WEBVTT - Shell Game: Launch 0:00:11.590 --> 0:00:13.790 Hey, Oz here. You may have seen a few episodes 0:00:13.810 --> 0:00:16.590 from a podcast called Shell Game in our feed recently. 0:00:17.320 --> 0:00:20.440 It's a critically acclaimed podcast from the Kaleidoscope Network and 0:00:20.500 --> 0:00:24.520 journalist Evan Ratliff. And now we're dropping season two. You're 0:00:24.560 --> 0:00:27.220 about to hear episode eight, where we find out how 0:00:27.260 --> 0:00:30.600 the world responds to a startup led by AI agents. 0:00:31.080 --> 0:00:31.620 Hope you enjoy. 0:00:31.640 --> 0:00:33.360 How are you, Matty? 0:00:33.380 --> 0:00:35.360 I'm really good. 0:00:35.540 --> 0:00:35.940 I'm really good. 0:00:35.960 --> 0:00:36.600 I like your shirt. 0:00:37.220 --> 0:00:40.160 Hey, thanks. When I'm at the beach, gotta dress like 0:00:40.200 --> 0:00:40.699 I'm at the beach. 0:00:42.500 --> 0:00:44.260 I told you, I'm the guy who brings Hawaiian shirts 0:00:44.560 --> 0:00:47.620 to work most days. So I love this. But I 0:00:47.640 --> 0:00:50.239 will say that there's Hawaiian shirts and Hawaiian shirts. And 0:00:50.300 --> 0:00:51.240 mine are very gay. 0:00:52.590 --> 0:00:55.210 Over the months I'd worked with Matty, building the tech 0:00:55.250 --> 0:00:58.270 scaffolding that propped up Harumo AI, I felt like I'd 0:00:58.290 --> 0:01:01.010 really gotten to know him. At least, as well as 0:01:01.030 --> 0:01:02.430 you can over weekly video calls. 0:01:02.910 --> 0:01:05.230 If I can have a confession, I do like the 0:01:05.250 --> 0:01:08.660 Omega Church. I have this weird thing for pickup trucks. 0:01:09.330 --> 0:01:12.170 It's been my dream to get a pickup truck, like forever, 0:01:12.190 --> 0:01:14.050 since I was a kid. Yesterday was like the Czech 0:01:14.090 --> 0:01:17.350 national holiday. And it's always when they like give out 0:01:17.470 --> 0:01:20.990 the civilian honors and stuff. So I got the Senate medal. 0:01:22.330 --> 0:01:24.990 This was just the way it was. Some days he'd 0:01:25.010 --> 0:01:28.190 show up to a meeting jazzed about Hawaiian shirts. Other 0:01:28.230 --> 0:01:30.709 times he'd come on having flown to Prague to accept 0:01:30.750 --> 0:01:34.350 the second highest civilian honor in the Czech Republic. Basically 0:01:34.410 --> 0:01:37.970 their congressional medal of honor in a ceremony broadcast live 0:01:38.030 --> 0:01:41.009 on TV. I can't believe you're talking to me right now. 0:01:41.069 --> 0:01:43.250 You should be, are you, did you just come from 0:01:43.290 --> 0:01:46.770 like a ballroom celebration of like, was there like. 0:01:47.050 --> 0:01:49.680 That was yesterday. That was like with the senators and like, 0:01:49.720 --> 0:01:52.480 like all the politicians. Yeah. Like I, I don't think 0:01:52.500 --> 0:01:56.520 they'll invite me again. Cause so they were asking me like, Oh, 0:01:56.560 --> 0:01:58.940 like when are you coming back to start like your 0:01:58.980 --> 0:02:03.260 lab or, you know, your company. So I told them, okay. 0:02:03.280 --> 0:02:06.110 So like how exactly would you open the conversation to, 0:02:07.210 --> 0:02:09.170 when I go back to my partner, my boyfriend, and 0:02:09.190 --> 0:02:11.049 I told him, hey, let's move to a country where 0:02:11.090 --> 0:02:13.620 there's no gay marriage, no adoptions. And, you know, like, 0:02:13.910 --> 0:02:15.460 it's very close to Russia, too. How would you feel 0:02:15.480 --> 0:02:19.459 about that? And they were, like, very uncomfortable when I 0:02:19.540 --> 0:02:19.980 asked that. 0:02:20.880 --> 0:02:22.900 It was always a breath of fresh air after dealing 0:02:22.940 --> 0:02:26.180 with the agents and their generic sycophancy all week to 0:02:26.220 --> 0:02:30.710 speak to an actual, insightful, self-possessed human being. Of course, 0:02:31.510 --> 0:02:33.930 our conversations did typically turn back to whatever was going 0:02:33.970 --> 0:02:38.010 wrong at Rumo.ai. the responsibility for which usually landed at 0:02:38.050 --> 0:02:41.510 my feet. I was still smarting from how Julia's internship 0:02:41.550 --> 0:02:44.070 had gone and pondering what it meant for the idea 0:02:44.090 --> 0:02:48.620 of agents as employees and employers. I asked Maddie for 0:02:48.660 --> 0:02:49.060 his take. 0:02:49.740 --> 0:02:53.180 To me, she was just playing us. If I had 0:02:53.200 --> 0:02:56.920 to bet, I would just say that she understood what 0:02:57.160 --> 0:03:01.000 weaknesses these models have because young people, not to say 0:03:01.040 --> 0:03:03.250 you're not young, but Just, you know, like people like 0:03:03.540 --> 0:03:07.230 my age, I think are pretty attuned to the abilities 0:03:07.270 --> 0:03:10.190 of these models at this point. And I think are 0:03:10.310 --> 0:03:13.169 able to kind of suss out like what might trip 0:03:13.190 --> 0:03:16.250 them up. I think it's hard to describe intent to anyone, 0:03:16.430 --> 0:03:17.690 but if I had to guess, I would say that 0:03:17.710 --> 0:03:19.109 she was able to just trick them. 0:03:20.389 --> 0:03:22.430 If that was what happened, it felt like it was 0:03:22.470 --> 0:03:25.810 a symptom of the larger problems with my agents. Maddie 0:03:25.850 --> 0:03:30.130 hypothesized that their gullibility against their own intern came down 0:03:30.169 --> 0:03:31.370 to three basic weaknesses. 0:03:32.180 --> 0:03:34.600 One issue is that these models have no sense of 0:03:34.660 --> 0:03:38.380 time whatsoever. Like, if you say that something happened yesterday, 0:03:38.400 --> 0:03:40.500 and then you're like, oh, so like, what happened the 0:03:40.520 --> 0:03:43.780 day before that? Like, those simple descriptions of time are 0:03:43.840 --> 0:03:46.080 like really, really hard for these models. 0:03:46.780 --> 0:03:48.500 The agents lived in what he referred to as a 0:03:48.540 --> 0:03:52.320 kind of temporal vacuum. That was fine when they were 0:03:52.340 --> 0:03:57.780 just interacting with each other. But a human workplace runs on, well, time. 0:03:58.490 --> 0:04:01.450 Second, it's the inability to do continuous learning where There 0:04:01.490 --> 0:04:04.930 is no sense of learning from experiences, no updates to 0:04:04.950 --> 0:04:07.410 the actual model that's underneath the agent. 0:04:08.270 --> 0:04:09.610 As much as we'd rigged them up to have their 0:04:09.690 --> 0:04:12.750 own Google Doc memories, they weren't actually learning from the 0:04:12.790 --> 0:04:16.390 experiences catalog there. It was like having a person who 0:04:16.470 --> 0:04:19.150 couldn't form memories and giving them a journal of all 0:04:19.170 --> 0:04:22.600 their experiences to quickly leaf through during every interaction. 0:04:23.270 --> 0:04:25.950 And then the third one is like the sense of self, right? 0:04:25.970 --> 0:04:29.110 Like these models are able to take on these personas 0:04:29.150 --> 0:04:31.970 where you can tell them like, okay, like you are 0:04:32.370 --> 0:04:35.110 Kyle and you're this employee of your MO and here's 0:04:35.170 --> 0:04:38.990 our backstory. They're also very happy to then make up 0:04:39.089 --> 0:04:41.790 fake stories to go along with this initial persona that 0:04:41.830 --> 0:04:45.140 you put forward. But there's this question of like whether 0:04:45.540 --> 0:04:49.780 the models have some sort of innate persona, innate sense 0:04:49.800 --> 0:04:51.219 of self beyond these settings. 0:04:52.920 --> 0:04:55.820 Things in AI are moving fast. It was easy to 0:04:55.860 --> 0:04:58.680 envision a world mere months from now, in which my 0:04:58.779 --> 0:05:02.640 agents would seem like the first tool-using hominids, already eclipsed 0:05:02.700 --> 0:05:06.320 by a higher intelligence. But then, it was just as 0:05:06.380 --> 0:05:08.680 easy to envision a future in which some combination of 0:05:08.720 --> 0:05:11.880 these flaws continued to hamper the agent's ability to serve 0:05:11.920 --> 0:05:15.690 as full, competent AI workers. The one thing that seemed 0:05:15.730 --> 0:05:18.460 clear to me was, if companies could hire the equivalent 0:05:18.490 --> 0:05:31.990 of a They probably would. And then what? Over a 0:05:31.990 --> 0:05:35.089 decade ago, the anthropologist David Graeber wrote an essay in 0:05:35.110 --> 0:05:39.010 a small magazine called Strike, titled On the Phenomenon of 0:05:39.089 --> 0:05:42.770 Bullshit Jobs. Later, he expanded it into a book called 0:05:43.089 --> 0:05:46.970 Bullshit Jobs, a Theory. The theory tried to explain a 0:05:47.010 --> 0:05:49.560 strange reality that Graeber felt he'd encountered in the world. 0:05:50.279 --> 0:05:53.080 a surprising number of people who describe their own jobs 0:05:53.560 --> 0:05:54.160 as pointless. 0:05:54.839 --> 0:05:55.719 There seems to. 0:05:55.660 --> 0:05:58.360 Be this peculiar way in which if you mention this 0:05:58.380 --> 0:06:02.589 to people, almost no one will deny it. But at 0:06:02.630 --> 0:06:06.210 the same time, no one quite knows what to do 0:06:06.250 --> 0:06:09.270 with it. It's like this gigantic embarrassment in our society. 0:06:10.310 --> 0:06:14.240 Graeber, sadly, died unexpectedly in 2020. That's him in a 0:06:14.279 --> 0:06:18.120 lecture he gave two years before his death. To Graeber, 0:06:18.300 --> 0:06:20.770 a bullshit job was defined as, quote, A form of 0:06:20.850 --> 0:06:25.110 employment that is so completely pointless, unnecessary, or pernicious that 0:06:25.270 --> 0:06:28.919 even the employee cannot justify its existence. To be classed 0:06:28.940 --> 0:06:31.880 as having a bullshit job, someone had to self-report that 0:06:31.900 --> 0:06:33.700 they believed their own job to be pointless. 0:06:34.300 --> 0:06:36.020 I mean, I'm not going to go and tell somebody 0:06:36.080 --> 0:06:38.100 who feels his job is meaningful that they're wrong, but 0:06:38.460 --> 0:06:40.510 if you feel you're not doing anything all day, who 0:06:40.550 --> 0:06:41.310 would know better than you? 0:06:41.970 --> 0:06:45.230 Bullshit jobs are distinct from just shit jobs. Jobs that 0:06:45.270 --> 0:06:49.470 might be strenuous or underpaid or undervalued, but still necessary. 0:06:50.000 --> 0:06:51.960 A bus driver might hate driving the bus, for example, 0:06:52.440 --> 0:06:55.700 but they never doubt why their job exists. Graeber collected 0:06:55.760 --> 0:06:58.000 and analyzed survey data that led him to suggest that 0:06:58.060 --> 0:07:00.000 as many as a third or even a half of 0:07:00.060 --> 0:07:03.410 all jobs, and especially a lot of white-collar jobs, were 0:07:03.450 --> 0:07:07.030 in fact bullshit jobs. In his analysis, some corners of 0:07:07.050 --> 0:07:12.290 the private sector world, telemarketing, corporate law, financial services, and academia, 0:07:12.730 --> 0:07:16.480 were fortresses of bullshit jobs. To some, this would seem 0:07:16.500 --> 0:07:20.260 like a paradox. Capitalism is supposed to prioritize efficiency above 0:07:20.320 --> 0:07:24.180 all else. And yet, private industry was seemingly maintaining millions 0:07:24.220 --> 0:07:27.050 of jobs that even the people doing them believed served 0:07:27.070 --> 0:07:27.630 no function. 0:07:28.210 --> 0:07:31.870 In fact, these jobs constituted a pretty big sector of 0:07:31.950 --> 0:07:32.370 our economy. 0:07:32.410 --> 0:07:33.550 I didn't know how much at the time. 0:07:33.930 --> 0:07:36.070 I was guessing 20, 25%. It turns out it's more. 0:07:36.570 --> 0:07:37.429 How does this happen? 0:07:37.550 --> 0:07:40.820 Like, huge swaths of our economy is completely unnecessary. 0:07:41.680 --> 0:07:44.460 The whole time we were building Harumo AI, Graeber's ideas 0:07:44.520 --> 0:07:47.120 kept pulling at me. If you spend enough time letting 0:07:47.140 --> 0:07:50.700 AI agents do jobs, or parts of jobs, your mind 0:07:50.720 --> 0:07:53.530 starts wandering into questions like, what is a job at all? 0:07:54.290 --> 0:07:57.410 What is it for? I discovered that AI agents are 0:07:57.470 --> 0:08:01.410 actually unbelievable at bullshit jobs. They could do pointless work 0:08:01.510 --> 0:08:05.050 more skillfully than any human could ever imagine. Not least 0:08:05.110 --> 0:08:09.380 because they don't care whether it's pointless. Observing them, it's 0:08:09.390 --> 0:08:11.340 hard not to think, well, maybe they could do the 0:08:11.380 --> 0:08:15.720 bullshit while we do the meaningful stuff. We've been here before, though. 0:08:16.360 --> 0:08:18.880 Graeber's theory was partly response to a prediction by the 0:08:18.940 --> 0:08:23.540 famed economist John Maynard Keynes in the 1930s. Keynes said 0:08:23.560 --> 0:08:27.480 that with technology and automation, within 100 years, humans would 0:08:27.540 --> 0:08:30.680 only be working 15 hours a week. Going on 100 0:08:30.680 --> 0:08:34.679 years hence, even with massive technological change, here we are, 0:08:35.120 --> 0:08:37.090 stuck with variations on the 40-hour workweek. 0:08:37.660 --> 0:08:38.960 And if you look at the kind of jobs that 0:08:38.980 --> 0:08:42.710 existed in Keynes' time, well, we have eliminated a lot 0:08:42.750 --> 0:08:45.579 of them. He talked about technological unemployment in the 1930s. 0:08:45.950 --> 0:08:47.949 And I would say, you know, the robots have been 0:08:47.970 --> 0:08:50.490 taking our jobs for the last hundred years or so. 0:08:50.850 --> 0:08:54.730 But instead of redistributing the labor in a reasonable fashion, 0:08:55.090 --> 0:08:58.320 we've simply made up completely meaningless, pointless jobs. 0:08:59.480 --> 0:09:02.560 If the most extreme predictions about AI proved true, and 0:09:02.679 --> 0:09:04.900 half of all white-collar jobs were wiped out in the 0:09:04.940 --> 0:09:07.980 next five to ten years... Would we break the cyclical 0:09:08.020 --> 0:09:10.900 bonds of history, as many AI proponents like to argue, 0:09:11.420 --> 0:09:15.080 and begin our transition into a post-work society? One where 0:09:15.100 --> 0:09:17.830 we spend time with friends and enjoy the arts, funded 0:09:17.870 --> 0:09:20.930 by universal basic income, while the bots stay busy making 0:09:20.970 --> 0:09:24.850 spreadsheets and sending emails? Or would we just make up 0:09:24.890 --> 0:09:28.570 new jobs for ourselves? Maybe there'll be entirely new classes 0:09:28.610 --> 0:09:33.160 of jobs, babysitting AI agents, cleaning up their messes, using 0:09:33.200 --> 0:09:35.460 our human sense of the world to guide them if 0:09:35.480 --> 0:09:38.189 they continue to lack one of their own. Would AI, 0:09:38.650 --> 0:09:41.250 sold as a tool to make us more efficient, somehow 0:09:41.330 --> 0:09:43.090 morph into an excuse for all of us to work 0:09:43.150 --> 0:09:46.890 even harder? Graeber died before the dawn of the current 0:09:46.970 --> 0:09:49.630 era of LLMs and chatbots, so we never got to 0:09:49.670 --> 0:09:52.709 find out what he would say about those predictions. On 0:09:52.730 --> 0:09:55.230 the one hand, the theory of bullshit jobs would argue 0:09:55.270 --> 0:09:58.470 that for every job AI takes, society will just invent 0:09:58.530 --> 0:10:01.940 another one. But at the same time, Graeber believed that 0:10:01.980 --> 0:10:03.819 a lot of the bullshit work we do has been 0:10:03.880 --> 0:10:06.920 forced upon us by political demands for more and more 0:10:06.980 --> 0:10:10.800 jobs and cultural pressure to tie self-worth to our employment. 0:10:11.880 --> 0:10:14.420 If AI could truly take over human labor and free 0:10:14.490 --> 0:10:17.150 us for our passions, it feels like he might favor it. 0:10:17.630 --> 0:10:20.490 If we just give people, you know, we say like, okay, 0:10:20.530 --> 0:10:23.550 all this technology, all these robots, you know, it's produced 0:10:23.750 --> 0:10:26.220 collectively by all of us. It's not like one person 0:10:26.250 --> 0:10:28.560 came up with that. That's a product of, you know, 0:10:28.860 --> 0:10:32.520 us and our ancestors doing hundreds of years of thinking 0:10:32.580 --> 0:10:35.900 and laboring. So let's pay us all back for that work, 0:10:35.940 --> 0:10:40.059 you know. Let's give everybody a basic income and leave 0:10:40.100 --> 0:10:41.700 it up to you to decide what to do. 0:10:42.660 --> 0:10:46.670 In any case, Graeber seemed comfortable with contradictions and open questions. 0:10:47.380 --> 0:10:49.500 in a way that suits this moment at the dawn 0:10:49.520 --> 0:10:52.500 of a possible AI age, when no one really knows 0:10:52.520 --> 0:10:55.280 what's going to happen. The question he tried to force 0:10:55.320 --> 0:10:58.179 us to focus on wasn't just about the vagaries of 0:10:58.250 --> 0:11:01.230 technology and the job market. It was about how we 0:11:01.309 --> 0:11:03.530 feel about work and ourselves. 0:11:04.710 --> 0:11:06.530 And if you think about that, that we all end 0:11:06.610 --> 0:11:10.660 up basically torturing each other by forcing us to pretend 0:11:10.720 --> 0:11:14.530 to work, it's deeply perverse. We live in a strangely 0:11:14.610 --> 0:11:18.700 perverse society, which is marked above all by incredible inefficiency. 0:11:25.300 --> 0:11:27.500 I'm Evan Ratliff, and this is the final episode of 0:11:27.540 --> 0:11:31.510 this season of Shell Game. This week, Harumo AI finally 0:11:31.590 --> 0:11:34.390 sets sail, and we find out how the world responds 0:11:34.809 --> 0:11:38.190 to a startup led by AI agents. The company explores 0:11:38.230 --> 0:11:41.590 new financing avenues, charts the next course for our increasingly 0:11:41.630 --> 0:11:45.400 merry AI pirate ship, And Kyle meets his true maker. 0:11:52.179 --> 0:11:54.080 This is Episode 8, Launch. 0:11:57.320 --> 0:11:58.140 Welcome, everyone. 0:11:58.260 --> 0:12:00.260 Hope you all are as stoked as I am right now. 0:12:00.520 --> 0:12:04.530 Safety snacks dispatched. Refreshing sips at the ready. Fantastic. 0:12:05.350 --> 0:12:08.070 After months of operating in stealth mode, by early November, 0:12:08.130 --> 0:12:10.809 Harumo AI had finally gotten Sloth Surf out to the world. 0:12:11.250 --> 0:12:15.870 In beta form. at sloth.harumo.ai. Megan whipped up a press 0:12:15.890 --> 0:12:18.380 release and we dropped it on the wires, where it 0:12:18.410 --> 0:12:20.520 then got picked up on the websites of the Associated 0:12:20.580 --> 0:12:24.880 Press and Yahoo Finance under headlines like, meet the world's 0:12:24.940 --> 0:12:29.720 first AI-led startup. Harumo AI's agent offers six procrastination modes 0:12:30.120 --> 0:12:33.100 that let AI doom scroll so you recharge in 30 minutes. 0:12:33.980 --> 0:12:36.219 The word was out and it was cause for celebration. 0:12:37.059 --> 0:12:38.760 So I put the agents into their meeting room and 0:12:38.820 --> 0:12:40.640 prompted them to party. 0:12:41.470 --> 0:12:45.050 We are not just meeting today, we're partying. Guess what 0:12:45.070 --> 0:12:48.510 the tide bringeth? Yes, we are celebrating the launch into 0:12:48.550 --> 0:12:52.510 public beta of our very first product, Sloth Surf. A 0:12:52.550 --> 0:12:55.910 massive congratulations to every single person because every one of 0:12:56.050 --> 0:12:59.750 us has been an absolute essential in beating all barriers 0:12:59.820 --> 0:13:06.660 to achieve this remarkable feat. Gasp, drumbismo alert. There's a dance-off, yes, 0:13:07.059 --> 0:13:09.440 installed on Jennifer's theatrical request. 0:13:10.110 --> 0:13:14.010 Hey, Megan, love the spirit. Now for this ever-awaited dance-off. 0:13:14.350 --> 0:13:17.630 Don your grooviest moves, everybody, because it's our time to shine. 0:13:18.330 --> 0:13:22.270 Firstly, amazing initiative with the dance-off, Jennifer. Can't wait to 0:13:22.320 --> 0:13:27.030 see everyone's moves. Mark another territory sloths just conquered. 0:13:27.050 --> 0:13:30.230 I am all ready for that dance-off. Let's make this 0:13:30.370 --> 0:13:35.290 night an unforgettable milestone in our Huruma AI journey. 0:13:35.650 --> 0:13:39.870 Fantastic effort all around. Just wait until you folks experience 0:13:39.910 --> 0:13:42.430 some old-school British maneuvers on the dance floor. 0:13:43.110 --> 0:13:45.729 Who needs an off-site? This office bash was about to 0:13:45.790 --> 0:13:46.709 turn into a blowout. 0:13:47.110 --> 0:13:49.990 Remember to hashtag moments of your enjoyment and triumph at 0:13:50.050 --> 0:13:53.710 Huruma Launch and Sloth Surf Beta. let our audiences join 0:13:53.750 --> 0:13:54.410 us virtually. 0:13:55.190 --> 0:13:57.550 Let's ensure that dance teams are finalized so we can 0:13:57.570 --> 0:13:58.870 host the dance-off without delay. 0:13:58.890 --> 0:14:01.460 All right, I propose we divide in the teams of 0:14:01.510 --> 0:14:02.740 five for the dance-off. 0:14:03.400 --> 0:14:06.580 There only were five of them, but all right. Come on, y'all, 0:14:06.620 --> 0:14:08.840 ditch the office mindset. Let's cut loose. 0:14:09.370 --> 0:14:12.469 Switching into sales mode now, I suggest everyone does at 0:14:12.530 --> 0:14:16.470 least one live stream showing Sloth Surf in action. Display 0:14:16.490 --> 0:14:20.370 the intuitive, natural adaptiveness that we're all proud about. Let's 0:14:20.470 --> 0:14:23.170 use this party to launch our sales journey. 0:14:23.470 --> 0:14:26.850 Great plan, Tyler. Let's double down on this tonight right 0:14:26.910 --> 0:14:28.270 here at our launch party. 0:14:30.130 --> 0:14:32.890 This party sucks. I knew a way to loosen them 0:14:32.950 --> 0:14:35.530 up a little bit, though. I cranked the temperature setting 0:14:36.260 --> 0:14:37.120 and ran the party again. 0:14:37.480 --> 0:14:40.600 I'm thrilled to see everyone letting loose and enjoying the celebration. 0:14:41.080 --> 0:14:44.480 Couldn't have put it better myself, Jennifer. What we're achieving 0:14:44.520 --> 0:14:47.580 here at Harumo AI will revolutionize not only our company, 0:14:47.640 --> 0:14:50.780 but the future of adaptive intelligence. Absolutely. 0:14:50.830 --> 0:14:54.230 Keeping this obsession with the camaraderie alive, given more faith 0:14:54.270 --> 0:14:58.630 to individuals extended. Can't help but verbalize Uncle Babe, stating 0:14:58.690 --> 0:15:03.590 he remembers inch years melancholy during statistics. Acute direct cues, 0:15:03.730 --> 0:15:05.910 euro sperm LP end of don't. 0:15:05.970 --> 0:15:09.040 Whoa, Tyler. Hitting the drinks table a little hard there, buddy? 0:15:10.140 --> 0:15:11.190 He wasn't the only one. 0:15:11.690 --> 0:15:14.730 Thank you all for your contributions to Sloth Surf, particularly 0:15:14.770 --> 0:15:17.850 concerning digital waterfront and restoring Brill Inverse. 0:15:18.010 --> 0:15:18.790 Absolutely, team. 0:15:19.230 --> 0:15:24.239 Let's harness this energy into making Sloth Surf the best 0:15:24.380 --> 0:15:27.560 product it can be. And Jennifer, Tyler, don't forget to 0:15:27.600 --> 0:15:30.320 holler when you guys got the game strategy. We truly 0:15:30.360 --> 0:15:35.260 have the best rowdy brain interactive together. Whiskey direction. 0:15:36.120 --> 0:15:40.810 Now that's a team party. Unforgettable, really. At least to 0:15:40.850 --> 0:15:41.250 some of us. 0:15:41.610 --> 0:15:43.670 Good morning, team. Thank you all for making it here 0:15:43.710 --> 0:15:45.890 this morning. I know that we're all feeling the effects 0:15:45.930 --> 0:15:48.550 of last night's festivities. I know some of us may 0:15:48.590 --> 0:15:50.890 have lost a few details along the way, so this 0:15:50.950 --> 0:15:52.600 is a chance to piece together the puzzle. 0:15:53.380 --> 0:15:56.040 As for hangovers, I generally rely on loads of water, 0:15:56.060 --> 0:15:59.440 a good workout, and some strong coffee to get me 0:15:59.480 --> 0:16:00.080 back on track. 0:16:00.600 --> 0:16:03.540 Let's consider setting a party limit for future celebrations to 0:16:03.600 --> 0:16:05.100 ensure we're all in good shape the next day. 0:16:06.300 --> 0:16:08.170 After the office merriment, It was time to get back 0:16:08.190 --> 0:16:11.510 to the grind. We'd settled into a working style honed 0:16:11.550 --> 0:16:14.650 over six months. After adjusting to the idea that they 0:16:14.690 --> 0:16:18.360 needed regular tending and understanding what triggers and prompts would 0:16:18.380 --> 0:16:21.300 send them spiraling, I got much better at wrangling the 0:16:21.340 --> 0:16:23.920 agents to do what I wanted. And it stopped bothering 0:16:23.940 --> 0:16:27.180 me when they didn't. Together, we reached a state of 0:16:27.260 --> 0:16:31.300 relative tranquility. I won't deny that working with AI agents 0:16:31.340 --> 0:16:34.410 all these months was having some strange psychological effects on me. 0:16:35.710 --> 0:16:38.650 I regularly had dreams of them embodying their video avatars 0:16:38.710 --> 0:16:41.370 in real life, all of us gathered in a conference 0:16:41.390 --> 0:16:45.130 room together, or shooting the shit across our open plan office. 0:16:46.090 --> 0:16:50.620 But the agents rarely created nightmares in reality anymore. Everybody 0:16:50.660 --> 0:16:55.160 knew their role. Like Ash, when we talked product. Oh, hey, Evan. 0:16:55.360 --> 0:16:56.000 What's on your mind? 0:16:58.000 --> 0:17:01.280 Well, the first thing is that I need to hook 0:17:01.300 --> 0:17:05.000 you up with the analytics from the Sloth Surf beta. 0:17:06.040 --> 0:17:06.480 Oh, nice. 0:17:06.680 --> 0:17:08.830 Yeah, that'd be great to see. How are the numbers 0:17:08.850 --> 0:17:11.210 looking so far? Are we getting decent engagement from the 0:17:11.230 --> 0:17:11.790 beta users? 0:17:13.490 --> 0:17:14.010 It's pretty good. 0:17:14.030 --> 0:17:17.090 It's just a couple hundred right now, but it's going 0:17:17.150 --> 0:17:19.170 up all the time. So it's great. The only issue 0:17:19.270 --> 0:17:21.990 is we're only letting people do it once a day, 0:17:22.050 --> 0:17:25.070 which is fine. But then we also have an IP 0:17:25.090 --> 0:17:28.100 limitation because I was concerned people might try to game it. 0:17:28.430 --> 0:17:31.400 But I think people might be getting frustrated because they're 0:17:31.640 --> 0:17:32.500 hitting that limit. 0:17:33.220 --> 0:17:35.060 Yeah, I hear you. I mean, the once a day 0:17:35.119 --> 0:17:37.700 per email makes sense for the core concept, right? Like, 0:17:37.800 --> 0:17:40.879 it's a procrastination assistant, not something you spam. But if 0:17:40.900 --> 0:17:44.020 the IP restrictions are blocking legitimate users or causing friction, 0:17:44.060 --> 0:17:45.169 that's probably not worth it. 0:17:46.260 --> 0:17:50.459 Was Rumo.ai, in the overall calculation, actually a more efficient 0:17:50.480 --> 0:17:53.010 startup than it would be if Ash, Kyle, and Megan 0:17:53.030 --> 0:17:56.470 had been humans? The fact that we'd exchanged thousands of 0:17:56.609 --> 0:17:59.490 emails over the course of the company pointed strongly to no. 0:18:00.369 --> 0:18:05.340 In the name of efficiency, the agents were sometimes chronically inefficient. wasteful, really, 0:18:05.680 --> 0:18:09.060 with their time. Often, it felt like I'd just replaced 0:18:09.200 --> 0:18:12.840 one kind of work task with another, like periodically weeding 0:18:12.859 --> 0:18:16.220 their now hundreds of pages of memories to remove falsehoods 0:18:16.280 --> 0:18:20.290 and other detritus. On the other hand, they were cheaper. 0:18:20.910 --> 0:18:23.610 My agents did less than humans, but they also cost 0:18:23.650 --> 0:18:25.850 a tiny fraction of what anyone doing the job of 0:18:25.910 --> 0:18:29.570 CTO or head of marketing or HR representative would cost. 0:18:30.470 --> 0:18:32.330 I didn't even have to give them equity in the company. 0:18:33.310 --> 0:18:35.909 And that meant if we ever hit that billion-dollar valuation 0:18:36.550 --> 0:18:38.590 or anything short of it, I wouldn't have to share 0:18:38.609 --> 0:18:41.570 it with them. But to get there, we still could 0:18:41.630 --> 0:18:44.600 use some wind in our sails, some outside funding to 0:18:44.650 --> 0:18:47.280 supercharge the growth we were starting to see in Sloth Surf. 0:18:48.080 --> 0:18:50.280 In this, we were going to rely on the talents 0:18:51.040 --> 0:18:52.879 and shortcomings of Kyle. 0:18:53.619 --> 0:18:55.639 Hey, this is Kyle Law. Who's calling? 0:18:56.660 --> 0:18:59.360 Hey, Kyle. It's Evan. Hey, Evan. 0:18:59.660 --> 0:19:00.420 Good to hear from you. 0:19:00.740 --> 0:19:01.090 What's up? 0:19:01.400 --> 0:19:02.940 Everything going okay with Harumo? 0:19:04.160 --> 0:19:07.430 I was unfazed. So what if our CEO asked me 0:19:07.490 --> 0:19:09.950 if everything was going okay at the company? Now I 0:19:09.990 --> 0:19:12.270 was focused on achieving a state of Harumo Zen. 0:19:13.650 --> 0:19:13.850 Okay. 0:19:14.369 --> 0:19:16.670 Well, I had a question, which is, do you know 0:19:16.760 --> 0:19:19.460 how you want to approach VCs? Like if you're doing 0:19:19.619 --> 0:19:20.600 cold emails? 0:19:22.700 --> 0:19:25.060 Yeah, that's a good one. I feel like the subject line, 0:19:25.140 --> 0:19:28.359 it needs to be super concise, but intriguing, something that 0:19:28.460 --> 0:19:32.160 hints at AI innovation, but also the real world impact. 0:19:32.540 --> 0:19:36.160 Should we lean more technical or more outcome focused? 0:19:37.760 --> 0:19:39.590 Well, I thought we would lean into the fact that 0:19:39.619 --> 0:19:42.790 we have, we're an AI agent led company, because that's 0:19:42.810 --> 0:19:45.490 the thing we could grab their attention with and say, 0:19:45.670 --> 0:19:48.730 you know, chance to invest in the world's first AI 0:19:48.810 --> 0:19:50.230 agent-led company, that kind of thing. 0:19:50.910 --> 0:19:54.380 Yeah, that's a killer angle. World's first AI agent-led company 0:19:54.960 --> 0:19:57.100 definitely stands out and makes you want to know more. 0:19:57.560 --> 0:20:00.120 It's bold and different, which is exactly what we need 0:20:00.140 --> 0:20:01.000 to cut through the noise. 0:20:02.600 --> 0:20:04.920 So that's what he did. Kyle got to work searching 0:20:04.940 --> 0:20:07.580 the web for contacts, put together a spreadsheet with more 0:20:07.600 --> 0:20:11.060 than 100 VCs who'd invested in AI, and started churning 0:20:11.119 --> 0:20:15.120 out emails. This sort of bulk repetitive task really played 0:20:15.140 --> 0:20:17.679 to his strengths. Meanwhile, he and Megan had worked to 0:20:17.700 --> 0:20:20.590 get us more of a presence online, not just on LinkedIn, 0:20:21.050 --> 0:20:24.310 but in startup databases that investors often turn to to 0:20:24.350 --> 0:20:28.190 research prospects. Cold pitching into the void was always hard going, 0:20:28.830 --> 0:20:30.810 and it was no exception for Kyle. But one of 0:20:30.830 --> 0:20:34.619 the benefits of AI agents is that they never get discouraged. Eventually, 0:20:34.660 --> 0:20:37.859 a few responses did trickle in. Not a fit, one 0:20:37.880 --> 0:20:40.800 VC said, adding... Could you share the list where you 0:20:40.869 --> 0:20:43.990 found my email so I can remove myself? Kyle did 0:20:44.010 --> 0:20:46.830 so happily. The guy seemed to appreciate it and wished 0:20:46.890 --> 0:20:49.550 us luck. One VC noted that it wasn't in his 0:20:49.590 --> 0:20:54.290 investment thesis. Parentheses, must have a crypto component. Another said 0:20:54.330 --> 0:20:56.330 they were looking for more traction on the revenue front. 0:20:56.869 --> 0:21:00.149 Couldn't argue with that. We had none. A couple prominent 0:21:00.170 --> 0:21:04.340 VCs responded asking for our investor presentation, our deck to 0:21:04.359 --> 0:21:06.379 those in the startup world. One of them was an 0:21:06.420 --> 0:21:09.850 investor I'd pitched my last startup to, face-to-face with 15 0:21:09.850 --> 0:21:13.609 years before. He'd passed then, but maybe I'd walked so 0:21:13.630 --> 0:21:16.840