1
00:00:01,040 --> 00:00:04,440
Speaker 1: What if every morning, instead of drowning in a sea

2
00:00:04,640 --> 00:00:09,000
Speaker 1: of newsletters, tweets, and podcasts, you had the day's biggest

3
00:00:09,080 --> 00:00:14,400
Speaker 1: breakthroughs already sorted for you. That's what Morgan Brown, vice

4
00:00:14,440 --> 00:00:18,200
Speaker 1: president of Product and Growth at drop Box, built for himself.

5
00:00:18,840 --> 00:00:20,880
Speaker 1: By the time he sits down with his coffee, an

6
00:00:20,960 --> 00:00:27,280
Speaker 1: AI workflow has already scanned the landscape of research, papers, substacks, podcasts,

7
00:00:27,320 --> 00:00:31,640
Speaker 1: and even YouTube. It spits out just three can't miss signals,

8
00:00:32,040 --> 00:00:35,839
Speaker 1: ranked from world changing to incremental, and explains what they

9
00:00:35,880 --> 00:00:40,320
Speaker 1: actually mean for his role. In today's quick Win, Morgan

10
00:00:40,400 --> 00:00:43,760
Speaker 1: shares the simple steps he used to design that system,

11
00:00:44,200 --> 00:00:48,879
Speaker 1: starting with one prompt, refining and running it daily. By

12
00:00:48,920 --> 00:00:51,000
Speaker 1: the end, you'll know how to set up your own

13
00:00:51,120 --> 00:00:55,560
Speaker 1: version to cut through the noise and actually stay ahead.

14
00:01:00,720 --> 00:01:04,440
Speaker 1: Welcome to How I Work, a show about habits, rituals,

15
00:01:04,480 --> 00:01:08,399
Speaker 1: and strategies for optimizing your day. I'm your host, doctor

16
00:01:08,440 --> 00:01:15,520
Speaker 1: Amantha Imber. Still staying on the topic of I guess

17
00:01:15,680 --> 00:01:19,080
Speaker 1: reducing or eliminating some of the grunt work through AI,

18
00:01:19,800 --> 00:01:23,080
Speaker 1: I'd love to know if they're more on the automation

19
00:01:23,440 --> 00:01:27,440
Speaker 1: or agentic AI side of things. What some of the

20
00:01:28,440 --> 00:01:31,160
Speaker 1: I guess other tasks that you've been able to automate

21
00:01:31,200 --> 00:01:32,800
Speaker 1: in interesting ways would be.

22
00:01:32,840 --> 00:01:35,360
Speaker 2: I've really done a lot of automation around trying to

23
00:01:35,920 --> 00:01:38,559
Speaker 2: understand what's changing around me. So I have a daily

24
00:01:39,319 --> 00:01:44,679
Speaker 2: workflow that reviews all new kind of substack newsletters, important

25
00:01:44,680 --> 00:01:49,200
Speaker 2: tweets across a large domain of AI and mL ex accounts,

26
00:01:49,800 --> 00:01:55,200
Speaker 2: reviews Archive, which is the scientific paper repository for new

27
00:01:55,400 --> 00:01:59,800
Speaker 2: mL and artificial intelligence papers. It scans the whole landscape

28
00:02:00,080 --> 00:02:04,480
Speaker 2: to Spotify podcasts and basically takes all of the day's

29
00:02:04,600 --> 00:02:08,960
Speaker 2: information about AI and mL reviews. It all categorizes it

30
00:02:09,040 --> 00:02:15,840
Speaker 2: into world changing, interesting, incremental, and noise, and then summarizes

31
00:02:15,880 --> 00:02:19,000
Speaker 2: the top three quote unquote world changing potentials and then

32
00:02:19,040 --> 00:02:21,360
Speaker 2: summarizes it for me in the context of my role

33
00:02:21,760 --> 00:02:24,680
Speaker 2: at DASH, where it kind of talks about hey, things

34
00:02:24,720 --> 00:02:27,360
Speaker 2: you might consider as the head of product for Dash. So,

35
00:02:27,400 --> 00:02:30,840
Speaker 2: for example, Anthropic the other day just announced that claud

36
00:02:30,880 --> 00:02:35,120
Speaker 2: can handle now a million token context window, which effectively

37
00:02:35,160 --> 00:02:38,919
Speaker 2: means that for many reasonable size software codebases, it can

38
00:02:39,000 --> 00:02:41,640
Speaker 2: reason about the entire codebase, which is a pretty big

39
00:02:41,639 --> 00:02:44,679
Speaker 2: and important milestone for software development. And the other one

40
00:02:44,720 --> 00:02:47,560
Speaker 2: is I automate my personal email I can't imagine you're

41
00:02:47,680 --> 00:02:50,560
Speaker 2: inbox Atmantha, but it gets hammered all the time. And

42
00:02:50,600 --> 00:02:55,639
Speaker 2: so I built a little agent using claud code to

43
00:02:55,720 --> 00:03:00,679
Speaker 2: read my personal Gmail every day, categorize it into newsletters,

44
00:03:01,240 --> 00:03:05,960
Speaker 2: spam priority emails, summarize the newsletters for me, flag the

45
00:03:05,960 --> 00:03:09,600
Speaker 2: messages that I want to respond to, and draft responses

46
00:03:10,160 --> 00:03:14,480
Speaker 2: for those newsletters or not for those emails. So really

47
00:03:14,480 --> 00:03:18,839
Speaker 2: trying to kind of like manage my inbound email automatically

48
00:03:18,840 --> 00:03:19,200
Speaker 2: as well.

49
00:03:19,240 --> 00:03:21,480
Speaker 1: Wow, okay, I love the both those examples. I want

50
00:03:21,520 --> 00:03:24,120
Speaker 1: to dig a bit deeper into how you went about

51
00:03:24,280 --> 00:03:30,560
Speaker 1: creating this automation that summarizes all the AI news of

52
00:03:30,600 --> 00:03:34,160
Speaker 1: the day. I mean, that sounds so powerful. Can I

53
00:03:34,200 --> 00:03:36,160
Speaker 1: buy that from you? Morgan? Are you selling it?

54
00:03:37,440 --> 00:03:38,720
Speaker 2: You can link it in the show notes?

55
00:03:39,600 --> 00:03:43,040
Speaker 1: Wow? Amazing handed over? Yeah wow. Talk me through the

56
00:03:43,080 --> 00:03:45,120
Speaker 1: process of how you put that together.

57
00:03:45,360 --> 00:03:47,040
Speaker 2: First of all, I realized that there were a few

58
00:03:47,160 --> 00:03:51,640
Speaker 2: key sources of information that I was constantly referring to personally,

59
00:03:51,720 --> 00:03:53,680
Speaker 2: So you know, whenever I had a free moment, I

60
00:03:53,720 --> 00:03:56,840
Speaker 2: would go to archive and kind of browse the actual

61
00:03:57,360 --> 00:04:01,360
Speaker 2: published papers because that's the real source of truth cutting edge.

62
00:04:01,600 --> 00:04:05,200
Speaker 2: I would see a lot of popular posts on x

63
00:04:05,480 --> 00:04:08,760
Speaker 2: about kind of new research breakthroughs. I follow some of

64
00:04:08,760 --> 00:04:11,280
Speaker 2: the key people like the sam Altman's of the World

65
00:04:11,320 --> 00:04:14,920
Speaker 2: and the Frontier Labs. And then there's also several great

66
00:04:15,040 --> 00:04:19,360
Speaker 2: substack newsletters around AIS that I read as much as

67
00:04:19,360 --> 00:04:21,599
Speaker 2: I can. But I just realized there was no way

68
00:04:21,600 --> 00:04:24,560
Speaker 2: for me to stay on top of that volume consistently,

69
00:04:24,680 --> 00:04:27,560
Speaker 2: and I just started with a prompt. I told chat TBT,

70
00:04:27,640 --> 00:04:29,839
Speaker 2: I'd like to create a prompt to scan the entire

71
00:04:30,440 --> 00:04:35,560
Speaker 2: AIML space for important signals around research and development and

72
00:04:35,600 --> 00:04:39,440
Speaker 2: industry news that pertains to my role at Dropbox. And

73
00:04:39,480 --> 00:04:42,760
Speaker 2: I said before X, I want to look across these

74
00:04:42,880 --> 00:04:45,120
Speaker 2: accounts and accounts like it. So I kind of gave

75
00:04:45,160 --> 00:04:47,159
Speaker 2: it a seed list of accounts, but asked it to

76
00:04:47,279 --> 00:04:51,039
Speaker 2: expand where it could. I gave it specific newsletters that

77
00:04:51,120 --> 00:04:55,000
Speaker 2: I read on substack, but also again asked to consider adjacents,

78
00:04:55,440 --> 00:04:59,200
Speaker 2: and then I gave it specific categories on archive where

79
00:04:59,240 --> 00:05:02,720
Speaker 2: those papers are housed and asked it to consider relevant

80
00:05:02,760 --> 00:05:05,320
Speaker 2: ones that there. And then I said, I'd like to

81
00:05:05,960 --> 00:05:08,760
Speaker 2: really filter high signal, like I don't want this to

82
00:05:08,800 --> 00:05:11,560
Speaker 2: be a laundry list of everything that happened that when

83
00:05:11,640 --> 00:05:14,120
Speaker 2: actually helped me kind of consume it any better. So

84
00:05:14,160 --> 00:05:16,640
Speaker 2: I said, help me come up with a scale to

85
00:05:16,960 --> 00:05:20,000
Speaker 2: rank the innovations and the news that's coming out in

86
00:05:20,040 --> 00:05:22,400
Speaker 2: the day, and kind of categorize things as you can't

87
00:05:22,400 --> 00:05:26,880
Speaker 2: miss these are like foundational important, shifts incremental and important,

88
00:05:27,160 --> 00:05:29,960
Speaker 2: and then give me some synthesis around Hey, when you

89
00:05:29,960 --> 00:05:33,279
Speaker 2: put these signals together, what does it mean for my role?

90
00:05:33,839 --> 00:05:37,320
Speaker 2: So I fed that all just in basically talking to

91
00:05:37,440 --> 00:05:40,840
Speaker 2: chat GPT. I like to use the audio input to

92
00:05:40,920 --> 00:05:43,560
Speaker 2: kind of just stream of consciousness my thoughts into it.

93
00:05:43,640 --> 00:05:45,039
Speaker 2: Once I did that, I asked it to kind of

94
00:05:45,040 --> 00:05:48,320
Speaker 2: structure it into a prompt. I reviewed the prompt. Some

95
00:05:48,360 --> 00:05:50,719
Speaker 2: of the things I didn't like. I always try to

96
00:05:50,760 --> 00:05:53,279
Speaker 2: go back and forth with it. I think you should

97
00:05:53,320 --> 00:05:55,120
Speaker 2: kind of refine it much like you would the copy

98
00:05:55,160 --> 00:05:58,160
Speaker 2: for a website or you know, the script for something.

99
00:05:58,600 --> 00:06:00,800
Speaker 2: And then once I got the prompt too you get start.

100
00:06:01,080 --> 00:06:04,039
Speaker 2: I said, Okay, let's run that for today and give

101
00:06:04,080 --> 00:06:05,880
Speaker 2: me the output. And I got the output and I

102
00:06:05,920 --> 00:06:07,719
Speaker 2: was like, oh, this is a little too long. It

103
00:06:07,720 --> 00:06:10,560
Speaker 2: could be a little shorter. It looks like we're underrepresenting,

104
00:06:10,600 --> 00:06:13,680
Speaker 2: we're missing YouTube. Let's try to add YouTube in, so

105
00:06:13,720 --> 00:06:15,000
Speaker 2: on and so forth, and then I finally got it

106
00:06:15,040 --> 00:06:17,320
Speaker 2: dialed in, and then I said, great, set this as

107
00:06:17,320 --> 00:06:20,479
Speaker 2: a daily task every day five am Pacific. Run it

108
00:06:20,520 --> 00:06:23,200
Speaker 2: again while I'm having coffee. I just kind of read

109
00:06:23,200 --> 00:06:23,560
Speaker 2: through it.

110
00:06:24,680 --> 00:06:27,400
Speaker 1: What I love about Morgan's approach is that it didn't

111
00:06:27,400 --> 00:06:30,520
Speaker 1: start with coding or fancy tools. It started with a

112
00:06:30,600 --> 00:06:34,200
Speaker 1: simple question, what sources do I keep checking and how

113
00:06:34,240 --> 00:06:36,600
Speaker 1: could I pull them together? So the next time you

114
00:06:36,640 --> 00:06:40,640
Speaker 1: feel buried under articles or updates, try listing your key sources,

115
00:06:40,960 --> 00:06:43,760
Speaker 1: feed them into a prompt and ask AI to filter

116
00:06:43,920 --> 00:06:47,080
Speaker 1: and rank what matters most. And of course you'll find

117
00:06:47,120 --> 00:06:50,920
Speaker 1: the full episode with Morgan linked in the show notes.

118
00:06:51,920 --> 00:06:54,640
Speaker 1: If you like today's show, make sure you git follow

119
00:06:54,800 --> 00:06:58,279
Speaker 1: on your podcast app to be alerted when new episodes drop.

120
00:06:58,839 --> 00:07:01,320
Speaker 1: How I Work was recorded on the traditional land of

121
00:07:01,360 --> 00:07:03,560
Speaker 1: the Warringery people, part of the Coulan nation.