1 00:00:00,000 --> 00:00:03,720 Eric Anderson: I think you have to be as good at the AI game as these frontier 2 00:00:03,720 --> 00:00:05,250 labs, but I think that's possible. 3 00:00:05,370 --> 00:00:07,410 Like they clearly don't have a lock. 4 00:00:07,410 --> 00:00:07,860 Hold on. 5 00:00:07,860 --> 00:00:10,110 Talent, you know, talent's just leaking everywhere. 6 00:00:15,330 --> 00:00:17,010 Corey Quinn: Welcome to Screaming in the Cloud. 7 00:00:17,160 --> 00:00:18,300 I'm Corey Quinn. 8 00:00:18,570 --> 00:00:22,260 My guest today is one of those rare breeds we don't see a 9 00:00:22,260 --> 00:00:24,240 lot of as guests on this show. 10 00:00:24,450 --> 00:00:29,310 Eric Anderson is a partner at Scale, which is a VC firm, Eric. 11 00:00:29,770 --> 00:00:30,730 Thank you for joining me. 12 00:00:30,880 --> 00:00:31,660 Eric Anderson: Thank you, Corey. 13 00:00:32,290 --> 00:00:35,380 Corey Quinn: This episode is sponsored in part by my day job Duck. 14 00:00:35,380 --> 00:00:38,590 Bill, do you have a horrifying AWS bill? 15 00:00:38,860 --> 00:00:40,750 That can mean a lot of things. 16 00:00:40,960 --> 00:00:46,030 Predicting what it's going to be, determining what it should be, negotiating 17 00:00:46,030 --> 00:00:51,490 your next long-term contract with AWS, or just figuring out why it increasingly 18 00:00:51,490 --> 00:00:55,990 resembles a phone number, but nobody seems to quite know why that is. 19 00:00:56,290 --> 00:00:56,980 To learn more. 20 00:00:56,980 --> 00:00:57,490 Visit dot. 21 00:00:57,610 --> 00:00:59,860 Bill hq.com. 22 00:01:00,160 --> 00:01:03,040 Remember, you can't duck the duck bill. 23 00:01:03,070 --> 00:01:07,870 Bill, which my CEO reliably informs me is absolutely not our slogan. 24 00:01:09,000 --> 00:01:12,390 Usually we talk a lot more to folks who are the engineering type, the founder 25 00:01:12,390 --> 00:01:15,600 type, occasionally the marketing type who weasel their way through, but 26 00:01:15,600 --> 00:01:20,280 we've only had a handful of VCs in the years this show has been running. 27 00:01:20,520 --> 00:01:25,949 So for those listening in the audience who might not be entirely clear on what the 28 00:01:25,949 --> 00:01:31,169 role of a general partner at a VC firm is and only be able to gather it contextually 29 00:01:31,169 --> 00:01:34,800 and badly from the platform formerly known as Twitter, what is it you'd say? 30 00:01:34,800 --> 00:01:35,850 It is you do here? 31 00:01:36,615 --> 00:01:39,075 Eric Anderson: We fund and support startups, it's a lot of 32 00:01:39,075 --> 00:01:42,675 trying to find what's the next big thing, who's building it. 33 00:01:42,795 --> 00:01:45,585 Uh, convince 'em to take your money and then make 'em 34 00:01:45,585 --> 00:01:47,324 successful to the best they can. 35 00:01:47,324 --> 00:01:48,675 Corey Quinn: It always seems counterintuitive for that to 36 00:01:48,675 --> 00:01:51,074 be the framing because like, please, please take my money. 37 00:01:51,074 --> 00:01:54,345 Sounds like it's, it doesn't sound like it's a real problem, if that 38 00:01:54,345 --> 00:01:58,245 makes much sense, but I've, I have a bunch of friends who've raised and 39 00:01:58,245 --> 00:02:01,185 talking to them about the process and seeing folks go through it. 40 00:02:01,505 --> 00:02:02,104 It's weird. 41 00:02:02,104 --> 00:02:03,455 It's it's feast or famine. 42 00:02:03,455 --> 00:02:05,795 It's either no one will fund you or everyone wants to fund you. 43 00:02:05,795 --> 00:02:11,555 And how do you decide which of the various economic suitors you decide to go with? 44 00:02:11,555 --> 00:02:14,795 So a lot of it comes back, channel references, track record history, 45 00:02:14,855 --> 00:02:18,125 similar to the same way that VCs wind up trying to pick the founders that 46 00:02:18,125 --> 00:02:20,795 they want to take bets on these days. 47 00:02:20,795 --> 00:02:25,595 It seems like it is difficult to separate out the world of VC and 48 00:02:25,595 --> 00:02:30,125 funding and startups from the monstrosity that has become ai. 49 00:02:30,705 --> 00:02:33,945 But you've been doing this longer than AI has been a thing. 50 00:02:33,975 --> 00:02:39,045 Historically, you ran a product team, I'm not sure which one, over at AWS, which 51 00:02:39,045 --> 00:02:40,695 I will accept apologies for in a moment. 52 00:02:40,935 --> 00:02:45,075 Uh, then you were at GCP doing similar things for a while and then decided, 53 00:02:45,075 --> 00:02:46,815 you know, building things seems hard. 54 00:02:46,815 --> 00:02:49,635 Let's go instead do the corporate version of betting on the 55 00:02:49,810 --> 00:02:50,805 ponies, on the horse track. 56 00:02:50,865 --> 00:02:51,915 Uh, what was the progression there? 57 00:02:52,195 --> 00:02:55,375 Eric Anderson: Spot instances was the, was the AWS thing. 58 00:02:55,495 --> 00:02:56,665 Corey Quinn: Ah, yes, yes, yes. 59 00:02:56,665 --> 00:02:56,935 Eric Anderson: Yeah. 60 00:02:56,935 --> 00:03:00,865 Everyone's favorite, like intellectual product to, to, to tinker on. 61 00:03:00,925 --> 00:03:01,675 And then it was. 62 00:03:03,045 --> 00:03:06,285 Uh, kind of BigQuery, but mostly this thing called data flow. 63 00:03:06,554 --> 00:03:09,495 But I, I say BigQuery just 'cause people know it a little bit better. 64 00:03:09,885 --> 00:03:13,274 And the progression was really, I don't know, it was incidental at the time. 65 00:03:13,274 --> 00:03:16,005 I was just, uh, I was interested in startups and I wanted to prove 66 00:03:16,035 --> 00:03:18,285 my, my metal in Silicon Valley. 67 00:03:18,345 --> 00:03:22,995 And so I was, I felt like the best way to do that was I going to Amazon or Google 68 00:03:22,995 --> 00:03:24,495 and working on the most technical thing. 69 00:03:24,975 --> 00:03:27,015 I didn't study cs, I started mechanical. 70 00:03:27,015 --> 00:03:30,825 And that, that was always a bit of like an uphill battle with these like hiring. 71 00:03:30,885 --> 00:03:31,545 Firms. 72 00:03:31,695 --> 00:03:33,975 I was like, oh, just stick me on EC2 and I can show you. 73 00:03:33,975 --> 00:03:35,864 I, I can, I'll survive. 74 00:03:36,075 --> 00:03:38,084 Corey Quinn: It kind of feels like that's almost the problem we have at 75 00:03:38,084 --> 00:03:42,795 Google as well, where you have a cash cow that is advertising and everything 76 00:03:42,795 --> 00:03:45,345 else is almost incidental to that, oh, you wanna build a moon base? 77 00:03:45,345 --> 00:03:45,584 Fine. 78 00:03:45,584 --> 00:03:46,394 Go build a moon base. 79 00:03:46,394 --> 00:03:46,755 Good luck. 80 00:03:46,785 --> 00:03:48,495 We're still selling ads primarily. 81 00:03:48,674 --> 00:03:52,245 I, I've always gotten the sense that the AWS world, that EC2 was 82 00:03:52,245 --> 00:03:53,745 kind of that 800 pound gorilla. 83 00:03:53,835 --> 00:03:54,674 Eric Anderson: Yes, yes. 84 00:03:54,674 --> 00:03:59,085 That was like, like you build products and then you monetize them via EC2. 85 00:03:59,385 --> 00:04:02,055 You know, when I was first there, it was before they broke out revenues. 86 00:04:02,115 --> 00:04:06,165 You know, so it was kind of, no one quite knew how interesting this thing was. 87 00:04:06,465 --> 00:04:08,685 Corey Quinn: Everyone thought the thing was losing money hand over fist and then 88 00:04:08,685 --> 00:04:12,495 one day they make an announcement and oh my God, those are damn near SaaS margins. 89 00:04:12,555 --> 00:04:14,445 Eric Anderson: And I would get this mini announcement, right? 90 00:04:14,445 --> 00:04:17,175 I got an email every Friday that told me how many cores we had sold. 91 00:04:17,175 --> 00:04:19,875 Like a little team summary, and I would get out my calculator 92 00:04:19,875 --> 00:04:20,960 and be like, this is crazy. 93 00:04:20,960 --> 00:04:21,440 They're. 94 00:04:21,700 --> 00:04:22,990 Printing cash. 95 00:04:23,349 --> 00:04:23,620 Corey Quinn: Yeah. 96 00:04:23,620 --> 00:04:27,430 And what I love about Spot is that from, I've heard this from multiple folks who 97 00:04:27,430 --> 00:04:32,920 were there at the time and afterwards, that it really is just unused capacity. 98 00:04:32,920 --> 00:04:36,849 It's not like they wind up building stuff specifically to shore up their spot. 99 00:04:37,030 --> 00:04:38,590 It's just stuff that would otherwise. 100 00:04:38,625 --> 00:04:42,075 Be going to waste sitting there as more or less air conditioner ballast. 101 00:04:42,195 --> 00:04:45,975 Suddenly they found a way to monetize this and they managed to do it in a way 102 00:04:46,125 --> 00:04:50,535 that doesn't completely destroy anyone ever paying for on demand anything. 103 00:04:50,715 --> 00:04:52,455 And I think that's kind of a neat approach. 104 00:04:52,485 --> 00:04:54,495 'cause there are sub use cases for which it's phenomenal. 105 00:04:54,755 --> 00:04:56,075 Others for, which is terrifying. 106 00:04:56,285 --> 00:05:00,545 Uh, you were there back in the days when it was inter, our wide 107 00:05:00,545 --> 00:05:04,415 swings in pricing before they, they stabilized it significantly, 108 00:05:04,415 --> 00:05:05,825 which frankly was for the best. 109 00:05:05,975 --> 00:05:08,855 I didn't really want to become a high frequency trader in this 110 00:05:08,855 --> 00:05:10,235 one incredibly niche thing. 111 00:05:10,505 --> 00:05:12,365 Eric Anderson: Apparently this was Bezos's idea. 112 00:05:12,365 --> 00:05:14,555 I mean, I never spoke with him, but 113 00:05:14,945 --> 00:05:16,145 Corey Quinn: I have no trouble believing that. 114 00:05:16,235 --> 00:05:16,445 Eric Anderson: Yeah. 115 00:05:16,445 --> 00:05:19,985 The banker guy was like, why aren't we doing this kind of marketplace? 116 00:05:20,955 --> 00:05:22,544 And yeah, the wild swings. 117 00:05:22,575 --> 00:05:26,025 I didn't appreciate the fact that there was this, you, you kind of imagine 118 00:05:26,025 --> 00:05:31,005 like unused capacity is this nebulous, singular blob, but they carve up all the 119 00:05:31,005 --> 00:05:34,275 instances into all these instance types, into all these regions availability. 120 00:05:34,275 --> 00:05:39,015 Suddenly you're like, we have 400 SKUs of EC2, right? 121 00:05:39,015 --> 00:05:41,570 And each one is a little tiny spot market. 122 00:05:42,450 --> 00:05:43,170 It's a mess. 123 00:05:43,470 --> 00:05:43,710 Corey Quinn: Yeah. 124 00:05:43,710 --> 00:05:45,360 You, you're definitely dating yourself with that reference. 125 00:05:45,360 --> 00:05:47,790 There's over 700 now in US East one alone. 126 00:05:47,820 --> 00:05:50,790 It's, I, I did the math on all of this where I wind up tracking a 127 00:05:50,790 --> 00:05:54,090 lot of, I, I, I had Claude build me some nonsense because why not? 128 00:05:54,270 --> 00:05:58,380 Where it tracks the, uh, the pricing pages for everything that gets dropped out. 129 00:05:58,870 --> 00:06:01,600 It's multiple gigabytes for all the pricing information 130 00:06:01,600 --> 00:06:02,710 broken down by service. 131 00:06:02,710 --> 00:06:06,850 But the EC2 specific one, I had to refactor some of the code 132 00:06:06,850 --> 00:06:08,080 because it kept timing out. 133 00:06:08,110 --> 00:06:08,500 Lambdas. 134 00:06:08,500 --> 00:06:10,000 There's no way you can grab the whole thing. 135 00:06:10,210 --> 00:06:14,530 IOOM killed an EC2 box a couple of times because yeah, that thing's 136 00:06:14,530 --> 00:06:18,820 enormous and it's, and I'm also on bad at programming, but that's okay. 137 00:06:18,820 --> 00:06:21,280 Gonna fix that particular problem for me rather nicely. 138 00:06:21,640 --> 00:06:26,200 And just being able to track all of this, it, it is a monstrous surface area. 139 00:06:26,955 --> 00:06:30,645 And not just tracking the pricing, let alone actually making the thing work. 140 00:06:30,795 --> 00:06:33,975 Eric Anderson: So, so keeping stock, you know, like, like basically a spot 141 00:06:33,975 --> 00:06:35,415 is like an inventory problem, right? 142 00:06:35,445 --> 00:06:38,565 And keeping inventory of one product is a lot different than 400. 143 00:06:38,565 --> 00:06:39,135 I mean, it's like. 144 00:06:39,735 --> 00:06:40,034 Corey Quinn: Right. 145 00:06:40,034 --> 00:06:42,344 My, my customers all tend to be extraordinarily large scale, which 146 00:06:42,344 --> 00:06:46,605 kind of puts the lie to a lot of the historical way the cloud was 147 00:06:46,605 --> 00:06:49,094 positioned and sold even in 2018. 148 00:06:49,094 --> 00:06:51,675 I had a client when the I threes came out like, okay, we're gonna 149 00:06:51,675 --> 00:06:53,895 spin up 1200 of those in Ohio. 150 00:06:54,045 --> 00:06:56,715 And the response from eight OS was, can you give us about 151 00:06:56,715 --> 00:06:58,065 six weeks on that, please? 152 00:06:58,065 --> 00:06:59,355 That would be great. 153 00:06:59,715 --> 00:07:02,205 There's the cloud does not scale. 154 00:07:02,205 --> 00:07:03,705 Infinitely, uh, source tried. 155 00:07:03,705 --> 00:07:04,395 It didn't go so well. 156 00:07:05,295 --> 00:07:05,685 Doing. 157 00:07:05,685 --> 00:07:09,405 The capacity planning comes back around at significant scale. 158 00:07:09,405 --> 00:07:12,255 It starts to resemble a lot of the old school data center stuff. 159 00:07:12,465 --> 00:07:15,435 It's, oh, just turn this thing on for an experiment and turn it off. 160 00:07:15,495 --> 00:07:16,185 That's still there. 161 00:07:16,215 --> 00:07:17,415 That's incredibly powerful. 162 00:07:17,415 --> 00:07:21,735 I think it has been a tremendous boon to getting companies from idea to 163 00:07:21,735 --> 00:07:26,475 start up to success, but also from idea to, oh wait, that won't work. 164 00:07:26,475 --> 00:07:29,745 Nevermind, turn it off, and I owe 24 cents at the end of the month. 165 00:07:30,075 --> 00:07:31,965 Both of those are incredibly powerful. 166 00:07:31,965 --> 00:07:32,590 Things. 167 00:07:32,859 --> 00:07:37,330 Uh, but as you continue to succeed and, and grow and grow and grow it, 168 00:07:37,330 --> 00:07:40,630 it starts to resemble multi-year capacity and contractual planning. 169 00:07:41,080 --> 00:07:44,049 So these days, what are you finding that's exciting you? 170 00:07:44,049 --> 00:07:48,669 What is it that you are, that you're looking at and saying, yes, that is 171 00:07:48,669 --> 00:07:49,900 something worth paying attention to? 172 00:07:50,910 --> 00:07:52,170 Eric Anderson: Certainly the coding agents. 173 00:07:52,470 --> 00:07:57,120 This is, I mean, there's all this talk about a GI, this, which is, is a 174 00:07:57,450 --> 00:07:59,700 poorly, it's an unhelpful phrase, right? 175 00:07:59,700 --> 00:08:02,880 I mean, what I, I, when we get there, does anything change? 176 00:08:03,540 --> 00:08:04,260 Probably not. 177 00:08:04,260 --> 00:08:08,050 And then the overlap between better than human or less than better. 178 00:08:09,180 --> 00:08:15,270 But regardless, whatever it is, I think coding agents are, maybe it, uh, yeah. 179 00:08:15,270 --> 00:08:18,690 Software has forever changed and it, I feel like it, it happened more. 180 00:08:19,065 --> 00:08:22,005 In the last three months than, than I guess before that 181 00:08:22,340 --> 00:08:24,345 Corey Quinn: I, I think that even looking at something like Claude Code as a 182 00:08:24,345 --> 00:08:28,245 software, as a coding agent is a bit of a misnomer and a bit of a weird approach. 183 00:08:28,425 --> 00:08:31,605 It can integrate with effectively everything, and the interaction 184 00:08:31,605 --> 00:08:34,455 model is human language, where. 185 00:08:35,179 --> 00:08:37,909 You can tell it to go out and grab a bunch of different APIs. 186 00:08:37,909 --> 00:08:41,089 Research the best way to do this, construct a research report. 187 00:08:41,089 --> 00:08:44,900 You can treat it just like you can the, the Claude Chatbot expression. 188 00:08:45,110 --> 00:08:50,089 When you have access to the entire CLI, when you have access to every API out 189 00:08:50,089 --> 00:08:55,189 there on the planet, suddenly it's starts to look a lot less like a coding agent 190 00:08:55,189 --> 00:08:59,089 and a lot more like an orchestrator where you can tie together all sorts of 191 00:08:59,089 --> 00:09:01,730 things with a, we're still at a point where you need a little bit of coding 192 00:09:01,730 --> 00:09:03,920 knowledge to make things work, but. 193 00:09:04,305 --> 00:09:08,295 Software is no longer the bottleneck for an awful lot of stuff. 194 00:09:08,655 --> 00:09:08,925 Eric Anderson: Yeah. 195 00:09:08,925 --> 00:09:12,825 I don't, like, I don't use slides anymore or PowerPoint. 196 00:09:13,005 --> 00:09:16,755 It's actually, I think, easier to just ask Claude to generate a presentation 197 00:09:16,755 --> 00:09:21,765 and it does it in like an HTML, you know, webpage ish thing, which is 198 00:09:21,765 --> 00:09:23,145 like, how would you ever edit it? 199 00:09:23,145 --> 00:09:27,225 You don't, you ask Claude to edit it or, or your, your agent of choice. 200 00:09:27,704 --> 00:09:28,725 Corey Quinn: Yeah, I, I do much the same. 201 00:09:28,725 --> 00:09:32,084 I use a slide dev theme that for my company branding and the rest, but I 202 00:09:32,084 --> 00:09:35,865 built an entire custom plugin that has a multiple, uh, different skills for how 203 00:09:35,865 --> 00:09:37,785 I do slides, how I think it should work. 204 00:09:37,995 --> 00:09:41,324 And I'll give it the, I'll give it an outline and Great. 205 00:09:41,324 --> 00:09:42,765 Turn this into a slide deck. 206 00:09:42,944 --> 00:09:47,439 Suddenly all the problems I had as a presenter, I, I. I'm a 207 00:09:47,439 --> 00:09:49,060 public speaker probably too much. 208 00:09:49,060 --> 00:09:51,760 I have this ongoing love affair with the sound of my own voice. 209 00:09:51,760 --> 00:09:57,579 He said on his own podcast where this became, or my biggest challenge was I 210 00:09:57,579 --> 00:10:00,730 would work on a part of the slide deck here, then part of the slide deck here. 211 00:10:00,880 --> 00:10:03,730 Then I'd go and give the thing, and I'm circling the same 0.3 212 00:10:03,730 --> 00:10:06,610 different times at different points throughout the presentation. 213 00:10:06,880 --> 00:10:11,500 It is a terrific editor for, okay, now go back and fix the narrative flow. 214 00:10:11,650 --> 00:10:14,770 Make sure that it does the things in the right order. 215 00:10:15,600 --> 00:10:16,800 It's almost, but not quite. 216 00:10:16,800 --> 00:10:18,840 At a point where I can have it just build my slide deck for me. 217 00:10:18,990 --> 00:10:22,260 It's great for a first attempt at that, but it'll just make things up. 218 00:10:22,319 --> 00:10:24,750 But it turns out, if you say things with a straight enough 219 00:10:24,750 --> 00:10:25,829 face, people will believe you. 220 00:10:26,910 --> 00:10:27,240 Eric Anderson: Yeah. 221 00:10:27,540 --> 00:10:31,020 Slop through the mouth of a human is like all the, all the content 222 00:10:31,020 --> 00:10:32,189 and all the credibility in one. 223 00:10:32,340 --> 00:10:33,000 Corey Quinn: Exactly. 224 00:10:33,060 --> 00:10:35,069 Now it's I, I think that it's an assistant. 225 00:10:35,069 --> 00:10:38,460 I think human in the loop is still gonna be required for the foreseeable future 226 00:10:38,490 --> 00:10:39,870 when it comes to most of these things. 227 00:10:39,960 --> 00:10:41,235 I think that as soon as you take. 228 00:10:42,180 --> 00:10:46,650 That judgment piece out of it and let it speak for you. 229 00:10:47,250 --> 00:10:48,570 There are problems. 230 00:10:48,720 --> 00:10:51,360 You are risking your own credibility every time you do it. 231 00:10:51,360 --> 00:10:55,770 Like I have a ea style bot that I built, Billy the Platypus, uh, whatever 232 00:10:55,770 --> 00:10:58,710 I wind up, uh, turning it loose on various pitches that people send me. 233 00:10:58,890 --> 00:11:02,790 It's technically professional, technically, but it, he is 234 00:11:02,790 --> 00:11:03,960 just basically a total jerk. 235 00:11:03,990 --> 00:11:06,150 That's sort of the entire persona that's built into it. 236 00:11:06,360 --> 00:11:09,870 There's a reason he sends as Billy the Platypus and not as me. 237 00:11:10,050 --> 00:11:12,060 The first time that gets even slightly wrong. 238 00:11:12,090 --> 00:11:14,820 I suddenly have a serious reputation problem. 239 00:11:15,120 --> 00:11:16,950 Eric Anderson: Things that get me excited along those lines 240 00:11:17,100 --> 00:11:19,500 are like, we won't look at code. 241 00:11:19,500 --> 00:11:20,790 I mean, I, I'm excited about this. 242 00:11:20,880 --> 00:11:23,490 Like there was a time when I thought we used the coding agents and then you review 243 00:11:23,490 --> 00:11:24,960 it and someone else does the code review. 244 00:11:25,020 --> 00:11:26,790 Like as long as you do the code review, you're safe. 245 00:11:26,790 --> 00:11:27,120 Right? 246 00:11:27,690 --> 00:11:31,290 And now we have the agents doing the code review, and now maybe the risk is 247 00:11:31,290 --> 00:11:32,820 like, well, what about like performance? 248 00:11:33,464 --> 00:11:36,615 You know, you get these like terribly, you know, no one's refactoring 249 00:11:36,615 --> 00:11:38,625 the code is just slop on top of. 250 00:11:38,775 --> 00:11:41,055 But I think we, I think we'll have agents refactoring the code. 251 00:11:41,714 --> 00:11:44,295 And so I'm excited about the idea, like, what, what ha what 252 00:11:44,295 --> 00:11:46,454 does the world look like when no one looks at the code anymore? 253 00:11:46,694 --> 00:11:49,305 Like, what, what emerges, what are the opportunity? 254 00:11:49,305 --> 00:11:53,564 And so I think there's some, maybe some cool concepts around like, yeah, you know, 255 00:11:53,564 --> 00:11:56,969 performance improvement bots or you know, someone that goes through and kind of. 256 00:11:57,765 --> 00:12:02,175 Refactor optimizes deploys this thing to, you know, constantly keeps it updated with 257 00:12:02,175 --> 00:12:04,395 latest libraries, patching, I don't know. 258 00:12:04,665 --> 00:12:04,875 Corey Quinn: Yeah. 259 00:12:04,875 --> 00:12:06,675 That sort of maintenance bot on it on some level. 260 00:12:06,705 --> 00:12:10,005 Uh, there's also, this, this is an early optimization in some ways too. 261 00:12:10,035 --> 00:12:12,765 Most of the stuff that I have it build and go nuts on only 262 00:12:12,765 --> 00:12:14,325 lives in my internal network. 263 00:12:14,445 --> 00:12:18,075 It's stuff that improves quality of life for the way that I do things. 264 00:12:18,075 --> 00:12:21,345 It improves my own workflows, but I don't. 265 00:12:22,199 --> 00:12:23,189 Make this public. 266 00:12:23,189 --> 00:12:27,689 I don't expose it on the internet and the performance issues of, for example, when 267 00:12:27,689 --> 00:12:31,079 I write my newsletter every week and I've, uh, got it the way I want, it goes ahead 268 00:12:31,079 --> 00:12:34,709 and does the rendering, the formatting, checks, all the links, et cetera, and 269 00:12:34,770 --> 00:12:38,219 they're small performance improvements, like, huh, you're checking 35 links. 270 00:12:38,219 --> 00:12:42,870 Maybe you could do that in parallel rather than sequentially, but even it's not. 271 00:12:43,219 --> 00:12:43,849 Okay. 272 00:12:43,849 --> 00:12:47,180 I, it doesn't, I, I'm not sitting here with a stopwatch waiting for this 273 00:12:47,180 --> 00:12:50,569 thing to finish on my stuff 'cause it's saving me a fair bit of time. 274 00:12:50,569 --> 00:12:51,979 Checking those links manually. 275 00:12:52,280 --> 00:12:54,890 I can grab a cup of coffee while it does it at some point. 276 00:12:54,890 --> 00:12:58,849 Yes, I'll do the easy optimizations, but performance on a lot of 277 00:12:58,849 --> 00:13:03,770 those back of house workflow style tools does not need to be. 278 00:13:04,015 --> 00:13:04,855 Top notch. 279 00:13:04,855 --> 00:13:08,095 In fact, one of the things I like with my own expression of how I think 280 00:13:08,095 --> 00:13:12,475 about things is I'll have, like all my development stuff with Claude Code now 281 00:13:12,475 --> 00:13:17,365 exists in an EC2 box where it has root, where it lives in its own AWS account 282 00:13:17,365 --> 00:13:21,085 where it has admin access and there's nothing of value in this AWS account. 283 00:13:21,085 --> 00:13:22,195 Let be very clear on that. 284 00:13:22,195 --> 00:13:25,765 It's just a bill risk where it can do anything that it wants, but it doesn't 285 00:13:25,765 --> 00:13:27,205 have access to anything sensitive. 286 00:13:27,295 --> 00:13:30,055 And I'll just go and I'll tab over to it and kick it to the 287 00:13:30,055 --> 00:13:32,485 next step, and then I'll go back to doing whatever I was doing. 288 00:13:32,485 --> 00:13:32,575 It's. 289 00:13:32,865 --> 00:13:35,535 It's sort of a drive by and now do the next step. 290 00:13:35,835 --> 00:13:38,235 And I'm sure that's that there's an orchestration story that's coming to, 291 00:13:38,235 --> 00:13:42,015 uh, as an overlayer on top of that anytime now everyone's trying to build 292 00:13:42,015 --> 00:13:45,135 one and get those funded, it seems this week, but there's gonna be something 293 00:13:45,135 --> 00:13:49,395 that emerges and is the next iteration of this, and we'll see how it goes. 294 00:13:49,455 --> 00:13:53,385 I like the fact that if you don't like how things work, give it a month. 295 00:13:53,775 --> 00:13:58,215 Now that said, I think it's really hard to come up with a durable pitch 296 00:13:58,275 --> 00:14:01,545 in the AI space right now that is. 297 00:14:02,085 --> 00:14:05,985 That is fundable just from the perspective of that's a feature release 298 00:14:05,985 --> 00:14:10,575 from Anthropic or open AI before suddenly you have to do a massive pivot. 299 00:14:10,605 --> 00:14:12,255 Like we saw this historically. 300 00:14:12,705 --> 00:14:13,155 Oh wow. 301 00:14:13,155 --> 00:14:16,694 Suddenly ChatGPT can speak to PDF, and suddenly a whole bunch 302 00:14:16,694 --> 00:14:17,895 of companies had a problem. 303 00:14:18,105 --> 00:14:20,805 But that was also relatively easy to predict that that was coming. 304 00:14:21,045 --> 00:14:21,825 How do you think about it? 305 00:14:21,885 --> 00:14:26,505 Eric Anderson: The thing that has proven the most defensible is like, I 306 00:14:26,505 --> 00:14:29,745 mean, I, I agree with you certainly, but I'm impressed that cursor. 307 00:14:30,255 --> 00:14:35,655 The, the way open AI and anthropic became so big is like be, became so scary. 308 00:14:35,655 --> 00:14:39,795 It's just, it's just the sheer growth rate and, and like cursor captured 309 00:14:39,795 --> 00:14:40,724 a little bit of that lightning. 310 00:14:40,724 --> 00:14:41,025 Right. 311 00:14:41,025 --> 00:14:44,665 And, and, and then became maybe threatening to Anthropics 312 00:14:44,689 --> 00:14:45,584 and Anthropics kind of. 313 00:14:45,885 --> 00:14:49,334 But like, I, I feel like when I talk to my portfolio, it's like. 314 00:14:49,770 --> 00:14:52,860 Yes, we should be afraid of them unless we can just grow faster than that. 315 00:14:52,860 --> 00:14:57,360 Like is there, is there a way you can kind of find the vein and shoot to like cursor 316 00:14:57,360 --> 00:15:02,160 scale to the point that, that you kind of can own something and, and some of these 317 00:15:02,160 --> 00:15:04,440 things are growing just incredibly fast? 318 00:15:04,620 --> 00:15:05,250 Corey Quinn: Oh yeah. 319 00:15:05,400 --> 00:15:08,940 I used to use Cursor a fair bit and then I pivoted to Claude Goad and I 320 00:15:08,940 --> 00:15:11,430 haven't gone back since just because. 321 00:15:11,925 --> 00:15:14,775 Cursor was great when I was looking at the code and, okay, now make this 322 00:15:14,775 --> 00:15:16,605 section, do this other thing instead. 323 00:15:16,845 --> 00:15:20,475 But increasingly I don't look at the code that this stuff puts out. 324 00:15:20,805 --> 00:15:23,714 Also, again, this is all backend stuff that I'm building 325 00:15:23,714 --> 00:15:25,425 for ease of use in my stuff. 326 00:15:25,545 --> 00:15:28,515 I suppose now is a good time to detour into Germane story that 327 00:15:28,515 --> 00:15:29,714 is, you know, our sponsor break. 328 00:15:29,775 --> 00:15:33,314 'cause my own company sponsors this at Duck Bill, we have a history as a 329 00:15:33,314 --> 00:15:37,814 consultancy helping companies fix their horrifying AWS and other cloud bills. 330 00:15:37,895 --> 00:15:42,335 By a contract negotiation and diving into finops strategy. 331 00:15:42,395 --> 00:15:44,165 Then now we're doing it with software as well. 332 00:15:44,165 --> 00:15:48,905 Our product is called Skyway, and yeah, we're using Cursor and Claude code 333 00:15:48,905 --> 00:15:53,704 and the rest to build this thing, but it is not itself an AI play directly. 334 00:15:53,704 --> 00:15:57,335 It's providing foundational, normalized data warehouse infrastructure for other 335 00:15:57,335 --> 00:16:01,235 things such as CPS to wind up talking to and getting that data out of it. 336 00:16:01,475 --> 00:16:06,185 But by and large, that is still a place that is relatively not. 337 00:16:06,750 --> 00:16:07,890 Where AI excels. 338 00:16:07,890 --> 00:16:10,290 And it's not because I have a bias in this perspective that I'm 339 00:16:10,290 --> 00:16:13,920 saying that I've done a number of experiments and continue to do them. 340 00:16:14,280 --> 00:16:18,089 It's not there yet for data sets of this scale and this sensitivity. 341 00:16:18,390 --> 00:16:23,400 So if you're interested in learning more, duck bill hq.com, please give us a shout 342 00:16:23,400 --> 00:16:26,490 and you might even have to deal with me, should we have that conversation. 343 00:16:26,520 --> 00:16:27,000 Don't worry. 344 00:16:27,030 --> 00:16:28,740 We have people who are actually good at this stuff. 345 00:16:29,099 --> 00:16:31,740 But yeah, there, there are some areas where it seems like. 346 00:16:32,245 --> 00:16:34,225 Everyone's like, oh, you're really building a B2B SaaS. 347 00:16:34,285 --> 00:16:36,355 Isn't this going to be disrupted by ai? 348 00:16:36,505 --> 00:16:39,295 Well, when you're, when you're talking about things like normalized 349 00:16:39,295 --> 00:16:42,865 infrastructure spend across a wide variety of providers, yeah, AI can 350 00:16:42,865 --> 00:16:47,065 help build the tooling and whatnot, but telling Claude to go out and hit your 351 00:16:47,065 --> 00:16:51,745 billing data for all of your providers and put it into a database for you sure 352 00:16:51,745 --> 00:16:53,905 would be terrific if that were to work. 353 00:16:53,965 --> 00:16:56,605 And it does in 80 to 90% of it. 354 00:16:56,875 --> 00:16:59,365 And then the edge and corner cases absolutely cut you to 355 00:16:59,365 --> 00:17:02,815 ribbons because that's why this is an area of enterprise concern. 356 00:17:03,355 --> 00:17:05,185 If it were simple, it wouldn't be worth doing. 357 00:17:05,485 --> 00:17:05,665 Eric Anderson: Yeah. 358 00:17:05,665 --> 00:17:09,625 I think maybe an analogy that, that your listeners might appreciate. 359 00:17:09,745 --> 00:17:11,454 You know, we, we've been through this before, right? 360 00:17:11,454 --> 00:17:16,135 When AWS was kind of in its early heyday, everyone was afraid of AWS. 361 00:17:16,135 --> 00:17:18,475 All the investors would go down to reinvent and. 362 00:17:18,795 --> 00:17:21,705 They'd announced this new database and a bunch of startups 363 00:17:21,705 --> 00:17:23,595 would, would die because of it. 364 00:17:23,955 --> 00:17:26,925 And, and so we all thought the cloud was gonna be vertical and, and Amazon 365 00:17:26,925 --> 00:17:28,935 was just taking all over all the things. 366 00:17:29,385 --> 00:17:34,335 And then yet, uh, you know, come years later, like four or five years later. 367 00:17:34,475 --> 00:17:35,855 Kind of late to the party. 368 00:17:35,855 --> 00:17:39,094 We got Datadog, kind of horizontal monitoring across all the stack. 369 00:17:39,094 --> 00:17:42,754 We got eventually, just a couple years ago, Wiz security monitoring across 370 00:17:42,754 --> 00:17:44,074 all, all, you know, all the clouds. 371 00:17:44,225 --> 00:17:46,564 We get the proprietary databases, snowflake and 372 00:17:46,564 --> 00:17:48,795 Databricks and, and ClickHouse. 373 00:17:48,814 --> 00:17:51,365 I think these are, these are the things people prefer to use. 374 00:17:51,665 --> 00:17:54,754 So I'm optimistic that, uh, and, and I guess I'm referring 375 00:17:54,754 --> 00:17:56,044 mostly the infrastructure stack. 376 00:17:56,534 --> 00:17:57,625 That, that it doesn't go off. 377 00:17:57,824 --> 00:17:59,070 Kubernetes was a big unlock 378 00:17:59,070 --> 00:17:59,820 Corey Quinn: for this as well. 379 00:17:59,820 --> 00:18:03,480 I mean, back when I started doing this, I have of opinion this is game 380 00:18:03,480 --> 00:18:07,469 point and match to AWS, the end and there's gonna be a bunch of also ran. 381 00:18:07,889 --> 00:18:09,899 I do not have that opinion these days. 382 00:18:09,929 --> 00:18:11,699 Uh, they're obviously not going away. 383 00:18:11,699 --> 00:18:13,409 They're not going anywhere, but. 384 00:18:13,670 --> 00:18:18,470 It's impossible to ignore Azure, GCP, and even Oracle Cloud, but all the value seems 385 00:18:18,470 --> 00:18:20,510 to be at just one level up the stack. 386 00:18:20,720 --> 00:18:24,230 Take Vercel, for example, for front end it, it does all the 387 00:18:24,230 --> 00:18:26,150 things that you can do on AWS. 388 00:18:26,150 --> 00:18:30,590 Clearly, Vercel runs on AWS and about a 20 to 30% markup on top 389 00:18:30,590 --> 00:18:34,310 of it, but I have a lot of stuff running on Vercel instead of on AWS. 390 00:18:34,540 --> 00:18:35,080 Why? 391 00:18:35,229 --> 00:18:38,020 Well, because I don't know anything about front end, but that's 392 00:18:38,020 --> 00:18:40,899 what the LLM picks and Okay. 393 00:18:40,899 --> 00:18:44,379 I, I don't have a strong enough opinion to override it on that space, so. 394 00:18:44,379 --> 00:18:44,800 Okay. 395 00:18:44,800 --> 00:18:46,149 I guess we're putting the front end there. 396 00:18:46,449 --> 00:18:48,879 Eric Anderson: Yeah, so I think, I think there's a chance, you know, there's a way 397 00:18:48,879 --> 00:18:51,590 to compete against open AI or Anthropic. 398 00:18:51,610 --> 00:18:57,129 I think certainly the thing that they're weak on is just the diversity of like, 399 00:18:57,129 --> 00:19:00,820 they can go into Claude Code, they can go into Claude Bot, they can go into. 400 00:19:01,005 --> 00:19:05,355 The Claude Cowork, they're fighting a battle on many fronts. 401 00:19:05,835 --> 00:19:09,915 And, and so if you can, if you can be laser focused, if you can fe realize 402 00:19:09,915 --> 00:19:13,305 what is the front that is actually the most valuable, like in the case of the 403 00:19:13,305 --> 00:19:17,685 cloud, it turned out to be like the data warehouse was the front to fight on. 404 00:19:18,150 --> 00:19:21,750 Um, that was the area to win where both Amazon was weak 405 00:19:21,750 --> 00:19:23,040 and the value would accrue. 406 00:19:23,100 --> 00:19:25,380 So if you can figure out the right front and then just be 407 00:19:25,380 --> 00:19:26,880 laser focused and be good. 408 00:19:26,880 --> 00:19:31,290 I mean, I think, I think you have to be as good at the AI game as these frontier 409 00:19:31,290 --> 00:19:32,820 labs, but I think that's possible. 410 00:19:32,970 --> 00:19:35,430 Like they clearly don't have a lock hold on. 411 00:19:35,430 --> 00:19:37,770 Talent, you know, talent's just leaking everywhere. 412 00:19:38,340 --> 00:19:41,700 So yeah, you find great talent, you figure out where value's gonna 413 00:19:41,700 --> 00:19:44,640 accrue in an interesting space, and then you're just laser focused. 414 00:19:44,820 --> 00:19:46,050 And if you can catch the growth. 415 00:19:46,500 --> 00:19:47,730 I think there's a viable path. 416 00:19:48,000 --> 00:19:48,210 Corey Quinn: Yeah. 417 00:19:48,240 --> 00:19:51,150 There's also the question of what are the underserved niches? 418 00:19:51,150 --> 00:19:54,750 I've always liked finding the expression of these things that, that works. 419 00:19:54,930 --> 00:19:59,280 There is no amount of money I can raise from anyone that is going to mean that 420 00:19:59,280 --> 00:20:03,450 I am now the third massive frontier lab that's building this stuff out. 421 00:20:03,450 --> 00:20:06,330 I'm discounting the ones at Google, for example, like the, that, that's 422 00:20:06,330 --> 00:20:07,650 not exactly the same thing here. 423 00:20:07,860 --> 00:20:12,360 Uh, but I, I'm not gonna ex, I'm not gonna outrun these players at that, but, and. 424 00:20:12,775 --> 00:20:14,905 The capabilities are growing by leaps and bounds. 425 00:20:14,965 --> 00:20:18,115 So where are the areas that I know well that I can bring some of these things to 426 00:20:18,115 --> 00:20:21,355 bear on, uh, industry specific expertise. 427 00:20:21,715 --> 00:20:22,015 Uh. 428 00:20:22,590 --> 00:20:26,399 Opportunity passes everywhere, and I think that that is the way to think 429 00:20:26,399 --> 00:20:29,189 about it to, to no small extent. 430 00:20:29,820 --> 00:20:34,620 I also don't necessarily know that I want to be building the exact same 431 00:20:34,620 --> 00:20:36,120 thing that everyone else is building. 432 00:20:36,389 --> 00:20:39,330 I, I like finding a direction to take things in. 433 00:20:39,720 --> 00:20:42,360 It's weird because I find myself for one of the first times in my 434 00:20:42,360 --> 00:20:45,960 life being something of a centrist on this because I don't believe 435 00:20:45,960 --> 00:20:49,470 the doomsayers say that we're going to build a GI and we're going to. 436 00:20:49,845 --> 00:20:53,295 At this point, trample everything out there because computers 437 00:20:53,295 --> 00:20:54,195 will think for themselves. 438 00:20:54,735 --> 00:20:56,355 We're not summoning God through JSO here. 439 00:20:56,775 --> 00:20:59,565 And I also am on the other side where I don't think it's just a 440 00:20:59,565 --> 00:21:03,105 jumped up auto complete because it is clearly far more than that. 441 00:21:03,315 --> 00:21:05,115 I'm, I'm between those two extremes. 442 00:21:05,115 --> 00:21:07,065 And it's a weird place to find myself, 443 00:21:07,305 --> 00:21:07,725 Eric Anderson: right. 444 00:21:08,055 --> 00:21:08,325 Yeah. 445 00:21:08,505 --> 00:21:12,615 To, because usually the world's just either really wrong or it's obvious. 446 00:21:12,615 --> 00:21:15,135 And in this case it's neither, it's, it's like. 447 00:21:15,570 --> 00:21:19,470 This thing is for real and it's, but you know, it, it's subject to to 448 00:21:19,470 --> 00:21:21,510 physical laws like, like everybody else. 449 00:21:21,540 --> 00:21:21,750 Yeah, 450 00:21:21,870 --> 00:21:22,110 Corey Quinn: yeah. 451 00:21:22,560 --> 00:21:25,200 OpenAI alone has, RA has committed to do more infrastructure spend 452 00:21:25,200 --> 00:21:28,590 between now and 2030 than there is deployable global VC capital. 453 00:21:29,130 --> 00:21:33,540 I have some questions about what that's going to look like because they're not the 454 00:21:33,540 --> 00:21:35,310 only lab that is doing this sort of thing. 455 00:21:35,550 --> 00:21:38,160 What does it look like five years from now? 456 00:21:38,190 --> 00:21:40,620 What is the economic story of this? 457 00:21:40,620 --> 00:21:42,600 We're clearly looking at something bubble shaped. 458 00:21:42,965 --> 00:21:44,705 What does the correction look like? 459 00:21:45,035 --> 00:21:45,215 It. 460 00:21:45,215 --> 00:21:46,385 I'll tell you what, it's not. 461 00:21:46,535 --> 00:21:49,895 It's not, and now we're going to act as if LLMs never existed. 462 00:21:49,985 --> 00:21:52,145 You're not putting that genie back in the bottle. 463 00:21:52,415 --> 00:21:54,845 Maybe this price of inference is going to skyrocket. 464 00:21:54,875 --> 00:21:58,175 Maybe the ability to run things that are almost as good locally 465 00:21:58,175 --> 00:21:59,585 is going to be the approach. 466 00:22:00,389 --> 00:22:03,330 Even with having coding assistance build this stuff, maybe I don't need 467 00:22:03,330 --> 00:22:07,290 the top tier frontier, bleeding edge state-of-the-art model to wind up 468 00:22:07,290 --> 00:22:11,760 doing what is effectively a fancy, uh, said string replacement in a file. 469 00:22:11,760 --> 00:22:14,790 Maybe that can be the local thing and the deep architecture planning 470 00:22:14,790 --> 00:22:16,080 is something attica's outsourced. 471 00:22:16,500 --> 00:22:20,129 These are all things that people way smarter than I am, are focused on. 472 00:22:20,370 --> 00:22:23,310 I'm just curious to see where it goes because the benefit as a developer 473 00:22:23,310 --> 00:22:25,620 myself is accruing rapidly and massively 474 00:22:25,889 --> 00:22:26,939 Eric Anderson: a bubble is inevitable. 475 00:22:27,145 --> 00:22:31,014 There's, there's no avoiding a bubble because the growth rates are so 476 00:22:31,044 --> 00:22:37,345 incredibly high and un like, like when, so anthropic went from one to 7 billion 477 00:22:37,345 --> 00:22:41,010 in revenue in a year, so they have to plan for another seven x increase. 478 00:22:41,985 --> 00:22:43,365 Or at least a five x increase. 479 00:22:43,365 --> 00:22:46,965 Like they can't just like, not buy the capacity they need. 480 00:22:47,385 --> 00:22:49,455 And, and could it be higher, could it be They have no idea. 481 00:22:49,455 --> 00:22:52,125 So they have to, they have to, they have to procure the 482 00:22:52,125 --> 00:22:56,475 capacity to satisfy at least some portion of that expected demand. 483 00:22:57,075 --> 00:23:01,125 And until these crazy growth rates give, we have to plan. 484 00:23:01,530 --> 00:23:05,190 We, we have to overbuy, you know, over like, until, like eventually they 485 00:23:05,190 --> 00:23:06,810 give, we won't know when they give. 486 00:23:07,470 --> 00:23:10,410 And, and then when they give, we'll have realized we have overbought, 487 00:23:10,470 --> 00:23:13,320 but until they give, we'll feel like we have under bought. 488 00:23:13,710 --> 00:23:17,100 So a bubble is inevitable, but I don't, and, and so I think in terms 489 00:23:17,100 --> 00:23:19,890 things, in terms of like bubble prediction isn't all that helpful. 490 00:23:20,565 --> 00:23:23,835 Unless you can kind of call the point at which we saturate, 491 00:23:23,865 --> 00:23:24,165 Corey Quinn: right? 492 00:23:24,165 --> 00:23:26,925 Economists have successfully produced, uh, predicted five 493 00:23:26,925 --> 00:23:28,005 of the last three recessions. 494 00:23:28,005 --> 00:23:29,265 I mean, this is always the problem. 495 00:23:29,265 --> 00:23:30,495 You smack into it. 496 00:23:30,495 --> 00:23:31,815 You can't timing the market. 497 00:23:31,815 --> 00:23:34,935 It can remain irrational longer than you can remain solvent, but 498 00:23:34,935 --> 00:23:36,255 there are, there are limits on this. 499 00:23:36,255 --> 00:23:39,475 I, I spend 200 bucks a month for the Claude Pro Max Plan 500 00:23:39,495 --> 00:23:40,605 with a smile on my face. 501 00:23:40,845 --> 00:23:44,145 I'm not gonna spend $5,000 a month on that because at some 502 00:23:44,145 --> 00:23:47,115 point there is a limit and. 503 00:23:48,120 --> 00:23:51,090 It's, there has to be something that gives, you're into population limits. 504 00:23:51,090 --> 00:23:53,490 People willing to drop that kind of money on these things, 505 00:23:53,700 --> 00:23:55,170 but where is that boundary? 506 00:23:55,200 --> 00:23:55,830 I don't know. 507 00:23:56,040 --> 00:24:00,810 A lot of the funding sort of acts and the messaging has been around that your boss 508 00:24:00,810 --> 00:24:05,760 is going to replace you with AI and then split your salary with the AI company. 509 00:24:05,790 --> 00:24:09,000 Like I, I don't think that that necessarily tracks 510 00:24:09,060 --> 00:24:09,390 Eric Anderson: Yeah. 511 00:24:09,390 --> 00:24:12,900 This, this idea that all pricing holds and like whoever deploys 512 00:24:12,900 --> 00:24:14,370 the AI gets to keep all the money. 513 00:24:14,800 --> 00:24:15,730 Is crazy. 514 00:24:15,790 --> 00:24:19,270 Like there's certainly gonna be some amount of commodification where people 515 00:24:19,270 --> 00:24:23,050 are like, oh, I don't have to, like, I wanna keep some of my money too. 516 00:24:23,050 --> 00:24:24,190 I'm not just gonna give it all to you. 517 00:24:24,190 --> 00:24:27,970 And you deploy the ai I'm expecting that all of the people I buy 518 00:24:27,970 --> 00:24:29,260 things from are deploying ai. 519 00:24:30,130 --> 00:24:32,140 They're gonna bargain, you know, they're, they're gonna 520 00:24:32,140 --> 00:24:33,580 compete in a marketplace where. 521 00:24:34,155 --> 00:24:38,265 Everyone lower lowers their prices because they can eventually, margins 522 00:24:38,325 --> 00:24:39,975 get to the point where they always were. 523 00:24:40,245 --> 00:24:42,765 You know, there's a certain amount of money people are willing to work for 524 00:24:43,125 --> 00:24:45,975 and if you have monopoly power, you get, you get to charge a little bit extra. 525 00:24:46,155 --> 00:24:48,615 Like, like the lawyers today, there's a lot of talk that, like 526 00:24:48,765 --> 00:24:50,205 lawyers used to bill by the hour. 527 00:24:50,205 --> 00:24:53,025 They can't do that anymore because now AI is doing the work. 528 00:24:53,025 --> 00:24:53,385 So they have to. 529 00:24:54,645 --> 00:24:56,715 Bill by the, by the project. 530 00:24:57,075 --> 00:24:59,895 And then they just keep the extra money, I think. 531 00:24:59,895 --> 00:25:02,325 I think we're all just gonna be like, no, you're not really doing 532 00:25:02,325 --> 00:25:03,855 any work, so I'm gonna pay you less. 533 00:25:03,855 --> 00:25:05,295 And then we get back to billing by the hour. 534 00:25:05,865 --> 00:25:07,695 Corey Quinn: Everyone acts like this changes everything, and 535 00:25:07,695 --> 00:25:09,045 I'm not convinced that it does. 536 00:25:09,225 --> 00:25:10,545 There are strong. 537 00:25:11,100 --> 00:25:13,770 Indications that this is, there are ways forward on this. 538 00:25:13,950 --> 00:25:14,700 Uh, take a couple. 539 00:25:14,700 --> 00:25:15,930 You're a board observer for Honeycomb. 540 00:25:15,930 --> 00:25:18,960 Uh, we've been working with them both with the client and other ways for a long time. 541 00:25:19,169 --> 00:25:23,370 I love the way that they do AI because they don't splatter AI all over 542 00:25:23,370 --> 00:25:24,810 their messaging and their marketing. 543 00:25:24,990 --> 00:25:26,879 They have built it in useful ways. 544 00:25:26,909 --> 00:25:28,560 Their MCP is a thing of beauty. 545 00:25:28,740 --> 00:25:31,440 The fact that you can ask in plain language what the hell is 546 00:25:31,440 --> 00:25:34,530 going on in your environment and it will tell you is glorious. 547 00:25:34,620 --> 00:25:38,040 But when, whenever someone talks about AI to the exclusion of all else. 548 00:25:38,115 --> 00:25:40,785 I'm sorry to break the hearts of marketers out there, but as 549 00:25:40,785 --> 00:25:42,495 a customer, that's off-putting. 550 00:25:42,735 --> 00:25:47,325 I don't care if you are using ai, incredibly smart if statements or 551 00:25:47,325 --> 00:25:49,035 just interns that type very quickly. 552 00:25:49,305 --> 00:25:52,575 I just care about the outcome that you are delivering for value. 553 00:25:52,845 --> 00:25:56,505 That's the important piece from where I sit and I'm not, I'm very far from 554 00:25:56,535 --> 00:25:59,685 alone in that it's similar to, I don't care how the sausage is made. 555 00:25:59,685 --> 00:26:00,615 I care that it tastes good. 556 00:26:00,795 --> 00:26:01,125 Eric Anderson: Yeah. 557 00:26:01,185 --> 00:26:01,455 Yeah. 558 00:26:01,455 --> 00:26:03,075 There's a kind of a second order. 559 00:26:03,405 --> 00:26:07,875 Wave, I think of AI use, like the first is the obvious, where we use it in the 560 00:26:07,875 --> 00:26:13,065 context of our current, what's a good example, you know, the, the AI workers. 561 00:26:13,125 --> 00:26:17,835 Like, oh, let's have an ai, SDR or an AI data scientist, because that's, 562 00:26:17,865 --> 00:26:21,045 that's like the current framework of our society, and we can plug them in in 563 00:26:21,045 --> 00:26:25,665 those holes, but presumably we should discover new ways of organizing society. 564 00:26:26,250 --> 00:26:28,080 That weren't possible until we had ai. 565 00:26:28,080 --> 00:26:30,750 And once we discover those new ways, then we'll have products that take shapes. 566 00:26:30,750 --> 00:26:32,790 We don't, we can't really imagine at this point. 567 00:26:33,330 --> 00:26:35,640 And so I think you're right, like the way Honeycomb feels like a, 568 00:26:35,640 --> 00:26:39,060 a tease towards this future where like, hey, maybe not, everything's a 569 00:26:39,060 --> 00:26:42,930 chat bot, actually, uh, like a side panel alongside the traditional app. 570 00:26:43,365 --> 00:26:46,485 Maybe it's like infused within applications in ways that we 571 00:26:46,485 --> 00:26:49,215 didn't really think possible before because it wasn't possible. 572 00:26:49,455 --> 00:26:51,825 Corey Quinn: Maybe I don't wanna alert your dumb proprietary SQL version 573 00:26:51,825 --> 00:26:53,445 to get value out of your platform. 574 00:26:53,445 --> 00:26:57,705 Maybe I, maybe your robot can do that for me, similar to when I have a problem, 575 00:26:57,705 --> 00:26:58,965 I need to reach out to a company. 576 00:26:59,085 --> 00:27:00,764 Don't make me talk to an AI bot. 577 00:27:01,240 --> 00:27:04,899 But have that AI bot provide valuable context to this human support agent that 578 00:27:04,899 --> 00:27:09,010 I'm talking to, to power through that a lot more quickly and provide context 579 00:27:09,010 --> 00:27:10,389 like, I had the last seven tickets. 580 00:27:10,389 --> 00:27:13,209 This guy either knows what he's talking about or is a complete 581 00:27:13,209 --> 00:27:14,679 buffoon, just accordingly. 582 00:27:15,010 --> 00:27:18,639 And they can provide, they can get to answers a lot more effectively 583 00:27:18,639 --> 00:27:22,209 that way, as opposed to making me run the AI gauntlet before. 584 00:27:22,209 --> 00:27:22,780 Finally. 585 00:27:22,840 --> 00:27:22,929 Huh. 586 00:27:23,290 --> 00:27:24,879 It looks like you can't solve this problem yourself. 587 00:27:24,879 --> 00:27:26,350 You're gonna have to talk to a human. 588 00:27:26,379 --> 00:27:27,340 No kidding. 589 00:27:27,970 --> 00:27:30,520 Uh, all these companies have been talking about chatbots, like it's 590 00:27:30,520 --> 00:27:32,560 somehow the pinnacle of user experience. 591 00:27:32,770 --> 00:27:32,949 No. 592 00:27:32,949 --> 00:27:36,879 People talk to chatbots or humans when the user experience has failed them. 593 00:27:36,939 --> 00:27:38,379 You're already starting a step behind. 594 00:27:38,500 --> 00:27:38,740 Eric Anderson: Yes. 595 00:27:38,745 --> 00:27:38,925 Yes. 596 00:27:39,250 --> 00:27:41,649 What if the people in the call center were three times more effective 597 00:27:41,830 --> 00:27:45,159 because they just solved their problem three times faster rather than talk 598 00:27:45,159 --> 00:27:48,669 to three customers at once, which is what I feel like most of the time is. 599 00:27:48,669 --> 00:27:50,770 They're like, yeah, let me check on that for you. 600 00:27:50,770 --> 00:27:51,490 Two minutes later. 601 00:27:52,000 --> 00:27:53,470 Like, how is it taking this long? 602 00:27:54,040 --> 00:27:54,430 Corey Quinn: Right. 603 00:27:54,460 --> 00:27:57,760 Or they wind up asking you questions you answered three messages ago. 604 00:27:57,760 --> 00:27:59,590 It's like, I thought I had a short context window. 605 00:27:59,800 --> 00:28:00,280 It's awful. 606 00:28:00,850 --> 00:28:01,270 Ugh. 607 00:28:01,450 --> 00:28:04,060 I wanna thank you for taking the time to speak with me about all this. 608 00:28:04,090 --> 00:28:07,000 If people wanna learn more, where's the best place for them to find you? 609 00:28:07,090 --> 00:28:08,650 Eric Anderson: You can find me on Scale's website. 610 00:28:08,740 --> 00:28:10,180 I'm fairly active on LinkedIn. 611 00:28:10,180 --> 00:28:12,700 I need to get my Twitter, what do we call it now? 612 00:28:12,700 --> 00:28:13,480 Game up. 613 00:28:13,480 --> 00:28:18,280 But yeah, I, I, LinkedIn or uh, scale website would be a good place to start. 614 00:28:18,665 --> 00:28:21,335 Corey Quinn: And we'll of course put links to that in the show notes. 615 00:28:21,665 --> 00:28:23,765 Thank you so much for taking the time to speak with me. 616 00:28:23,795 --> 00:28:24,575 I appreciate it. 617 00:28:24,875 --> 00:28:25,355 Eric Anderson: Thanks, Corey. 618 00:28:25,535 --> 00:28:27,545 Corey Quinn: Eric Anderson, partner at scale. 619 00:28:27,965 --> 00:28:31,205 I'm cloud Economist Corey Quinn, and this is Screaming in the Cloud. 620 00:28:31,475 --> 00:28:32,825 You've enjoyed this podcast. 621 00:28:32,825 --> 00:28:36,035 Please leave a five star review on your podcast platform of choice, 622 00:28:36,215 --> 00:28:39,365 whereas if you've hated this podcast, please, we have a five star review 623 00:28:39,365 --> 00:28:43,175 on your podcast platform of choice, along with an angry comment talking 624 00:28:43,175 --> 00:28:47,315 about how your minor incremental feature to an AI foundation model. 625 00:28:47,530 --> 00:28:51,520 Is way different than the others and no one could ever possibly compete with you.