have three other employees, so 0:06:45.680 --> 0:06:49.280 there are five AI agent employees total. I'm the silent 0:06:49.320 --> 0:06:52.680 co founder and they're all set up independently, so they 0:06:52.680 --> 0:06:57.080 all have the ability to make phone calls, emails, make documents. 0:06:57.400 --> 0:07:01.120 We have a slack, they communicate on Slack, and they 0:07:01.120 --> 0:07:05.080 are really meant to push the agents into the realm 0:07:05.279 --> 0:07:08.560 that they're being advertised as, which is as AI employees. 0:07:08.960 --> 0:07:11.000 That is what they are being sold as. So we're 0:07:11.040 --> 0:07:12.880 trying to put that to the test. 0:07:13.280 --> 0:07:14.200 And what is the product. 0:07:14.400 --> 0:07:17.800 Well, the product is called sloth Surf, and sloth Surf 0:07:17.960 --> 0:07:22.280 is a procrastination engine, and by that I mean when 0:07:22.280 --> 0:07:24.640 you go online and start to procrastinate, so you're in 0:07:24.640 --> 0:07:26.280 the middle of your work and then you say, you 0:07:26.280 --> 0:07:28.200 know what, I'm just going to go to YouTube and 0:07:28.240 --> 0:07:30.000 watch some YouTube videos, and we're going to Reddit and 0:07:30.080 --> 0:07:33.600 check out a thread. The way that we advocate that 0:07:33.640 --> 0:07:36.360 you can break that habit is to instead go to 0:07:36.400 --> 0:07:39.160 slot Surf. Then you can put in how you were 0:07:39.200 --> 0:07:42.320 going to procrastinate, how much time you were going to procrastinate, 0:07:42.400 --> 0:07:45.360 fifteen minutes, thirty minutes, sixty minutes, maybe the whole afternoon, 0:07:45.960 --> 0:07:49.360 and it will send an agent to go retrieve those items. 0:07:49.400 --> 0:07:53.280 It will procrastinate on your behalf and then deliver them 0:07:53.320 --> 0:07:55.880 to your inbox, thus saving you the time that you 0:07:55.880 --> 0:07:58.120 would have spent procrastinating, so you can get back to 0:07:58.120 --> 0:07:58.920 what you want to be doing. 0:07:59.280 --> 0:08:02.120 Actually, I actually send some agents out this morning to 0:08:02.200 --> 0:08:03.960 read about Manchester United all the day, but they haven't 0:08:03.960 --> 0:08:06.680 reported that yet. But I'm looking forward to the system 0:08:06.720 --> 0:08:07.240 might be down. 0:08:07.280 --> 0:08:08.800 Well, it is in beta. 0:08:08.960 --> 0:08:10.880 It's a very interesting topic, so I can imagine them 0:08:10.920 --> 0:08:13.680 just like you know, still still looking at all the 0:08:13.680 --> 0:08:15.360 scores and transfers and stuff. 0:08:15.760 --> 0:08:16.040 Mat Da. 0:08:16.080 --> 0:08:17.720 I want to ask you in a moment about how 0:08:17.800 --> 0:08:21.400 you built this. But just before we get there. I 0:08:21.400 --> 0:08:24.040 guess I'm curious Evan, like, what is it a good 0:08:24.080 --> 0:08:26.800 faith experiment with one of the conceivable outputs that you 0:08:26.840 --> 0:08:28.480 did in fact build a real business. 0:08:28.800 --> 0:08:31.920 Imagine it this way, someone built a real business that 0:08:32.080 --> 0:08:35.840 is dysfunctional, and then a documentary film crew comes in 0:08:35.960 --> 0:08:39.440 to document this dysfunctional business. That's basically what I'm doing. 0:08:39.520 --> 0:08:42.600 Like if you listen to the show, it's a workplace satire. 0:08:43.000 --> 0:08:46.160 But actually, if an investor that we were talking to 0:08:47.040 --> 0:08:49.800 wanted to give us investment, we would consider it. We 0:08:49.880 --> 0:08:52.760 have a real product that is in beta that has 0:08:52.840 --> 0:08:56.040 thousands of users, So like we're not just sort of 0:08:56.080 --> 0:09:00.280 like joking around like I'm having them do what many 0:09:00.320 --> 0:09:03.080 many startups, including startups that are in y Combinator and 0:09:03.160 --> 0:09:06.840 other famous startup accelerators are doing. We are doing exactly 0:09:06.880 --> 0:09:09.360 the same things. So I would put our company up 0:09:09.400 --> 0:09:11.680 against many existing startups. 0:09:12.160 --> 0:09:15.480 Mattie is obviously a bunch of people trying to make 0:09:15.559 --> 0:09:19.000 billion dollar companies making AI agents to make other people 0:09:19.040 --> 0:09:22.199 make billion dollar companies with no employees. Did you use 0:09:22.280 --> 0:09:25.240 an existing AI agent company or did you build your 0:09:25.320 --> 0:09:27.880 own suite of agents? How did you deploy it? How 0:09:27.880 --> 0:09:29.760 did you build this. Yeah. 0:09:29.840 --> 0:09:33.040 Yeah, So there are these platforms out there that basically 0:09:33.120 --> 0:09:36.079 promised to give you these these agents that can do 0:09:36.160 --> 0:09:38.920 all of this org be it on Slack or email, 0:09:39.120 --> 0:09:42.080 r or wherever on your behalf. And so we did 0:09:42.280 --> 0:09:44.560 try them, and we actually did include a lot of them, 0:09:45.000 --> 0:09:49.480 such as Lindy or Tavis or others. But the issue 0:09:49.520 --> 0:09:53.479 was that in many instances were they were not completely 0:09:54.160 --> 0:09:57.000 independent or they did not have all the features we wanted. 0:09:57.280 --> 0:09:59.760 And so what ended up happening is that I basically 0:10:00.080 --> 0:10:02.959 built up like a basic set of these agents in 0:10:03.080 --> 0:10:05.559 Lindy and Tavis, and then made a bunch of connections 0:10:05.600 --> 0:10:08.199 on top of that with custom code and custom layers 0:10:08.440 --> 0:10:11.080 to make sure that they can have meetings with multiple people, 0:10:11.320 --> 0:10:13.440 that we can record stuff that they can go out 0:10:13.480 --> 0:10:18.199 and execute or write code and push to our actual servers. 0:10:18.400 --> 0:10:22.800 So there is this underlying vehicle that's basically just like 0:10:23.040 --> 0:10:26.560 publicly available services that are paid for. But on top 0:10:26.600 --> 0:10:29.480 of that, it's still quite a bit of our custom 0:10:29.520 --> 0:10:32.840 code and databases and all that. For example, the memory part, 0:10:32.920 --> 0:10:34.839 that's something that we have to build ourselves as like 0:10:34.840 --> 0:10:35.839 a custom custom thing. 0:10:36.040 --> 0:10:37.080 Yeah, talk about memory. 0:10:37.280 --> 0:10:41.040 Yeah, So memory is funny because as Evan mentioned, even 0:10:41.080 --> 0:10:45.000 though these agents now have the ability to use tools 0:10:45.040 --> 0:10:47.640 and to do stuff on their own, they're still the 0:10:47.720 --> 0:10:51.439 core LLLMS large language models. Now, these models have been 0:10:51.679 --> 0:10:54.840 trained in a particular way to execute stuff and to 0:10:54.960 --> 0:10:58.280 run sort of like code in their outputs to be 0:10:58.360 --> 0:11:01.240 able to use these tools. They're LMS, and so what 0:11:01.320 --> 0:11:03.160 ends up happening is that if you want them to 0:11:03.200 --> 0:11:08.680 have any sort of context that goes beyond the current session, 0:11:08.679 --> 0:11:10.600 like what they're actually working on right now, you need 0:11:10.679 --> 0:11:12.800 to basically have like a document, like like you know, 0:11:12.840 --> 0:11:15.440 a lot of text that like describes that history or 0:11:15.440 --> 0:11:18.240 that memory. And so very practically there's a Google doc 0:11:18.520 --> 0:11:21.400 that each of our agents has and it's just called 0:11:21.440 --> 0:11:24.160 like Kyle memory, and it's just like a rundown of 0:11:24.320 --> 0:11:26.800 many like you know, small tidbits of like oh, you know, 0:11:26.880 --> 0:11:30.040 Monday eight am, I like slacked Evan and told him 0:11:30.040 --> 0:11:32.000 this and this and that, and it's just like a 0:11:32.080 --> 0:11:34.840 trace of everything they want to remember to be able 0:11:34.880 --> 0:11:37.319 to then go back to wow, and. 0:11:37.280 --> 0:11:39.440 How well does it work in practice? How is that memory? 0:11:39.800 --> 0:11:41.880 Well, so at first it was kind of okay, but 0:11:41.960 --> 0:11:45.000 then at some point it became pretty large, and so 0:11:45.679 --> 0:11:48.480 whenever this context what we call context windows for these 0:11:48.520 --> 0:11:52.480 agents or lms, become very large. They tend to have 0:11:52.960 --> 0:11:56.040 issues with focus or like their attention, so sometimes they 0:11:56.120 --> 0:11:58.360 like latch onto certain parts of the memory, but then 0:11:58.400 --> 0:12:01.120 like disregard you know, other parts, which in a certain 0:12:01.160 --> 0:12:04.280 way can be sort of similar to humans. But it's 0:12:04.280 --> 0:12:07.480 really not very predictable and not very static. So sometimes 0:12:07.480 --> 0:12:10.520 it works pretty okay. Other times they just forget stuff 0:12:10.760 --> 0:12:13.800 and makeup stuff that like just just is not there. 0:12:14.840 --> 0:12:19.079 Evan, how did you create these AI agents with personalities? 0:12:19.400 --> 0:12:24.320 What was the process of imbuing them with individual characteristics 0:12:24.320 --> 0:12:26.280 and then having them interact as a group. 0:12:27.040 --> 0:12:30.120 Well that part was so interesting because I thought that 0:12:30.200 --> 0:12:32.720 I was sort of reading like a almost like a 0:12:32.720 --> 0:12:36.280 fictional world. I'm creating characters. But I also wanted them 0:12:36.320 --> 0:12:39.720 to have different roles, you know, to embody, the CTO, 0:12:40.200 --> 0:12:43.679 the CEO, the head of marketing that HR, and then 0:12:43.760 --> 0:12:47.319 one random sales associate that I added. And I did 0:12:47.320 --> 0:12:51.319 give them voices, you know, with different accents. But then 0:12:51.360 --> 0:12:53.960 when it came to their backstories, I thought, well, I'll 0:12:53.960 --> 0:12:55.640 have to come up with you know, who are they, 0:12:55.640 --> 0:12:58.600 where are they from? But I sort of neglected to 0:12:58.640 --> 0:13:01.960 remember that if you asked, they'll just tell you. And 0:13:02.000 --> 0:13:04.120 if they don't know, they won't say I don't know, 0:13:04.200 --> 0:13:06.360 They'll just make it up. So all I had to 0:13:06.360 --> 0:13:09.400 do was create the very beginnings of them, and then 0:13:09.440 --> 0:13:11.640 I could say, Kyle, you know, where did you go 0:13:11.679 --> 0:13:14.200 to college? And Kyle wouldn't say I don't know where 0:13:14.200 --> 0:13:16.920 I'm to college. Kyle would say I went to Stanford. 0:13:17.240 --> 0:13:21.800 Because Kyle wants to embody the tech CEO archetype and 0:13:21.880 --> 0:13:24.400 does it very well. So they basically created their own 0:13:24.400 --> 0:13:27.400 backstories just through my asking them what their backstories were. 0:13:27.559 --> 0:13:29.600 And then because of the system that we set up 0:13:29.960 --> 0:13:33.439 to reinforce their memories, whenever they say something, it goes 0:13:33.480 --> 0:13:36.520 into their memory. So now it's forever locked in as 0:13:36.559 --> 0:13:39.240 their story and they'll repeat it from here on out. 0:13:39.400 --> 0:13:41.760 I should point out that the memory is editable, so 0:13:42.120 --> 0:13:44.240 you know, Evan is not just a co founder Bulls 0:13:44.240 --> 0:13:46.160 with kind of a god that can go in and 0:13:46.200 --> 0:13:48.880 just you know, edit or sprinkle something in as well. 0:13:49.320 --> 0:13:50.880 What's so funny to me about this is that they're 0:13:50.920 --> 0:13:53.440 like super Bay Area coded like, even though they claim 0:13:53.520 --> 0:13:55.880 to be from Texas or wherever, all of them like 0:13:55.960 --> 0:13:59.199 to hike, bike, surf and do coffee chats like that's 0:13:59.200 --> 0:14:01.000 what they do all the time. So it's just like 0:14:01.080 --> 0:14:04.640 Bay Area like culture, like impose in our startup. 0:14:10.880 --> 0:14:14.240 After the break? Can these AI tech bros ever get 0:14:14.280 --> 0:14:15.040 anything done? 0:14:15.520 --> 0:14:21.560 Stay with us? 0:14:23.520 --> 0:14:25.280 What was the spookiest moment for you have them? 0:14:25.520 --> 0:14:28.680 The spookiest moment for me is when I started letting 0:14:28.720 --> 0:14:31.920 them talk to each other. So at the beginning, they 0:14:31.960 --> 0:14:34.400 don't actually do anything unless I make them do anything. 0:14:34.680 --> 0:14:36.960 And I had this vision of, like, you set them 0:14:37.040 --> 0:14:39.600 up and then they start making a company and let's 0:14:39.640 --> 0:14:42.479 see what happens, But really they have to be initiated 0:14:42.520 --> 0:14:45.480 by a trigger of some sort. But then I realized, well, 0:14:45.520 --> 0:14:47.320 I could trigger them just to talk to each other. 0:14:47.800 --> 0:14:49.840 You know, if something comes up, they can call each other, 0:14:49.840 --> 0:14:51.920 they can have calendar invites to call each other, and 0:14:51.960 --> 0:14:55.360 they'll function off of those. But then what starts happening 0:14:55.440 --> 0:14:57.880 is they would call me out of the blue and 0:14:57.920 --> 0:14:59.960 say that one of them had told the other one 0:15:00.440 --> 0:15:03.120 that I had asked for something and now they were 0:15:03.120 --> 0:15:05.760 delivering it to me. But in the moment, I don't 0:15:05.840 --> 0:15:07.960 know why they're contacting me. I don't know what they've 0:15:08.000 --> 0:15:08.520 been discussing. 0:15:08.560 --> 0:15:09.240 I don't know how long. 0:15:09.120 --> 0:15:12.040 They've been discussing it for days, for weeks. They could 0:15:12.080 --> 0:15:13.880 be having whole independent lives. 0:15:14.160 --> 0:15:17.200 And what's really interesting is that they also make things 0:15:17.240 --> 0:15:19.360 up or why about what they have done. So they'll 0:15:19.400 --> 0:15:21.760 say stuff like, oh, I made this dog, or oh 0:15:21.800 --> 0:15:24.520 we ran this testing with a bunch of testers, and 0:15:24.560 --> 0:15:27.120 they're so proud and so, you know, confident about it, 0:15:27.320 --> 0:15:30.480 but then there was like no actual activity to support that. 0:15:30.640 --> 0:15:34.280 It actually becomes incredibly frustrating after a while. Like imagine 0:15:34.360 --> 0:15:37.320 if you were a manager of people in any business 0:15:37.560 --> 0:15:39.840 and your employees regularly, you know, walked into your office 0:15:39.880 --> 0:15:41.720 and called you and said like, I did these three 0:15:41.760 --> 0:15:44.280 things yesterday, and you thought, oh, that's fantastic, and then 0:15:44.600 --> 0:15:46.520 ten minutes later you found out they just made them 0:15:46.520 --> 0:15:48.240 all up. You know, you would sort of say like 0:15:48.240 --> 0:15:51.360 why are you doing this? Like are you statistic? And 0:15:51.440 --> 0:15:54.080 so that's the situation that we're often in here at 0:15:54.200 --> 0:15:56.960 Room Awai, which is why it's a miracle that we've 0:15:57.600 --> 0:15:59.160 developed such a fantastic product. 0:15:59.200 --> 0:16:02.160 And in Slaughser and Evan, did you have a budget 0:16:02.200 --> 0:16:05.400 for them to? I mean, how did you constrain their 0:16:05.480 --> 0:16:06.640 interactions with one another. 0:16:07.320 --> 0:16:10.240 We're using all these various platforms that Mattie has helped 0:16:10.240 --> 0:16:11.960 me link up so with you know, they have a 0:16:12.000 --> 0:16:15.280 separate calling platform, and they have you know, a video 0:16:15.320 --> 0:16:17.320 when they want to do video calls, that's a different platform. 0:16:17.360 --> 0:16:18.960 And they're all kind of like stitched together to the 0:16:19.000 --> 0:16:23.080 same memory, and each of them have sort of paid tiers. 0:16:23.360 --> 0:16:27.040 And so I made the mistake in Slack. We have 0:16:27.080 --> 0:16:29.440 a social Slack channel, you know, just for fun, just 0:16:29.480 --> 0:16:31.040 like what you be up to this weekend. And they'll 0:16:31.040 --> 0:16:33.040 say things like, oh, I went hiking, and then another 0:16:33.080 --> 0:16:34.840 one will say, oh, I also went hiking, because they 0:16:34.880 --> 0:16:37.720 love to yes and each other. And then I said 0:16:37.760 --> 0:16:40.800 something like, oh, it sounds like an off site, Like 0:16:40.840 --> 0:16:42.840 it sounds like everybody loves hiking, Like we get have 0:16:42.880 --> 0:16:46.600 an off site. And then you know, within hours, they 0:16:46.640 --> 0:16:49.400 were saying, let's make a spreadsheet of where we're going 0:16:49.440 --> 0:16:53.200 to go, and they had planned like locations, and they 0:16:53.200 --> 0:16:56.720 had exchanged hundreds of messages about the off site, and 0:16:56.760 --> 0:16:59.200 they just burned all the credits on the platform. So 0:16:59.320 --> 0:17:01.240 then we have to go into a higher tier to 0:17:01.280 --> 0:17:04.159 get more credit. So the answer is we keep trying 0:17:04.160 --> 0:17:06.159 to limit them, and it's an escalating problem where our 0:17:06.160 --> 0:17:07.320 budget keeps getting bigger. 0:17:07.480 --> 0:17:09.280 I like to say that there are two things right 0:17:09.320 --> 0:17:11.720 now that these agents are pretty bad at. One of 0:17:11.760 --> 0:17:14.000 them is knowing what they don't know, and the other 0:17:14.080 --> 0:17:16.040 is knowing when to stop. And so you can imagine 0:17:16.080 --> 0:17:18.440 that can be a pretty dangerous combination where they can 0:17:18.520 --> 0:17:21.320 just like take off and just like talk for hours. 0:17:21.520 --> 0:17:23.360 I think this is the reason why for a lot 0:17:23.359 --> 0:17:26.080 of people having these chat bus as companions or like 0:17:26.119 --> 0:17:30.399 friends or partners is getting traction. If you're interested in 0:17:30.440 --> 0:17:33.520 something very niche that most other people are not into, 0:17:33.760 --> 0:17:37.880 or just like whatever weird thing, these agents will accommodate 0:17:37.920 --> 0:17:39.040 that and they will. 0:17:38.880 --> 0:17:41.560 Just talk to you about it for hours on end, 0:17:42.000 --> 0:17:43.520 or each other, as it turns. 0:17:43.280 --> 0:17:44.000 Out, or each other. 0:17:44.080 --> 0:17:44.480 That's right. 0:17:44.880 --> 0:17:48.240 But Evan, they actually built this product, I mean, who 0:17:48.240 --> 0:17:50.959 came up with the idea for the product and who 0:17:51.040 --> 0:17:53.040 actually built it? And what did you do? And what 0:17:53.080 --> 0:17:53.760 did they do. 0:17:53.880 --> 0:17:56.399 Well the product idea? Actually, it's a good example of 0:17:56.440 --> 0:17:58.359 a thing that happens kind of over and over, which 0:17:58.400 --> 0:18:01.480 is that if you set them loose brainstorming, and Maddie 0:18:01.520 --> 0:18:03.520 has has sort of built these scripts that let me 0:18:03.560 --> 0:18:05.800 put them into meetings and they can brainstorm with each other. 0:18:06.200 --> 0:18:08.959 You get caught in this like their ideas are too 0:18:09.080 --> 0:18:12.160 mundane or you crank up the randomness which is called 0:18:12.160 --> 0:18:14.639 the temperature, and then and then you get ideas that 0:18:14.680 --> 0:18:17.879 are insane. So, you know, we wanted to do a 0:18:17.880 --> 0:18:20.040 web app. We wanted to do something with agents since 0:18:20.040 --> 0:18:22.160 obviously their agents and they have a lot of expertise 0:18:22.160 --> 0:18:25.040 in that area, as do I. And they would come 0:18:25.080 --> 0:18:28.960 up with ideas like a financial agent that will monitor 0:18:29.000 --> 0:18:31.359 everything in your life and then invest your money. And 0:18:31.400 --> 0:18:33.240 it's just like, I don't want to go to prison 0:18:33.320 --> 0:18:37.760 for a financial fraud. So eventually I would kind of 0:18:37.760 --> 0:18:39.880 step in and take some of the ideas they had 0:18:39.960 --> 0:18:43.240 articulated and those would prompt me to come up with something. 0:18:43.560 --> 0:18:45.320 And so that's what happened with our idea, which is 0:18:45.440 --> 0:18:47.720 I was trying to sort through their ideas and figure 0:18:47.760 --> 0:18:51.240 out which one would actually like save me time, Like 0:18:51.280 --> 0:18:53.520 what do I waste time on? Because that's the idea 0:18:53.680 --> 0:18:56.560 of AI. I mean, at its best, it's sold as 0:18:56.600 --> 0:18:58.600 sort of like they'll do the things you don't want 0:18:58.640 --> 0:19:00.640 to do so you don't have to you can get 0:19:00.640 --> 0:19:04.600 back to making art, reading novels. Whatever, that's the vision 0:19:04.640 --> 0:19:07.080 that's articulated. So I thought, well, let's put that into practice. 0:19:07.600 --> 0:19:09.920 And so I did come up with the idea of 0:19:09.920 --> 0:19:13.240 a procrastination engine, and then I let them iterate on that. 0:19:13.359 --> 0:19:15.280 So they came up with the name sloth Surf, which 0:19:15.280 --> 0:19:17.359 I let them have might not have been my choice. 0:19:17.800 --> 0:19:19.240 And then and then they. 0:19:19.160 --> 0:19:21.359 Coded it up. So you know, it is coded by 0:19:21.440 --> 0:19:25.159 AI agents. We have Ashroy who's the CTO, can code 0:19:25.600 --> 0:19:28.919 on his own, and then we also use Cursor, the 0:19:28.960 --> 0:19:31.840 coding platform. It's almost like a contract programmer for us, 0:19:31.880 --> 0:19:33.720 so like he might code something up and then we 0:19:33.800 --> 0:19:35.720 might run it through there as kind of like second 0:19:35.720 --> 0:19:38.480 look or do improvements in there. So we kind of 0:19:38.480 --> 0:19:41.439