Meeting Title: Gabe <> Sam - Internal AI Platform Sync Date: 2026-01-21 Meeting participants: Samuel Roberts, Gabriel Lam


WEBVTT

1 00:00:39.290 00:00:40.220 Samuel Roberts: A.

2 00:00:42.070 00:00:43.250 Gabriel Lam: Hello!

3 00:00:45.260 00:00:46.730 Samuel Roberts: How are you doing?

4 00:00:46.730 00:00:49.890 Gabriel Lam: I’m doing well. I am…

5 00:00:50.730 00:00:57.220 Gabriel Lam: My schedule is crazy, so I’m, like, on-site, like, 3 days a week for another commitment.

6 00:00:57.430 00:01:00.080 Gabriel Lam: And, like, really hard.

7 00:01:00.080 00:01:00.420 Samuel Roberts: Oh.

8 00:01:00.420 00:01:01.830 Gabriel Lam: juggle things.

9 00:01:02.290 00:01:04.900 Samuel Roberts: Jeez, yeah, where… where about… where are you right now?

10 00:01:05.200 00:01:07.649 Gabriel Lam: I’m in Boston. Okay. Okay.

11 00:01:08.880 00:01:20.489 Gabriel Lam: But, yeah, I told Lutamin, I’m like, hey, my hours are gonna go down, and we’re working it out the exact schedule, and it’s like… I’m, like, jumping into meeting rooms sometimes, or just trying to find a place.

12 00:01:20.490 00:01:27.499 Samuel Roberts: That’s crazy. Oh, man. So, like, I, like, in person, on site, is what you said? Is that what you said?

13 00:01:27.500 00:01:28.940 Gabriel Lam: Some days, yeah.

14 00:01:28.940 00:01:32.690 Samuel Roberts: Jeez, oh man, I haven’t gone into an office in so long.

15 00:01:33.560 00:01:38.610 Gabriel Lam: But yeah, I sort of just wanted to…

16 00:01:39.140 00:01:48.430 Gabriel Lam: sort of do a little brainstorming session. Also, like, I had… sorry, one second…

17 00:01:53.630 00:01:57.919 Gabriel Lam: Yeah, sort of like a quick demo or workflow. I think the main…

18 00:01:58.410 00:02:04.320 Gabriel Lam: thing that I already have so far are… I have, like, a PRD written out.

19 00:02:04.780 00:02:11.489 Gabriel Lam: For… Like, this mini-podcast idea.

20 00:02:12.380 00:02:18.510 Gabriel Lam: And basically, like, being able to extract information from…

21 00:02:18.680 00:02:23.980 Gabriel Lam: You know, a resources folder, whether it’s, like, articles, whether it’s…

22 00:02:25.700 00:02:31.000 Gabriel Lam: Transcripts, whether it’s research, and just be able to sort of digest it and put it in a format.

23 00:02:31.300 00:02:34.870 Gabriel Lam: We already have done… a…

24 00:02:35.350 00:02:42.329 Gabriel Lam: you know, our case study assistant, and stuff like that there is a sort of voice component that we’ve done before, but I guess my question there would be, like.

25 00:02:42.850 00:02:46.930 Gabriel Lam: In terms of a sort of final output, do you imagine it to be…

26 00:02:47.230 00:02:54.010 Gabriel Lam: a sort of additional, you know, tab on the platform, where… like…

27 00:02:55.990 00:02:58.780 Gabriel Lam: Sort of, like, how our marketing assets already has.

28 00:02:58.970 00:03:03.879 Gabriel Lam: Everything in a certain structure that, like, gets displayed, or do you sort of imagine it as, like, a…

29 00:03:04.690 00:03:15.120 Gabriel Lam: plug and… not a plug and play, but like a drag and drop, or… I guess maybe, on your end, imagining what would be the easiest lift, given what we’ve already accomplished so far.

30 00:03:16.760 00:03:24.410 Samuel Roberts: Yeah, we… I think… So… what…

31 00:03:24.590 00:03:29.839 Samuel Roberts: So, I mean, I guess talk me through the PRD a little bit, like, what is the… what are we summarizing? Like, what is the source?

32 00:03:30.540 00:03:33.270 Samuel Roberts: For the most part. Like, meetings, transcripts?

33 00:03:34.370 00:03:35.200 Gabriel Lam: So there’s…

34 00:03:35.290 00:03:37.629 Samuel Roberts: Go ahead. Yeah, yeah, there’s, there’s…

35 00:03:37.630 00:03:41.019 Gabriel Lam: It’s currently 3 use cases, I’m just trying to bring it up.

36 00:03:41.020 00:03:41.420 Samuel Roberts: Sure.

37 00:03:41.430 00:03:50.610 Gabriel Lam: There’s currently 3 use cases. The main use case seems to be for…

38 00:03:52.500 00:03:55.660 Gabriel Lam: Like, articles or, or, or research?

39 00:03:55.790 00:03:59.440 Gabriel Lam: That I think UTAM or, like, other people might be interested in.

40 00:03:59.880 00:04:10.650 Gabriel Lam: I think another part to me that seems, like, lower priority, but something that he brought up, was being able to then also digest, like, meeting transcripts, or…

41 00:04:10.770 00:04:16.670 Gabriel Lam: meeting, like, agendas or summaries, just because he doesn’t have a lot of time. I personally feel like

42 00:04:16.810 00:04:24.870 Gabriel Lam: Fat is maybe… More of, like, a preference, as opposed to, like, truly… urgent.

43 00:04:25.550 00:04:29.189 Gabriel Lam: I don’t know how it is, like, you, yourself, like to…

44 00:04:29.690 00:04:39.730 Gabriel Lam: you know, ingest information, or if, like, you’re like, hey, I need to prepare for this meeting coming up, like, I’ll look on the signup assistant, or I’ll look on, like, you know, the last couple meetings.

45 00:04:39.900 00:04:44.360 Gabriel Lam: And bring them into context, and then try to find something from there, right?

46 00:04:44.490 00:04:50.210 Gabriel Lam: But I do think That sort of research component is something that we could… Create.

47 00:04:53.210 00:05:00.820 Samuel Roberts: Okay, yeah. Yeah, sorry, the reason I’m asking is because I was just trying to think, like, from, like, a UX, like, side, is it going to be, like, you know.

48 00:05:00.920 00:05:07.360 Samuel Roberts: generate me some… like, is there just, like, a button on a meeting, or is it, like, drag and drop, kind of the way you were saying?

49 00:05:08.520 00:05:10.529 Samuel Roberts: And yeah, those are kind of different…

50 00:05:12.660 00:05:18.700 Samuel Roberts: I mean, the underlying thing would be the same, just, like, taking in a bunch of… Text, I guess.

51 00:05:19.150 00:05:19.760 Samuel Roberts: Right.

52 00:05:20.300 00:05:21.840 Gabriel Lam: And so when it’s.

53 00:05:21.840 00:05:23.700 Samuel Roberts: Rousing to a… yeah, okay.

54 00:05:23.900 00:05:35.230 Gabriel Lam: Yeah, would it be, sort of, like, if you had a bunch of, I don’t know, let’s say there’s multiple formats, right? And so, like, is that also going to be a problem where we have to convert,

55 00:05:36.140 00:05:44.399 Gabriel Lam: like… web pages, or, like, PDFs into, like, markdowns to be read, or…

56 00:05:44.660 00:05:47.930 Gabriel Lam: Do you feel like that’s not gonna be as much of a problem?

57 00:05:50.100 00:05:58.490 Samuel Roberts: I mean, some of the models can handle PDFs, so it might not be a huge issue there. Web pages, I’m sure there’s ways to…

58 00:05:59.180 00:06:02.330 Samuel Roberts: Well, depending on how they, you know, protect their…

59 00:06:03.540 00:06:07.229 Samuel Roberts: Articles and stuff, you know, like, sometimes it only loads, like, half an article or something.

60 00:06:07.870 00:06:12.589 Samuel Roberts: But there are ways to, like, sort out what the article is from the page.

61 00:06:14.020 00:06:19.509 Samuel Roberts: Not too worried about that. The PDF stuff, maybe, depending on which model we go with.

62 00:06:21.790 00:06:30.900 Samuel Roberts: Yeah. Yeah, I like the idea of dropping in Whatever it is.

63 00:06:32.920 00:06:36.710 Samuel Roberts: As, like, an initial… like, V1?

64 00:06:37.780 00:06:42.330 Samuel Roberts: You know, if we wanted to, say, like, import these meetings, we could do something like that, maybe?

65 00:06:43.170 00:06:46.840 Samuel Roberts: You know, because I think…

66 00:06:47.130 00:06:54.870 Samuel Roberts: Yeah, I think there’s a, you know, you kind of outlined, like, a few different… ways… and sources…

67 00:06:59.030 00:07:00.780 Samuel Roberts: I mean, I think another… like, I don’t…

68 00:07:01.200 00:07:04.530 Samuel Roberts: Have a great sense of how best to generate a, like.

69 00:07:04.800 00:07:09.690 Samuel Roberts: output conversation. I really… I guess it’s really just, like, a transcript that has to get

70 00:07:10.260 00:07:12.839 Samuel Roberts: Ex-speech the right way?

71 00:07:13.240 00:07:13.810 Gabriel Lam: Right.

72 00:07:14.540 00:07:22.139 Gabriel Lam: Right, so it’s… there’s the, like, okay, what’s the information from the content that we’re trying to turn into

73 00:07:23.110 00:07:28.700 Gabriel Lam: And then, how do you turn, you know, if it’s an article, then, like, turn that into a conversation?

74 00:07:29.530 00:07:30.709 Samuel Roberts: Yeah, I mean, so this is, this is…

75 00:07:30.710 00:07:31.990 Gabriel Lam: Text to speech.

76 00:07:31.990 00:07:35.930 Samuel Roberts: Yeah, that’s where I… and I don’t know how that will sound. I don’t know how, like.

77 00:07:36.920 00:07:47.959 Samuel Roberts: you know, obviously, like, text-to-speech is pretty easy if you’re just, like, getting it to read something, but as a conversation, I know there’s, like, ums and ahs and all kinds of little things to make it sound more real. I’m wondering…

78 00:07:50.090 00:07:54.590 Samuel Roberts: So this is Notebook LN that was the original kind of inspiration for this, right?

79 00:07:54.590 00:07:55.410 Gabriel Lam: Yeah.

80 00:07:55.410 00:07:56.560 Samuel Roberts: Yeah, I’m wondering…

81 00:07:56.560 00:07:58.479 Gabriel Lam: so… Yeah, go ahead, Doc.

82 00:07:58.690 00:08:01.230 Samuel Roberts: No, I was just wondering what’s out there that does this.

83 00:08:01.230 00:08:04.190 Gabriel Lam: Yeah, I was also thinking through, like, we have…

84 00:08:04.880 00:08:08.170 Gabriel Lam: For example, ChatGPT has a voice mode that I’ve been using.

85 00:08:08.480 00:08:08.940 Samuel Roberts: Oh, yeah.

86 00:08:08.940 00:08:11.560 Gabriel Lam: Where I sort of just, like, chat with it, and…

87 00:08:11.560 00:08:12.260 Samuel Roberts: Yeah, yeah.

88 00:08:12.260 00:08:19.269 Gabriel Lam: if I’m, like, not at my desk, or I’m, like, hands-free, and then, like, something that I was thinking is… is that…

89 00:08:19.810 00:08:23.459 Gabriel Lam: like, the reason why we’re using cursor is for all the…

90 00:08:23.960 00:08:26.570 Gabriel Lam: all the sort of RAG capabilities.

91 00:08:27.440 00:08:30.560 Gabriel Lam: And… I believe…

92 00:08:31.520 00:08:42.319 Gabriel Lam: I’ve actually never fully tested this, so please correct me if I’m wrong, but I don’t know if ChatGPT has that same… like, if I input a bunch of files, it’s able to talk over all those files in real time.

93 00:08:44.460 00:08:47.649 Samuel Roberts: Yeah, I bet there’s… I mean, there’s definitely…

94 00:08:47.750 00:08:51.610 Samuel Roberts: you can upload things, and it can analyze images and things like that. I don’t know…

95 00:08:52.170 00:08:58.059 Samuel Roberts: What the limit is when you hit that, because obviously, like, parsing those are probably a lot more tokens than just text.

96 00:08:59.170 00:09:03.720 Samuel Roberts: And so for, like, ChatGPT, I don’t have a ton of insight into, like, how they’re doing that versus, like, the API.

97 00:09:03.840 00:09:04.680 Samuel Roberts: But…

98 00:09:09.170 00:09:15.559 Samuel Roberts: Sorry, I’m a little distracted, because I found a generate podcast API method on the Notebook LM.

99 00:09:15.880 00:09:16.960 Gabriel Lam: Oh, interesting.

100 00:09:16.960 00:09:18.440 Samuel Roberts: documentation.

101 00:09:22.420 00:09:23.690 Samuel Roberts: Interesting.

102 00:09:25.080 00:09:26.790 Samuel Roberts: That might be a worth…

103 00:09:26.950 00:09:32.670 Samuel Roberts: trying thing, because that just is… you know, I mean, I’m not… I’m sorry, I’m getting distracted here, just…

104 00:09:32.970 00:09:34.129 Gabriel Lam: They’re gonna talk about that.

105 00:09:34.130 00:09:40.649 Samuel Roberts: Because, like, you know, there’s a lot that could go into this, and there’s a lot of, like, refining to make it sound…

106 00:09:40.980 00:09:42.790 Samuel Roberts: good, and I’m wondering…

107 00:09:42.790 00:09:43.380 Gabriel Lam: Yeah.

108 00:09:43.380 00:09:45.630 Samuel Roberts: How much that’s already baked into something like this.

109 00:09:46.420 00:09:48.160 Samuel Roberts: But anyway,

110 00:09:52.950 00:09:55.740 Samuel Roberts: Sorry, we were talking cursor, chat GPT…

111 00:09:56.540 00:10:09.829 Gabriel Lam: So there’s, like, the… there’s the LLM… there’s a notebook LM voice podcast, which was something we’re looking at. I think Pranav had brought up this idea of, like, the sort of ChatGPT voice mode.

112 00:10:09.830 00:10:17.020 Samuel Roberts: Oh, yes, yes, yes, okay, I was only… okay, I forgot about that, that’s why I was throwing up. Okay, yes. I think that’s an interesting idea, too.

113 00:10:19.640 00:10:24.479 Samuel Roberts: The thing about the podcast is, like, okay, so it’s like a one-time generation, it has all the context.

114 00:10:25.050 00:10:30.950 Samuel Roberts: The thing about… chatting with it is… well, I guess not.

115 00:10:31.170 00:10:33.220 Samuel Roberts: Sorry, I’m just thinking of it, but…

116 00:10:33.220 00:10:45.259 Gabriel Lam: Yeah, no, no, it’s like, it needs to, like, at least go through the first, sort of, embedding process of, like, okay, I’m just gonna read everything, and then I’ll begin the conversation so that I have everything I need, as opposed to, like, having to call

117 00:10:46.290 00:10:49.260 Gabriel Lam: your files every time it happens.

118 00:10:49.260 00:10:53.209 Samuel Roberts: Yeah, I mean, part of it is also how big are these inputs gonna be?

119 00:10:54.160 00:10:54.700 Gabriel Lam: Yeah.

120 00:10:54.700 00:10:59.489 Samuel Roberts: Because, like, if you… if you just go to… I just go to ChatGPT, which I don’t…

121 00:10:59.740 00:11:09.170 Samuel Roberts: No… I’ve used the, like, the conversation on the phone. Does it do it here? You can dictate…

122 00:11:09.330 00:11:12.529 Samuel Roberts: use voice mode, there it is. So if I click Use Voice Mode…

123 00:11:13.010 00:11:16.639 Samuel Roberts: Say hello. Okay, I can start a new chat.

124 00:11:19.390 00:11:22.020 Samuel Roberts: I don’t know if this is gonna work on Mike right now, but…

125 00:11:22.180 00:11:24.740 Samuel Roberts: Okay. Oh, it is working.

126 00:11:27.200 00:11:29.350 Samuel Roberts: No, it’s not. It’s just letting me dictate stuff.

127 00:11:30.620 00:11:37.480 Samuel Roberts: Oh, there it is. Okay, yep, I’m all ears, you’re dictating. So if I add a file here… sorry, I’m just… I don’t know how this works.

128 00:11:37.810 00:11:40.190 Samuel Roberts: Here, let’s see,

129 00:11:42.850 00:11:49.510 Samuel Roberts: Oh, it’s listening to me. Oh, I forgot about this. Alright, let me just give it Where this?

130 00:11:50.390 00:11:51.480 Gabriel Lam: Creative…

131 00:11:51.880 00:11:55.899 Samuel Roberts: Oh, no, okay, so it kills it as I add a file, so, okay.

132 00:11:56.670 00:11:59.980 Samuel Roberts: I upload this file… oh, there we go, okay. Tell me about this image.

133 00:12:06.180 00:12:07.540 Samuel Roberts: Hmm, alright, yeah.

134 00:12:07.830 00:12:10.500 Samuel Roberts: Okay, so that kind of works. Interesting.

135 00:12:18.790 00:12:20.130 Samuel Roberts: Okay. Oh.

136 00:12:23.770 00:12:25.479 Samuel Roberts: I am the host now, okay.

137 00:12:39.340 00:12:44.819 Gabriel Lam: Around that, I think I got, disconnected, but you were talking.

138 00:12:44.820 00:12:45.260 Samuel Roberts: Yeah.

139 00:12:45.260 00:12:49.499 Gabriel Lam: you had justified something, and it was dictating, and then I sort of…

140 00:12:49.940 00:13:04.180 Samuel Roberts: Yeah, yeah, no, I was just… I was, I tried to dictate on, like, ChatGPT in the browser to see if I could add files, and it worked. So, I mean, there might be a way to do that without having to do all the pre-embedding and everything.

141 00:13:05.390 00:13:10.619 Samuel Roberts: The, like, context window of these models has gotten so big that, you know, you don’t necessarily need to, like.

142 00:13:11.440 00:13:23.229 Samuel Roberts: if it’s just, like, an article, or a few articles, we could probably just pass that right in and chat over it. The trick is going to be making sure that it stays within the context window as conversations go on, if that’s the case.

143 00:13:23.980 00:13:28.349 Gabriel Lam: Yeah. Or the other alternative is, like… If, you know.

144 00:13:28.500 00:13:35.350 Gabriel Lam: if I send you an article, and then I send, like, Mustafa an article, then you both have to add it to your context windows, and…

145 00:13:35.590 00:13:36.310 Gabriel Lam: I was like…

146 00:13:36.310 00:13:37.300 Samuel Roberts: Oh, yeah.

147 00:13:37.300 00:13:43.549 Gabriel Lam: what is… I was like, is there an alternative that exists that… We’re not aware of, or…

148 00:13:47.400 00:13:48.150 Gabriel Lam: Yeah, I think.

149 00:13:48.150 00:13:48.780 Samuel Roberts: for…

150 00:13:48.780 00:13:54.599 Gabriel Lam: ways to look into it. I always just, like, kind of bounce ideas off of… Off of you.

151 00:13:55.080 00:14:01.039 Samuel Roberts: Yeah, no, I like the… I mean, the podcast thing, I think, is a… I like that idea, because it’s definitely, like, a nice way to…

152 00:14:01.460 00:14:03.770 Samuel Roberts: digest something quickly.

153 00:14:03.880 00:14:09.709 Samuel Roberts: And people like podcasts. You know, I’ve sometimes had things just, like, read the articles to me.

154 00:14:09.860 00:14:16.029 Samuel Roberts: While I’m doing something else anyway. I’m a big audiobook guy to begin with, more than podcasts, even.

155 00:14:17.070 00:14:27.110 Samuel Roberts: The conversation over it is… Perhaps more interesting, certainly more complex, but… maybe worth it.

156 00:14:30.810 00:14:33.629 Samuel Roberts: I mean, they’re almost two different things, though.

157 00:14:34.080 00:14:36.100 Gabriel Lam: Yeah, yeah, I do realize that.

158 00:14:36.100 00:14:38.700 Samuel Roberts: Yeah, I don’t… I mean, there could be a way where it, like.

159 00:14:39.090 00:14:45.939 Samuel Roberts: you give it something, it gives you a quick summary, and then it’s like, got any questions? But I think I would keep them separate for now.

160 00:14:46.140 00:14:51.690 Gabriel Lam: Because, like, if you can just, like, upload a few things, get a few podcasts to listen to, that seems like a good…

161 00:14:52.630 00:14:56.289 Samuel Roberts: Starting point, whether it’s meetings or articles or whatever.

162 00:14:56.780 00:14:57.899 Samuel Roberts: Yeah.

163 00:14:59.830 00:15:03.890 Samuel Roberts: I think… It’s definitely worth exploring the…

164 00:15:04.270 00:15:10.840 Samuel Roberts: podcast-generating API endpoint, which is a crazy thing to say, but…

165 00:15:11.400 00:15:17.659 Samuel Roberts: exists. This is the world we live in now. Where did I put that? Oh, here it is, yeah.

166 00:15:18.130 00:15:20.800 Samuel Roberts: Gemini Enterprise, I don’t know.

167 00:15:22.400 00:15:27.480 Samuel Roberts: All you need to enable Google Cloud project in the podcast API user role. Yeah, so we could probably just spin this up

168 00:15:27.710 00:15:29.119 Samuel Roberts: And try this out.

169 00:15:30.380 00:15:32.970 Samuel Roberts: Relatively easy. Here, let me send you this.

170 00:15:33.620 00:15:34.810 Gabriel Lam: Thank you.

171 00:15:35.920 00:15:41.599 Samuel Roberts: Copy link… You want it here in Zoom, or should I Slack it? What would be easier?

172 00:15:41.970 00:15:43.250 Gabriel Lam: You can Slack it from wherever you.

173 00:15:43.250 00:15:43.840 Samuel Roberts: Okay.

174 00:15:44.130 00:15:47.100 Samuel Roberts: There’s so many places to send them things, it’s.

175 00:15:47.100 00:15:48.190 Gabriel Lam: All good.

176 00:15:48.190 00:15:51.279 Samuel Roberts: Too much. Where are you? How are you gonna be here?

177 00:15:59.210 00:16:07.090 Samuel Roberts: So, I think, like, certainly worth exploring that if it does, you know, Close to what we want.

178 00:16:08.680 00:16:13.719 Samuel Roberts: Replace the fall… project ID, text content, inline data… oh, wow, okay.

179 00:16:16.200 00:16:17.870 Samuel Roberts: Title for the podcast…

180 00:16:21.210 00:16:27.510 Samuel Roberts: Okay, yeah, I would say let’s try that, see how it is, because that might even give us good direction, even if it’s not, like, 100%.

181 00:16:28.150 00:16:33.620 Samuel Roberts: What we’re looking for… It will give us… Ayy.

182 00:16:34.960 00:16:36.400 Samuel Roberts: starting point.

183 00:16:37.110 00:16:40.709 Samuel Roberts: I’m trying to see what other things, for example, coming up.

184 00:16:41.610 00:16:42.790 Samuel Roberts: Yeah.

185 00:16:43.640 00:16:51.500 Samuel Roberts: The chat thing… chat thing is honestly, like, Not much different than the…

186 00:16:52.360 00:16:55.889 Samuel Roberts: Case study, except we would be giving it

187 00:16:56.020 00:16:58.260 Samuel Roberts: Different context to start, wouldn’t it?

188 00:16:58.260 00:16:58.910 Gabriel Lam: Right.

189 00:16:59.110 00:17:00.030 Gabriel Lam: Right.

190 00:17:00.030 00:17:05.449 Samuel Roberts: And then it’s not about generating the transcript so much as it is about just being a voice chat agent.

191 00:17:06.300 00:17:07.130 Gabriel Lam: Yeah.

192 00:17:12.040 00:17:22.130 Gabriel Lam: I think I… yeah, I think having this conversation, I’m like, maybe we can sort of shelve the chat thing for a little later, because I do think there are, like, good enough stopgap

193 00:17:23.119 00:17:27.240 Gabriel Lam: Sort of use, or, like, the tools out there.

194 00:17:27.470 00:17:29.760 Samuel Roberts: Yeah. Where I’m not too worried about that.

195 00:17:30.040 00:17:32.939 Gabriel Lam: I guess the other thing that I wanted to run by you was…

196 00:17:34.610 00:17:42.709 Gabriel Lam: the sort of linear integration, and so I had been… I don’t know if I was just gonna share screen and show you, but it’s not working.

197 00:17:42.710 00:17:43.050 Samuel Roberts: Okay.

198 00:17:44.000 00:17:50.359 Gabriel Lam: So, I’m gonna try to find a… see if I have an old chat.

199 00:17:50.570 00:17:51.660 Gabriel Lam: exists.

200 00:17:56.880 00:18:01.829 Gabriel Lam: Yeah, I don’t know why it’s doing that, give me a second. I’m just gonna close cursor and reopen it.

201 00:18:01.990 00:18:05.570 Samuel Roberts: Oh, never mind, I just saw the top level here, it says the…

202 00:18:05.920 00:18:09.929 Samuel Roberts: Podcast API is available to select Google Cloud customers.

203 00:18:09.930 00:18:11.170 Gabriel Lam: I see.

204 00:18:11.360 00:18:13.249 Samuel Roberts: So that might not be…

205 00:18:19.300 00:18:23.170 Samuel Roberts: Contact your Google Cloud sales representative. Oh boy, okay.

206 00:18:24.890 00:18:28.990 Samuel Roberts: Well, that’s… I mean, I’ll keep digging a little bit, see if there’s anything else out there.

207 00:18:29.130 00:18:31.060 Samuel Roberts: Those, you know, even if it’s, like, a…

208 00:18:31.220 00:18:36.270 Samuel Roberts: Library that helps us manage the conversational element or something, or just the…

209 00:18:36.730 00:18:41.279 Samuel Roberts: I guess it does need to be a conversation, too. It could just be a person reporting something.

210 00:18:41.840 00:18:42.730 Gabriel Lam: Yeah.

211 00:18:45.070 00:18:46.820 Samuel Roberts: Conversation element is nice.

212 00:18:49.480 00:18:51.690 Samuel Roberts: Anyway, linear, yes, sorry.

213 00:18:53.220 00:19:00.780 Gabriel Lam: Yeah, so I guess, like, I’ll maybe split it into two parts. I think the first part is… the…

214 00:19:01.080 00:19:03.409 Gabriel Lam: Existing flow that happens.

215 00:19:03.850 00:19:11.200 Gabriel Lam: After… A meeting is ingested into the web platform, like, we’re able to currently you know.

216 00:19:11.490 00:19:18.570 Gabriel Lam: if I just bring up… let me just try to bring it up… Like, today’s stand-up, for example.

217 00:19:39.420 00:19:42.859 Gabriel Lam: So we sort of see how it already is able to

218 00:19:43.790 00:19:46.730 Gabriel Lam: At least do a pretty good job at

219 00:19:46.810 00:20:04.880 Gabriel Lam: editing and making sure that, you know, the body is what we want it to be. One of the problems that we’ve been facing so far, and something that I’ve been thinking of whether, like, this flow is worth keeping, is this only happens when the stand-up gets

220 00:20:05.770 00:20:10.040 Gabriel Lam: ingested.

221 00:20:10.580 00:20:11.220 Samuel Roberts: Yes.

222 00:20:11.220 00:20:15.769 Gabriel Lam: sometimes takes… sometimes it’s faster, sometimes it’s longer. The next thing is that…

223 00:20:15.900 00:20:19.980 Gabriel Lam: We aren’t yet able to update existing tickets.

224 00:20:20.370 00:20:22.109 Gabriel Lam: Through this method.

225 00:20:22.110 00:20:23.140 Samuel Roberts: Hmm…

226 00:20:23.140 00:20:24.710 Gabriel Lam: And so…

227 00:20:25.140 00:20:29.480 Gabriel Lam: with cursor and with an MCP. I was like, hey, is there a way to bring this

228 00:20:31.180 00:20:35.039 Gabriel Lam: Off the platform, or some way to sort of augment this?

229 00:20:35.630 00:20:40.030 Gabriel Lam: One example would be… Like, if I…

230 00:20:41.190 00:20:46.770 Gabriel Lam: I’m just letting it run in the background. One thing is we have…

231 00:20:47.780 00:20:57.630 Gabriel Lam: been… I’ve been able to just connect to Supabase directly via an MCP. I see UTOM’s also… planning to…

232 00:20:57.830 00:21:03.470 Gabriel Lam: Create some kind of, like, cron or some kind of… Supabase, like.

233 00:21:04.880 00:21:13.490 Gabriel Lam: call or hook into GitHub directly, so I think that might be something there that I was like, hey, is this a time to revisit it?

234 00:21:13.810 00:21:20.780 Gabriel Lam: And instead of having to do all this, we could just have that live.

235 00:21:21.610 00:21:30.150 Gabriel Lam: on cursor, for now. And the reason why was… We’re also able to add like…

236 00:21:30.280 00:21:33.380 Gabriel Lam: There’s enough tools where you can… find…

237 00:21:34.100 00:21:37.650 Gabriel Lam: Tickets, update tickets, create new tickets, you can add comments.

238 00:21:38.630 00:21:48.060 Gabriel Lam: Which seems to be… In my opinion, something that… might be…

239 00:21:49.680 00:21:58.769 Gabriel Lam: a possibility, but wanted to see, like, any thoughts you had so far, or if there’s ways that we can leverage what we have now existing on this web platform instead.

240 00:22:00.830 00:22:01.610 Samuel Roberts: Mmm…

241 00:22:01.610 00:22:05.799 Gabriel Lam: I’m just trying to think what, like, might be a lighter lift, at least for now.

242 00:22:05.800 00:22:14.069 Samuel Roberts: Yeah, I mean, the light list is definitely, I think, cursor, getting the transcripts in there and having that. Problem is that’s, like, very,

243 00:22:15.330 00:22:17.510 Samuel Roberts: ad hoc, I feel like. It’s gonna be…

244 00:22:17.510 00:22:18.080 Gabriel Lam: out.

245 00:22:18.580 00:22:24.700 Samuel Roberts: you know, I mean, we would… we could get it to run on some kind of cadence and generate tickets and everything, but, like.

246 00:22:25.250 00:22:27.709 Samuel Roberts: Automating cursor feels like it’s…

247 00:22:29.000 00:22:39.179 Samuel Roberts: not the right way to do it, basically. The benefit of Cursor is, like, all these MCP tools are there, all the models are there, all the GitHub repos are there, but it’s not really a…

248 00:22:40.150 00:22:43.170 Samuel Roberts: Tool to automate, it’s a tool for… Yeah.

249 00:22:44.120 00:22:46.070 Samuel Roberts: interacting with.

250 00:22:46.070 00:22:46.690 Gabriel Lam: Yeah.

251 00:22:47.760 00:22:58.170 Samuel Roberts: So, I mean, maybe as, like, a first version, especially to, like, see what the MCPs can, you know, are capable of, and which models are better at doing it, like, that might be a pretty good way to…

252 00:22:58.770 00:23:01.179 Samuel Roberts: Really test this out better.

253 00:23:02.790 00:23:07.499 Samuel Roberts: The thing about, like, generating these linear tickets on the platform and everything is, like, we’re kind of re…

254 00:23:08.080 00:23:11.099 Samuel Roberts: building a lot of the UI that Linear has, which feels a little…

255 00:23:11.930 00:23:17.599 Gabriel Lam: Yeah, I don’t want to, like, you know, build something that someone’s already spent a lot of time and money in doing.

256 00:23:17.600 00:23:22.069 Samuel Roberts: Exactly, so I think the best… Thing is probably…

257 00:23:22.200 00:23:27.460 Samuel Roberts: Well, so this is all generating new tickets, for the most part, right? That’s what you’re saying? Yes.

258 00:23:27.460 00:23:28.240 Gabriel Lam: Yes.

259 00:23:28.400 00:23:37.799 Samuel Roberts: focusing on… Accessing previous tickets… Is probably worth… Putting some time into?

260 00:23:38.240 00:23:42.789 Samuel Roberts: Or, or something…

261 00:23:45.470 00:23:51.180 Samuel Roberts: I mean, okay, this UI, forget this UI for a second, let’s just think through, like, what is the ideal, maybe, for a second?

262 00:23:52.200 00:23:54.700 Samuel Roberts: So, you know, a meeting happens.

263 00:23:55.280 00:24:06.290 Samuel Roberts: We get a transcript, we get a video file, whatever. Something… looks at that, looks at linear…

264 00:24:07.280 00:24:10.529 Samuel Roberts: Tries to tie things together to update tickets, maybe?

265 00:24:11.090 00:24:11.690 Gabriel Lam: Yeah.

266 00:24:11.840 00:24:15.269 Samuel Roberts: And then potentially generate new tickets from that discussion.

267 00:24:15.530 00:24:19.530 Samuel Roberts: Yes. It doesn’t really matter where that happens, I suppose, whether that’s, like.

268 00:24:19.770 00:24:22.400 Samuel Roberts: Happening in the back end, and someone’s getting

269 00:24:22.560 00:24:33.529 Samuel Roberts: a message somewhere, or a Slack, you know what I mean? Like, whatever it is, if it’s in the platform. I’m just trying to think, like, you know, everything we want Cursor to do is effectively, like, using the MCP tools, using the models, using the transcripts.

270 00:24:33.640 00:24:37.170 Samuel Roberts: do something with that. And so, I think…

271 00:24:39.010 00:24:41.249 Samuel Roberts: It’s not a bad idea in terms of, like.

272 00:24:42.170 00:24:48.539 Samuel Roberts: not ideation, but just, like, experimentation, to keep going with cursor for now.

273 00:24:48.670 00:24:54.420 Samuel Roberts: I could, like, this was a little,

274 00:24:56.470 00:24:58.600 Samuel Roberts: I don’t… like, we’re using the…

275 00:24:59.450 00:25:01.810 Samuel Roberts: API here for Linear, I believe.

276 00:25:02.410 00:25:03.080 Gabriel Lam: Yes.

277 00:25:03.080 00:25:09.449 Samuel Roberts: we’re not using the MCP. So, like, if we want to back this up a little bit more out of here and say, like, okay.

278 00:25:09.790 00:25:16.800 Samuel Roberts: when a transcript comes in, let’s trigger a MOSTRA agent that runs, looks at the linear tickets.

279 00:25:17.060 00:25:20.049 Samuel Roberts: You know, has a human in the loop kind of thing.

280 00:25:24.700 00:25:27.170 Gabriel Lam: Yeah, or for now, it might even just be, like.

281 00:25:28.250 00:25:33.959 Gabriel Lam: It has the transcript, it looks at the transcript, it looks at all the existing linear tickets.

282 00:25:34.090 00:25:34.800 Samuel Roberts: Yeah.

283 00:25:35.130 00:25:38.970 Gabriel Lam: And then it tries to map relevant.

284 00:25:39.950 00:25:43.810 Gabriel Lam: Information or updates to existing tickets.

285 00:25:44.270 00:25:44.910 Samuel Roberts: Yeah, exactly.

286 00:25:45.380 00:25:45.950 Samuel Roberts: active.

287 00:25:45.950 00:25:50.700 Gabriel Lam: Either in the body or as a comment, which I think the MCP can both do at the moment.

288 00:25:51.080 00:25:51.460 Samuel Roberts: Yeah.

289 00:25:51.460 00:25:52.590 Gabriel Lam: I mean, I think…

290 00:25:53.170 00:25:57.660 Samuel Roberts: I think this sounds like the right kind of strat… like, the right end goal, certainly.

291 00:25:57.920 00:25:58.760 Gabriel Lam: Yay.

292 00:25:59.020 00:26:05.410 Samuel Roberts: And… to be… well, I don’t know.

293 00:26:09.120 00:26:11.200 Samuel Roberts: I think there’s…

294 00:26:12.300 00:26:20.759 Samuel Roberts: So the question of, like, what does the user interface, user experience seem like? Because, like, this is nice on, like, a per-meeting basis, but it’s not really…

295 00:26:22.040 00:26:27.689 Samuel Roberts: like, I… hold on, let me see. So if I… I have the same thing open, so if I go to Create Linear Tickets.

296 00:26:28.440 00:26:28.840 Gabriel Lam: Yep.

297 00:26:28.840 00:26:30.110 Samuel Roberts: So it… it…

298 00:26:30.440 00:26:37.530 Samuel Roberts: generated these when you asked it to? I don’t remember how the… the trigger… It generates upon…

299 00:26:37.530 00:26:39.000 Gabriel Lam: Upon ingestion.

300 00:26:39.400 00:26:42.780 Samuel Roberts: Oh, wait, so it is, so it’s doing… it’s doing all these, like, temp.

301 00:26:43.680 00:26:49.290 Samuel Roberts: Linear tickets. Yeah. Okay, I couldn’t remember which way we did that. So, I…

302 00:26:50.140 00:26:54.570 Samuel Roberts: There’s a… well, actually, I don’t remember the code specifically, but…

303 00:26:54.700 00:26:56.249 Samuel Roberts: There’s a way where we could…

304 00:26:56.580 00:26:58.780 Samuel Roberts: Change the way we’re doing that.

305 00:26:59.990 00:27:06.480 Samuel Roberts: to… Also, look at… previous tickets.

306 00:27:06.760 00:27:15.030 Samuel Roberts: That might be the highest leverage thing here that we’ve discussed in general, because, like.

307 00:27:15.030 00:27:15.770 Gabriel Lam: Yeah.

308 00:27:15.960 00:27:21.789 Samuel Roberts: I think if that’s really the hang-up here, is that these are just new, and it’s not updating old things, at the very least.

309 00:27:22.150 00:27:28.790 Samuel Roberts: we could… And I don’t know exactly where that… Work…

310 00:27:29.330 00:27:39.389 Samuel Roberts: falls into, but if we even just say, like, okay, forget the creating linear tickets bought what we made, let’s just make a new agent that has

311 00:27:40.440 00:27:44.950 Samuel Roberts: instructions to… use the linear MCP,

312 00:27:45.080 00:27:52.859 Samuel Roberts: in a, you know, read-only mode for now, for example, just to see what it finds. Like, does that do a good job searching linear tickets?

313 00:27:53.100 00:28:04.460 Samuel Roberts: From the transcript. Does it need more context for the client? Does it need more context? You know what I mean? Like, we could do a pretty quick and dirty, like, monster agent, maybe, that can, like.

314 00:28:05.220 00:28:09.480 Samuel Roberts: have access to the linear MCP, do this.

315 00:28:09.710 00:28:11.910 Samuel Roberts: See how that compares to…

316 00:28:12.640 00:28:20.120 Samuel Roberts: I guess I wouldn’t compare to this, because this is just generating, but eventually generating. Yeah.

317 00:28:21.190 00:28:26.450 Samuel Roberts: Okay. I, I think, I think there’s something to be said for… It happening kind of…

318 00:28:28.990 00:28:36.359 Samuel Roberts: Automatically like it is, but also with the updates, and probably with some kind of approval process for now, too, right?

319 00:28:38.110 00:28:47.299 Samuel Roberts: So, like, kind of like how this is, like, here are the tickets, would you like to approve and send them to Linear? Also, like, here are a bunch of the changes I want to make to linear tickets, do you like that?

320 00:28:47.300 00:28:47.980 Gabriel Lam: Yeah.

321 00:28:48.150 00:28:50.070 Samuel Roberts: I think we could probably add that…

322 00:28:51.370 00:28:55.749 Samuel Roberts: I don’t want to say easily, because I don’t like using that word anymore, but…

323 00:28:58.180 00:29:05.829 Samuel Roberts: I think I have to look at exactly how we did this. I want to say, given the fact that, like, the linear MCP

324 00:29:06.230 00:29:11.539 Samuel Roberts: works pretty well in Cursor, you said? I haven’t actually done a lot with it, so I don’t know.

325 00:29:12.230 00:29:13.440 Gabriel Lam: It does, it…

326 00:29:14.140 00:29:20.739 Gabriel Lam: It… it… it has. I don’t know if I was trying today, it hasn’t been very good, but when I’m, like…

327 00:29:21.610 00:29:23.840 Gabriel Lam: You know, I want you to…

328 00:29:23.980 00:29:26.860 Gabriel Lam: Like, find all the open tickets.

329 00:29:27.960 00:29:31.559 Gabriel Lam: Like, it usually returns something that’s

330 00:29:31.880 00:29:34.640 Gabriel Lam: more or less pretty accurate, and I’ve been able to…

331 00:29:34.960 00:29:38.330 Gabriel Lam: Use, like, plan mode to…

332 00:29:39.400 00:29:45.680 Gabriel Lam: I think plan mode or agent mode only are the only two that actually work with the MCPs, so…

333 00:29:45.950 00:29:49.420 Gabriel Lam: that’s actually had some pretty good responses. I think the main thing was

334 00:29:52.580 00:29:53.510 Gabriel Lam: I was…

335 00:29:53.630 00:30:05.349 Gabriel Lam: wondering… I just, like, noticed as I was looking through, like, Lilo or, like, ABC, I was just like, oh, everyone’s gotten back to, like, their old ways, and I’m like, oh, there’s… there’s gotta be a way to, like, bridge.

336 00:30:05.470 00:30:06.560 Gabriel Lam: that gap.

337 00:30:06.860 00:30:17.139 Gabriel Lam: So I was just trying to brainstorm what that would look like, and when I tried using Cursor to just, like, do it on an ad hoc basis, I was like, hey, maybe there’s something here.

338 00:30:17.300 00:30:21.839 Gabriel Lam: So yeah, I just thought I’d bring it up.

339 00:30:23.930 00:30:28.369 Samuel Roberts: No, you’re absolutely right, because I even haven’t been thinking about it.

340 00:30:29.350 00:30:39.989 Samuel Roberts: Okay, let’s… I think if we make a plan for me or someone to play around with, like, a…

341 00:30:40.580 00:30:49.779 Samuel Roberts: Well, I don’t know, I want to take a look at the code and see how we’re doing what we did, maybe, and figure out if we can patch that easily, or just say, like, no, let’s point people to cursor for now.

342 00:30:50.320 00:30:51.060 Gabriel Lam: Yeah.

343 00:30:51.190 00:30:57.869 Gabriel Lam: I think now it’s, I don’t… I don’t remember if it is a master agent, or… I know there’s a line fuse.

344 00:30:58.020 00:30:59.070 Gabriel Lam: connection.

345 00:30:59.670 00:31:10.140 Samuel Roberts: Yeah, yeah, no, I think it might be, but I think we were using the API, not the MCP, so I gotta look at that. But the other thing I would say is, there was talk way back when of just a grooming bot.

346 00:31:10.780 00:31:17.340 Samuel Roberts: Just, like, analyze linear tickets and be like, this isn’t good enough. That might also be…

347 00:31:17.340 00:31:21.310 Gabriel Lam: Add a bunch and then groom it, as opposed to, like, trying to mix and match.

348 00:31:21.710 00:31:28.659 Samuel Roberts: Or just, like, if people aren’t gonna, like, lean into the platform, even if we leave this, like, make sure that, like, something… something else is enforcing…

349 00:31:28.840 00:31:40.240 Samuel Roberts: you know, you can make tickets however you want, but you gotta make them right. You know, you want to use a tool? That’s fine. You want to put them in linear manually? That’s fine too. We’ll make sure it has all these things, or it’s gonna get, like, kicked somewhere else, you know?

350 00:31:40.530 00:31:41.360 Gabriel Lam: Yeah.

351 00:31:41.360 00:31:43.239 Samuel Roberts: So that’s another thought, too.

352 00:31:43.820 00:31:53.650 Samuel Roberts: Yeah, exactly, exactly. I think, you know, getting the platform to be useful is good, and getting it to be better than just linear is probably a good idea, because otherwise people will just make tickets in linear.

353 00:31:53.650 00:31:57.080 Gabriel Lam: But if and when they make tickets in linear.

354 00:31:57.330 00:32:00.909 Samuel Roberts: Something that is enforcing this… this structure is probably not a bad idea.

355 00:32:00.910 00:32:04.760 Gabriel Lam: Yeah, so just briefly, just to share, it’s like.

356 00:32:05.550 00:32:19.870 Gabriel Lam: for some reason, it wasn’t connecting earlier, but, like, how… if I say I wanted you to find all the open tickets on… via the MCP, I think I specifically have to call out the MCP, because it usually will prioritize, like, any files I have, and I have a couple markdown files.

357 00:32:19.870 00:32:20.430 Samuel Roberts: Sure, sure.

358 00:32:20.430 00:32:27.920 Gabriel Lam: So, but as long as I call it out, for example, here’s the AI team, and it does call out everything that’s open. So I do think there is…

359 00:32:28.230 00:32:29.330 Gabriel Lam: something.

360 00:32:30.440 00:32:34.480 Gabriel Lam: There’s something here. I don’t know exactly what it is, but there is something.

361 00:32:34.800 00:32:43.799 Samuel Roberts: Yeah, let me… let me maybe do a little thinking about that and test and look at the code and try to get back to you tomorrow with, like, if I think we could patch that easily or not, and…

362 00:32:43.800 00:32:50.729 Gabriel Lam: Okay. Okay. Yeah. There’s no rush. I think I’m also just doing a bunch of brainstorming, and I’ll also get back to you with, like, you know.

363 00:32:51.120 00:32:55.170 Gabriel Lam: maybe the Mastra version is the best way, or maybe there’s, like.

364 00:32:57.690 00:33:08.390 Gabriel Lam: after, you know, at the end of the day, whoever’s running, like, a CSO or a service lead will just be like, oh yeah, I’ll just do that after, because now that the roles are a little more defined.

365 00:33:08.450 00:33:11.809 Samuel Roberts: It could be more ad hoc. I’m sort of thinking through that lens as well.

366 00:33:11.980 00:33:18.069 Samuel Roberts: That’s not a bad way to think about it, too, yeah. Or, like, a Slack message to that person who has to do something, rather than go to the platform or something.

367 00:33:18.300 00:33:25.369 Gabriel Lam: Yeah, so there’s a couple ways to think about it, so I’ll try to fit that all into a PRE, and then we’ll see what people think.

368 00:33:25.560 00:33:26.679 Samuel Roberts: Alright, yeah, sounds good.

369 00:33:26.680 00:33:28.800 Gabriel Lam: Yeah. Awesome. Appreciate it.

370 00:33:28.800 00:33:34.959 Samuel Roberts: Alright, yeah, no problem. Sorry, I gotta… there’s another meeting happening now, I gotta… I gotta bounce that everywhere. Alright, I’ll talk to you later.

371 00:33:35.380 00:33:36.449 Gabriel Lam: Talk to you later. Bye.