1
00:00:02,520 --> 00:00:07,000
Speaker 1: Bloomberg Audio Studios, Podcasts, radio news.

2
00:00:08,160 --> 00:00:11,879
Speaker 2: You're listening to Bloomberg business Week with Carol Masser and

3
00:00:11,960 --> 00:00:14,400
Speaker 2: Tim Stenoveek on Bloomberg Radio.

4
00:00:14,920 --> 00:00:16,360
Speaker 3: All right, so we kind of want to stick with

5
00:00:16,520 --> 00:00:19,800
Speaker 3: AI and trying to get to I mean, it's part

6
00:00:19,800 --> 00:00:23,000
Speaker 3: of almost every story that we get to, and a

7
00:00:23,040 --> 00:00:25,680
Speaker 3: fascinating story. It's a most read on the Bloomberg. It's

8
00:00:25,680 --> 00:00:28,120
Speaker 3: buy our business Week team about how the reality of

9
00:00:28,160 --> 00:00:31,040
Speaker 3: AI and its impact and AI reckoning you might say,

10
00:00:31,280 --> 00:00:34,199
Speaker 3: has already arrived for software engineers, and we want to

11
00:00:34,200 --> 00:00:36,680
Speaker 3: get to this story. Mark Millian wrote it. He's Bloomberg

12
00:00:36,680 --> 00:00:39,559
Speaker 3: BusinessWeek deputy editor, joining Tim and me right here in studio.

13
00:00:40,159 --> 00:00:42,479
Speaker 3: You know, we talk about first of all, great story,

14
00:00:42,800 --> 00:00:44,960
Speaker 3: loved it. We caught our attention this morning on our

15
00:00:45,000 --> 00:00:49,839
Speaker 3: reading we talk about ROI and productivity gains. Coders, software

16
00:00:49,880 --> 00:00:53,599
Speaker 3: engineers are already feeling the impact of what AI can

17
00:00:53,680 --> 00:00:56,800
Speaker 3: do for their world. Maybe there's an upside, maybe there's

18
00:00:56,800 --> 00:00:58,840
Speaker 3: a downside, but walk us through what's going on.

19
00:00:59,480 --> 00:01:02,640
Speaker 1: Yeah, I used to live in San Francisco. I covered

20
00:01:02,640 --> 00:01:07,479
Speaker 1: the tech industry for more than fifteen years. I'm living

21
00:01:07,520 --> 00:01:09,640
Speaker 1: here now working on BusinessWeek. But I had just been

22
00:01:09,680 --> 00:01:12,400
Speaker 1: hearing for the last six months from sources and old

23
00:01:12,400 --> 00:01:15,560
Speaker 1: friends out there that this is like the biggest thing

24
00:01:15,800 --> 00:01:18,080
Speaker 1: that's happened in their industry in at least twenty years.

25
00:01:18,920 --> 00:01:20,560
Speaker 1: And I felt like I just had to go out

26
00:01:20,560 --> 00:01:24,080
Speaker 1: there and see and just really understand what was going on,

27
00:01:24,120 --> 00:01:27,440
Speaker 1: because you hear a lot of doomers talking about how

28
00:01:27,440 --> 00:01:29,680
Speaker 1: AI is going to take all the jobs, right and

29
00:01:30,920 --> 00:01:33,720
Speaker 1: here they're saying that because it's taking their job or

30
00:01:34,360 --> 00:01:38,440
Speaker 1: I leave drastically changing their jobs. And so I went

31
00:01:38,440 --> 00:01:41,440
Speaker 1: into anthropic and open AI. I spent time at the

32
00:01:41,560 --> 00:01:44,479
Speaker 1: labs and just met with people who have been doing

33
00:01:44,480 --> 00:01:46,600
Speaker 1: this their whole live, you know, their whole adult lives,

34
00:01:46,600 --> 00:01:49,640
Speaker 1: code learning to code, coding, and now you know, they're

35
00:01:49,640 --> 00:01:54,040
Speaker 1: not writing code anymore. They're just having conversations with claude.

36
00:01:53,880 --> 00:01:57,639
Speaker 2: Meaning they're vibe coding or something different, well, vibe coding

37
00:01:57,760 --> 00:01:59,880
Speaker 2: because like it's not honestarily a name for this. Yeah,

38
00:02:00,040 --> 00:02:02,800
Speaker 2: they look at as a little of a derogatory term.

39
00:02:02,880 --> 00:02:05,160
Speaker 2: So that's more elementary. That's for the folks like me

40
00:02:05,200 --> 00:02:07,440
Speaker 2: who don't actually know how to code exactly, okay.

41
00:02:07,840 --> 00:02:10,160
Speaker 1: And so there have been different terms for most of them.

42
00:02:10,240 --> 00:02:12,720
Speaker 1: Just call it coding, even though they're not actually writing

43
00:02:12,760 --> 00:02:17,120
Speaker 1: the code themselves. But yeah, it's like it's like basically

44
00:02:17,400 --> 00:02:19,760
Speaker 1: telling it what you would have spent the next six

45
00:02:19,800 --> 00:02:22,880
Speaker 1: hours doing. It's like, here's how I want the app

46
00:02:22,919 --> 00:02:23,640
Speaker 1: to be structured.

47
00:02:23,680 --> 00:02:25,560
Speaker 2: Here's okay, So then what do they spend the six

48
00:02:25,560 --> 00:02:26,120
Speaker 2: hours doing?

49
00:02:26,160 --> 00:02:31,519
Speaker 1: Now they spend the six hours building ten different features

50
00:02:31,560 --> 00:02:34,520
Speaker 1: for their app. And so they'll they'll open separate windows

51
00:02:34,520 --> 00:02:36,360
Speaker 1: and saying like, you work on this, you work on this,

52
00:02:36,400 --> 00:02:38,840
Speaker 1: you work in the It's sort of like a product

53
00:02:38,840 --> 00:02:41,320
Speaker 1: manager with which is a job that exists, but like

54
00:02:41,360 --> 00:02:44,760
Speaker 1: those people don't have people who are you know, going

55
00:02:44,840 --> 00:02:47,519
Speaker 1: to produce things as quickly, you know, within ten minutes.

56
00:02:47,560 --> 00:02:49,560
Speaker 1: That would take a person, you know, a couple of days.

57
00:02:49,680 --> 00:02:52,680
Speaker 3: Mark tell us about Boris Journey out in San Francisco.

58
00:02:52,760 --> 00:02:55,840
Speaker 3: He created the software cloud Code runs it at Ananthropic,

59
00:02:56,200 --> 00:02:58,840
Speaker 3: and in your reporting you say that at Anthropic the

60
00:02:58,880 --> 00:03:01,160
Speaker 3: expectation that everybody at least a little bit of a

61
00:03:01,200 --> 00:03:03,640
Speaker 3: coder has become endemic, and that mindset is beginning to

62
00:03:03,639 --> 00:03:05,640
Speaker 3: infect much of the tech industry. Tell us little bit

63
00:03:05,639 --> 00:03:07,920
Speaker 3: more about I'm assuming you chatted with him. You talked

64
00:03:07,919 --> 00:03:08,800
Speaker 3: with him, Yeah.

65
00:03:08,639 --> 00:03:10,200
Speaker 1: I spent a bunch of time with Boris. He's a

66
00:03:10,240 --> 00:03:13,240
Speaker 1: really interesting guy. He's like a coder's coder, and so

67
00:03:13,360 --> 00:03:15,560
Speaker 1: he's like this a little bit of a guru that

68
00:03:15,639 --> 00:03:18,919
Speaker 1: like people follow him on their podcast and his ex

69
00:03:19,000 --> 00:03:22,040
Speaker 1: post just like figure out, like, how is my job

70
00:03:22,120 --> 00:03:25,080
Speaker 1: going to change tomorrow? Like Anthropic is releasing stuff almost

71
00:03:25,120 --> 00:03:28,040
Speaker 1: every single day. But he has, you know, like a

72
00:03:28,120 --> 00:03:31,400
Speaker 1: long background in the industry. He was the head, the

73
00:03:31,480 --> 00:03:34,560
Speaker 1: leader of code quality at Meta, so like he kind

74
00:03:34,600 --> 00:03:37,160
Speaker 1: of set the standards for how Meta developed Facebook and

75
00:03:37,160 --> 00:03:41,200
Speaker 1: Instagram and worked on the server architecture and Instagram and

76
00:03:41,360 --> 00:03:45,720
Speaker 1: he's like, you know, just sort of articulated what this

77
00:03:45,800 --> 00:03:48,920
Speaker 1: new world looks like. And it's changing even for him

78
00:03:49,160 --> 00:03:51,560
Speaker 1: every week, you like, instead of actually telling it what

79
00:03:51,640 --> 00:03:55,120
Speaker 1: to do now, he just like creates his broad framework

80
00:03:55,200 --> 00:03:57,320
Speaker 1: for his AI and it's just like go spend the

81
00:03:57,360 --> 00:04:00,000
Speaker 1: next like several days working on this challenge.

82
00:04:00,000 --> 00:04:02,880
Speaker 2: Well, what does this mean for the what we've seen

83
00:04:02,920 --> 00:04:05,839
Speaker 2: with the talent pipeline in Silicon Valley over the last

84
00:04:05,840 --> 00:04:07,600
Speaker 2: twenty or thirty years or so, It comes to mind.

85
00:04:07,640 --> 00:04:10,200
Speaker 2: I was actually speaking with a Bloomberg intern. We have

86
00:04:10,240 --> 00:04:12,480
Speaker 2: a great class of interns here he's not on the

87
00:04:12,480 --> 00:04:15,160
Speaker 2: media side, and I asked what he was studying. He said, well,

88
00:04:15,600 --> 00:04:18,360
Speaker 2: I'm now studying cybersecurity, but I spent my first year

89
00:04:18,560 --> 00:04:21,560
Speaker 2: majoring in computer science and programming, and he saw the

90
00:04:21,600 --> 00:04:24,080
Speaker 2: writing on the wall he decided to switch his major

91
00:04:24,080 --> 00:04:24,680
Speaker 2: as a result.

92
00:04:25,000 --> 00:04:30,480
Speaker 1: Yes, it's been a c change within especially entry level

93
00:04:30,760 --> 00:04:33,200
Speaker 1: jobs in this industry, and it's still so new that

94
00:04:33,240 --> 00:04:35,159
Speaker 1: it's hard to know how it's really going to shake out.

95
00:04:35,320 --> 00:04:37,760
Speaker 1: There's plenty of optimists who say there will just be

96
00:04:37,839 --> 00:04:41,160
Speaker 1: more demand than ever for software and people who understand

97
00:04:41,560 --> 00:04:44,919
Speaker 1: how software fits together. I talked to somebody at Carnegie

98
00:04:44,960 --> 00:04:50,839
Speaker 1: Mellon who's working on helping restructure their computer science program.

99
00:04:51,120 --> 00:04:55,760
Speaker 1: But the early science have not been great. So since

100
00:04:56,000 --> 00:04:59,400
Speaker 1: chat CHIPAD came out in twenty twenty two, coding jobs

101
00:04:59,400 --> 00:05:04,039
Speaker 1: for people in twenties is down twenty percent, and it's

102
00:05:04,080 --> 00:05:06,200
Speaker 1: a lot of the first in, first out. At the

103
00:05:06,200 --> 00:05:10,000
Speaker 1: big companies, they're they're doing mass layoffs. Some of them

104
00:05:10,080 --> 00:05:12,760
Speaker 1: are blaming AI when it's really maybe not AI's fault,

105
00:05:12,839 --> 00:05:14,960
Speaker 1: they just maybe over hired during COVID or trying to

106
00:05:14,960 --> 00:05:18,839
Speaker 1: correct that. But for a lot of companies, they're true

107
00:05:18,880 --> 00:05:21,440
Speaker 1: believers that this is a new way of working.

108
00:05:21,640 --> 00:05:26,080
Speaker 3: Measuring productivity or ROI, though, is tricky even among these

109
00:05:26,120 --> 00:05:29,720
Speaker 3: folks who maybe are finding they can, you know, basically

110
00:05:30,080 --> 00:05:32,120
Speaker 3: use AI and talk to it and say go for

111
00:05:32,160 --> 00:05:33,920
Speaker 3: five or six hours and do something and then maybe

112
00:05:34,000 --> 00:05:37,120
Speaker 3: do another one somewhere else. But it's still tricky.

113
00:05:37,600 --> 00:05:41,359
Speaker 1: It's true everything based on my reporting leads me to

114
00:05:41,400 --> 00:05:46,240
Speaker 1: believe that this dramatically increases productivity for a software engineer,

115
00:05:46,600 --> 00:05:49,960
Speaker 1: but finding actual data to support that has been very difficult.

116
00:05:49,960 --> 00:05:53,560
Speaker 1: There's one study that was highly respected that took place

117
00:05:53,600 --> 00:05:57,200
Speaker 1: early last year before this was like real, before AI

118
00:05:57,320 --> 00:05:59,600
Speaker 1: coding was really a common thing, and that the models

119
00:05:59,600 --> 00:06:02,960
Speaker 1: weren't as good and they found that well, people thought

120
00:06:03,000 --> 00:06:05,680
Speaker 1: they were more productive, but they're not actually more productive.

121
00:06:06,000 --> 00:06:09,119
Speaker 1: But there was this moment in time last November when

122
00:06:09,160 --> 00:06:13,800
Speaker 1: when Claude released a new model that really was capable

123
00:06:12,960 --> 00:06:17,640
Speaker 1: of replacing parts of people's jobs, and Chatchab defouled with

124
00:06:17,680 --> 00:06:21,120
Speaker 1: a similar model. They couldn't run the experiment again because

125
00:06:21,200 --> 00:06:23,200
Speaker 1: all the software most of the software engineers that went

126
00:06:23,240 --> 00:06:25,479
Speaker 1: to like, we don't want to handcode anymore, Like we

127
00:06:25,800 --> 00:06:27,360
Speaker 1: got a better way. We're not going to waste our

128
00:06:27,400 --> 00:06:28,680
Speaker 1: time for your experiment.

129
00:06:29,839 --> 00:06:32,040
Speaker 2: What's the thing now that you saw in Silicon Valley

130
00:06:32,040 --> 00:06:33,800
Speaker 2: that still can't be done by Claude, that people are

131
00:06:33,839 --> 00:06:36,560
Speaker 2: having to do that maybe a year from now Claude

132
00:06:36,560 --> 00:06:37,719
Speaker 2: will be able to do. You know.

133
00:06:37,839 --> 00:06:41,520
Speaker 1: There were some examples of like people who are coding

134
00:06:41,800 --> 00:06:47,359
Speaker 1: on very like outdated hardware that had severe limitations that Claude,

135
00:06:47,480 --> 00:06:51,760
Speaker 1: which is trained on incredible amounts of code bases to

136
00:06:51,880 --> 00:06:55,560
Speaker 1: understand the best practices. There are like very niche types

137
00:06:55,600 --> 00:06:59,120
Speaker 1: of electronics that require a human touch. But for the

138
00:06:59,120 --> 00:07:02,000
Speaker 1: most part, those people, they would use Claude, they would

139
00:07:02,000 --> 00:07:04,200
Speaker 1: find errors, they'd fix the errors themselves, but they're still

140
00:07:04,279 --> 00:07:05,360
Speaker 1: using Claude and Codex.

141
00:07:05,400 --> 00:07:07,440
Speaker 3: It's almost shocking how quickly we are here. I know

142
00:07:07,480 --> 00:07:09,280
Speaker 3: we've been talking about three years, but it does still

143
00:07:09,279 --> 00:07:11,360
Speaker 3: feel like pretty shocking.

144
00:07:11,640 --> 00:07:13,960
Speaker 1: Mark glad we're still here right now.

145
00:07:14,760 --> 00:07:19,000
Speaker 3: Sorry, Mark, thank you, great story. Mark Millian Bloomberg BusinessWeek

146
00:07:19,320 --> 00:07:22,360
Speaker 3: deputy editor. Check out his story. It's on the terminal