1
00:00:02,720 --> 00:00:14,000
Speaker 1: Bloomberg Audio Studios, Podcasts, radio News.

2
00:00:18,360 --> 00:00:21,840
Speaker 2: Hello and welcome to another episode of the Odd Lots podcast.

3
00:00:21,880 --> 00:00:25,400
Speaker 2: I'm Joe Wassental and I'm Tracy Alloway. So, Tracy, you're

4
00:00:25,400 --> 00:00:28,880
Speaker 2: cool like if I, like, you know, just start doing

5
00:00:28,880 --> 00:00:32,000
Speaker 2: this part time as I like build out my software business, right, Like,

6
00:00:32,040 --> 00:00:32,720
Speaker 2: you're cool about that.

7
00:00:32,960 --> 00:00:36,440
Speaker 3: I was gonna say, I've been thinking about AI and productivity,

8
00:00:36,680 --> 00:00:40,720
Speaker 3: and so far your productivity has gone down, Joe, because

9
00:00:40,760 --> 00:00:44,080
Speaker 3: instead of doing Odd Lots things, you're coding your own software.

10
00:00:44,560 --> 00:00:47,639
Speaker 2: Except that I'm creating content for the Odd Lots newsletter

11
00:00:47,680 --> 00:00:53,120
Speaker 2: about coding, and that is productivity a creative debatable debatable,

12
00:00:53,159 --> 00:00:55,880
Speaker 2: But but you're cool with that. You're cool with like

13
00:00:55,960 --> 00:00:57,800
Speaker 2: me like, oh, I'm just gonna like check in part

14
00:00:57,800 --> 00:00:59,480
Speaker 2: time on Odd Lots when we have a recording.

15
00:00:59,520 --> 00:01:01,680
Speaker 3: Well, so of course not okay, good, of course not.

16
00:01:01,840 --> 00:01:03,600
Speaker 2: Yeah, that's the right answer. I want you to yes,

17
00:01:03,680 --> 00:01:05,640
Speaker 2: I want you to be really sad. But like a

18
00:01:05,680 --> 00:01:08,360
Speaker 2: few other people, you know, I have like caught the

19
00:01:08,400 --> 00:01:10,959
Speaker 2: sort of like bug of like AI coding and I'm

20
00:01:11,000 --> 00:01:13,680
Speaker 2: totally blown away. I've like played with it from the

21
00:01:13,720 --> 00:01:16,040
Speaker 2: big I started playing around with it last year, but

22
00:01:16,080 --> 00:01:18,200
Speaker 2: then over the holidays. I've been writing about this in

23
00:01:18,240 --> 00:01:20,839
Speaker 2: the newsletter. Suddenly, like my Twitter feed is like clud code,

24
00:01:20,840 --> 00:01:24,320
Speaker 2: clug code, clud code. I do just cursor before, which

25
00:01:24,319 --> 00:01:27,560
Speaker 2: I was very impressed by at the time. And so

26
00:01:27,680 --> 00:01:29,800
Speaker 2: when I got home from vacation, one of the first

27
00:01:29,840 --> 00:01:31,559
Speaker 2: things I did is like figure out how to install

28
00:01:31,640 --> 00:01:34,920
Speaker 2: clug code on my computer. And I was like, oh,

29
00:01:35,080 --> 00:01:39,080
Speaker 2: I am like hooked. This is actually like I see

30
00:01:39,120 --> 00:01:41,200
Speaker 2: why I have my Twitter feed is just like people

31
00:01:41,200 --> 00:01:41,880
Speaker 2: posting about this.

32
00:01:42,160 --> 00:01:42,520
Speaker 4: All right.

33
00:01:42,560 --> 00:01:44,520
Speaker 3: So I have to say I have not tried it

34
00:01:44,760 --> 00:01:47,319
Speaker 3: because I only have a work computer and I can't

35
00:01:47,360 --> 00:01:51,960
Speaker 3: install new software, and I probably definitely cannot install new

36
00:01:52,000 --> 00:01:53,760
Speaker 3: software that then makes changes to.

37
00:01:53,720 --> 00:01:54,920
Speaker 2: Your existing software.

38
00:01:55,080 --> 00:01:57,680
Speaker 3: I don't think Bloomberg would like that, but I have

39
00:01:57,800 --> 00:02:00,760
Speaker 3: seen the hype. Lots of people talk talking about it.

40
00:02:01,200 --> 00:02:05,000
Speaker 3: Have you seen claud Cowork Have you heard.

41
00:02:04,920 --> 00:02:06,080
Speaker 2: Of Oh yeah, yeah, yeah yeah.

42
00:02:06,160 --> 00:02:08,680
Speaker 3: So one of the criticisms of claud code was that

43
00:02:08,800 --> 00:02:11,560
Speaker 3: you know, like, okay code, but you still need some

44
00:02:11,680 --> 00:02:14,880
Speaker 3: background knowledge in coding, because like, you know, the interface

45
00:02:14,960 --> 00:02:18,200
Speaker 3: is kind of like it pies and all of that

46
00:02:18,360 --> 00:02:22,800
Speaker 3: or nineteen nineties Cowork apparently like goes a step further

47
00:02:23,000 --> 00:02:27,600
Speaker 3: for normal people in quoting. It makes it super super easy.

48
00:02:27,720 --> 00:02:32,400
Speaker 3: And the funniest thing is that apparently cloud code actually

49
00:02:32,440 --> 00:02:33,720
Speaker 3: coded Cowork.

50
00:02:34,080 --> 00:02:37,399
Speaker 2: So this is like really relates to my experience last

51
00:02:37,480 --> 00:02:40,320
Speaker 2: year and then this year, which is that even last year,

52
00:02:40,440 --> 00:02:43,760
Speaker 2: like trying to use the AI coding tools, it was

53
00:02:43,800 --> 00:02:47,280
Speaker 2: an annoying process because there are various things that you

54
00:02:47,320 --> 00:02:49,280
Speaker 2: had to do in the actual command line of the

55
00:02:49,280 --> 00:02:52,320
Speaker 2: computer that were like I didn't I don't know command

56
00:02:52,400 --> 00:02:55,040
Speaker 2: line vernacular, and you have to install these libraries and

57
00:02:55,080 --> 00:02:56,880
Speaker 2: stuff like that. So there was this sort of like

58
00:02:57,440 --> 00:03:03,680
Speaker 2: barrier that existed. But what's really changed in the last

59
00:03:03,760 --> 00:03:06,400
Speaker 2: year or with the with Claude code, which has actually

60
00:03:06,400 --> 00:03:07,920
Speaker 2: been around for a while and I should have like

61
00:03:07,919 --> 00:03:11,080
Speaker 2: played with it before, is that like, because it sits

62
00:03:11,160 --> 00:03:14,480
Speaker 2: on your computer, it sort of takes away it de

63
00:03:14,560 --> 00:03:15,840
Speaker 2: abstracts it and so when you talk.

64
00:03:15,680 --> 00:03:17,880
Speaker 3: About like that it actually does the stuff.

65
00:03:17,560 --> 00:03:19,520
Speaker 2: It does it it just like oh, it's like, oh,

66
00:03:19,520 --> 00:03:22,280
Speaker 2: we're gonna need to install this open source natural language

67
00:03:22,320 --> 00:03:25,160
Speaker 2: processing library. It just does it automatically. Instead of me

68
00:03:25,240 --> 00:03:27,240
Speaker 2: trying like figure out like what are the right keystrokes

69
00:03:27,240 --> 00:03:29,480
Speaker 2: to pull that in or why is this not going

70
00:03:29,480 --> 00:03:32,640
Speaker 2: into the right file? Folder or whatever, and so like

71
00:03:32,639 --> 00:03:35,440
Speaker 2: what like Cowork, it's like all like all of these

72
00:03:35,480 --> 00:03:38,200
Speaker 2: sort of like little frictions, like these technical things like

73
00:03:38,400 --> 00:03:43,280
Speaker 2: command line user very rapidly are like dissipating and so

74
00:03:43,320 --> 00:03:45,120
Speaker 2: that like then you have something like Cowork where it's

75
00:03:45,200 --> 00:03:47,320
Speaker 2: just like they know they're taking care of that, and

76
00:03:47,440 --> 00:03:50,120
Speaker 2: so you get this like user interface that's just like

77
00:03:50,280 --> 00:03:53,960
Speaker 2: it's just getting easier and friendlier. There's almost no technical

78
00:03:54,000 --> 00:03:55,240
Speaker 2: frictions at all anymore.

79
00:03:55,280 --> 00:03:58,040
Speaker 3: Also, it feels very iterative, like the code is improving

80
00:03:58,120 --> 00:04:00,400
Speaker 3: upon itself at this point I think was one of

81
00:04:00,440 --> 00:04:01,680
Speaker 3: Claude's main selling points.

82
00:04:01,880 --> 00:04:04,720
Speaker 2: Well, this is like you've seen like people talk about like,

83
00:04:04,760 --> 00:04:07,720
Speaker 2: oh is AGI here? And this is like part of

84
00:04:07,760 --> 00:04:10,400
Speaker 2: the debate because the premp one of the ideas I

85
00:04:10,400 --> 00:04:12,800
Speaker 2: guess behind AGI is like, well what happens when you

86
00:04:12,840 --> 00:04:15,680
Speaker 2: have software that can train itself and so forth? And

87
00:04:15,760 --> 00:04:17,960
Speaker 2: I don't really know if I buy that, but you

88
00:04:18,040 --> 00:04:20,520
Speaker 2: do just see like how fast the iteration cycles are,

89
00:04:20,880 --> 00:04:23,719
Speaker 2: and I think we want to get into this. In part,

90
00:04:24,000 --> 00:04:26,480
Speaker 2: they're fast because a bunch of people are suddenly getting excited.

91
00:04:26,480 --> 00:04:29,200
Speaker 2: So then the human provides this sort of like we're

92
00:04:29,240 --> 00:04:31,400
Speaker 2: sowing the seeds of our own demise because we're so

93
00:04:31,560 --> 00:04:35,800
Speaker 2: enthusiastically participating in the evolution. But I just like it's

94
00:04:35,800 --> 00:04:38,640
Speaker 2: suddenly clear, like, oh, this is going to change I

95
00:04:38,680 --> 00:04:42,000
Speaker 2: think computing. And the other thing is the code works,

96
00:04:42,240 --> 00:04:44,840
Speaker 2: like it creates code that like this is like there's

97
00:04:44,839 --> 00:04:46,520
Speaker 2: no bugs. You know it works.

98
00:04:47,600 --> 00:04:50,360
Speaker 3: Did you see speaking of automating yourself, you see there

99
00:04:50,400 --> 00:04:52,320
Speaker 3: was a post on Reddit from a lawyer who said

100
00:04:52,360 --> 00:04:55,760
Speaker 3: he's basically used claud code to automate like his entire job,

101
00:04:55,800 --> 00:04:56,960
Speaker 3: and he hasn't told anyone.

102
00:04:57,440 --> 00:04:59,800
Speaker 2: I'm not exactly surprised because the other thing that I

103
00:05:00,000 --> 00:05:04,080
Speaker 2: experimented with is and I haven't one hundred percent verified this,

104
00:05:04,160 --> 00:05:07,120
Speaker 2: but on jobs Day last week, I downloaded the full

105
00:05:07,160 --> 00:05:09,960
Speaker 2: pdf and I just typed into the cloud code like

106
00:05:10,480 --> 00:05:13,560
Speaker 2: find the most interesting details and make some charts based

107
00:05:13,600 --> 00:05:15,440
Speaker 2: on and it did it in like a couple of minutes.

108
00:05:15,480 --> 00:05:17,400
Speaker 2: I have no like ability to like I've never like

109
00:05:17,440 --> 00:05:21,599
Speaker 2: built charts myself or whatever, like designing or whatever. And

110
00:05:21,680 --> 00:05:24,320
Speaker 2: I didn't totally confirm yet that the data was all correct,

111
00:05:24,320 --> 00:05:26,559
Speaker 2: but I'm pretty sure it was because everything I spot checked,

112
00:05:26,680 --> 00:05:28,440
Speaker 2: so I didn't just that crucial I didn't tell you

113
00:05:28,600 --> 00:05:30,200
Speaker 2: I know I didn't. That's why I didn't want to like, oh,

114
00:05:30,320 --> 00:05:32,960
Speaker 2: like here's what, here's the today's jobs reporting charts.

115
00:05:33,080 --> 00:05:33,400
Speaker 5: But I did.

116
00:05:34,200 --> 00:05:37,320
Speaker 3: So what application did it actually build it in the charts?

117
00:05:37,520 --> 00:05:39,200
Speaker 2: I don't know. I just had a file like that's

118
00:05:39,200 --> 00:05:41,120
Speaker 2: the thing. I had a file on my computer at

119
00:05:41,120 --> 00:05:43,920
Speaker 2: that point. What kind of file, like a P and

120
00:05:44,000 --> 00:05:46,479
Speaker 2: G file, like an image file, that's the crazy thing.

121
00:05:46,800 --> 00:05:48,880
Speaker 2: I don't know. And so there was just this image

122
00:05:48,920 --> 00:05:50,760
Speaker 2: that had a bunch of charts, and my spot checks

123
00:05:50,800 --> 00:05:54,360
Speaker 2: did suggest like I didn't see anything off. And people

124
00:05:54,440 --> 00:05:56,839
Speaker 2: get paid money to like build that kind of stuff

125
00:05:56,839 --> 00:05:58,360
Speaker 2: for like analysts and stuff like that.

126
00:05:58,480 --> 00:06:01,160
Speaker 3: And right, so this is the other big question. If

127
00:06:01,200 --> 00:06:05,120
Speaker 3: everyone can build their own software, what actually happens to software?

128
00:06:05,160 --> 00:06:07,640
Speaker 3: And I was reading something I forget who it was by,

129
00:06:07,960 --> 00:06:10,800
Speaker 3: but someone used claud code to create they wanted a

130
00:06:10,839 --> 00:06:13,719
Speaker 3: website that would basically make the money for doing nothing,

131
00:06:13,839 --> 00:06:15,880
Speaker 3: and that was the prompt And did.

132
00:06:15,720 --> 00:06:16,160
Speaker 2: They do it?

133
00:06:16,440 --> 00:06:16,760
Speaker 6: Yeah.

134
00:06:16,839 --> 00:06:20,240
Speaker 3: So the idea that the model came up with was

135
00:06:20,680 --> 00:06:23,920
Speaker 3: you can sell prompts, packages of good prompts and sell

136
00:06:23,960 --> 00:06:26,560
Speaker 3: them for like forty bucks and you'll make tons of money.

137
00:06:27,160 --> 00:06:29,680
Speaker 3: And I was thinking about that like, Okay, it's possible

138
00:06:29,760 --> 00:06:32,760
Speaker 3: to make money that way, but also why wouldn't I

139
00:06:32,839 --> 00:06:35,400
Speaker 3: just use claud code to do the same thing.

140
00:06:36,400 --> 00:06:38,800
Speaker 2: There are many big questions that we use an economy

141
00:06:38,800 --> 00:06:40,200
Speaker 2: are going to have to think about, and I think

142
00:06:40,279 --> 00:06:42,400
Speaker 2: my main takeaway is we're gonna have to think about

143
00:06:42,400 --> 00:06:44,560
Speaker 2: these sooner red than later. But what is clud code?

144
00:06:44,560 --> 00:06:46,640
Speaker 2: Why is everyone so hyped about it? Like what is

145
00:06:46,680 --> 00:06:50,240
Speaker 2: it about this particular piece of software versus what exists

146
00:06:50,240 --> 00:06:52,800
Speaker 2: from open Aye and Gemini and all this stuff, Like

147
00:06:52,800 --> 00:06:55,719
Speaker 2: why is this captured everyone's imagination? We really do have

148
00:06:55,720 --> 00:06:59,200
Speaker 2: the perfect guess because it's someone who, unlike me, has

149
00:06:59,240 --> 00:07:01,440
Speaker 2: been getting their hands and the stuff for longer. One

150
00:07:01,480 --> 00:07:03,800
Speaker 2: of the few people that I know who is into

151
00:07:04,000 --> 00:07:08,240
Speaker 2: lollms before chat GPT existed and was actually using them

152
00:07:08,320 --> 00:07:11,520
Speaker 2: via the API, and was actually talking about their technical

153
00:07:11,560 --> 00:07:14,880
Speaker 2: capacity to do things like coding even before November of

154
00:07:14,920 --> 00:07:17,640
Speaker 2: twenty twenty two. So truly the perfect guests We're going

155
00:07:17,720 --> 00:07:19,280
Speaker 2: to be speaking with Noah Bryer. He is the co

156
00:07:19,360 --> 00:07:22,720
Speaker 2: founder of Alefic, which is a consultancy that helps big

157
00:07:22,760 --> 00:07:25,920
Speaker 2: companies deal with AI stuff. So, uh, Noah, thank you

158
00:07:25,920 --> 00:07:27,160
Speaker 2: so much for coming on odd lots.

159
00:07:27,240 --> 00:07:28,200
Speaker 5: Thank you for having me.

160
00:07:28,680 --> 00:07:31,160
Speaker 2: What is eleven? What's the deal? How are you like

161
00:07:31,320 --> 00:07:34,960
Speaker 2: using llms before chat GPT existed? I don't know. I

162
00:07:35,000 --> 00:07:36,280
Speaker 2: know very few people who were doing that.

163
00:07:36,880 --> 00:07:39,679
Speaker 6: I had the good fortune of shutting down a startup

164
00:07:39,720 --> 00:07:41,200
Speaker 6: in twenty twenty two, and so I had a lot

165
00:07:41,240 --> 00:07:42,280
Speaker 6: of free time on my hands.

166
00:07:42,400 --> 00:07:43,720
Speaker 2: And then how are you using it though?

167
00:07:43,760 --> 00:07:43,840
Speaker 3: Like?

168
00:07:43,880 --> 00:07:45,720
Speaker 2: How did you look at your like how did you

169
00:07:45,880 --> 00:07:48,080
Speaker 2: aware that there was this thing that could be of

170
00:07:48,160 --> 00:07:49,000
Speaker 2: potential used to.

171
00:07:49,080 --> 00:07:51,000
Speaker 6: What was So my very first thing I was doing

172
00:07:51,120 --> 00:07:53,880
Speaker 6: was using gethub Copilot, which at the time was built

173
00:07:53,920 --> 00:07:57,240
Speaker 6: into VS code, and it was autocomplete inside VS code.

174
00:07:57,240 --> 00:07:59,800
Speaker 6: So it was a nice and pretty immediately realized that

175
00:07:59,800 --> 00:08:03,920
Speaker 6: there were certain coding tasks that it could just handle completely.

176
00:08:04,960 --> 00:08:08,040
Speaker 6: Anything that was very pattern based. So if you write code,

177
00:08:08,080 --> 00:08:10,120
Speaker 6: you write a lot of tests. If you write tests,

178
00:08:10,200 --> 00:08:12,880
Speaker 6: every test kind of follows the same pattern, and you

179
00:08:12,920 --> 00:08:14,520
Speaker 6: want it to follow the same pattern. You're looking for

180
00:08:14,560 --> 00:08:17,160
Speaker 6: that structure, and over time, because it was looking at

181
00:08:17,200 --> 00:08:19,200
Speaker 6: your code base, it was able to basically autocomplete it.

182
00:08:20,000 --> 00:08:24,240
Speaker 6: I also started playing with the GPT three API, which

183
00:08:24,240 --> 00:08:25,880
Speaker 6: had come out. I think that came out in November

184
00:08:25,920 --> 00:08:28,240
Speaker 6: of twenty twenty one, and that was the first time

185
00:08:28,280 --> 00:08:31,640
Speaker 6: it was publicly available to everybody, and they had a

186
00:08:31,720 --> 00:08:35,040
Speaker 6: large language, Boddel as we know it today, available to them.

187
00:08:35,080 --> 00:08:38,000
Speaker 6: So I was just testing and building things, and I

188
00:08:38,040 --> 00:08:41,840
Speaker 6: pretty immediately realized the very first thing I did where

189
00:08:42,000 --> 00:08:44,720
Speaker 6: it just blew my mind was I built a web scraper.

190
00:08:44,800 --> 00:08:47,240
Speaker 6: So I was just trying to pull pricing data from

191
00:08:47,240 --> 00:08:49,520
Speaker 6: a website. And I've done a lot of this in

192
00:08:49,559 --> 00:08:52,199
Speaker 6: my career. It's maybe the most annoying task you have

193
00:08:52,320 --> 00:08:54,360
Speaker 6: to do in all of coding, because HTML is the

194
00:08:54,360 --> 00:08:58,920
Speaker 6: most miserable language to have to parse. And I just

195
00:08:59,000 --> 00:09:00,600
Speaker 6: had this thing where I took the page, I took

196
00:09:00,600 --> 00:09:02,200
Speaker 6: the content, I took the text, and I gave it

197
00:09:02,240 --> 00:09:04,040
Speaker 6: to the AI and I asked it to give me

198
00:09:04,080 --> 00:09:06,600
Speaker 6: back the pricing table and give me back the pricing table.

199
00:09:06,920 --> 00:09:09,480
Speaker 6: And I just thought, I'll never do it the other

200
00:09:09,520 --> 00:09:09,880
Speaker 6: way again.

201
00:09:09,920 --> 00:09:10,320
Speaker 5: That's it.

202
00:09:10,400 --> 00:09:16,040
Speaker 3: Yeah, that HTML mentioned just brought up like memories of

203
00:09:16,080 --> 00:09:19,080
Speaker 3: me in like the mid nineties on HTML goodies. Do

204
00:09:19,080 --> 00:09:21,840
Speaker 3: you remember that side? Yeah? I wonder if it's still

205
00:09:22,200 --> 00:09:26,640
Speaker 3: Is it still up? That would be wild? Does cloud code?

206
00:09:26,679 --> 00:09:29,520
Speaker 3: Does that count as AGI? This seems to be the debate,

207
00:09:29,600 --> 00:09:30,560
Speaker 3: right is it AGI?

208
00:09:31,160 --> 00:09:33,720
Speaker 6: I try not to wait into what's AGI and what's not.

209
00:09:33,880 --> 00:09:36,360
Speaker 6: I think my guess on AGI, for what it's worth

210
00:09:36,480 --> 00:09:39,240
Speaker 6: is that it's probably going to be a conversation like

211
00:09:39,280 --> 00:09:41,640
Speaker 6: the Turing Test, where everybody thought it was really really

212
00:09:41,679 --> 00:09:43,640
Speaker 6: important for a really long time. We thought the Turing

213
00:09:43,640 --> 00:09:46,800
Speaker 6: Test was the biggest thing for seventy years or whatever,

214
00:09:46,880 --> 00:09:50,280
Speaker 6: and then CHATCHYBT very clearly passed the Turing test, and

215
00:09:50,280 --> 00:09:53,360
Speaker 6: now everybody pretends, like we it's not just that they forgot,

216
00:09:53,400 --> 00:09:56,920
Speaker 6: they pretend that it never mattered. Oh and so I

217
00:09:56,960 --> 00:09:59,440
Speaker 6: am kind of guessing that that's going to be what

218
00:09:59,480 --> 00:10:00,960
Speaker 6: the conversation. It's like, it's just going to be a

219
00:10:00,960 --> 00:10:04,120
Speaker 6: sort of forever moving goalpost, because it turns out that

220
00:10:04,160 --> 00:10:07,000
Speaker 6: the idea we had for what general intelligence looks like

221
00:10:07,720 --> 00:10:11,200
Speaker 6: is not quite that. But I also think, you know,

222
00:10:11,360 --> 00:10:14,600
Speaker 6: the computer scientists and the sort of serious AI researchers

223
00:10:14,800 --> 00:10:17,600
Speaker 6: would say that much of what's going on inside claud

224
00:10:17,640 --> 00:10:20,720
Speaker 6: code is not the model itself. It's the model paired

225
00:10:20,760 --> 00:10:23,720
Speaker 6: with a human and I think that is a pretty

226
00:10:23,760 --> 00:10:24,480
Speaker 6: important distinction.

227
00:10:24,559 --> 00:10:25,760
Speaker 5: But I don't know about AGI.

228
00:10:41,600 --> 00:10:46,640
Speaker 2: Well, okay, so you were using GPT to code prior

229
00:10:46,720 --> 00:10:49,920
Speaker 2: to the release of jag GUPT. So therefore, coding models

230
00:10:49,920 --> 00:10:53,120
Speaker 2: have been around a long time. So what is for

231
00:10:53,200 --> 00:10:56,120
Speaker 2: those who haven't played around with it? What is claude code?

232
00:10:56,160 --> 00:10:58,760
Speaker 2: Because again, coding models have been around for a long time.

233
00:11:00,040 --> 00:11:02,599
Speaker 2: Well maybe have heard of cursor, copilot or some of

234
00:11:02,679 --> 00:11:05,760
Speaker 2: these other harnesses, et cetera. What is cloud code?

235
00:11:05,840 --> 00:11:08,920
Speaker 5: So if we back up first and we go to copilot.

236
00:11:09,160 --> 00:11:12,560
Speaker 6: So Copilot was the first sort of commercial application of

237
00:11:12,559 --> 00:11:17,120
Speaker 6: a largelanguage model by most accounts, And what Copilot did

238
00:11:17,160 --> 00:11:19,360
Speaker 6: in its initial instantiation was just.

239
00:11:19,440 --> 00:11:20,600
Speaker 2: Auto Microsoft product.

240
00:11:20,640 --> 00:11:23,400
Speaker 6: It's a Microsoft product, so Microsoft doones, get hub, getthub,

241
00:11:23,520 --> 00:11:28,120
Speaker 6: develop copilot. It was Microsoft had the partnership with open Ai,

242
00:11:28,280 --> 00:11:30,120
Speaker 6: and so they built it in and what it was

243
00:11:30,160 --> 00:11:33,280
Speaker 6: doing was doing autocomplete. So if you're writing code, a

244
00:11:33,320 --> 00:11:35,679
Speaker 6: lot of writing code is boiler plate or trying to

245
00:11:35,760 --> 00:11:38,320
Speaker 6: remember the name of a function. And you know, the

246
00:11:38,400 --> 00:11:41,400
Speaker 6: reason stack overflow existed was because you can never remember

247
00:11:41,480 --> 00:11:45,000
Speaker 6: the exact name of that function or the exact rejects

248
00:11:45,080 --> 00:11:47,320
Speaker 6: that you need to use in order to find and

249
00:11:47,360 --> 00:11:50,520
Speaker 6: replace something, and so you would go search for it.

250
00:11:50,559 --> 00:11:53,160
Speaker 6: And they realized that you could just build that into

251
00:11:53,200 --> 00:11:56,959
Speaker 6: the ID your code editor and and have it autocomplete

252
00:11:57,000 --> 00:12:02,640
Speaker 6: for you, and it was pretty amazing. Then, uh, Chatchipt

253
00:12:02,800 --> 00:12:04,640
Speaker 6: came out, and even before that, I had built a

254
00:12:04,679 --> 00:12:07,199
Speaker 6: simple chatbot for myself because I realized that, hey, I

255
00:12:07,200 --> 00:12:09,360
Speaker 6: could just ask this and instead of going and searching

256
00:12:09,400 --> 00:12:13,559
Speaker 6: stack overflow. It was totally capable of answering code questions,

257
00:12:13,559 --> 00:12:15,800
Speaker 6: and it was capable of writing rejects or doing these

258
00:12:15,800 --> 00:12:18,480
Speaker 6: things and didn't make mistakes. Yes, but like there's famous

259
00:12:18,520 --> 00:12:22,120
Speaker 6: mistakes on stack overflow of incorrect rejects that now exists

260
00:12:22,120 --> 00:12:25,520
Speaker 6: in every code base in the world, and so you know,

261
00:12:25,559 --> 00:12:26,920
Speaker 6: I think there were a lot of us just kind

262
00:12:26,920 --> 00:12:29,480
Speaker 6: of playing with these things and realizing they were a

263
00:12:29,559 --> 00:12:33,000
Speaker 6: huge boon. And so I think, really the next step

264
00:12:33,040 --> 00:12:35,080
Speaker 6: is Cursor comes out. And the thing Curser realized that

265
00:12:35,240 --> 00:12:37,719
Speaker 6: co Pilot didn't was that it wasn't good enough to

266
00:12:37,720 --> 00:12:40,160
Speaker 6: have autocomplete. You also needed the Q and A because

267
00:12:40,160 --> 00:12:42,640
Speaker 6: you have these things that you can't just auto complete.

268
00:12:42,640 --> 00:12:43,880
Speaker 6: You want to be able to ask the question and

269
00:12:43,920 --> 00:12:46,079
Speaker 6: answer it. And then Chatchipt came out and everybody was

270
00:12:46,080 --> 00:12:50,280
Speaker 6: switching between ID and then I think, really the next

271
00:12:50,280 --> 00:12:52,840
Speaker 6: big piece is that claud code came out. And what

272
00:12:52,920 --> 00:12:56,040
Speaker 6: claud Code did that was so remarkable was they took

273
00:12:56,080 --> 00:12:59,079
Speaker 6: the same set of models really and they took them

274
00:12:59,080 --> 00:13:01,600
Speaker 6: out of the chatbot and they really just gave it

275
00:13:01,679 --> 00:13:06,880
Speaker 6: some very basic functionality to operate within your machine.

276
00:13:06,960 --> 00:13:07,160
Speaker 5: Right.

277
00:13:07,200 --> 00:13:08,760
Speaker 6: And so you know, if you really look at kind

278
00:13:08,760 --> 00:13:11,079
Speaker 6: of what exists within claud code, you're calling out to

279
00:13:11,120 --> 00:13:13,760
Speaker 6: a model, and they gave it capability around sort of

280
00:13:13,880 --> 00:13:16,000
Speaker 6: two big things. One is you can read and write

281
00:13:16,000 --> 00:13:19,320
Speaker 6: files on your computer, and then two is that you

282
00:13:19,400 --> 00:13:24,040
Speaker 6: can operate Unix the base commands, the bashed commands that

283
00:13:24,200 --> 00:13:29,240
Speaker 6: exist in your environment. And again, because these models were

284
00:13:29,280 --> 00:13:31,600
Speaker 6: trained on the Internet, and there's no greater source of

285
00:13:31,600 --> 00:13:33,840
Speaker 6: information on the Internet than how to make the Internet,

286
00:13:34,080 --> 00:13:38,040
Speaker 6: they know how to use Unix commands incredibly well, right,

287
00:13:38,040 --> 00:13:40,680
Speaker 6: because Unix has existed for whatever it is, sixty years,

288
00:13:40,679 --> 00:13:43,120
Speaker 6: and the way these commands were designed, they're all designed

289
00:13:43,120 --> 00:13:45,920
Speaker 6: to be very very simple. There's a find command, and

290
00:13:46,000 --> 00:13:47,640
Speaker 6: you know, there's some thing called re rep and it

291
00:13:47,679 --> 00:13:50,520
Speaker 6: can search to a code base. And Unix has this

292
00:13:50,600 --> 00:13:53,240
Speaker 6: sort of beautiful way of tying one command to another,

293
00:13:53,280 --> 00:13:55,120
Speaker 6: so you can take the output of one command and

294
00:13:55,120 --> 00:13:58,120
Speaker 6: send it to another. And they kind of just gave

295
00:13:58,200 --> 00:14:00,839
Speaker 6: the model access to these two or three very simple things,

296
00:14:01,400 --> 00:14:03,800
Speaker 6: and it kind of turned out that it unlocked a

297
00:14:03,840 --> 00:14:06,720
Speaker 6: whole bunch of functionality. I don't think even the people

298
00:14:06,800 --> 00:14:10,320
Speaker 6: who built it fully realized. Like one example that I

299
00:14:10,320 --> 00:14:13,199
Speaker 6: think about a lot is just the challenge you have

300
00:14:13,360 --> 00:14:15,040
Speaker 6: with all of these.

301
00:14:14,880 --> 00:14:16,880
Speaker 5: AI models is that they're stateless.

302
00:14:17,200 --> 00:14:20,920
Speaker 6: So every time you talk to chat GPT, it's sending

303
00:14:20,960 --> 00:14:23,960
Speaker 6: your entire conversation history back to chat GPT because it

304
00:14:23,960 --> 00:14:28,600
Speaker 6: has no saved history of that chat, right, And that's fine,

305
00:14:28,640 --> 00:14:30,560
Speaker 6: it's the way it works. It's just fact. But it

306
00:14:30,680 --> 00:14:32,960
Speaker 6: means that you know, it forgets things. It doesn't no

307
00:14:33,040 --> 00:14:37,160
Speaker 6: conversation to conversation. And one very easy way to save

308
00:14:37,280 --> 00:14:41,080
Speaker 6: your state is just write it to a file. And

309
00:14:41,160 --> 00:14:44,560
Speaker 6: so you give it right access and it can create files.

310
00:14:44,560 --> 00:14:47,440
Speaker 6: And now all of a sudden you've overcome this, like

311
00:14:47,680 --> 00:14:51,520
Speaker 6: probably the single biggest challenge that exists inside these large

312
00:14:51,560 --> 00:14:54,600
Speaker 6: language models, which is that they're fundamentally stateless.

313
00:14:55,000 --> 00:14:58,400
Speaker 3: So Claude writes itself little like memory notes right to

314
00:14:58,920 --> 00:15:01,960
Speaker 3: remember the entire ca context of the conversation, and that's

315
00:15:01,960 --> 00:15:03,200
Speaker 3: how it solved that problem.

316
00:15:03,480 --> 00:15:05,760
Speaker 6: No, so there's sort of two things going on in

317
00:15:05,840 --> 00:15:06,560
Speaker 6: claude code.

318
00:15:06,600 --> 00:15:07,280
Speaker 5: Beneath the hood.

319
00:15:07,520 --> 00:15:10,080
Speaker 6: There's one thing that works exactly like chetchipt or any

320
00:15:10,080 --> 00:15:13,240
Speaker 6: of these other ones, which is it's maintaining a conversation history.

321
00:15:13,280 --> 00:15:18,840
Speaker 6: So every message you send it and every action it takes,

322
00:15:18,880 --> 00:15:22,200
Speaker 6: it's recording to a log, which is just one big file.

323
00:15:22,400 --> 00:15:25,240
Speaker 6: That's really no different than what chet ChiPT can do.

324
00:15:25,520 --> 00:15:28,160
Speaker 6: Where it gets really interesting though, is it can also

325
00:15:28,440 --> 00:15:32,600
Speaker 6: write files that it can then read, so whereas that

326
00:15:32,640 --> 00:15:36,280
Speaker 6: conversation history is all saved off, and eventually that conversation

327
00:15:36,320 --> 00:15:37,600
Speaker 6: gets too long and it needs to do a thing

328
00:15:37,640 --> 00:15:41,120
Speaker 6: called compaction. And when it compacts it, it tries to

329
00:15:41,160 --> 00:15:44,560
Speaker 6: sort of just remember the bits because there the total.

330
00:15:44,320 --> 00:15:44,880
Speaker 5: Window is.

331
00:15:47,440 --> 00:15:50,160
Speaker 6: Large. But I mean it's like one hundred thousand tokens

332
00:15:50,840 --> 00:15:51,160
Speaker 6: I mean.

333
00:15:51,040 --> 00:15:54,320
Speaker 3: By memory notes, right. It compacts the information into the

334
00:15:54,400 --> 00:15:56,400
Speaker 3: important stuff, then retreat.

335
00:15:56,520 --> 00:15:56,960
Speaker 5: It does that.

336
00:15:57,720 --> 00:15:59,800
Speaker 6: It only does that at the end, like once it

337
00:15:59,840 --> 00:16:02,680
Speaker 6: runs out of space, once it runs out of context window.

338
00:16:02,880 --> 00:16:04,600
Speaker 6: So it has two hundred thousand tokens I think, and

339
00:16:04,600 --> 00:16:06,840
Speaker 6: two hundred thousand tokens in rough terms is probably one

340
00:16:06,880 --> 00:16:10,920
Speaker 6: hundred and fifty thousand words. It says, okay, it's time

341
00:16:10,960 --> 00:16:13,600
Speaker 6: for me to compact all of this stuff, and so

342
00:16:13,640 --> 00:16:16,280
Speaker 6: it still saves your whole history.

343
00:16:15,960 --> 00:16:16,600
Speaker 5: On your computer.

344
00:16:16,640 --> 00:16:19,320
Speaker 6: You still have the entire message, but for that session,

345
00:16:19,720 --> 00:16:23,280
Speaker 6: it just compacts it down to this, you know, maybe

346
00:16:23,360 --> 00:16:27,920
Speaker 6: twenty five thousand token memory of what it was.

347
00:16:29,400 --> 00:16:32,240
Speaker 2: And is this like something that was not obvious before

348
00:16:32,360 --> 00:16:36,400
Speaker 2: as a solution like this compaction? How important is it

349
00:16:36,920 --> 00:16:40,320
Speaker 2: for this being? Like okay? As a human I can

350
00:16:40,320 --> 00:16:42,600
Speaker 2: work on this on a project for a long time,

351
00:16:42,680 --> 00:16:44,920
Speaker 2: like how much of an an unlock was there?

352
00:16:45,120 --> 00:16:47,640
Speaker 6: I'm not sure compaction was the okay walk. I think

353
00:16:47,680 --> 00:16:52,280
Speaker 6: the compaction functionality is helpful. The way chat GPT does it,

354
00:16:52,320 --> 00:16:54,320
Speaker 6: for what it's worth is they don't do compaction. They

355
00:16:54,360 --> 00:16:57,600
Speaker 6: just forget your messages eventually. So if you're in one chat,

356
00:16:58,000 --> 00:17:01,000
Speaker 6: eventually your oldest message is going to fall off the back.

357
00:17:02,200 --> 00:17:06,680
Speaker 6: For coding, that's probably less helpful, but there are trade offs.

358
00:17:06,720 --> 00:17:10,399
Speaker 6: I both techniques work. I think fundamentally, the thing that

359
00:17:10,520 --> 00:17:14,440
Speaker 6: is special about cloud code is not the compaction. It's

360
00:17:14,680 --> 00:17:17,199
Speaker 6: the ability to write and read files on your computer,

361
00:17:17,560 --> 00:17:19,520
Speaker 6: which means you can always write off memories.

362
00:17:19,960 --> 00:17:21,600
Speaker 2: And then what does that mean? Write off memory?

363
00:17:21,760 --> 00:17:24,359
Speaker 6: So you could say, hey, it's really important that I

364
00:17:24,400 --> 00:17:26,800
Speaker 6: remember this thing for future sessions. I want to always

365
00:17:26,880 --> 00:17:30,040
Speaker 6: work this way. So in a code base of mine,

366
00:17:30,720 --> 00:17:34,119
Speaker 6: I have a set of documentation that explains how I

367
00:17:34,200 --> 00:17:37,200
Speaker 6: like to do things. And cloud code makes a mistake,

368
00:17:37,240 --> 00:17:39,320
Speaker 6: and so the next time I can write a memory,

369
00:17:39,680 --> 00:17:42,159
Speaker 6: essentially it's written as a thing they call a skill,

370
00:17:43,000 --> 00:17:44,640
Speaker 6: and you can write it off and you say, hey,

371
00:17:44,760 --> 00:17:46,840
Speaker 6: whenever you run into this, I want you to operate

372
00:17:46,880 --> 00:17:50,800
Speaker 6: in this kind of way. And that existing across every

373
00:17:50,840 --> 00:17:53,960
Speaker 6: session is really a thing you can only do when

374
00:17:53,960 --> 00:17:56,600
Speaker 6: you can store it as a file. It's a thing

375
00:17:56,640 --> 00:17:58,760
Speaker 6: you can't do in quite the same way when you're

376
00:17:58,800 --> 00:18:00,840
Speaker 6: operating in this environment or it's just going back and

377
00:18:00,880 --> 00:18:03,840
Speaker 6: forth to the API. So this access to the file

378
00:18:03,920 --> 00:18:06,000
Speaker 6: system is one really big piece. And then the second

379
00:18:06,040 --> 00:18:09,640
Speaker 6: is is just the Unix commands, I mean computers. Every

380
00:18:09,640 --> 00:18:13,359
Speaker 6: computer program lives on top of these sort of baseline

381
00:18:14,400 --> 00:18:18,720
Speaker 6: functions and the way that the designers of Unix built

382
00:18:18,760 --> 00:18:22,600
Speaker 6: them is really elegant, and they're very small. They all

383
00:18:22,640 --> 00:18:27,480
Speaker 6: do one thing, and they're all composable, and in coding terms,

384
00:18:27,480 --> 00:18:30,359
Speaker 6: composable means they can be chained together, right, and so

385
00:18:31,240 --> 00:18:34,920
Speaker 6: you can say, hey, look for files that mentioned this word,

386
00:18:35,119 --> 00:18:37,560
Speaker 6: and then from those files, I want you to take

387
00:18:37,600 --> 00:18:40,280
Speaker 6: this second action, and then from the output of that action,

388
00:18:40,320 --> 00:18:41,400
Speaker 6: I want you to take a third action.

389
00:18:41,440 --> 00:18:43,119
Speaker 5: And that's just built into Unix.

390
00:18:43,160 --> 00:18:45,760
Speaker 6: You literally just put a little pipe in between and

391
00:18:45,840 --> 00:18:48,320
Speaker 6: you just pipe them from one to another and that's it.

392
00:18:48,400 --> 00:18:51,199
Speaker 6: And so you give it access to write these commands,

393
00:18:51,200 --> 00:18:52,960
Speaker 6: and all of a sudden, it gets these sort of

394
00:18:52,960 --> 00:18:55,600
Speaker 6: second and third order effects that are just incredibly powerful

395
00:18:55,600 --> 00:18:57,720
Speaker 6: and built over a really long time.

396
00:18:58,320 --> 00:19:01,159
Speaker 3: So how much of claud code, the way it's different

397
00:19:01,200 --> 00:19:04,400
Speaker 3: to other models, how much of that was overcoming technological

398
00:19:04,480 --> 00:19:08,760
Speaker 3: challenges versus just having a good idea, because hearing you

399
00:19:08,840 --> 00:19:12,280
Speaker 3: describe it, I mean, giving access to a computer seems

400
00:19:12,359 --> 00:19:15,760
Speaker 3: like kind of obvious, like let's just do that.

401
00:19:17,640 --> 00:19:19,480
Speaker 6: I don't have a good answer to that. I think

402
00:19:19,560 --> 00:19:22,000
Speaker 6: that it was kind of just a good idea. Yeah,

403
00:19:22,200 --> 00:19:25,040
Speaker 6: I think they did some patterns really well. They're clearly

404
00:19:25,119 --> 00:19:30,119
Speaker 6: incredibly talented, not just engineers, but kind of thinkers about

405
00:19:30,119 --> 00:19:33,680
Speaker 6: how to structure it. Like the primitives inside cloud code

406
00:19:33,720 --> 00:19:36,280
Speaker 6: are just smart. And then the thing that they've done

407
00:19:36,640 --> 00:19:41,880
Speaker 6: and Boris Cherney, who's the lead developer on cloud coded anthropic,

408
00:19:41,920 --> 00:19:44,920
Speaker 6: he talks about light and demand a lot right, and

409
00:19:45,000 --> 00:19:47,280
Speaker 6: Layton Demand is basically just, hey, look at the ways

410
00:19:47,320 --> 00:19:49,520
Speaker 6: people are using these systems and then figure out ways

411
00:19:49,520 --> 00:19:52,080
Speaker 6: to make that a part of the product itself. I

412
00:19:52,119 --> 00:19:54,600
Speaker 6: think what they've done brilliantly, and this is kind of

413
00:19:54,640 --> 00:19:56,800
Speaker 6: easy when you have a community of developers who are

414
00:19:56,840 --> 00:19:58,440
Speaker 6: nerds who want to go talk about all the ways

415
00:19:58,480 --> 00:20:01,399
Speaker 6: that they're using these things. Is they have I am

416
00:20:01,480 --> 00:20:03,679
Speaker 6: amazed at the speed in which you know, I have

417
00:20:03,720 --> 00:20:07,199
Speaker 6: a small community of fifteen CTOs who all use this

418
00:20:07,240 --> 00:20:10,600
Speaker 6: stuff religiously. And you know, when we first started that community,

419
00:20:10,720 --> 00:20:12,879
Speaker 6: it took them a month to I would see it

420
00:20:12,920 --> 00:20:14,480
Speaker 6: in the chat, and then a month later it would

421
00:20:14,480 --> 00:20:17,399
Speaker 6: get built into cloud code, and then increasingly.

422
00:20:16,920 --> 00:20:18,959
Speaker 5: It's like a day later. It feels like they're just

423
00:20:19,160 --> 00:20:19,960
Speaker 5: they're just listening to it.

424
00:20:20,000 --> 00:20:22,040
Speaker 6: But I think they're just not only tapped in, but

425
00:20:22,040 --> 00:20:26,080
Speaker 6: they're really fundamentally, you know, they're dog fooding it. They

426
00:20:26,200 --> 00:20:29,080
Speaker 6: use their own products when you you know, they talk

427
00:20:29,119 --> 00:20:33,480
Speaker 6: about the productivity engineering productivity at Anthropic. You know, despite

428
00:20:33,560 --> 00:20:36,600
Speaker 6: growing at a crazy clip, it continues to go up.

429
00:20:36,640 --> 00:20:40,639
Speaker 6: And you know, anybody who's built had to manage large

430
00:20:40,680 --> 00:20:44,720
Speaker 6: scale pieces of software, large scale code basis knows that's

431
00:20:44,880 --> 00:20:46,119
Speaker 6: not the norm.

432
00:20:46,840 --> 00:20:52,000
Speaker 2: So VS code and cursor, these are IDEs. Cloud code

433
00:20:52,040 --> 00:20:54,520
Speaker 2: is not an ID. What it's called a CLI is.

434
00:20:54,520 --> 00:20:56,160
Speaker 6: A CLI a command line interface.

435
00:20:56,320 --> 00:20:59,600
Speaker 2: Got it and the other labs now they also have CLIs.

436
00:21:00,080 --> 00:21:02,800
Speaker 2: So why are we all talking about claud code and

437
00:21:02,640 --> 00:21:06,520
Speaker 2: I chattypts is called codex. I don't know what Geminis

438
00:21:06,560 --> 00:21:06,960
Speaker 2: is called.

439
00:21:07,080 --> 00:21:08,000
Speaker 5: I think it's just called the gem.

440
00:21:08,760 --> 00:21:11,720
Speaker 2: Why are we all talking about claud code rather than

441
00:21:11,840 --> 00:21:14,480
Speaker 2: the other cla is that kind of have the same thing? Like,

442
00:21:14,520 --> 00:21:15,320
Speaker 2: what is the difference?

443
00:21:15,880 --> 00:21:18,040
Speaker 6: I think, first and foremost they were first, okay, so

444
00:21:18,560 --> 00:21:20,600
Speaker 6: and I think they've they've had a lot more and

445
00:21:20,920 --> 00:21:24,399
Speaker 6: you know, from my very personal opinion, I think they've

446
00:21:24,680 --> 00:21:28,879
Speaker 6: done something smarter and better as far as the permissioning model.

447
00:21:29,000 --> 00:21:30,720
Speaker 6: So you know, one of the really dangerous things is

448
00:21:30,720 --> 00:21:32,360
Speaker 6: you've got the same running on your computer. You don't

449
00:21:32,400 --> 00:21:36,080
Speaker 6: want it to go and delete everything. And they have

450
00:21:36,119 --> 00:21:38,679
Speaker 6: a very fine grain permissioning model where you can say, hey,

451
00:21:38,720 --> 00:21:40,520
Speaker 6: I want to allow this just this one time. I

452
00:21:40,520 --> 00:21:41,560
Speaker 6: want to always allow it.

453
00:21:41,680 --> 00:21:43,840
Speaker 2: You know, I always click always allow. I'm living on

454
00:21:43,840 --> 00:21:44,240
Speaker 2: the edge.

455
00:21:44,480 --> 00:21:47,040
Speaker 6: You can you can next time you run it. You

456
00:21:47,040 --> 00:21:50,080
Speaker 6: can just do a flag that says dangerously skip permissions

457
00:21:50,920 --> 00:21:54,840
Speaker 6: and and it'll just They call it yolo mode. I

458
00:21:54,880 --> 00:21:58,560
Speaker 6: think I think more fundamentally though, if I look at

459
00:21:58,760 --> 00:22:02,480
Speaker 6: at Codex versus Code, I think it's a difference in

460
00:22:02,480 --> 00:22:07,760
Speaker 6: philosophy around what you want AI to do to me. Codex,

461
00:22:08,400 --> 00:22:12,800
Speaker 6: which is excellent, is very focused on building an agent

462
00:22:12,880 --> 00:22:15,199
Speaker 6: that you could just give something to and it'll just

463
00:22:15,280 --> 00:22:17,600
Speaker 6: go do it. So I want to give it that task.

464
00:22:17,680 --> 00:22:19,560
Speaker 6: I don't want to intercede. I don't want to give

465
00:22:19,600 --> 00:22:24,680
Speaker 6: it any more feedback. And claud code is much more

466
00:22:24,720 --> 00:22:28,159
Speaker 6: designed to be kind of a pair programmer. And so

467
00:22:28,320 --> 00:22:31,040
Speaker 6: you know, in engineering, pair programming has existed for a while.

468
00:22:31,280 --> 00:22:34,320
Speaker 6: It's a really weird sort of productivity thing where you

469
00:22:34,400 --> 00:22:36,480
Speaker 6: put two engineers on the same problem and it turns

470
00:22:36,480 --> 00:22:40,120
Speaker 6: out that you can get better code and multiplier Yeah,

471
00:22:40,160 --> 00:22:41,800
Speaker 6: and it sort of makes up for the fact that

472
00:22:41,800 --> 00:22:44,240
Speaker 6: obviously you know you're doubling the staff on it. But

473
00:22:44,359 --> 00:22:47,240
Speaker 6: because of how many fewer bugs because you've both said svies,

474
00:22:47,720 --> 00:22:50,200
Speaker 6: it has seemed to work out for many folks. Most

475
00:22:50,280 --> 00:22:53,840
Speaker 6: companies don't practice it, but I think cloud code fundamentally

476
00:22:53,880 --> 00:22:56,120
Speaker 6: is much more designed in that way.

477
00:22:56,200 --> 00:22:58,280
Speaker 5: It's a pair programmer. It's they. You know.

478
00:22:58,320 --> 00:23:01,560
Speaker 6: Whenever I start a project, I start in plan mode.

479
00:23:01,600 --> 00:23:03,440
Speaker 6: So you start in plan mode. You put together a plan.

480
00:23:03,480 --> 00:23:04,639
Speaker 6: I really, I mean it has spent a lot of

481
00:23:04,720 --> 00:23:07,159
Speaker 6: time in plan mode. You go through, it gives you

482
00:23:07,160 --> 00:23:09,000
Speaker 6: a plan, back, it asks you how you feel you

483
00:23:09,000 --> 00:23:11,320
Speaker 6: can give it a whole bunch of direction, and then

484
00:23:11,359 --> 00:23:13,760
Speaker 6: it's only then that it goes off and it goes

485
00:23:13,800 --> 00:23:16,359
Speaker 6: into it. So you know, we're working together. And I

486
00:23:16,359 --> 00:23:18,560
Speaker 6: actually have a whole system now that I've designed where

487
00:23:19,680 --> 00:23:22,280
Speaker 6: I use a task management system called Linear, So I

488
00:23:22,280 --> 00:23:25,359
Speaker 6: have claud Code write tasks off to Linear, and then

489
00:23:25,640 --> 00:23:29,520
Speaker 6: I've worked with claud Code to write a document that

490
00:23:29,600 --> 00:23:31,919
Speaker 6: helps sort of decide a set of heuristics to decide

491
00:23:31,920 --> 00:23:34,240
Speaker 6: when you should assign it to Codex versus when you

492
00:23:34,280 --> 00:23:36,120
Speaker 6: should give it to claud Code. And so if it's

493
00:23:36,119 --> 00:23:38,240
Speaker 6: tightly defined enough and simple enough, I just send it

494
00:23:38,240 --> 00:23:41,200
Speaker 6: off to Codex and it does it totally independently. And

495
00:23:41,240 --> 00:23:44,280
Speaker 6: then if it's complicated enough that I think it requires

496
00:23:44,320 --> 00:23:47,879
Speaker 6: my time and attention, then it saves it for me

497
00:23:48,920 --> 00:23:52,840
Speaker 6: us to do together, and we'll work on it together.

498
00:23:52,880 --> 00:23:55,520
Speaker 6: And so if it's sort of touching a kind of

499
00:23:55,600 --> 00:23:58,520
Speaker 6: important enough if it's changing some part of the data model.

500
00:23:58,880 --> 00:24:01,440
Speaker 6: There's these other kind of you know, fairly basic extet

501
00:24:01,480 --> 00:24:05,080
Speaker 6: of criteria that I use, But that to me is

502
00:24:05,080 --> 00:24:09,480
Speaker 6: the fundamental distinction. And you know, I find cloud code

503
00:24:09,480 --> 00:24:12,800
Speaker 6: in that way to be just it sort of fits

504
00:24:12,920 --> 00:24:15,359
Speaker 6: what I want to do and how I want to

505
00:24:15,400 --> 00:24:17,080
Speaker 6: work much better.

506
00:24:17,640 --> 00:24:21,320
Speaker 3: Talk a little bit more about how it actually impacts

507
00:24:21,440 --> 00:24:24,960
Speaker 3: the workflow of an engineer, because you know, my impression

508
00:24:25,160 --> 00:24:29,760
Speaker 3: was people can code, right, Like, the coding problem is

509
00:24:29,880 --> 00:24:31,919
Speaker 3: kind of solved at this point, and even if you

510
00:24:32,000 --> 00:24:34,760
Speaker 3: can't code, even if you're not a professional engineer, you

511
00:24:34,800 --> 00:24:38,880
Speaker 3: can hire someone from like India or Indonesia or wherever

512
00:24:39,040 --> 00:24:40,920
Speaker 3: to just write you a code. Maybe it'll take them

513
00:24:40,920 --> 00:24:44,120
Speaker 3: a week instead of like two days with claud code.

514
00:24:44,240 --> 00:24:49,040
Speaker 3: But how much does this actually change the workflow for

515
00:24:49,080 --> 00:24:50,399
Speaker 3: an engineer.

516
00:24:50,880 --> 00:24:53,399
Speaker 6: As completely as it could be changed. I mean, I

517
00:24:53,440 --> 00:24:58,879
Speaker 6: would say that over the last three months, I've written, personally,

518
00:24:59,640 --> 00:25:01,679
Speaker 6: I don't know a few hundred lines of code. Like

519
00:25:01,840 --> 00:25:05,920
Speaker 6: I am mostly a manager of a set of agents

520
00:25:05,960 --> 00:25:08,879
Speaker 6: who are writing code on my behalf. And you know,

521
00:25:08,960 --> 00:25:11,680
Speaker 6: increasingly what I think is interesting. I've been thinking about

522
00:25:11,680 --> 00:25:14,400
Speaker 6: this a bunch lately, is like, in some ways, it's

523
00:25:14,400 --> 00:25:16,479
Speaker 6: just bringing me back to the core challenge that has

524
00:25:16,480 --> 00:25:19,639
Speaker 6: always existed in software development, which is how do you

525
00:25:19,720 --> 00:25:23,320
Speaker 6: manage all large scale software.

526
00:25:23,000 --> 00:25:24,680
Speaker 5: Development project acrossation.

527
00:25:25,200 --> 00:25:28,680
Speaker 6: It has become a coordination problem, and I spend a

528
00:25:28,760 --> 00:25:33,000
Speaker 6: lot of time sort of now designing my CLAUD code

529
00:25:33,040 --> 00:25:35,879
Speaker 6: system to ensure that code goes through all the proper

530
00:25:35,960 --> 00:25:39,280
Speaker 6: spec checks and that it has all these things. The

531
00:25:39,359 --> 00:25:43,040
Speaker 6: other thing that you know makes code a particularly good

532
00:25:43,080 --> 00:25:45,480
Speaker 6: place to do this is that code is verifiable in

533
00:25:45,520 --> 00:25:47,879
Speaker 6: a way that you know most other work is not.

534
00:25:48,440 --> 00:25:51,720
Speaker 6: So you know, with code, you can verify that the

535
00:25:51,800 --> 00:25:53,840
Speaker 6: build works right. So you can say, hey, I want

536
00:25:53,880 --> 00:25:56,040
Speaker 6: to build this, I want to build this package. I

537
00:25:56,040 --> 00:25:57,560
Speaker 6: want to make sure that it's actually going to build

538
00:25:57,560 --> 00:25:59,240
Speaker 6: and that there's going to be no failures. That's a

539
00:25:59,320 --> 00:26:01,520
Speaker 6: very easy check. It's either true or it's not true.

540
00:26:01,880 --> 00:26:06,639
Speaker 6: There's also coders use linting, and so linting is a

541
00:26:06,680 --> 00:26:10,280
Speaker 6: way to kind of look at it's static code analysis,

542
00:26:10,280 --> 00:26:11,960
Speaker 6: so it basically tries to sort.

543
00:26:11,760 --> 00:26:12,680
Speaker 5: Of find.

544
00:26:15,200 --> 00:26:17,119
Speaker 6: Things in your code base that are not going to

545
00:26:17,160 --> 00:26:19,840
Speaker 6: work ahead of time where you can predict that. Obviously,

546
00:26:19,880 --> 00:26:23,840
Speaker 6: you can't predict Alan Turing prove that you can't predict

547
00:26:24,560 --> 00:26:26,280
Speaker 6: with certainty whether code is going to run, but there's

548
00:26:26,280 --> 00:26:28,879
Speaker 6: certain patterns and things that it can find. It's essentially

549
00:26:28,960 --> 00:26:32,159
Speaker 6: does static pattern analysis, and so you know, you have

550
00:26:32,200 --> 00:26:35,000
Speaker 6: it run all these things. But the more kind of

551
00:26:35,000 --> 00:26:38,040
Speaker 6: opinionated you can be about that, and the more steps

552
00:26:38,080 --> 00:26:39,800
Speaker 6: you can have it go through, so I find, you know,

553
00:26:39,880 --> 00:26:42,359
Speaker 6: now I'm kind of the designer, which, honestly, as an

554
00:26:42,440 --> 00:26:45,360
Speaker 6: entrepreneur and as a CEO of company is like that's

555
00:26:45,480 --> 00:26:48,240
Speaker 6: kind of always been my job. Like I've been not

556
00:26:48,440 --> 00:26:50,160
Speaker 6: I have less and less been a person who writes

557
00:26:50,160 --> 00:26:52,240
Speaker 6: code and more and more been a person who designs

558
00:26:52,640 --> 00:26:54,440
Speaker 6: a system in that case of company with a bunch

559
00:26:54,440 --> 00:26:56,840
Speaker 6: of people who write code.

560
00:26:57,440 --> 00:27:00,199
Speaker 2: One of the funny things it seems to me is

561
00:27:00,240 --> 00:27:05,280
Speaker 2: that setting aside Claud code, Claude itself has a reputation

562
00:27:05,560 --> 00:27:08,720
Speaker 2: for it's a nicer chat bot to talk to. People

563
00:27:08,840 --> 00:27:13,280
Speaker 2: find it, and you know, chat gbt seems to really

564
00:27:13,359 --> 00:27:16,040
Speaker 2: be psycho fantic. I still think it's I know it's improved,

565
00:27:16,040 --> 00:27:17,600
Speaker 2: but I actually don't think it's improved or not. I

566
00:27:17,680 --> 00:27:22,719
Speaker 2: still people like the prose style of Claud Claude and

567
00:27:22,760 --> 00:27:26,360
Speaker 2: I'm curious that in the pair trading pair trading, I'm

568
00:27:26,359 --> 00:27:30,199
Speaker 2: thinking about finance, the pair engineering model, whether there is

569
00:27:30,320 --> 00:27:33,159
Speaker 2: also an edge there which is like, here is a

570
00:27:33,280 --> 00:27:36,440
Speaker 2: chat bot that is not annoying to talk to while

571
00:27:36,440 --> 00:27:39,280
Speaker 2: you're iterating, and whether that is like a meaningful distinction

572
00:27:39,720 --> 00:27:43,000
Speaker 2: between you know, coding with codex or whatever.

573
00:27:43,840 --> 00:27:47,800
Speaker 6: Yeah, I I don't know. It still can be very annoying,

574
00:27:47,840 --> 00:27:52,560
Speaker 6: I can tell you. And it'll still sometimes be overly,

575
00:27:54,480 --> 00:27:57,239
Speaker 6: overly effusive with me about a design choice, I mean,

576
00:27:57,680 --> 00:28:00,640
Speaker 6: or sort of notice something which I could live without.

577
00:28:01,000 --> 00:28:04,160
Speaker 2: So I work on this project that's doing this linguistic

578
00:28:04,200 --> 00:28:07,040
Speaker 2: things and I eventually had to say, like, give it

579
00:28:07,040 --> 00:28:09,399
Speaker 2: to me straight, how bad is this? And then so

580
00:28:09,440 --> 00:28:12,000
Speaker 2: I said, I said, Actually, what I said was assume

581
00:28:12,080 --> 00:28:14,359
Speaker 2: for a moment that you are a quantitative linguistics for

582
00:28:14,440 --> 00:28:17,840
Speaker 2: the PhD. Give me your honest assessment of where we

583
00:28:17,880 --> 00:28:21,480
Speaker 2: are with this. And it said like you've developed a

584
00:28:21,560 --> 00:28:24,680
Speaker 2: nice toy and there's no evidence that it actually does.

585
00:28:24,800 --> 00:28:26,639
Speaker 2: And I was like, okay, that's nice to hear. I

586
00:28:26,680 --> 00:28:29,040
Speaker 2: actually like, you know, I appreciate that, and I thank

587
00:28:29,040 --> 00:28:32,280
Speaker 2: you very blunt, not you know, it's still like polite,

588
00:28:32,480 --> 00:28:35,320
Speaker 2: but it was like, this, doesn't you haven't really shown anything.

589
00:28:35,400 --> 00:28:37,560
Speaker 2: You haven't really established at all that your saltware does

590
00:28:37,600 --> 00:28:38,280
Speaker 2: what it claims to.

591
00:28:38,600 --> 00:28:42,560
Speaker 6: Yeah, I think so. I think stylistically I kind of

592
00:28:42,600 --> 00:28:46,920
Speaker 6: personally agree. My theory by the way, on Claude versus

593
00:28:47,920 --> 00:28:50,800
Speaker 6: Opening Eye Chat GPT models is I think Claude is

594
00:28:50,840 --> 00:28:54,000
Speaker 6: actually better at sort of reflecting what you give it.

595
00:28:54,640 --> 00:28:56,240
Speaker 6: And so I think part of why we think it's

596
00:28:56,280 --> 00:28:57,920
Speaker 6: better is it it's better at.

597
00:28:57,800 --> 00:28:59,080
Speaker 5: Pretending it's us.

598
00:28:59,280 --> 00:29:01,520
Speaker 6: Yeah, And so we tend to like that is this

599
00:29:01,600 --> 00:29:05,160
Speaker 6: is purely speculation, but that's always been my theory on so.

600
00:29:05,200 --> 00:29:06,440
Speaker 2: It flatters you in a different way.

601
00:29:06,720 --> 00:29:09,520
Speaker 5: I think it's flattering you in a much more way.

602
00:29:09,600 --> 00:29:09,840
Speaker 1: Yeah.

603
00:29:09,960 --> 00:29:15,360
Speaker 6: Interest But for a long time, just Anthropic has been

604
00:29:15,400 --> 00:29:18,040
Speaker 6: producing the best coding models, you know. I mean there's

605
00:29:18,240 --> 00:29:21,160
Speaker 6: there can be some debate there now, but you know,

606
00:29:21,200 --> 00:29:24,280
Speaker 6: there's a great curse story from Cursor actually where Cursor

607
00:29:24,480 --> 00:29:27,920
Speaker 6: basically wasn't that good, and then SunNet three point five

608
00:29:27,960 --> 00:29:30,760
Speaker 6: came out and all of a sudden, Cursor was amazing,

609
00:29:30,960 --> 00:29:33,760
Speaker 6: and Cursor became a tool that everybody started using. But

610
00:29:33,800 --> 00:29:35,840
Speaker 6: it wasn't until this other model came out and they

611
00:29:35,880 --> 00:29:38,800
Speaker 6: made that the default model, and you know, I for

612
00:29:38,840 --> 00:29:40,720
Speaker 6: what it's worth. I think the other takeaway from that,

613
00:29:40,840 --> 00:29:43,480
Speaker 6: which is a kind of big theme we see in

614
00:29:43,520 --> 00:29:46,160
Speaker 6: the market as a thing that the cloud Code team

615
00:29:46,200 --> 00:29:49,160
Speaker 6: has talked about, is you constantly have to be building

616
00:29:49,320 --> 00:29:52,600
Speaker 6: ahead with AI in a way that is very unique

617
00:29:52,640 --> 00:29:55,280
Speaker 6: in the world of software, where you kind of always

618
00:29:55,320 --> 00:29:58,160
Speaker 6: want to build things that are working at like seventy

619
00:29:58,240 --> 00:30:00,480
Speaker 6: or eighty percent, because if you really spend the time

620
00:30:00,520 --> 00:30:02,640
Speaker 6: to get it up to ninety or one hundred, you're

621
00:30:02,680 --> 00:30:04,240
Speaker 6: going to lose all the games you get when the

622
00:30:04,280 --> 00:30:07,160
Speaker 6: next model comes out, and the you know, with the

623
00:30:07,200 --> 00:30:09,560
Speaker 6: amount of capex being spent on these models, like there's

624
00:30:09,600 --> 00:30:11,480
Speaker 6: a next model that's going to come out that's going

625
00:30:11,520 --> 00:30:13,240
Speaker 6: to be awesome, and you just kind of want to

626
00:30:13,240 --> 00:30:15,480
Speaker 6: be downstream from that, and you don't want to waste

627
00:30:15,520 --> 00:30:18,400
Speaker 6: six months getting an extra three percent when that new

628
00:30:18,400 --> 00:30:19,960
Speaker 6: model is going to give you an extra seven.

629
00:30:20,840 --> 00:30:23,200
Speaker 3: Yeah, this is the only certainty with AI is like

630
00:30:23,280 --> 00:30:24,680
Speaker 3: there's always going to be a new model.

631
00:30:25,440 --> 00:30:27,920
Speaker 2: The worst model ever use is the one that we're using.

632
00:30:28,240 --> 00:30:40,080
Speaker 4: That's right, That's right.

633
00:30:44,560 --> 00:30:47,800
Speaker 3: Are we all going to become coding illiterate? Are we

634
00:30:47,880 --> 00:30:50,360
Speaker 3: just going to forget how to code. If everyone's using

635
00:30:50,640 --> 00:30:52,440
Speaker 3: you know, general language to do.

636
00:30:52,480 --> 00:30:54,600
Speaker 2: Forget, I've never learned. Yeah, okay, you know what I've

637
00:30:54,600 --> 00:30:55,240
Speaker 2: been thinking about.

638
00:30:55,560 --> 00:30:56,040
Speaker 1: You know that.

639
00:30:57,680 --> 00:31:01,160
Speaker 2: Scott Karp, the CEO Palanteer, here's that line. He's like,

640
00:31:01,400 --> 00:31:03,120
Speaker 2: when I was young, I was too poor to have

641
00:31:03,160 --> 00:31:04,760
Speaker 2: a car, or so I didn't get a so I

642
00:31:04,800 --> 00:31:06,960
Speaker 2: never learned to drive. And now I'm too rich, so

643
00:31:07,000 --> 00:31:09,240
Speaker 2: I never learned to drive. I feel like when I

644
00:31:09,280 --> 00:31:11,240
Speaker 2: was young, I was too dumb to learn to code,

645
00:31:11,400 --> 00:31:12,320
Speaker 2: and now.

646
00:31:12,480 --> 00:31:13,280
Speaker 3: You leaped ahead.

647
00:31:13,440 --> 00:31:16,080
Speaker 2: Yeah, now I'm too smart to learn Python or HTML

648
00:31:16,160 --> 00:31:16,560
Speaker 2: or whatever.

649
00:31:17,440 --> 00:31:19,920
Speaker 6: I have a couple of takes on this person one personally.

650
00:31:19,960 --> 00:31:23,880
Speaker 6: So first one is I just think like this is

651
00:31:23,920 --> 00:31:25,160
Speaker 6: the worry of all technology.

652
00:31:25,240 --> 00:31:25,440
Speaker 5: Ever.

653
00:31:25,600 --> 00:31:28,440
Speaker 6: There was a paper that came out that showed that

654
00:31:28,720 --> 00:31:32,400
Speaker 6: people were, uh, you know, they were forgetting more things

655
00:31:32,480 --> 00:31:34,880
Speaker 6: or something because they were using chat GBD. But you know, uh,

656
00:31:35,800 --> 00:31:39,440
Speaker 6: in Phadrious Plato was worried that people were gonna forget

657
00:31:39,480 --> 00:31:42,440
Speaker 6: things because they started writing things down. And you know,

658
00:31:42,480 --> 00:31:44,160
Speaker 6: I think the trade off there was pretty good. We

659
00:31:44,240 --> 00:31:47,840
Speaker 6: got the scientific revolution a couple of other things, So, uh,

660
00:31:48,400 --> 00:31:51,040
Speaker 6: you know, I think that's the sort of natural knee jerk.

661
00:31:52,720 --> 00:31:57,200
Speaker 6: With that said, it is. I it's very strange when

662
00:31:57,240 --> 00:32:00,160
Speaker 6: you have people, you know, the Cloud Code team is

663
00:32:00,200 --> 00:32:04,080
Speaker 6: talking about how little code they write. Now, I draw

664
00:32:04,080 --> 00:32:06,800
Speaker 6: a distinction between the sort of Vibe coding and the

665
00:32:07,080 --> 00:32:10,479
Speaker 6: kind of amateur people who have never written code. And

666
00:32:10,840 --> 00:32:12,800
Speaker 6: I think that is amazing, by the way, And I

667
00:32:12,800 --> 00:32:15,800
Speaker 6: think there's a lot of software developers who are really

668
00:32:15,800 --> 00:32:19,680
Speaker 6: mad about that because there they claim it's for safety

669
00:32:19,680 --> 00:32:21,800
Speaker 6: reasons or whatever, but I think fundamentally it's just they've

670
00:32:21,800 --> 00:32:25,480
Speaker 6: got people on their turf. But I think that's incredible.

671
00:32:25,480 --> 00:32:29,240
Speaker 6: I mean, my hat, my my nine year old Vibe

672
00:32:29,240 --> 00:32:33,880
Speaker 6: coded a website wo and for Secret Santa. She's now ten.

673
00:32:34,080 --> 00:32:35,520
Speaker 6: She would get bad at me if I called her nine,

674
00:32:35,560 --> 00:32:37,600
Speaker 6: but I think she vibe good at when she was nine.

675
00:32:38,040 --> 00:32:40,440
Speaker 6: But that that's awesome, right, I don't know, that's amazing.

676
00:32:40,480 --> 00:32:43,000
Speaker 6: That's a way for people to express themselves in a

677
00:32:43,040 --> 00:32:46,440
Speaker 6: way that they couldn't before you did your your linguistics process.

678
00:32:46,480 --> 00:32:51,480
Speaker 6: That's that's fun and interesting. But yeah, I also think

679
00:32:51,520 --> 00:32:54,720
Speaker 6: the other the the thing that's happening with professional software

680
00:32:54,720 --> 00:32:56,960
Speaker 6: developers when you hear from anthropic or you know, when

681
00:32:56,960 --> 00:33:00,600
Speaker 6: I'm talking about it's you know, the code going through

682
00:33:00,600 --> 00:33:04,360
Speaker 6: this process, and you know, all the code still gets

683
00:33:04,400 --> 00:33:06,080
Speaker 6: reviewed by people. We're not letting it get out the

684
00:33:06,120 --> 00:33:08,400
Speaker 6: door if it's not at the same level as human

685
00:33:08,560 --> 00:33:11,480
Speaker 6: and it's just But what's amazing is I'm I'm running

686
00:33:11,520 --> 00:33:13,400
Speaker 6: five of these sessions at a time, right, and so

687
00:33:13,440 --> 00:33:16,000
Speaker 6: I've got like software being developed in parallel in a

688
00:33:16,040 --> 00:33:19,280
Speaker 6: way that is unimaginable. And you know the other thing

689
00:33:19,320 --> 00:33:23,080
Speaker 6: is just now, the best software engineers wrote the least

690
00:33:23,080 --> 00:33:26,800
Speaker 6: code anyway. You know the sort of classic story of

691
00:33:26,840 --> 00:33:29,640
Speaker 6: like the difference between a junior developer or senior developers

692
00:33:29,640 --> 00:33:32,720
Speaker 6: that a junior developer gets a problem and they sit

693
00:33:32,760 --> 00:33:34,320
Speaker 6: down and they put their fingers on the keyboard and

694
00:33:34,320 --> 00:33:37,320
Speaker 6: they start writing code. And a senior developer gets a

695
00:33:37,320 --> 00:33:39,920
Speaker 6: problem and sits there for three hours and tries to

696
00:33:39,920 --> 00:33:41,360
Speaker 6: figure out what the best way to solve it is

697
00:33:41,400 --> 00:33:43,720
Speaker 6: and then spends five minutes writing code to get it done.

698
00:33:43,960 --> 00:33:46,400
Speaker 3: True elegance is restraint. That's what I say.

699
00:33:46,840 --> 00:33:47,280
Speaker 4: What are you.

700
00:33:47,280 --> 00:33:50,440
Speaker 2: Seeing in the companies you're working for? Like, I find

701
00:33:50,480 --> 00:33:53,200
Speaker 2: it hard to believe and I was maybe skeptical of this,

702
00:33:53,320 --> 00:33:55,800
Speaker 2: but it feels like right now we're here with technology,

703
00:33:55,800 --> 00:33:58,640
Speaker 2: where like if viral like company is like, like I said,

704
00:33:58,640 --> 00:34:00,920
Speaker 2: you can build charts of data in a way that

705
00:34:01,040 --> 00:34:02,800
Speaker 2: used to be like someone would have had to get

706
00:34:02,800 --> 00:34:04,760
Speaker 2: their hands dirty, ear, et cetera. And the companies that

707
00:34:04,840 --> 00:34:07,720
Speaker 2: you talk to is right now that having an effect

708
00:34:07,800 --> 00:34:09,920
Speaker 2: on how they think about what positions they're hiring for

709
00:34:09,960 --> 00:34:11,680
Speaker 2: and the skills they're looking for and so forth.

710
00:34:12,200 --> 00:34:17,360
Speaker 6: I think that it's hard to answer, right, really, I

711
00:34:17,440 --> 00:34:22,239
Speaker 6: think that certainly. I do think I personally think if

712
00:34:22,280 --> 00:34:24,440
Speaker 6: I look at the sort of layoffs in the technology

713
00:34:24,480 --> 00:34:26,080
Speaker 6: industry of the last couple of years, I think some

714
00:34:26,200 --> 00:34:28,600
Speaker 6: part of that is just looking at the output of

715
00:34:28,640 --> 00:34:33,000
Speaker 6: these models and saying, hey, these models are able to

716
00:34:33,000 --> 00:34:35,880
Speaker 6: produce it, you know, the median, and I have a

717
00:34:35,880 --> 00:34:38,120
Speaker 6: whole bunch of sort of middle managers who are producing

718
00:34:38,160 --> 00:34:41,040
Speaker 6: it the sixty fifth percentile. And it's like I can

719
00:34:41,120 --> 00:34:44,440
Speaker 6: produce median for a dollar fifty per million tokens, or

720
00:34:44,480 --> 00:34:48,440
Speaker 6: I can produce sixty fifth percentile for hover many hundreds

721
00:34:48,480 --> 00:34:50,160
Speaker 6: of thousands dollars a year. It's it's a sort of

722
00:34:50,160 --> 00:34:53,399
Speaker 6: fairly simple trade off, I think. So I do think

723
00:34:53,400 --> 00:34:55,000
Speaker 6: there's a lot of downstream effects. I think the other

724
00:34:55,000 --> 00:34:57,480
Speaker 6: thing that's happening, is is kind of like middle management

725
00:34:57,560 --> 00:35:00,239
Speaker 6: is under threat because it's the realization that hey, like,

726
00:35:00,520 --> 00:35:02,880
Speaker 6: part of what these models are amazing at is is

727
00:35:03,360 --> 00:35:05,080
Speaker 6: I think of them as like a fuzzy interface. They

728
00:35:05,120 --> 00:35:07,920
Speaker 6: can sort of turn any data into any other data, right.

729
00:35:07,960 --> 00:35:11,200
Speaker 6: You can sort of transform data from one format to another.

730
00:35:11,360 --> 00:35:13,000
Speaker 6: You can take a PDF and you can turn it

731
00:35:13,040 --> 00:35:15,279
Speaker 6: into charts, right, And there's whole people who exist or

732
00:35:15,400 --> 00:35:17,120
Speaker 6: you know, if you think about what product managers do

733
00:35:17,160 --> 00:35:19,000
Speaker 6: a lot of what product managers do is they take

734
00:35:19,239 --> 00:35:21,040
Speaker 6: how people are using a product and they try to

735
00:35:21,080 --> 00:35:25,319
Speaker 6: transform it into a format that engineers can then use

736
00:35:25,400 --> 00:35:27,000
Speaker 6: to figure out what to do. And I think a

737
00:35:27,040 --> 00:35:31,080
Speaker 6: lot of those kind of a lot of those pieces

738
00:35:31,239 --> 00:35:35,880
Speaker 6: that used to just be kind of transferring knowledge.

739
00:35:36,080 --> 00:35:39,520
Speaker 2: I've always said, Tracy, I think one of the most

740
00:35:39,560 --> 00:35:42,920
Speaker 2: important roles in any organization is essentially translation work. And

741
00:35:42,960 --> 00:35:46,120
Speaker 2: you see it in the newsroom where it's like here's

742
00:35:46,160 --> 00:35:50,919
Speaker 2: a team specialized in emerging market currencies, and then they

743
00:35:50,960 --> 00:35:53,280
Speaker 2: have to like they have to then tell the senior

744
00:35:53,400 --> 00:35:56,040
Speaker 2: editors what they're working on. But the senior editors, who

745
00:35:56,040 --> 00:35:58,920
Speaker 2: are maybe more generalists, don't really know like why like

746
00:35:59,000 --> 00:36:02,360
Speaker 2: some sort of like you know, one yen carry is

747
00:36:02,400 --> 00:36:05,600
Speaker 2: important and that a really important role within any organization

748
00:36:05,719 --> 00:36:08,680
Speaker 2: is essentially the team that can translate between the generalist

749
00:36:08,680 --> 00:36:11,319
Speaker 2: team and the specialist team. Absolutely, and so that's an

750
00:36:11,320 --> 00:36:14,120
Speaker 2: interesting observation and the sort of engineering world of like, okay,

751
00:36:14,120 --> 00:36:17,000
Speaker 2: these are tools that are in sometimes translation tools.

752
00:36:17,480 --> 00:36:19,840
Speaker 3: So we talked I agree completely, by the way, but

753
00:36:20,160 --> 00:36:23,359
Speaker 3: we talked about vibe coding and Joe has this application

754
00:36:24,160 --> 00:36:26,879
Speaker 3: that I don't think you're looking to monetize.

755
00:36:26,400 --> 00:36:27,960
Speaker 2: No, it's I'm just trying to make it for the

756
00:36:28,000 --> 00:36:30,160
Speaker 2: good of the world, right, Okay, so when did that

757
00:36:30,160 --> 00:36:33,640
Speaker 2: become a crown? I'm not a lotizing it, But.

758
00:36:33,719 --> 00:36:37,400
Speaker 3: Like this opens up massive questions for software as a service,

759
00:36:37,480 --> 00:36:41,239
Speaker 3: right versas because if everyone can write their own software,

760
00:36:42,480 --> 00:36:45,320
Speaker 3: you can replicate anything that's out there that is currently

761
00:36:45,480 --> 00:36:47,400
Speaker 3: charging money. What's going to happen to software?

762
00:36:48,200 --> 00:36:50,839
Speaker 6: I think software is pretty screwed. A lot of it,

763
00:36:50,880 --> 00:36:53,120
Speaker 6: at least not all of it. You know, you still

764
00:36:53,480 --> 00:36:55,880
Speaker 6: it depends on whether you call that cloud provider software

765
00:36:55,960 --> 00:36:57,640
Speaker 6: or not. You know, you still need to run this

766
00:36:57,640 --> 00:37:00,360
Speaker 6: stuff somewhere. And I think there's there's certain kind of

767
00:37:00,400 --> 00:37:02,920
Speaker 6: software that you know, you just don't really want to

768
00:37:02,960 --> 00:37:06,160
Speaker 6: be in the business of writing, you know, I as

769
00:37:06,200 --> 00:37:09,040
Speaker 6: someone who's tried to build a project management system, I'd

770
00:37:09,080 --> 00:37:11,840
Speaker 6: really rather I don't think anybody should be in that business.

771
00:37:13,400 --> 00:37:15,960
Speaker 6: But I do think fundamentally, I mean, we see this

772
00:37:16,120 --> 00:37:19,759
Speaker 6: every day inside enterprises. The sort of build versus buy

773
00:37:20,840 --> 00:37:23,960
Speaker 6: pendulum has just swung. And you know, I mean I

774
00:37:24,120 --> 00:37:26,960
Speaker 6: used to run a SaaS company and we sold to enterprises,

775
00:37:27,000 --> 00:37:29,919
Speaker 6: and you know, for a long time that I think

776
00:37:29,960 --> 00:37:32,520
Speaker 6: that made a lot of sense, right because like, hey,

777
00:37:32,880 --> 00:37:34,440
Speaker 6: it just didn't make sense to try to build this

778
00:37:34,480 --> 00:37:37,400
Speaker 6: thing on your own. And so but the price of

779
00:37:37,400 --> 00:37:39,960
Speaker 6: that was, you know, won the price, right, like, and

780
00:37:40,040 --> 00:37:41,759
Speaker 6: it got to be more and more expensive. The other

781
00:37:41,800 --> 00:37:43,320
Speaker 6: price was that you were paying for a lot of

782
00:37:43,360 --> 00:37:45,600
Speaker 6: stuff you didn't need, right, because the whole job of

783
00:37:45,640 --> 00:37:49,719
Speaker 6: building SaaS is you need to generalize problems, and so

784
00:37:49,920 --> 00:37:52,719
Speaker 6: you build things that are going to work for everybody.

785
00:37:53,040 --> 00:37:54,680
Speaker 5: And that means either you.

786
00:37:54,600 --> 00:37:56,279
Speaker 6: Have to sort of adapt or you have to build

787
00:37:56,320 --> 00:38:00,880
Speaker 6: this sort of very configurable software. And I think, and

788
00:38:01,400 --> 00:38:03,560
Speaker 6: what I see you just you know firsthand, is that

789
00:38:05,080 --> 00:38:10,440
Speaker 6: inside these organizations you can now solve very specific problems

790
00:38:10,440 --> 00:38:14,680
Speaker 6: that are highly valuable and not only can you solve

791
00:38:14,680 --> 00:38:18,279
Speaker 6: them better than generic software, but you can actually, in

792
00:38:18,320 --> 00:38:20,360
Speaker 6: a lot of ways do it for less money because

793
00:38:20,360 --> 00:38:23,000
Speaker 6: you're trying to tackle less stuff. You didn't need the

794
00:38:23,080 --> 00:38:25,319
Speaker 6: sixteen other features. You bought it for the one that

795
00:38:25,400 --> 00:38:29,480
Speaker 6: you really really cared about. And so I think that

796
00:38:29,640 --> 00:38:31,400
Speaker 6: part of it, you know, I don't like there's I

797
00:38:31,680 --> 00:38:34,120
Speaker 6: definitely think there are pieces of the software industry that

798
00:38:34,239 --> 00:38:36,200
Speaker 6: are gonna, you know, come out the other side. You're

799
00:38:36,200 --> 00:38:38,799
Speaker 6: gonna nobody wants to deal with payroll, right like you know,

800
00:38:38,840 --> 00:38:41,480
Speaker 6: somebody you're still gonna buy some payroll software and you're

801
00:38:41,520 --> 00:38:44,000
Speaker 6: still gonna have that. But you know, I do think

802
00:38:44,000 --> 00:38:46,840
Speaker 6: there are a lot of pieces where the software existed

803
00:38:46,960 --> 00:38:50,000
Speaker 6: essentially as a kind of wrapper around a database, and

804
00:38:50,040 --> 00:38:53,120
Speaker 6: now you're just gonna, you know, with just the database,

805
00:38:53,160 --> 00:38:55,120
Speaker 6: you can do that. And then you know, the other

806
00:38:55,160 --> 00:38:57,040
Speaker 6: piece I'd say here is it's this is not this

807
00:38:57,160 --> 00:38:59,239
Speaker 6: is a kind of confluence of circumstances where it's not

808
00:38:59,360 --> 00:39:03,880
Speaker 6: just the coding, it's also the fact that you have

809
00:39:04,080 --> 00:39:06,200
Speaker 6: ai to do a whole bunch of work. So you know,

810
00:39:06,200 --> 00:39:08,719
Speaker 6: if we pick on CRM for a second, right, like

811
00:39:09,120 --> 00:39:13,960
Speaker 6: you know, Salesforce dot Com Salesforce dot com. We can

812
00:39:14,239 --> 00:39:16,400
Speaker 6: you know, you look at what the interface of that is.

813
00:39:16,440 --> 00:39:20,120
Speaker 6: And essentially it has existed to get salespeople to take

814
00:39:20,239 --> 00:39:22,839
Speaker 6: unstructured data, which is sales meetings, and turn it into

815
00:39:22,880 --> 00:39:25,839
Speaker 6: structured data that so can be stored into database. And

816
00:39:25,920 --> 00:39:29,960
Speaker 6: now you have AI, and AI is very capable of

817
00:39:30,040 --> 00:39:33,680
Speaker 6: taking unstructured data directly from the source, so you have

818
00:39:33,719 --> 00:39:37,160
Speaker 6: people recording meetings, and then it can structure it into

819
00:39:37,200 --> 00:39:39,400
Speaker 6: any data that you want. This is one of the

820
00:39:39,480 --> 00:39:42,320
Speaker 6: very first sort of mind blowing moments I had was

821
00:39:42,360 --> 00:39:46,399
Speaker 6: that I could give it a Jason interface. I could

822
00:39:46,400 --> 00:39:49,480
Speaker 6: describe exactly what I wanted the data structure to be,

823
00:39:49,920 --> 00:39:52,480
Speaker 6: and it would give me back that information in that

824
00:39:52,560 --> 00:39:55,080
Speaker 6: data structure. And we've just basically been having a bunch

825
00:39:55,080 --> 00:39:56,799
Speaker 6: of humans do that work for a very long time,

826
00:39:56,840 --> 00:39:58,880
Speaker 6: whether it's in CRM or project management or any of

827
00:39:58,880 --> 00:40:00,800
Speaker 6: these other places. And the ability to just kind of

828
00:40:00,840 --> 00:40:02,640
Speaker 6: get rid of that whole thing, I think it really

829
00:40:02,640 --> 00:40:05,480
Speaker 6: does bring into question the value of a lot of

830
00:40:05,480 --> 00:40:06,400
Speaker 6: these software companies.

831
00:40:06,520 --> 00:40:08,840
Speaker 2: Well, so we have seen like a lot of software sucks.

832
00:40:08,880 --> 00:40:11,960
Speaker 2: They look like melting ice cubes right now. Maybe they

833
00:40:12,440 --> 00:40:14,799
Speaker 2: So what is it I want to talk. I mean,

834
00:40:14,840 --> 00:40:17,000
Speaker 2: this is like you know, our listeners who are investors,

835
00:40:17,080 --> 00:40:19,800
Speaker 2: there's a pretty high stakes question of like what residual

836
00:40:19,960 --> 00:40:22,960
Speaker 2: value there is. But talk a little bit more about Salesforce.

837
00:40:23,000 --> 00:40:24,480
Speaker 2: Maybe there would be a time to learn what sale

838
00:40:24,840 --> 00:40:28,040
Speaker 2: what it actually does as it's massively being disrupted. Now

839
00:40:28,040 --> 00:40:30,560
Speaker 2: we get around to learning what Salesforce is. But I

840
00:40:30,600 --> 00:40:32,560
Speaker 2: know it's like many things there are apps that people

841
00:40:32,560 --> 00:40:35,279
Speaker 2: built onto Salesforce. But this sounds like we're hitting on

842
00:40:35,360 --> 00:40:37,160
Speaker 2: when I think probably one of the crucial questions for

843
00:40:37,239 --> 00:40:39,600
Speaker 2: like the future of the software industry. So talk a

844
00:40:39,600 --> 00:40:42,120
Speaker 2: little bit more about like the current approach and what

845
00:40:42,160 --> 00:40:45,879
Speaker 2: people are buying when they buy a package or subscribe

846
00:40:45,960 --> 00:40:49,000
Speaker 2: to a service from Salesforce, and then what the unlocked

847
00:40:49,000 --> 00:40:53,319
Speaker 2: opportunity is from having AI like live in the same

848
00:40:53,320 --> 00:40:54,520
Speaker 2: world as all your files.

849
00:40:55,040 --> 00:40:58,759
Speaker 6: Yeah, so I think if we take CRM as the

850
00:40:58,800 --> 00:41:01,640
Speaker 6: general category, know the biggest players there are.

851
00:41:01,640 --> 00:41:02,719
Speaker 2: That's customer relationship.

852
00:41:02,719 --> 00:41:05,920
Speaker 6: Customer relationship management, that's like what you know, Salesforce does it,

853
00:41:05,960 --> 00:41:08,600
Speaker 6: SVP does it, HubSpot does it for the mid market.

854
00:41:10,160 --> 00:41:12,040
Speaker 6: You know, when I think about that product and I

855
00:41:12,040 --> 00:41:15,600
Speaker 6: think about the way we've used it inside enterprise sales organizations. Essentially,

856
00:41:15,680 --> 00:41:18,280
Speaker 6: you know, it's a database of companies, it's a database

857
00:41:18,280 --> 00:41:20,360
Speaker 6: of contacts, it's a database of deals you have in

858
00:41:20,400 --> 00:41:22,799
Speaker 6: the pipeline, and it's a way to track all those

859
00:41:22,840 --> 00:41:25,200
Speaker 6: deals you guys hit on something before that I think

860
00:41:25,239 --> 00:41:30,680
Speaker 6: is really it, which is like inside companies, there is

861
00:41:30,719 --> 00:41:33,400
Speaker 6: a huge group of people and who exist to answer

862
00:41:33,440 --> 00:41:36,000
Speaker 6: the question from management of what is the status of

863
00:41:36,080 --> 00:41:38,720
Speaker 6: something right? And you know that can be sales management,

864
00:41:38,760 --> 00:41:40,360
Speaker 6: it can be product management, it doesn't matter, right, it

865
00:41:40,400 --> 00:41:43,520
Speaker 6: could be within a newsroom. Somebody wants to know what

866
00:41:43,560 --> 00:41:45,879
Speaker 6: the status is and somebody else exists to go figure

867
00:41:45,920 --> 00:41:49,200
Speaker 6: out what the answer to that question is. And so fundamentally,

868
00:41:49,719 --> 00:41:52,880
Speaker 6: I think those CRM tools are bought first and foremost

869
00:41:53,120 --> 00:41:55,760
Speaker 6: to answer what is the status right? What's my pipeline

870
00:41:55,760 --> 00:41:58,040
Speaker 6: look like? And to answer what your pipeline looks like,

871
00:41:58,040 --> 00:42:01,319
Speaker 6: you need a bunch of salespeople putting deals in and

872
00:42:01,360 --> 00:42:03,520
Speaker 6: those deals are associated with contacts and companies and they

873
00:42:03,520 --> 00:42:07,240
Speaker 6: say when is that deal going to close? And essentially

874
00:42:07,239 --> 00:42:09,880
Speaker 6: you were asking the sales people to make the updates

875
00:42:09,880 --> 00:42:14,560
Speaker 6: in the system to do that and just very tactically,

876
00:42:14,680 --> 00:42:17,319
Speaker 6: I mean, you know, I run a company. Now we

877
00:42:17,440 --> 00:42:18,880
Speaker 6: talk to a lot of we have a lot of

878
00:42:18,880 --> 00:42:23,400
Speaker 6: sales calls. We record those calls and they get transcribed,

879
00:42:23,440 --> 00:42:26,440
Speaker 6: and the AI then looks through them and makes decisions

880
00:42:26,480 --> 00:42:30,440
Speaker 6: about where this deal should be in the process. And

881
00:42:31,400 --> 00:42:34,080
Speaker 6: it's much better than having somebody try to go update it,

882
00:42:34,120 --> 00:42:36,600
Speaker 6: because those people never updated anywhere. The secret of all

883
00:42:36,640 --> 00:42:39,239
Speaker 6: of this enterprise software is that nobody was using it

884
00:42:39,239 --> 00:42:42,520
Speaker 6: the way that anybody wanted to anyway. And so you know,

885
00:42:43,120 --> 00:42:46,680
Speaker 6: I think that that is sort of you know, a

886
00:42:46,680 --> 00:42:48,359
Speaker 6: lot of what's happening there. Again, it's sort of some

887
00:42:48,400 --> 00:42:50,640
Speaker 6: of it's the coding, some of it's just the core capabilities.

888
00:42:50,680 --> 00:42:52,759
Speaker 6: And then you know, you still need databases, right, so

889
00:42:52,800 --> 00:42:54,320
Speaker 6: it's like, you know, you look at what data bricks

890
00:42:54,320 --> 00:42:56,400
Speaker 6: and stuff like. You know, I think those folks are

891
00:42:56,440 --> 00:42:58,840
Speaker 6: still sort of genuinely sitting in a pretty good place

892
00:42:58,880 --> 00:43:01,919
Speaker 6: where you know, all software has to sit on side

893
00:43:02,320 --> 00:43:05,000
Speaker 6: on top of some database that you can sort of

894
00:43:05,040 --> 00:43:08,080
Speaker 6: read and write to. But you know, I think some

895
00:43:08,120 --> 00:43:13,239
Speaker 6: of those categories that were specifically focused on kind of

896
00:43:13,320 --> 00:43:16,799
Speaker 6: like human input. Now, of course, you know, Salesforce has

897
00:43:16,800 --> 00:43:18,880
Speaker 6: a whole AI thing and they're saying, hey, you shouldn't

898
00:43:18,880 --> 00:43:22,320
Speaker 6: have humans inputting in salesforce. You know, at sales is

899
00:43:22,400 --> 00:43:24,799
Speaker 6: just one small piece. They have a whole customer support thing,

900
00:43:25,080 --> 00:43:28,680
Speaker 6: which obviously also has an interesting implication where you know

901
00:43:28,680 --> 00:43:31,160
Speaker 6: you're doing support with AI agents and so some of

902
00:43:31,200 --> 00:43:32,880
Speaker 6: it comes back to seats. I mean, you know, it

903
00:43:32,920 --> 00:43:36,000
Speaker 6: gets to be fairly complicated, but I do think I

904
00:43:36,000 --> 00:43:40,280
Speaker 6: think the fundamental underlying thing is anybody who buys software

905
00:43:41,080 --> 00:43:44,920
Speaker 6: that is, you know, uh SaaS, you're always buying for

906
00:43:44,960 --> 00:43:48,480
Speaker 6: a subset of the functionality that's nobody is using one

907
00:43:48,520 --> 00:43:50,839
Speaker 6: hundred percent of the functionality of SaaS, and so there's

908
00:43:50,840 --> 00:43:52,880
Speaker 6: always a trade off that's happening there where you know

909
00:43:53,000 --> 00:43:55,399
Speaker 6: you're spending more money than you need to because you're

910
00:43:55,440 --> 00:43:58,040
Speaker 6: not using all of these pieces. And so you know,

911
00:43:58,160 --> 00:44:00,600
Speaker 6: if you can more narrowly focus that, that is where

912
00:44:00,680 --> 00:44:02,560
Speaker 6: you could say, hey, we could solve this kind of

913
00:44:02,560 --> 00:44:04,400
Speaker 6: more narrow problem. And not only can we solve it

914
00:44:04,400 --> 00:44:07,320
Speaker 6: more narrowly, we can solve it way more effectively because

915
00:44:07,640 --> 00:44:10,080
Speaker 6: you know, the trick with AI is that the more

916
00:44:10,520 --> 00:44:13,520
Speaker 6: specific you are with it, the better the output is. Right,

917
00:44:13,560 --> 00:44:15,879
Speaker 6: So it's like, if you know, if outside of coding,

918
00:44:15,920 --> 00:44:18,480
Speaker 6: if you just ask chet gpz to write you a story.

919
00:44:18,600 --> 00:44:21,680
Speaker 6: It's going to write you a very very median story, right,

920
00:44:22,200 --> 00:44:25,600
Speaker 6: sort of exactly the median. But if you work with

921
00:44:25,640 --> 00:44:28,439
Speaker 6: it and you you know, then you're going to get it.

922
00:44:28,560 --> 00:44:31,640
Speaker 6: The more of your own expertise, you imbuing it, the

923
00:44:32,280 --> 00:44:34,319
Speaker 6: further up above the media, and it's going to be

924
00:44:34,360 --> 00:44:35,840
Speaker 6: and it's going to be you know. Of course that

925
00:44:35,920 --> 00:44:39,399
Speaker 6: also means it's less where the line is between what's

926
00:44:39,440 --> 00:44:41,640
Speaker 6: AI and what's not AI is going to continue to

927
00:44:41,640 --> 00:44:43,759
Speaker 6: get lorier, Joe, how.

928
00:44:43,719 --> 00:44:45,760
Speaker 3: Much does Claude Code actually cost?

929
00:44:46,280 --> 00:44:46,640
Speaker 4: Do you know?

930
00:44:47,280 --> 00:44:47,640
Speaker 5: Well?

931
00:44:48,600 --> 00:44:51,920
Speaker 2: I paid for the two two hundred dollars a month version,

932
00:44:52,360 --> 00:44:56,080
Speaker 2: but like high roller, yeah, I know, but uh, you know,

933
00:44:56,160 --> 00:44:57,640
Speaker 2: I think it's you can get it with the pro

934
00:44:57,760 --> 00:44:59,920
Speaker 2: version of like or whatever. The sub version of that

935
00:45:00,040 --> 00:45:02,200
Speaker 2: blow twenty dollars. But I hit a limit fairly quickly,

936
00:45:02,280 --> 00:45:03,719
Speaker 2: and I was like, I didn't have my website up

937
00:45:03,800 --> 00:45:06,600
Speaker 2: so like, and then I bought the five. Then I

938
00:45:06,640 --> 00:45:09,520
Speaker 2: paid five dollars for the extra compute, and I was like,

939
00:45:09,560 --> 00:45:10,160
Speaker 2: this is dumb.

940
00:45:10,239 --> 00:45:13,640
Speaker 3: I think, yeah, okay, so we're going out to.

941
00:45:13,800 --> 00:45:16,560
Speaker 2: Two nice dinners right month, that's not you know, when

942
00:45:16,560 --> 00:45:18,080
Speaker 2: I think about that way, it doesn't see that big.

943
00:45:17,920 --> 00:45:19,680
Speaker 3: Of a deal, it's worth it to you. Yeah, okay,

944
00:45:19,719 --> 00:45:21,600
Speaker 3: so I think we can all agree this is like

945
00:45:21,680 --> 00:45:26,319
Speaker 3: a valuable service that claud Code is providing. But we

946
00:45:26,400 --> 00:45:29,080
Speaker 3: touched on this in the intro. It seems like the

947
00:45:29,160 --> 00:45:33,320
Speaker 3: models just keep replicating themselves really really quickly. So anything

948
00:45:33,320 --> 00:45:36,520
Speaker 3: that claud Code can do, I would expect another model

949
00:45:36,560 --> 00:45:39,120
Speaker 3: will come in in like a month maybe less and

950
00:45:39,160 --> 00:45:41,880
Speaker 3: do the exact same thing. What does that mean for

951
00:45:41,960 --> 00:45:46,279
Speaker 3: the actual like valuations of these companies and the models, Like,

952
00:45:46,360 --> 00:45:48,360
Speaker 3: how are they going to monetize it when it seems

953
00:45:48,440 --> 00:45:53,239
Speaker 3: so difficult to actually differentiate yourself, especially for like a

954
00:45:53,280 --> 00:45:55,719
Speaker 3: substantial portion of time.

955
00:45:55,800 --> 00:45:58,279
Speaker 6: Yeah, well, so again here I think we have to

956
00:45:58,440 --> 00:46:02,480
Speaker 6: distinguish between claud Code and the Claude model. So in

957
00:46:02,760 --> 00:46:06,040
Speaker 6: claud Code's case, if you're using you know, the latest version,

958
00:46:06,040 --> 00:46:08,680
Speaker 6: you're using Opus four point five, which is the model.

959
00:46:08,760 --> 00:46:11,640
Speaker 6: Opus four point five has a price of something in

960
00:46:11,680 --> 00:46:14,480
Speaker 6: the dollar fifty to two dollars for a million input

961
00:46:14,560 --> 00:46:16,880
Speaker 6: tokens and whatever it is on the output, which is

962
00:46:16,920 --> 00:46:19,640
Speaker 6: like roughly the going rate for cutting edge modles. Gemini

963
00:46:19,800 --> 00:46:25,160
Speaker 6: three pro is the same price open a Chachipt five

964
00:46:25,200 --> 00:46:27,560
Speaker 6: point two is. They're all the same price. So the

965
00:46:27,560 --> 00:46:29,759
Speaker 6: first thing is is you have to differentiate between those.

966
00:46:30,760 --> 00:46:32,960
Speaker 6: And so I think a big part of what Anthropic

967
00:46:33,040 --> 00:46:34,880
Speaker 6: is trying to do is they're trying to lock people

968
00:46:34,880 --> 00:46:39,200
Speaker 6: into claud Code. In fact, there's just some controversy amongst

969
00:46:39,480 --> 00:46:43,560
Speaker 6: some nerds where open Code, which is a competitor to

970
00:46:43,640 --> 00:46:47,879
Speaker 6: claud Code, used to let you use your Claude Max

971
00:46:47,880 --> 00:46:50,120
Speaker 6: two hundred dollars. So the trick with the Claude Max

972
00:46:50,160 --> 00:46:53,320
Speaker 6: plan is if you're just buying those that number of tokens,

973
00:46:53,320 --> 00:46:56,000
Speaker 6: it would cost you significantly more than two hundred dollars.

974
00:46:56,040 --> 00:47:00,400
Speaker 6: It is a super super discounted plan. So like you,

975
00:47:00,400 --> 00:47:03,560
Speaker 6: you are probably you have the access, I have the

976
00:47:03,600 --> 00:47:06,239
Speaker 6: access to use I would guess in the thousand or

977
00:47:06,239 --> 00:47:10,640
Speaker 6: two thousand dollars of tokens for my two hundred dollars

978
00:47:10,640 --> 00:47:13,359
Speaker 6: a month. So it's a very very heavily subsidized plan.

979
00:47:13,640 --> 00:47:15,840
Speaker 6: And open Code, which is an open source version of

980
00:47:16,000 --> 00:47:19,920
Speaker 6: Claude code, a sort of competitor. They had found a

981
00:47:19,920 --> 00:47:22,600
Speaker 6: way that they would let you use your Claude Max

982
00:47:22,640 --> 00:47:27,680
Speaker 6: plan with open code, and Anthropic last week shut that down,

983
00:47:28,360 --> 00:47:31,520
Speaker 6: and some open Code people got very upset because they said,

984
00:47:31,560 --> 00:47:34,160
Speaker 6: like this is not what you're supposed to do or

985
00:47:34,520 --> 00:47:37,400
Speaker 6: I'm not sure exactly what they said. I never felt

986
00:47:37,440 --> 00:47:39,400
Speaker 6: like I got a particularly good argument out of it.

987
00:47:40,280 --> 00:47:42,600
Speaker 6: But you know, I do think part of what they're

988
00:47:42,600 --> 00:47:47,040
Speaker 6: trying to get at because is that, you know, at

989
00:47:47,040 --> 00:47:49,839
Speaker 6: the very top, models like these are all amazing, like

990
00:47:50,040 --> 00:47:56,040
Speaker 6: the Google Opening and Anthropic, their best models are all

991
00:47:56,080 --> 00:47:58,319
Speaker 6: on par with each other. I mean, I would move

992
00:47:58,320 --> 00:48:00,920
Speaker 6: them around a little bit. I still think Opus four

993
00:48:01,000 --> 00:48:03,120
Speaker 6: point five is the best model out there, but you know,

994
00:48:03,160 --> 00:48:08,439
Speaker 6: I mean that might change tomorrow. Like and that's where

995
00:48:08,480 --> 00:48:10,560
Speaker 6: something like cloud code is really interesting because it's a

996
00:48:11,200 --> 00:48:14,080
Speaker 6: product that is very it's just theirs. It's a piece

997
00:48:14,080 --> 00:48:17,120
Speaker 6: of software. It's not an AI model, and so it's

998
00:48:17,160 --> 00:48:21,000
Speaker 6: sort of it's less able to be disrupted. Now again,

999
00:48:21,120 --> 00:48:24,080
Speaker 6: I think if if somebody else wanted to copy that exactly,

1000
00:48:24,800 --> 00:48:28,319
Speaker 6: they could. Codex has one, Gemini has one. I just

1001
00:48:28,320 --> 00:48:30,439
Speaker 6: think they take a very different tact with it where

1002
00:48:30,440 --> 00:48:32,360
Speaker 6: it's much less and so, you know, I think what

1003
00:48:32,360 --> 00:48:34,120
Speaker 6: they're trying to do is get developers like me to

1004
00:48:34,160 --> 00:48:36,440
Speaker 6: feel very comfortable inside that so that when we go

1005
00:48:36,560 --> 00:48:39,640
Speaker 6: open I still open Codex or tried Gemini, or I

1006
00:48:39,719 --> 00:48:42,840
Speaker 6: was playing with open code the other day, and it

1007
00:48:42,960 --> 00:48:44,960
Speaker 6: just doesn't feel familiar in the same way that you know,

1008
00:48:45,000 --> 00:48:47,120
Speaker 6: if you're trying to move somebody from a PC to

1009
00:48:47,160 --> 00:48:48,480
Speaker 6: a MAC, it doesn't feel familiar. Right.

1010
00:48:48,480 --> 00:48:50,680
Speaker 3: They want to own like the ecosystem.

1011
00:48:50,080 --> 00:48:54,160
Speaker 2: The environment, that environment or what a world. Noah, thank

1012
00:48:54,200 --> 00:48:56,200
Speaker 2: you so much for coming on odd Laws. I was

1013
00:48:56,239 --> 00:48:58,560
Speaker 2: like dying to do an episode about this topic.

1014
00:48:58,600 --> 00:48:59,359
Speaker 5: Thanks for having me.

1015
00:49:00,080 --> 00:49:03,160
Speaker 2: Way, I don't have AI psychosis. I have a claud complex.

1016
00:49:03,600 --> 00:49:05,239
Speaker 3: Why is everyone making that joke?

1017
00:49:05,360 --> 00:49:06,439
Speaker 2: Wait, which joke?

1018
00:49:07,000 --> 00:49:08,080
Speaker 3: The psychosis joke?

1019
00:49:08,600 --> 00:49:10,160
Speaker 2: I think you were going to be proud of me

1020
00:49:10,200 --> 00:49:11,480
Speaker 2: for saying Claude complex.

1021
00:49:11,760 --> 00:49:13,040
Speaker 3: Oh oh that is very good.

1022
00:49:13,680 --> 00:49:16,400
Speaker 2: I do one pun finally for Tracy And why was

1023
00:49:16,400 --> 00:49:17,239
Speaker 2: it for making that joke?

1024
00:49:17,320 --> 00:49:18,640
Speaker 5: Well, I was a joke.

1025
00:49:18,840 --> 00:49:20,759
Speaker 2: I was handing your sirt. I finally make a pun,

1026
00:49:20,880 --> 00:49:21,879
Speaker 2: and you just jump right over.

1027
00:49:22,160 --> 00:49:25,520
Speaker 3: Everyone keeps saying that claud code is AI psychosis for

1028
00:49:25,600 --> 00:49:28,960
Speaker 3: smart people, right, Like, how did that beck come a thing? Yeah?

1029
00:49:29,080 --> 00:49:30,439
Speaker 2: All right, but there's a good pun.

1030
00:49:30,560 --> 00:49:33,160
Speaker 3: Also very pro coded. I find you think so all

1031
00:49:33,160 --> 00:49:34,400
Speaker 3: of AI is pro coded.

1032
00:49:34,880 --> 00:49:37,279
Speaker 2: Uh, this is true. We should talk more about this,

1033
00:49:37,320 --> 00:49:39,120
Speaker 2: you know, we should have David Shorn he's been doing

1034
00:49:39,120 --> 00:49:41,319
Speaker 2: a lot of polling about various demographics and how they

1035
00:49:41,320 --> 00:49:45,319
Speaker 2: feel about AI. We should need some interesting that. Yeah,

1036
00:49:45,320 --> 00:49:47,719
Speaker 2: we should do that anyway. No, thank you so much

1037
00:49:47,719 --> 00:49:48,480
Speaker 2: for coming on offline.

1038
00:49:48,520 --> 00:49:49,319
Speaker 5: Thanks for having me.

1039
00:50:01,480 --> 00:50:03,960
Speaker 2: Well that was fun, Tracy. I really like I It's

1040
00:50:04,200 --> 00:50:07,399
Speaker 2: obvious to anyone who's been within five minutes five feet

1041
00:50:07,440 --> 00:50:09,320
Speaker 2: of me for the last two weeks. I'm like totally

1042
00:50:09,400 --> 00:50:12,720
Speaker 2: addicted and gone down, I know, gone down the rabbit

1043
00:50:12,760 --> 00:50:16,840
Speaker 2: hole and stuff. But like I for the first time, unironically,

1044
00:50:17,000 --> 00:50:21,160
Speaker 2: I'm like, Okay, this is transformative technology beyond being very

1045
00:50:21,160 --> 00:50:22,120
Speaker 2: impressive technology.

1046
00:50:22,200 --> 00:50:26,000
Speaker 3: Right, So I've been coming to a conclusion, which is

1047
00:50:26,040 --> 00:50:31,040
Speaker 3: that you know, AI can be both under hyped and

1048
00:50:31,160 --> 00:50:34,920
Speaker 3: overvalued simultaneously. Like and I feel like that's kind of

1049
00:50:34,920 --> 00:50:36,480
Speaker 3: where we are at the moment.

1050
00:50:36,239 --> 00:50:37,560
Speaker 2: That were you're making your stock call.

1051
00:50:37,840 --> 00:50:42,040
Speaker 3: Yeah no, but seriously, like it it's a big deal.

1052
00:50:42,360 --> 00:50:45,400
Speaker 3: It's going to change the way we work, But is

1053
00:50:45,400 --> 00:50:48,839
Speaker 3: it monetizable? Can you differentiate the actual models? The better

1054
00:50:48,920 --> 00:50:52,440
Speaker 3: the technology gets, like, the easier it is to just

1055
00:50:52,480 --> 00:50:54,960
Speaker 3: do what everyone else is doing. And also like the

1056
00:50:55,040 --> 00:50:57,520
Speaker 3: compute gets cheaper and cheaper. So I just don't know

1057
00:50:57,560 --> 00:50:58,879
Speaker 3: how you monetize this.

1058
00:50:59,040 --> 00:51:01,799
Speaker 2: Well, so that's very interesting his point, which is that

1059
00:51:01,840 --> 00:51:05,560
Speaker 2: it's the tokens are heavily subsidized still, and so that

1060
00:51:05,560 --> 00:51:08,080
Speaker 2: if you're paying and actually using that two hundred dollars

1061
00:51:08,080 --> 00:51:10,880
Speaker 2: max program and you actually use it to the limit,

1062
00:51:11,400 --> 00:51:14,640
Speaker 2: Claude is going to lose money on this and then

1063
00:51:14,680 --> 00:51:18,040
Speaker 2: the prices keep dropping. And I know, like claud code

1064
00:51:18,200 --> 00:51:21,400
Speaker 2: is okay, they're attempting to create something that resembles a

1065
00:51:21,440 --> 00:51:24,400
Speaker 2: traditional software ecosystem that you feel as a user that

1066
00:51:24,440 --> 00:51:28,719
Speaker 2: you're locked into. But so far, in my various like

1067
00:51:29,000 --> 00:51:32,200
Speaker 2: since November twenty twenty two when I started playing with AI,

1068
00:51:32,680 --> 00:51:35,920
Speaker 2: it hasn't felt like anyone has established lock in with anything.

1069
00:51:36,160 --> 00:51:41,120
Speaker 2: And it's very it's very movable, and I suspect even

1070
00:51:41,160 --> 00:51:43,319
Speaker 2: though I have this file now on my desktop that

1071
00:51:43,400 --> 00:51:47,400
Speaker 2: has a file called claud md that gives instructions, et cetera,

1072
00:51:47,800 --> 00:51:50,320
Speaker 2: I'm certain that if I open this file with Codex

1073
00:51:50,400 --> 00:51:52,800
Speaker 2: or Googles, I could probably just pick it up the same.

1074
00:51:52,920 --> 00:51:55,640
Speaker 3: Yeah. I also think there's a fundamental issue with the

1075
00:51:55,640 --> 00:51:58,840
Speaker 3: lock in strategy because when you're talking about technology on

1076
00:51:58,880 --> 00:52:02,719
Speaker 3: the Internet, it just feels very against the grain to

1077
00:52:02,920 --> 00:52:06,600
Speaker 3: try to lock people into anything, and we've seen various

1078
00:52:06,600 --> 00:52:10,319
Speaker 3: projects over the years, and it's a lot harder than

1079
00:52:10,320 --> 00:52:10,840
Speaker 3: it looks.

1080
00:52:11,840 --> 00:52:13,840
Speaker 2: Yeah, I mean, I guess I would say it's a

1081
00:52:13,840 --> 00:52:15,600
Speaker 2: lot harder than it looks. But then we also know

1082
00:52:15,640 --> 00:52:17,680
Speaker 2: the flip side, which is that tons of people are

1083
00:52:17,719 --> 00:52:19,920
Speaker 2: locked into software that they hate, right, Yeah, people are

1084
00:52:20,200 --> 00:52:22,120
Speaker 2: Oh I hate people. How many times have you Oh,

1085
00:52:22,320 --> 00:52:26,080
Speaker 2: I hate Outlook right? Or I hate Microsoft teams and

1086
00:52:26,160 --> 00:52:28,040
Speaker 2: I hate this and I spend money on it every

1087
00:52:28,040 --> 00:52:30,239
Speaker 2: month and my organization can't move off of it, or

1088
00:52:30,239 --> 00:52:32,200
Speaker 2: we can't migrate off of it. So I do think

1089
00:52:32,239 --> 00:52:35,160
Speaker 2: that cuts both ways. I do think he offered the

1090
00:52:35,200 --> 00:52:40,520
Speaker 2: best explanation I've heard of why the AI coding models

1091
00:52:40,960 --> 00:52:43,840
Speaker 2: are a threat to a lot of pretty big software businesses,

1092
00:52:44,160 --> 00:52:47,680
Speaker 2: especially especially the point about how the user never uses

1093
00:52:47,760 --> 00:52:50,680
Speaker 2: all of the features that they actually that the software

1094
00:52:50,680 --> 00:52:53,759
Speaker 2: got built for, and therefore maybe the build versus by

1095
00:52:53,920 --> 00:52:57,560
Speaker 2: calculation really starts to shift when they can just design

1096
00:52:57,600 --> 00:52:58,920
Speaker 2: that one feature very quickly.

1097
00:52:59,560 --> 00:53:02,200
Speaker 3: I totally agree on the software side, it seems like

1098
00:53:02,239 --> 00:53:05,720
Speaker 3: an existential threat, but just like the locked in ecosystem

1099
00:53:05,960 --> 00:53:09,480
Speaker 3: of a particular model. I know he said it's not

1100
00:53:09,520 --> 00:53:12,360
Speaker 3: actually a model, but that seems like a bigger issue

1101
00:53:12,360 --> 00:53:14,120
Speaker 3: to me. I don't know. I guess we'll see.

1102
00:53:14,280 --> 00:53:15,799
Speaker 2: We're going to see and I don't know. I kind

1103
00:53:15,840 --> 00:53:16,880
Speaker 2: of think we're going to see quickly.

1104
00:53:17,200 --> 00:53:20,680
Speaker 3: Yeah, that's again, that's the only certainty is like stuff

1105
00:53:20,719 --> 00:53:21,160
Speaker 3: is happening.

1106
00:53:21,360 --> 00:53:22,120
Speaker 2: What is happening now?

1107
00:53:22,239 --> 00:53:22,439
Speaker 4: Yeah?

1108
00:53:22,520 --> 00:53:23,520
Speaker 3: Okay, shall we leave it there?

1109
00:53:23,600 --> 00:53:24,359
Speaker 2: Let's leave it there.

1110
00:53:24,480 --> 00:53:26,760
Speaker 3: This has been another episode of the Odd Lots podcast.

1111
00:53:26,840 --> 00:53:30,040
Speaker 3: I'm Tracy Alloway. You can follow me at Tracy Alloway.

1112
00:53:29,680 --> 00:53:31,799
Speaker 2: And I'm Jolly Isn't all. You can follow me at

1113
00:53:31,840 --> 00:53:34,719
Speaker 2: the Stalwart. Follow our guest Noah Brier, He's at Hey,

1114
00:53:34,760 --> 00:53:38,600
Speaker 2: It's Noah. Follow our producers Kerman, Rodriguez at Carman Arman, Dashill,

1115
00:53:38,600 --> 00:53:41,400
Speaker 2: ben Att at Dashbot, and Kilbrooks at Kilbrooks. And for

1116
00:53:41,440 --> 00:53:43,960
Speaker 2: more odd Laws content, go to Bloomberg dot com slash

1117
00:53:43,960 --> 00:53:46,640
Speaker 2: odd Lots with a daily newsletter and all of our episodes,

1118
00:53:46,800 --> 00:53:48,680
Speaker 2: and you can chat about all of these topics twenty

1119
00:53:48,719 --> 00:53:51,800
Speaker 2: four to seven in our discord Discord dot gg slash

1120
00:53:51,800 --> 00:53:52,280
Speaker 2: odd Lots.

1121
00:53:52,520 --> 00:53:54,560
Speaker 3: And if you enjoy odd Lots, if you like it

1122
00:53:54,600 --> 00:53:57,719
Speaker 3: when we talk about advances in AI, then please leave

1123
00:53:57,800 --> 00:54:01,400
Speaker 3: us a positive review on your favorite podcast platform and remember,

1124
00:54:01,440 --> 00:54:03,960
Speaker 3: if you are a Bloomberg subscriber, you can listen to

1125
00:54:04,000 --> 00:54:07,080
Speaker 3: all of our episodes absolutely ad free. All you need

1126
00:54:07,120 --> 00:54:09,800
Speaker 3: to do is find the Bloomberg channel on Apple Podcast

1127
00:54:09,880 --> 00:54:28,400
Speaker 3: and follow the instructions there. Thanks for listening.