Meeting Title: Arcturus Deck Creation Demo Date: 2025-08-28 Meeting participants: Rafay’s Circleback, Uttam Kumaran, Mustafa Raja, Rico Rejoso, Samuel Roberts, Matthew Good
WEBVTT
1 00:00:54.450 ⇒ 00:00:55.200 Mustafa Raja: 8.
2 00:00:56.340 ⇒ 00:00:57.440 Uttam Kumaran: Hello.
3 00:00:57.810 ⇒ 00:00:58.499 Mustafa Raja: How are you?
4 00:01:01.340 ⇒ 00:01:03.120 Uttam Kumaran: I am good, how are you?
5 00:01:03.610 ⇒ 00:01:04.839 Mustafa Raja: Yeah, doing good.
6 00:02:11.310 ⇒ 00:02:11.880 Matthew Good: Hey, guys.
7 00:02:11.880 ⇒ 00:02:12.650 Uttam Kumaran: Yeah.
8 00:02:13.150 ⇒ 00:02:14.019 Matthew Good: How’s it going?
9 00:02:14.260 ⇒ 00:02:15.830 Uttam Kumaran: World traveler, bro.
10 00:02:15.990 ⇒ 00:02:24.409 Matthew Good: Dude, I severely underestimated how long of a flight Istanbul to LA is direct. I was just frickin’ dying for 14 hours.
11 00:02:24.410 ⇒ 00:02:26.679 Uttam Kumaran: Was it for a wedding, or what was the… how are you.
12 00:02:26.680 ⇒ 00:02:31.720 Matthew Good: No, no, it was just the boys. It was just six of the boys.
13 00:02:31.720 ⇒ 00:02:42.760 Uttam Kumaran: either you have a client in Turkey, which is… which is, like, kind of weird. I don’t know, maybe it’s, like, OpenAI or somebody, or then I’m like, maybe if they’re both, like, homies and they’re going to a wedding or something, or….
14 00:02:42.760 ⇒ 00:02:50.749 Matthew Good: No, we just pulled up, it’s, like, me, Rafei, and then, like, four of our other friends. One guy is Turkish, thank God, because, you know, like, I mean, I don’t know. For me, I’m like.
15 00:02:51.220 ⇒ 00:02:52.040 Matthew Good: virus thing.
16 00:02:52.040 ⇒ 00:02:54.579 Uttam Kumaran: Turkish is a really different language, I don’t know any….
17 00:02:54.710 ⇒ 00:02:55.269 Matthew Good: You know what I mean?
18 00:02:55.270 ⇒ 00:03:02.040 Uttam Kumaran: A lot of Turkish food, dude, I love Turkish food, and I love Turkish coffee, but I don’t know any… I don’t have any Turkish friends.
19 00:03:02.040 ⇒ 00:03:07.950 Matthew Good: even, like, listen and kind of pick up, like, oh, yeah, yeah, yeah, this is what they’re talking about, you’re just like, dude, I don’t know what the fuck is going on right now.
20 00:03:10.090 ⇒ 00:03:17.400 Matthew Good: But yeah, man, I, … yeah, and I got back, and then the next night, I was wide awake at 3.30 in the morning, just like… just like that. So, …
21 00:03:17.520 ⇒ 00:03:22.260 Matthew Good: But it’s been, it’s been good. And then I came back, and I moved 2 days later, so….
22 00:03:22.260 ⇒ 00:03:24.660 Uttam Kumaran: No way. Where in town are you?
23 00:03:24.660 ⇒ 00:03:34.779 Matthew Good: I’m in Playa Vista, so, like, same area, but yeah, I rented the U-Haul and got the whole thing, which is also something you always underestimate, how, like, heavy of a lift, literally, it’s gonna be.
24 00:03:34.970 ⇒ 00:03:44.079 Matthew Good: But yeah, man, let me ping Raph, he should be popping in here, but if not, we can just get rolling. But yeah, and sorry for the, like.
25 00:03:44.380 ⇒ 00:03:47.040 Matthew Good: Probably what felt like a lack of responsiveness for the past….
26 00:03:47.040 ⇒ 00:03:57.640 Uttam Kumaran: No, you’re good. I mean, I… I sort of got us to, like, a good… a good point. I mean, for us, for my end, it’s like, we’re kind of at the end of that initial month, so I want to show you.
27 00:03:57.640 ⇒ 00:03:59.830 Matthew Good: So we got, I think we’ve….
28 00:03:59.900 ⇒ 00:04:14.819 Uttam Kumaran: like, I’m pretty impressed, like, I think kind of a couple of spots where I think there’s still a good amount of improvement. I would say, like, we’re… I’ll talk through it, we’ll do the demo, but I feel pretty good. I mean, I think, one, it’s sort of like…
29 00:04:15.370 ⇒ 00:04:23.870 Uttam Kumaran: it’s easy enough for, like, probably anyone on your team to kind of use it. I think definitely there’s probably some adjustments we can… we need… we kind of need to see
30 00:04:24.220 ⇒ 00:04:35.710 Uttam Kumaran: it get used for us to make some smaller tweaks. But, Mustafa, do you want to share, or do you want to have Matthew, like, just go through Slack and share, or what do you think’s best?
31 00:04:36.350 ⇒ 00:04:49.649 Mustafa Raja: Yeah, we can do any of those. I’d say the first iteration takes a few minutes, 2 or 3 minutes, so I have, I have first iteration ready, so we can go through how we can save it in Notion and all.
32 00:04:49.840 ⇒ 00:05:00.819 Uttam Kumaran: Yeah, maybe you… maybe you can share, and then, let’s just show Matthew where it is in Slack, in our Slack channel. Let’s kick… let’s kick one off, and then let’s look through a past example, like, while that’s processing.
33 00:05:00.820 ⇒ 00:05:06.610 Mustafa Raja: Yeah, yeah, yeah, yeah. Let me share… Okay, so…
34 00:05:06.820 ⇒ 00:05:13.370 Mustafa Raja: Here we have some… some past examples, but, in the interlude channel, let’s… let’s kick off a new one.
35 00:05:13.750 ⇒ 00:05:17.870 Mustafa Raja: So we’d mention this bot, and then we’d add…
36 00:05:19.510 ⇒ 00:05:24.890 Mustafa Raja: the relevant files to the deck we want to create, so this is for Arcturus.
37 00:05:25.370 ⇒ 00:05:28.449 Mustafa Raja: … Anya, let’s kick it off.
38 00:05:29.700 ⇒ 00:05:30.630 Mustafa Raja: Send.
39 00:05:30.800 ⇒ 00:05:43.820 Mustafa Raja: Yeah, so it’s going to create the deck for us over here. We can see that, it’s going to say that it’s working on it. While it’s working, let’s look into a previous example.
40 00:05:44.860 ⇒ 00:05:48.110 Mustafa Raja: Yeah, so this is something, …
41 00:05:48.810 ⇒ 00:05:51.460 Mustafa Raja: So this is something that it’s going to come up with.
42 00:05:51.610 ⇒ 00:06:03.820 Mustafa Raja: Sources, rationale, and, and a steelman review of that particular slide. And then when… when it’s done with the slides over here…
43 00:06:07.080 ⇒ 00:06:15.169 Mustafa Raja: Yeah, over here is going to give us a narrative flow, and then some questions, and then an overall feedback for the deck.
44 00:06:15.530 ⇒ 00:06:15.950 Matthew Good: Okay.
45 00:06:15.950 ⇒ 00:06:28.489 Mustafa Raja: And then what we can do is we can mention the bot and ask it to, revisit anything we want. Let’s revisit the feedback.
46 00:06:28.800 ⇒ 00:06:35.929 Mustafa Raja: … Let’s ask it to make it shorter, or anything else in your mind we can do?
47 00:06:36.290 ⇒ 00:06:39.519 Mustafa Raja: Yeah, Matthew, maybe think of, like, yeah, some sample feedback.
48 00:06:39.840 ⇒ 00:06:42.679 Matthew Good: Yeah, yeah, key questions.
49 00:06:43.070 ⇒ 00:06:44.090 Matthew Good: ….
50 00:06:46.040 ⇒ 00:06:48.819 Uttam Kumaran: Or you could take a second, read through it, and actually think about.
51 00:06:50.080 ⇒ 00:06:50.840 Matthew Good: Yeah, I’m just looking.
52 00:06:50.840 ⇒ 00:06:55.779 Uttam Kumaran: Oh, I’m gonna… I’ll pull it up… I’ll pull it up on my side, too, and maybe, like, I’ll… I can think about something as well.
53 00:06:56.030 ⇒ 00:06:56.450 Mustafa Raja: Okay.
54 00:06:56.450 ⇒ 00:06:58.990 Matthew Good: Rationale, assess the stage, yeah, problem.
55 00:06:59.360 ⇒ 00:07:00.320 Matthew Good: ….
56 00:07:02.610 ⇒ 00:07:04.990 Mustafa Raja: Let me know if I need to scroll or something.
57 00:07:05.150 ⇒ 00:07:07.320 Matthew Good: Yeah, maybe keep scrolling down a little bit.
58 00:07:09.600 ⇒ 00:07:16.440 Matthew Good: Negative. Okay, so positive and negative is just, like… It’s, it’s evaluating itself.
59 00:07:16.940 ⇒ 00:07:17.600 Mustafa Raja: Hmm.
60 00:07:17.600 ⇒ 00:07:18.440 Uttam Kumaran: Yes.
61 00:07:18.440 ⇒ 00:07:21.609 Matthew Good: Okay, market forces, major technology forces…
62 00:07:22.050 ⇒ 00:07:27.379 Matthew Good: Narrative flow, sources make matters worse. Oh, and it has sources too, okay, that’s nice.
63 00:07:27.380 ⇒ 00:07:28.180 Mustafa Raja: Yeah.
64 00:07:28.920 ⇒ 00:07:29.960 Matthew Good: the transcript.
65 00:07:35.510 ⇒ 00:07:37.150 Matthew Good: Okay, …
66 00:07:38.730 ⇒ 00:07:48.440 Matthew Good: We could say, … I mean, even just, like, random feedback, we’d just, like, come up with, like, different, like, variations of headlines, or something like that. A couple of variations of headlines.
67 00:07:48.780 ⇒ 00:07:53.440 Uttam Kumaran: Yeah, so just say, like, can we generate a couple different headlines for the first, like, 4 slides?
68 00:07:53.440 ⇒ 00:07:54.100 Matthew Good: Yeah.
69 00:07:55.560 ⇒ 00:07:57.180 Mustafa Raja: … hmm.
70 00:07:58.350 ⇒ 00:08:09.689 Mustafa Raja: One of those… Oh, so… Wait, wait a minute, so… It feels overwhelmed.
71 00:08:11.000 ⇒ 00:08:12.380 Mustafa Raja: And this looks good, right.
72 00:08:12.680 ⇒ 00:08:13.600 Matthew Good: Yeah.
73 00:08:14.370 ⇒ 00:08:31.450 Mustafa Raja: So how it’s going to work is it’s going to return us the whole, thing again, but it’s going to change, the headlines. Now, it’s not going to give us multiple headlines, because it wants to, keep the structure of the deck same.
74 00:08:31.450 ⇒ 00:08:31.860 Matthew Good: Yeah.
75 00:08:31.860 ⇒ 00:08:35.620 Mustafa Raja: So it’s going to return as a whole lick, again, which is going.
76 00:08:35.620 ⇒ 00:08:36.049 Uttam Kumaran: Fine.
77 00:08:36.059 ⇒ 00:08:40.899 Mustafa Raja: Same thing, but the headlines, are going to be redone.
78 00:08:41.159 ⇒ 00:08:41.769 Uttam Kumaran: That’s fine.
79 00:08:43.840 ⇒ 00:08:52.779 Matthew Good: And this is, and I’m sorry if this was already stated, but this is, like, the, I guess, the model or, like, the foundation layer of this. Is this… is this Claude Opus?
80 00:08:52.890 ⇒ 00:08:53.610 Matthew Good: Or Claude’s song.
81 00:08:53.610 ⇒ 00:08:55.150 Mustafa Raja: Oh, no, it’s Sonnet.
82 00:08:55.590 ⇒ 00:08:57.449 Matthew Good: It’s on… okay, that’s fine.
83 00:08:57.710 ⇒ 00:08:58.860 Mustafa Raja: We can use Opish.
84 00:08:58.860 ⇒ 00:09:01.860 Matthew Good: Okay, yeah, I figured that would be probably an easy change, but….
85 00:09:01.860 ⇒ 00:09:02.610 Mustafa Raja: Yeah.
86 00:09:03.270 ⇒ 00:09:11.369 Uttam Kumaran: The other thing is, like, based on the types of changes, so this is where, like, based on the types of tweaks that are common, we can have it, like.
87 00:09:11.790 ⇒ 00:09:26.970 Uttam Kumaran: just return the feedback, just return the headlines, or, like, sort of iterate over itself, versus, like, right now we’re having it… Do the whole thing. Yeah, because I guess our initial assumption was, like, you give, like, here’s, like, 10 pieces of feedback.
88 00:09:26.970 ⇒ 00:09:27.680 Matthew Good: Yeah.
89 00:09:27.680 ⇒ 00:09:30.380 Uttam Kumaran: Read, like, kind of, like, Try again.
90 00:09:30.580 ⇒ 00:09:31.430 Uttam Kumaran: Well….
91 00:09:31.430 ⇒ 00:09:33.869 Matthew Good: That… so what would happen is…
92 00:09:34.260 ⇒ 00:09:41.999 Matthew Good: we would get feedback from the client, typically, in a Figma comment or whatever, and what I would do is just, like, take that… Yes. …and swap it in.
93 00:09:42.000 ⇒ 00:09:44.489 Mustafa Raja: Yeah, exactly. So, so….
94 00:09:44.490 ⇒ 00:09:47.820 Uttam Kumaran: That’s kind of, like, what we’re… we kind of built towards, basically.
95 00:09:47.820 ⇒ 00:09:49.410 Matthew Good: Yes, yes, yes, yes, okay.
96 00:09:49.410 ⇒ 00:10:03.279 Mustafa Raja: Okay, so we’ll see that for the first four slides, we will have some sort of a different headline, than the previous ones, so we can double-check that if you want.
97 00:10:03.650 ⇒ 00:10:10.270 Matthew Good: Yeah, I mean, I agree with it, it’s saying the negative, but yeah, I mean, I like the first headline better. Positive, negative, rationale, okay.
98 00:10:10.440 ⇒ 00:10:12.080 Matthew Good: Yeah, awesome. So the, yeah, the flow….
99 00:10:12.080 ⇒ 00:10:17.889 Mustafa Raja: All the things are going… yeah, the rest of the things are going to be the same, it’s going to be only the headline that’s going to be altered.
100 00:10:18.250 ⇒ 00:10:18.670 Matthew Good: God.
101 00:10:18.690 ⇒ 00:10:27.870 Mustafa Raja: Yeah, and like that, we can, target any part of this, deck, with a review for it to look through.
102 00:10:28.850 ⇒ 00:10:30.770 Matthew Good: Got it. Okay, so you could say, like.
103 00:10:31.350 ⇒ 00:10:38.470 Matthew Good: So what do you mean by that? Like, if you’re saying, like, revise the problem slide, like, we could say, like, revise the problem slide to actually include
104 00:10:38.930 ⇒ 00:10:40.669 Matthew Good: X, Y, and Z bullet points instead.
105 00:10:40.670 ⇒ 00:10:57.249 Mustafa Raja: Yeah, we can say, we can say, for this slide 6, we do not like these bullets, or this headline isn’t good enough, we need to work on rationale, or this steel man, any of those sort of things, we can do.
106 00:10:57.560 ⇒ 00:10:57.920 Matthew Good: God.
107 00:10:57.920 ⇒ 00:11:01.249 Uttam Kumaran: The output generated is just gonna be the entire output.
108 00:11:01.260 ⇒ 00:11:02.600 Matthew Good: Versus, like.
109 00:11:02.600 ⇒ 00:11:09.020 Uttam Kumaran: okay, I changed this, here’s what it was. So, like, every time you’re gonna get the whole output, basically.
110 00:11:09.020 ⇒ 00:11:10.700 Matthew Good: Oh, okay, okay.
111 00:11:11.190 ⇒ 00:11:13.700 Uttam Kumaran: The, the alternative, you know, this is, again, like.
112 00:11:13.920 ⇒ 00:11:17.760 Uttam Kumaran: We just kind of have to see you use it for a bit to kind of get a sense of, like.
113 00:11:17.950 ⇒ 00:11:24.760 Uttam Kumaran: what the… we want the kind of user experience is. Yeah. But alternatively, we could generate it first.
114 00:11:24.990 ⇒ 00:11:26.399 Uttam Kumaran: We could save that.
115 00:11:26.620 ⇒ 00:11:30.919 Uttam Kumaran: You suggest edits, we just return the affected pieces, and then you say go.
116 00:11:31.160 ⇒ 00:11:32.009 Uttam Kumaran: I like that.
117 00:11:32.010 ⇒ 00:11:32.660 Matthew Good: Hmm.
118 00:11:32.660 ⇒ 00:11:33.710 Uttam Kumaran: Those are all, like, totally.
119 00:11:33.710 ⇒ 00:11:34.840 Matthew Good: That’s possible.
120 00:11:34.840 ⇒ 00:11:36.950 Uttam Kumaran: Totally possible, it’s… yeah.
121 00:11:36.950 ⇒ 00:11:38.440 Matthew Good: Okay, got it.
122 00:11:38.440 ⇒ 00:11:43.150 Uttam Kumaran: That’s more of, like, a… Yeah, whatever we need it to be, that’s, like, yeah, not….
123 00:11:43.150 ⇒ 00:11:48.249 Matthew Good: Got it. But this is, like, the… okay, this is great. So this is in Slack.
124 00:11:48.450 ⇒ 00:11:52.150 Uttam Kumaran: So yeah, you wanna show the, yeah, show the next part of the integration.
125 00:11:52.970 ⇒ 00:11:55.310 Mustafa Raja: Okay, so we want to save it in Notion, right?
126 00:11:55.840 ⇒ 00:11:56.380 Matthew Good: out.
127 00:11:56.580 ⇒ 00:12:01.190 Mustafa Raja: Yeah, okay, so we can say, approved for Notion.
128 00:12:01.350 ⇒ 00:12:04.250 Mustafa Raja: It’s going to save it in Notion, and it’ll send us a link for that.
129 00:12:08.370 ⇒ 00:12:17.200 Mustafa Raja: It’s going to… it’s going to look for the latest deck that was generated in the thread, and it’s going to save that over here.
130 00:12:19.740 ⇒ 00:12:20.890 Mustafa Raja: Oh, you see that?
131 00:12:21.370 ⇒ 00:12:22.670 Matthew Good: Hmm. Oh….
132 00:12:24.460 ⇒ 00:12:25.150 Mustafa Raja: Nope.
133 00:12:25.850 ⇒ 00:12:28.110 Mustafa Raja: And this is the whole, whole thing.
134 00:12:28.260 ⇒ 00:12:31.110 Mustafa Raja: Okay, perfect. Over in Notion.
135 00:12:31.300 ⇒ 00:12:33.420 Matthew Good: That was… okay, so Slack to Notion, got it.
136 00:12:33.420 ⇒ 00:12:36.450 Mustafa Raja: Yeah, and his database, and we’re saving it in.
137 00:12:37.780 ⇒ 00:12:38.560 Matthew Good: Got it.
138 00:12:39.040 ⇒ 00:12:44.480 Uttam Kumaran: Okay, that makes sense. The only thing I would say… so there’s still, like, the manual work of, like, of going through….
139 00:12:44.970 ⇒ 00:12:54.899 Matthew Good: And just, like, spacing things out. The one thing, so I’ve been doing in the… in terms of, like, the past, probably, one, two and a half weeks, is I’ve been using Claude, but, like, the workbench.
140 00:12:55.030 ⇒ 00:12:56.850 Matthew Good: Featured, or whatever.
141 00:12:57.030 ⇒ 00:13:07.220 Matthew Good: And the output that I get, the final output that I get, when I copy and paste into Notion, it’s already all formatted into, like, H1s, H2, H3s.
142 00:13:07.220 ⇒ 00:13:07.570 Mustafa Raja: Hmm….
143 00:13:07.570 ⇒ 00:13:14.309 Matthew Good: bullets, etc. So that, and that, which is, like, a small thing, but, like, that actually does take a lot of time to manually go through and, like, hit enter.
144 00:13:14.310 ⇒ 00:13:14.720 Mustafa Raja: Yeah.
145 00:13:14.720 ⇒ 00:13:15.979 Matthew Good: You know? ….
146 00:13:15.980 ⇒ 00:13:16.530 Mustafa Raja: Weird.
147 00:13:16.680 ⇒ 00:13:22.969 Matthew Good: So that just… I’m sure that’s probably solvable in some way, … But that’s one thing. ….
148 00:13:23.510 ⇒ 00:13:24.120 Uttam Kumaran: Cool.
149 00:13:24.340 ⇒ 00:13:33.170 Matthew Good: Which would… which, again, like, this would, you know, if we kept going, this would, like, be a thing to iterate on. … Yeah. But that’s… that’s great. So where… and just in, like, can I…
150 00:13:33.610 ⇒ 00:13:36.690 Matthew Good: in Notion, could I… does it know…
151 00:13:36.930 ⇒ 00:13:41.070 Matthew Good: to go to a specific place in Notion to put it into? Like, if we have, like, a….
152 00:13:41.430 ⇒ 00:13:42.480 Mustafa Raja: Yeah, we have this….
153 00:13:42.480 ⇒ 00:13:43.000 Matthew Good: and….
154 00:13:43.260 ⇒ 00:13:47.970 Uttam Kumaran: So we just have our AI database. Right now, like, I think…
155 00:13:48.150 ⇒ 00:13:58.520 Uttam Kumaran: we… we found that we could do that, I just didn’t want to spam that page. But, for example, if you have an Arcturus page, we can go in there and append it.
156 00:13:58.570 ⇒ 00:14:01.669 Matthew Good: Gotcha. It just already exists, so we can add… that’s….
157 00:14:01.670 ⇒ 00:14:07.360 Uttam Kumaran: Fairly easy to add, I just didn’t want to sort of add that there as we were testing this whole thing.
158 00:14:07.530 ⇒ 00:14:13.160 Matthew Good: Yeah, yeah, absolutely. Okay, … Awesome. …
159 00:14:14.770 ⇒ 00:14:27.120 Matthew Good: I mean, this looks great. What… in terms of… so that was the other question. So in terms of, like, feeding, I think in… when you showed the initial part of the demo, you, like, added in some, like.txt files and stuff like that. Like, what does that flow…
160 00:14:27.430 ⇒ 00:14:28.230 Matthew Good: Because, again….
161 00:14:28.230 ⇒ 00:14:29.820 Mustafa Raja: Yeah, so those are… yeah, go ahead.
162 00:14:30.390 ⇒ 00:14:36.180 Mustafa Raja: Yeah, so those are the files that, … this would be the questionnaire, this would be our transcript.
163 00:14:36.180 ⇒ 00:14:36.500 Matthew Good: Yeah.
164 00:14:36.500 ⇒ 00:14:44.350 Mustafa Raja: So yeah, we just need to give it the files, that are relevant to this, deck creation.
165 00:14:44.350 ⇒ 00:14:44.680 Matthew Good: Okay.
166 00:14:44.680 ⇒ 00:14:46.310 Mustafa Raja: More than two, or could be only one.
167 00:14:46.310 ⇒ 00:14:50.970 Matthew Good: So, end-to-end, so let’s say, like, I pull up, I have a new deck, right, whatever, we closed on this week, you know….
168 00:14:50.970 ⇒ 00:14:56.930 Uttam Kumaran: You could do it, yeah, you could even just do it right now if you want to. If it’s… is it live in the… in that interlude channel?
169 00:14:57.380 ⇒ 00:15:01.230 Mustafa Raja: Yeah, let me see… yeah, it did reply.
170 00:15:01.230 ⇒ 00:15:04.400 Uttam Kumaran: You could try, yeah, you could try it if you want to just go do it.
171 00:15:04.400 ⇒ 00:15:06.799 Matthew Good: Okay, so I just add the channel, and then…
172 00:15:07.300 ⇒ 00:15:17.169 Matthew Good: Let’s say… Okay, so… so I just have to go and collect, kind of, the data, so let’s say, like, the transcript from the kickoff call, I would go to Circleback and just copy…
173 00:15:17.580 ⇒ 00:15:22.250 Matthew Good: The text from, let’s say, this call here…
174 00:15:22.890 ⇒ 00:15:29.610 Matthew Good: And then I hit copy transcript, so then if I just copied and paste… well, it just… so it pastes… here, I guess let me screen share.
175 00:15:30.970 ⇒ 00:15:31.500 Matthew Good: ….
176 00:15:31.520 ⇒ 00:15:41.170 Mustafa Raja: So it should be a file for now. We can, do the, pasting the text later, but for now, it’s .txt files that needs to be there.
177 00:15:41.710 ⇒ 00:15:43.200 Matthew Good: Okay, great. So…
178 00:15:44.920 ⇒ 00:15:49.490 Matthew Good: So, okay, so I just forgot in Notion, or in Circleback, or whatever, just how to export as a .txt.
179 00:15:49.490 ⇒ 00:15:55.419 Uttam Kumaran: Yeah, you can just create a, like, you can just put a little notepad and then export as a text. It’s pretty easy.
180 00:15:55.420 ⇒ 00:15:56.330 Matthew Good: Okay, got it.
181 00:15:56.460 ⇒ 00:15:59.880 Uttam Kumaran: Yeah, Matt, do you want to share on your side, and then we can just see it?
182 00:15:59.880 ⇒ 00:16:00.910 Matthew Good: Sure, yeah.
183 00:16:01.760 ⇒ 00:16:02.520 Matthew Good: Yeah.
184 00:16:04.120 ⇒ 00:16:07.880 Matthew Good: Let’s see… Okay.
185 00:16:08.830 ⇒ 00:16:10.120 Matthew Good: Can you guys see my screen?
186 00:16:10.350 ⇒ 00:16:10.920 Mustafa Raja: Yep.
187 00:16:12.470 ⇒ 00:16:13.390 Matthew Good: Can you see Slack?
188 00:16:14.080 ⇒ 00:16:14.960 Uttam Kumaran: Yes.
189 00:16:14.960 ⇒ 00:16:19.140 Matthew Good: Okay, cool. So I would go, let’s say, randomly to this, like, site structure thing.
190 00:16:19.360 ⇒ 00:16:20.760 Matthew Good: This call we had.
191 00:16:21.060 ⇒ 00:16:22.579 Matthew Good: Copy transcript.
192 00:16:22.860 ⇒ 00:16:24.739 Matthew Good: And then I would go to…
193 00:16:25.310 ⇒ 00:16:29.549 Matthew Good: … when I go to Note? What’d you say? Notepad?
194 00:16:30.310 ⇒ 00:16:37.019 Uttam Kumaran: Yeah, if you just go to Notepad or Notes, and then you can just literally paste whatever in.
195 00:16:38.080 ⇒ 00:16:40.909 Uttam Kumaran: Paste the transcript, and then you can just go to File, Explorer.
196 00:16:43.400 ⇒ 00:16:45.310 Matthew Good: export as….
197 00:16:45.310 ⇒ 00:16:47.420 Uttam Kumaran: Can you do PDF, Mustafa?
198 00:16:47.880 ⇒ 00:16:50.910 Mustafa Raja: No, it’s only fixed for now.
199 00:16:51.330 ⇒ 00:16:53.130 Mustafa Raja: But you can add in….
200 00:16:53.130 ⇒ 00:16:56.269 Uttam Kumaran: You can also, just type in, just,
201 00:16:56.780 ⇒ 00:17:00.559 Uttam Kumaran: Click on, whatever the spotlight is, and you just type in text edit.
202 00:17:00.670 ⇒ 00:17:02.679 Matthew Good: This is actually gonna be easier. Yeah.
203 00:17:03.000 ⇒ 00:17:08.990 Uttam Kumaran: Just paste it into here, or just do File, New, yeah, and paste it into there.
204 00:17:10.540 ⇒ 00:17:11.409 Matthew Good: Oh, fault.
205 00:17:12.329 ⇒ 00:17:16.159 Uttam Kumaran: Yeah, document. Yeah, paste it here, and then you can go file.
206 00:17:17.099 ⇒ 00:17:24.359 Uttam Kumaran: … Save as… And then you can change it to…
207 00:17:25.319 ⇒ 00:17:27.419 Uttam Kumaran: I think you can just do RTF.
208 00:17:29.340 ⇒ 00:17:30.450 Matthew Good: Archie Africa.
209 00:17:30.450 ⇒ 00:17:31.070 Uttam Kumaran: Yeah.
210 00:17:31.360 ⇒ 00:17:33.839 Uttam Kumaran: And then you can just save it wherever.
211 00:17:36.050 ⇒ 00:17:37.340 Matthew Good: Okay, so now it’s here.
212 00:17:37.540 ⇒ 00:17:39.840 Matthew Good: So then I go back to Slack.
213 00:17:45.370 ⇒ 00:17:47.170 Matthew Good: Okay, so it’s an RTF?
214 00:17:47.170 ⇒ 00:17:47.790 Uttam Kumaran: Yeah.
215 00:17:48.310 ⇒ 00:17:52.049 Matthew Good: Okay, so this would work, and then I go… And then just hit enter.
216 00:17:52.240 ⇒ 00:17:53.140 Uttam Kumaran: Yeah, that’s.
217 00:17:53.140 ⇒ 00:17:54.270 Mustafa Raja: Yeah, they should work.
218 00:17:54.270 ⇒ 00:17:58.130 Matthew Good: Okay, and this is, I mean, this is not a deck call, so it’s not gonna be perfect, but….
219 00:18:02.680 ⇒ 00:18:04.000 Mustafa Raja: It’s going to take some time.
220 00:18:04.560 ⇒ 00:18:05.650 Matthew Good: Interesting, okay.
221 00:18:10.530 ⇒ 00:18:16.249 Matthew Good: Okay, so the flow is, like, Copy into text, RTF.
222 00:18:17.920 ⇒ 00:18:18.250 Uttam Kumaran: Yeah.
223 00:18:18.250 ⇒ 00:18:20.209 Mustafa Raja: We can adjust this, ….
224 00:18:20.210 ⇒ 00:18:22.979 Uttam Kumaran: You can adjust it so you can paste in the transcript pretty easily.
225 00:18:23.720 ⇒ 00:18:24.510 Matthew Good: Okay.
226 00:18:24.510 ⇒ 00:18:25.760 Uttam Kumaran: It takes 2 seconds.
227 00:18:25.910 ⇒ 00:18:26.560 Matthew Good: Okay.
228 00:18:27.420 ⇒ 00:18:32.519 Matthew Good: Yeah, I’ve kind of, like, since we last talked, I’ve, like, evolved, I don’t know, evolved the flow.
229 00:18:32.790 ⇒ 00:18:34.210 Matthew Good: To just do this.
230 00:18:35.360 ⇒ 00:18:48.279 Matthew Good: So, like, I made a bunch of different, like, templates. You guys can see my screen. So, like, a template web for SaaS, template for HealthTech, template for generic, template for a seed deck, so I’ll just go in here, and I already meta-prompted the system prompt.
231 00:18:48.390 ⇒ 00:18:50.909 Uttam Kumaran: And then the user prompt, and I’ll just go….
232 00:18:51.350 ⇒ 00:18:52.520 Matthew Good: make a copy.
233 00:18:52.690 ⇒ 00:18:57.750 Matthew Good: And then I’ll just fill in everything here, but I’ll just copy and paste the text, just, you know.
234 00:18:58.140 ⇒ 00:19:05.599 Uttam Kumaran: Yeah, so this is, like, this would, again, like, having this would just allow us to… we just add this to the N8N flow, and so you could basically put in this.
235 00:19:05.770 ⇒ 00:19:08.470 Uttam Kumaran: the type, And then hit enter.
236 00:19:08.730 ⇒ 00:19:14.560 Matthew Good: Yeah, because when we… when we first… when I first came to you guys with this, like, I didn’t realize that, like, I was gonna need different ones for each specific.
237 00:19:14.560 ⇒ 00:19:23.350 Uttam Kumaran: Yeah, yeah, yeah. Well, that’s right. When you start talking to us, you’re like, we start asking some hard questions about, like, what is the flow here, and you’re like, oh, shit, okay. So that’s exactly it. Yeah.
238 00:19:23.350 ⇒ 00:19:29.639 Matthew Good: 100%. Well, that’s what pushed me to be like, oh, I should get into Workbench and, like, spin up these, like, mini… Anyway, …
239 00:19:29.750 ⇒ 00:19:31.350 Matthew Good: Okay, so it’s still….
240 00:19:34.920 ⇒ 00:19:36.780 Uttam Kumaran: I think the last one took, like, ….
241 00:19:37.010 ⇒ 00:19:45.210 Matthew Good: Yeah, that’s fine. And the prompt that it’s using, the, like, deck prompt, was the one that I gave you guys from that Notion page on intake.
242 00:19:45.660 ⇒ 00:19:48.919 Uttam Kumaran: Yeah, I guess Mustafa, like, I don’t know if we may have changed it.
243 00:19:48.920 ⇒ 00:19:55.199 Mustafa Raja: Yeah, so for that, we are using multiple agents to cater this whole thing, so….
244 00:19:55.200 ⇒ 00:19:56.179 Uttam Kumaran: Yeah, so we have, like.
245 00:19:56.180 ⇒ 00:19:56.650 Mustafa Raja: to leave….
246 00:19:56.650 ⇒ 00:20:06.059 Uttam Kumaran: something that dictates the narrative, something that writes it, something that, like, steel mans. So, it would be like if you were to copy-paste it into the next
247 00:20:06.230 ⇒ 00:20:09.339 Uttam Kumaran: Workbench, and then copy and paste that into the next workbench.
248 00:20:09.340 ⇒ 00:20:11.060 Matthew Good: It’s doing that, like.
249 00:20:11.610 ⇒ 00:20:15.050 Uttam Kumaran: Like, kind of like you’re attacking it with multiple prompts.
250 00:20:15.180 ⇒ 00:20:16.450 Matthew Good: Okay, got it.
251 00:20:16.450 ⇒ 00:20:18.610 Uttam Kumaran: Yeah, go ahead.
252 00:20:18.870 ⇒ 00:20:27.659 Matthew Good: I was gonna say, just so I understand, to play back to you, there’s a prompt for, like, writing style, there’s a prompt for, like, structure of the product, there’s a prompt for how to process.
253 00:20:27.660 ⇒ 00:20:37.129 Uttam Kumaran: It’s like, is it a VC or LP? From their perspective, what would you say? Give it feedback, rewrite it. So it, like… and again, these are all things where…
254 00:20:37.410 ⇒ 00:20:40.010 Uttam Kumaran: It just has to be a little bit of an iteration, where.
255 00:20:40.290 ⇒ 00:20:40.699 Matthew Good: Yeah, yeah.
256 00:20:40.700 ⇒ 00:20:52.829 Uttam Kumaran: You see some of these, you’re like, okay, it’s, like, mostly there, but it would be better if it’s this. Okay, cool. We either have a couple options. We tweak prompts, we tweak the inputs, we tweak the human in the loop, so then we start to iterate and get this to, like.
257 00:20:53.010 ⇒ 00:20:54.450 Matthew Good: Yeah. Idaho State.
258 00:20:54.880 ⇒ 00:20:55.600 Matthew Good: Yeah, yeah.
259 00:20:56.070 ⇒ 00:20:58.609 Uttam Kumaran: But… Okay, great.
260 00:20:58.610 ⇒ 00:20:59.769 Matthew Good: There we go, okay.
261 00:21:00.330 ⇒ 00:21:01.090 Matthew Good: Interesting.
262 00:21:06.980 ⇒ 00:21:08.369 Matthew Good: Size of the stage.
263 00:21:08.750 ⇒ 00:21:09.779 Uttam Kumaran: Who is this with?
264 00:21:10.150 ⇒ 00:21:12.869 Matthew Good: This is, like, a neurotech company.
265 00:21:12.870 ⇒ 00:21:13.730 Uttam Kumaran: Sick.
266 00:21:14.350 ⇒ 00:21:18.770 Matthew Good: … Quantifiable details.
267 00:21:18.770 ⇒ 00:21:21.569 Uttam Kumaran: Kind of, like, similar to, Elon’s company?
268 00:21:22.270 ⇒ 00:21:29.649 Matthew Good: Dude, they have, like, some hardware thing that they’re building that, like, is a smaller diameter, so, like, but they’ve compacted more power than the smaller diameter.
269 00:21:29.650 ⇒ 00:21:30.890 Uttam Kumaran: Like, kind of, like, same thing.
270 00:21:30.890 ⇒ 00:21:31.710 Matthew Good: Yeah, yes.
271 00:21:31.710 ⇒ 00:21:32.150 Uttam Kumaran: Okay, okay.
272 00:21:32.150 ⇒ 00:21:36.999 Matthew Good: The dude’s, like, a big brain guy from Columbia. He’s, like, a prof, like, in his 60s. ….
273 00:21:37.240 ⇒ 00:21:44.080 Uttam Kumaran: then reading this, I mean, like, I mean, I watched all, like, the monkey demos that Elon did, and so this seems, like, kind of, like, in….
274 00:21:44.420 ⇒ 00:21:45.080 Matthew Good: Yeah.
275 00:21:45.410 ⇒ 00:21:46.670 Uttam Kumaran: Seems pretty in line.
276 00:21:46.670 ⇒ 00:21:51.220 Matthew Good: It definitely is in line. Competitive advantage, clinical validation, product strategy.
277 00:21:51.580 ⇒ 00:22:04.759 Matthew Good: clinical applications. I mean, yeah, this is solid. This is what I would put in a deck. Like, maybe I would move around the order or something like that based on his talk track, but all the pieces are here, which is what I’m trying to solve for with the V1. Market strategy, investment opportunity….
278 00:22:04.760 ⇒ 00:22:07.190 Uttam Kumaran: Yeah, so basically, another thing is, like, we…
279 00:22:07.460 ⇒ 00:22:13.290 Uttam Kumaran: I think at this point, we’re kind of, like, probably more debating on what pieces get the deck and the formatting.
280 00:22:13.460 ⇒ 00:22:13.870 Matthew Good: Yeah.
281 00:22:13.870 ⇒ 00:22:33.170 Uttam Kumaran: That is, like, dude, that’s, like, the easiest part of this whole project, frankly. So, my answer is gonna be, like, yes, yes, yes, like, that’s where we would need feedback. I think more important to me is, like, we talked about, like, I want this to not just replicate your process, I want this to be faster and have a better output.
282 00:22:33.230 ⇒ 00:22:51.409 Uttam Kumaran: Right? So it’s not just important to me that, like, okay, instead of copy and pasting the clot, I can copy and paste it here. The… what you get here on a product-wise, or how fast you can iterate, needs to be better. So a couple things were, like, we put in the feedback. So again, instead of you being like, why is this good, why is this bad?
283 00:22:51.410 ⇒ 00:22:53.070 Matthew Good: Yeah. You can see on each….
284 00:22:53.590 ⇒ 00:22:58.950 Uttam Kumaran: on each slide, why it could be good, why it’s bad. There’s also the rationale, which is, like.
285 00:22:59.090 ⇒ 00:23:06.990 Uttam Kumaran: Okay, you have a rationale, we could… we have a rationale for each slide, but at the end, also, there’s a rationale on, like, the entire thing.
286 00:23:06.990 ⇒ 00:23:07.700 Matthew Good: Yeah, I like that.
287 00:23:07.700 ⇒ 00:23:08.290 Uttam Kumaran: Boom.
288 00:23:08.700 ⇒ 00:23:09.940 Uttam Kumaran: That way…
289 00:23:10.080 ⇒ 00:23:18.350 Uttam Kumaran: for me, you know, and I think about my business, this is more about, like, when you hand this off to someone else to do, how close is the target they’re gonna get, you know?
290 00:23:18.350 ⇒ 00:23:21.279 Matthew Good: Right, right, exactly. Exactly. You know?
291 00:23:21.670 ⇒ 00:23:26.359 Matthew Good: Yeah, yeah. … Awesome, yeah, I’d have to read through this, like.
292 00:23:27.280 ⇒ 00:23:29.419 Uttam Kumaran: You’ll have to actually read through it, yeah.
293 00:23:29.420 ⇒ 00:23:33.070 Matthew Good: I know. It is, I think, like, the most technical product we have, too.
294 00:23:33.880 ⇒ 00:23:36.519 Matthew Good: Awesome. Okay, but no, this is great. …
295 00:23:37.340 ⇒ 00:23:43.620 Matthew Good: Anything else to, like, review, or, like… call out… ….
296 00:23:44.220 ⇒ 00:23:49.449 Uttam Kumaran: I think this is… I mean, probably, like, my questions… my question for you is gonna be…
297 00:23:49.480 ⇒ 00:23:55.279 Uttam Kumaran: Yeah. Like, do you want… like, there’s a couple ways we can keep going. One is, like, if you think, A, there’s…
298 00:23:55.290 ⇒ 00:24:11.820 Uttam Kumaran: there’s some more… there’s a little bit more tweaks we need to this, we need to do to this, and I can kind of give you, like, what it’s gonna be. Also, if you’re like, okay, we solved this, let’s move on to this other part of our business, where I know we talked a little bit about the other Notion thing, or if you’re like, hey, this was a great start.
299 00:24:12.000 ⇒ 00:24:15.129 Uttam Kumaran: Can we iterate on, like, what a V2 of this is gonna look like?
300 00:24:15.130 ⇒ 00:24:15.520 Matthew Good: Yeah.
301 00:24:15.520 ⇒ 00:24:17.360 Uttam Kumaran: You just kind of keep going at the…
302 00:24:17.520 ⇒ 00:24:30.769 Uttam Kumaran: we were going. I mean, I would say… I would say what you can expect from us are, one, we got through a lot of, like, the initial just, like, access of stuff, understanding of the whole business. Also here, we’ve solved
303 00:24:30.880 ⇒ 00:24:40.010 Uttam Kumaran: the Slack, the Notion, so if we were to take on, like, another agent or a second version of this, we’ve, like, built… like, that’s, like, all ready, so….
304 00:24:40.010 ⇒ 00:24:40.740 Matthew Good: Hmm,
305 00:24:41.220 ⇒ 00:24:45.889 Uttam Kumaran: it, like, the time, probably the week we spent on figuring out, getting into Notion, and….
306 00:24:46.050 ⇒ 00:24:46.400 Matthew Good: Yeah.
307 00:24:46.400 ⇒ 00:24:48.160 Uttam Kumaran: That’s actually kind of done.
308 00:24:48.410 ⇒ 00:24:53.400 Uttam Kumaran: So, like, probably good ways to use this are, one, if there’s another part in your business, like, it kind of rhymes with this.
309 00:24:53.560 ⇒ 00:24:54.130 Matthew Good: Yeah.
310 00:24:54.420 ⇒ 00:24:57.570 Uttam Kumaran: Rip through that, or if you’re like, this is great.
311 00:24:57.840 ⇒ 00:25:02.580 Uttam Kumaran: Now, after the deck outline, I need the email I’m gonna send to the client.
312 00:25:02.760 ⇒ 00:25:11.759 Uttam Kumaran: I need, like, maybe you want us to try to use, like, Gamma or the new Gemini model to generate you the slides.
313 00:25:11.930 ⇒ 00:25:17.079 Uttam Kumaran: like, or generate sample versions, like, that’s how I would push on us, if I was in your spot.
314 00:25:17.080 ⇒ 00:25:17.750 Matthew Good: Yeah.
315 00:25:18.010 ⇒ 00:25:19.220 Uttam Kumaran: So that’s, like….
316 00:25:19.490 ⇒ 00:25:20.100 Matthew Good: Yeah.
317 00:25:20.100 ⇒ 00:25:21.950 Uttam Kumaran: Those are, like, two good ways to…
318 00:25:22.670 ⇒ 00:25:24.969 Uttam Kumaran: You know, that you can continue to leverage us.
319 00:25:24.970 ⇒ 00:25:39.040 Matthew Good: Totally, totally. And I… yeah, and I definitely… I definitely want to, because now I feel like we’ve got kind of the scaffolding in place. I have a better sense… a couple things. I have a better sense of, like, we need multiple different… like, a C deck is very different than a Series B deck, and I, like, knew that in my head, and I applaud….
320 00:25:39.040 ⇒ 00:25:40.050 Uttam Kumaran: V2, yeah.
321 00:25:40.050 ⇒ 00:25:51.500 Matthew Good: Yeah, I intuitively applied that thinking when I was building these decks, but now I’m like, oh yeah, you have… that’s, like, a different agent, realistically, right? And we have, like, 6 seed decks now, right? And, like, 3 Series B decks. Like, we have…
322 00:25:51.670 ⇒ 00:25:59.509 Matthew Good: Point one. Point two is that we have now more and more of these, like, finished decks in, like, PDF form that are, like, we had one
323 00:26:00.140 ⇒ 00:26:13.439 Matthew Good: YC Demo Day was, like, today or yesterday or something. Like, it was produced and shipped there, so we have better… kind of, like, that spreadsheet that you guys gave me, which I didn’t fill in as well as I wanted to, was because a lot of those projects were in progress.
324 00:26:13.440 ⇒ 00:26:13.780 Uttam Kumaran: Yeah.
325 00:26:13.780 ⇒ 00:26:23.140 Matthew Good: like, V1s, and now we’re like, we have those PDFs now that are done. So there’s, like, better just, like, input data on, like, what’s good and why.
326 00:26:23.450 ⇒ 00:26:30.089 Matthew Good: And yeah, 3. I think my focus… I would want our focus to be on that, just because I think it’s the…
327 00:26:30.230 ⇒ 00:26:39.069 Matthew Good: it’s the most value additive for, like, my time, and then also thinking about the business as, like, an asset. Like, be… being able to, like, have…
328 00:26:39.210 ⇒ 00:26:41.949 Matthew Good: Whatever, those kind of, like, proprietary, proprietary, whatever.
329 00:26:41.950 ⇒ 00:26:42.390 Uttam Kumaran: Yeah.
330 00:26:42.390 ⇒ 00:26:44.899 Matthew Good: tech flows around it. …
331 00:26:45.110 ⇒ 00:26:50.790 Matthew Good: is someone that I really want to do. I’m bringing on, like, in the interim, and I don’t know, interim, whatever.
332 00:26:51.040 ⇒ 00:27:00.020 Matthew Good: someone to, like, help me with, like, the stuff that I’m doing in Cloud right now, like, do the manual work, and, like, intake, and, like, get… basically, like.
333 00:27:00.380 ⇒ 00:27:04.590 Matthew Good: you know, ideally it’s an agent, but, like, this is a person that’s now gonna hand me the V1.
334 00:27:04.940 ⇒ 00:27:10.009 Matthew Good: 90% of the way done. So I’m gonna have that person at least, like, a contract basis, but I want to push towards being able to just
335 00:27:10.260 ⇒ 00:27:11.339 Matthew Good: Me or him.
336 00:27:11.340 ⇒ 00:27:13.309 Uttam Kumaran: That person you should have use this, dude.
337 00:27:13.310 ⇒ 00:27:17.619 Matthew Good: I was gonna say, I’m gonna have him now, because he’s starting this coming week, I’m gonna be like.
338 00:27:17.620 ⇒ 00:27:28.170 Uttam Kumaran: I mean, for us, like, what we typically find is, like, and this is the challenge, especially because, look, I’m also, like, own the business, is that for us to have time directly to give feedback to, like, a team like ours is tough.
339 00:27:28.290 ⇒ 00:27:36.339 Uttam Kumaran: But for us, it’s like, if we can… if you have someone internally that’s gonna start to use these and give us feedback, we iterate as fast as, like, we can get the feedback.
340 00:27:36.340 ⇒ 00:27:37.170 Matthew Good: Right. Back.
341 00:27:37.210 ⇒ 00:27:49.089 Uttam Kumaran: Right. So, that’s really helpful. Second is, like, I would tell you to just dream big. Like, think about where in the whole flow the time really sinks, but…
342 00:27:49.340 ⇒ 00:27:53.539 Uttam Kumaran: And, like, maybe start there. Whether it can be done or not.
343 00:27:53.770 ⇒ 00:27:56.330 Uttam Kumaran: I’m happy to think through that, but….
344 00:27:56.330 ⇒ 00:27:57.830 Matthew Good: Right. I feel like….
345 00:27:57.980 ⇒ 00:28:04.530 Uttam Kumaran: the real innovation here is gonna… is… like, this is a great start, but what I hope is I’ve shown you that, like.
346 00:28:04.750 ⇒ 00:28:20.950 Uttam Kumaran: we can take this really well-defined… we can set some well-defined process, define it, get you to an output where you are. The things that are really, really crazy in this world now is, one, what is the piece after this? The second is also images, or, like, design guidelines.
347 00:28:20.950 ⇒ 00:28:28.709 Uttam Kumaran: And, like, how you would enable either that person or, like, the next contractor to do them plus AI. Like, I think the….
348 00:28:28.710 ⇒ 00:28:29.290 Matthew Good: Yeah.
349 00:28:29.290 ⇒ 00:28:38.410 Uttam Kumaran: AI doing the whole thing, yes, I think it’s a good… it’s something that will happen, but in the short term, where you’re gonna get alpha in your business is someone junior.
350 00:28:38.410 ⇒ 00:28:39.020 Matthew Good: Yeah.
351 00:28:39.020 ⇒ 00:28:48.770 Uttam Kumaran: plus this is gonna get you, like, basically, you’re gonna be like, tweak, tweak, tweak, done. Versus that person doing it without AI is… it’s gonna take a while to get there.
352 00:28:48.770 ⇒ 00:28:49.270 Matthew Good: Take a long time.
353 00:28:49.270 ⇒ 00:28:51.490 Uttam Kumaran: Also, the AI… like…
354 00:28:51.830 ⇒ 00:28:56.429 Uttam Kumaran: It’s just, like, we actually could be okay if it’s, like, 20… if it’s 80% there.
355 00:28:56.430 ⇒ 00:29:00.080 Matthew Good: the person will fill out the rest, and you can ship these, you know? So that’s….
356 00:29:00.080 ⇒ 00:29:12.289 Uttam Kumaran: But also, do you need to just sit with us for an hour and think through those? Like, that’s a great way to use us, too, because I’m sort of putting it back onto you, but it’s hard… it’s like, if I can ask tough questions, get you to ideate on that.
357 00:29:12.470 ⇒ 00:29:15.930 Uttam Kumaran: like, that’s where I would do it. That’s how it works.
358 00:29:15.930 ⇒ 00:29:30.260 Matthew Good: Yeah, no, it’s super helpful, because it’s like, it’s not my world, like, I don’t know what is or isn’t possible. I’m just kind of, like, freestyle. I’m like, I have, like, a general intuition, like, this is the direction we need to push into, but, like, beyond that, like, I’m not 100% sure. ….
359 00:29:30.560 ⇒ 00:29:32.769 Uttam Kumaran: But I’m telling you, if, like, if you’re, like.
360 00:29:33.270 ⇒ 00:29:40.720 Uttam Kumaran: designing the slides is an opportunity, so how can… for me, I would be like, cool, can we start to encroach on lo-fi?
361 00:29:40.840 ⇒ 00:29:45.100 Uttam Kumaran: So at least your design team gets lo-fi guidelines of stuff.
362 00:29:45.200 ⇒ 00:29:49.079 Matthew Good: then you start to encroach into that layer. But again, I’m just spitballing, like.
363 00:29:49.080 ⇒ 00:30:03.740 Uttam Kumaran: If that’s not the most important thing, and instead you’re like, hey, I want to get… another example is, like, I want to give all of our clients a podcast version of their deck presentation alongside our deliverables, so they can listen to, like.
364 00:30:03.850 ⇒ 00:30:06.359 Uttam Kumaran: What the presentation could sound like.
365 00:30:07.450 ⇒ 00:30:18.449 Uttam Kumaran: That is another thing, right? So there’s these, like, we just have to have… we just have to think through whether it is either augmenting your existing workflows, or you’re like, I can have AI add something
366 00:30:18.650 ⇒ 00:30:34.940 Uttam Kumaran: basically penny cost that you can then upsell or add as an existing offering, those are also opportunities. Whether it’s audio, video, still images, dude, like, it’s all so possible right now for, like, very, very cheap.
367 00:30:34.940 ⇒ 00:30:35.839 Matthew Good: Yeah, yeah.
368 00:30:35.840 ⇒ 00:30:36.510 Uttam Kumaran: You know?
369 00:30:36.730 ⇒ 00:30:37.340 Matthew Good: Yeah.
370 00:30:37.600 ⇒ 00:30:39.570 Matthew Good: Cool. …
371 00:30:40.150 ⇒ 00:30:45.319 Matthew Good: This is awesome. I mean, this is, like, exactly what I wanted to get to for, like, the V1 of this. …
372 00:30:45.500 ⇒ 00:30:58.800 Matthew Good: So, I’m gonna bring that guy on this coming week, and I agree, what I wanna… the ideal state is, like, where I’m reviewing stuff before it goes out, but he’s, like, working with these tools, and, like, generating that output, and the tools continue to get smarter, because that’s the thing, is, like.
373 00:30:59.340 ⇒ 00:31:13.929 Matthew Good: as you know, obviously, a net new Cloud chat. It’s not gonna get smarter based on the past decks, and that was the big thing that was, like, bugging me. So now… I mean, and we could talk later about, like, how that process would work, of, like, do I send you a PDF of the deck, and you just, like, add to the knowledge base? That’s something that, like, we do? Is that, like.
374 00:31:13.930 ⇒ 00:31:30.890 Uttam Kumaran: So where we didn’t get to in this process, which, like, I was kind of bummed, was we didn’t get to the feedback loop sort of part of it, where you would be interacting with this, and you would say, hey, you could comment and say it’s not good for these reasons. That feedback loop gets back into our scoring.
375 00:31:30.930 ⇒ 00:31:46.879 Uttam Kumaran: So when we score, the LLM that scores the output is using your sample feedback to give better scores, and ideally, we get higher scores over time. So there has to be a kind of direct thumbs up, thumbs-down qualitative feedback loop that
376 00:31:47.040 ⇒ 00:31:52.749 Uttam Kumaran: comes right from as many iterations as we do. You know, so you get that… you basically get that immediately.
377 00:31:53.030 ⇒ 00:32:03.499 Uttam Kumaran: And if it’s just for just this, like, deck creation, as we deploy more agents, there’s more opportunities for that feedback, so that just gets smarter and smarter over time.
378 00:32:04.400 ⇒ 00:32:05.619 Matthew Good: Got it, okay.
379 00:32:05.810 ⇒ 00:32:13.659 Matthew Good: So, and that… okay, so that… that kind of, like, that would happen in Slack, where I would just be like, hey, this is… change this, this feedback, whatever, blah blah blah, and it would… and it would…
380 00:32:13.790 ⇒ 00:32:17.009 Matthew Good: automatically take that Slack message and add it to, like, the….
381 00:32:17.010 ⇒ 00:32:29.289 Uttam Kumaran: So there’s two things, like, one is, like, we have a database of what we call, like, our control… the golden data set, right? So that database, we want to get better and better over time. Ideally, we don’t want to…
382 00:32:29.550 ⇒ 00:32:42.779 Uttam Kumaran: it would… we don’t want to populate that with LLM-generated results, because then it’s sort of, like, eating… it’s sort of going to judge itself very highly. So as much as that is, like, control that the humans develop, that’s a good judge.
383 00:32:42.780 ⇒ 00:32:59.169 Uttam Kumaran: Second is, exactly, like, the feedback that you give in Slack can go back, and then we improve the prompts. So naturally, it’s like, if we’re finding very key things, like the tone or stuff is different, we go to modify the prompts. We use AI, we use your feedback
384 00:32:59.210 ⇒ 00:33:14.419 Uttam Kumaran: in our own prompts to then improve the existing prompts. But the prompt is only, like, part of the solution. We can add more to the multi-agent system. We can change the ordering of where things go. We can ask you for feedback before doing the whole thing.
385 00:33:14.420 ⇒ 00:33:33.160 Uttam Kumaran: Like, instead, we’re like, hey, this is a rough outline I’m gonna start with, do you have any feedback? Great. Then go into this next larger mode, then go larger. So it’s just gonna have… it’s just gonna be, like, UX we develop with you on. For example, if ripping the whole deck to start with is not a great place to start, then we should start maybe with, like, here’s the 10 slides, here’s, like, the gist.
386 00:33:33.690 ⇒ 00:33:36.759 Uttam Kumaran: What do you think? Okay, then I’m gonna take this and…
387 00:33:36.910 ⇒ 00:33:56.650 Uttam Kumaran: go further. Even… even beyond this, you could have an agent for every single slide. If you are very prescriptive on what the impact of these 10 slides are and what each slide is, then we just build even towards that narrow use case. What you’re gonna have to find out is, like, does the ROI plateau? Like, does doing that get you, like.
388 00:33:56.810 ⇒ 00:33:58.260 Uttam Kumaran: 5% better results.
389 00:33:58.260 ⇒ 00:33:59.090 Matthew Good: better, yeah.
390 00:33:59.090 ⇒ 00:34:05.039 Uttam Kumaran: Yeah, so that’s why I’m like, I’ll give you the feedback on that. It would be, like, kind of OD, I feel like.
391 00:34:05.040 ⇒ 00:34:09.999 Matthew Good: That feels OD, because clients also, they invariably have this story in their head.
392 00:34:10.000 ⇒ 00:34:23.650 Uttam Kumaran: Like, I don’t think this is very prescriptive. There are other use cases, though, where it actually is very deterministic in that way, and you can get that narrow. But, like, if this is 90% on point, plus the human, it’s gonna get you to the shippable thing, you know?
393 00:34:23.650 ⇒ 00:34:36.149 Matthew Good: Yeah, exactly, because you need… it’s like, to get something shippable… V1 is never perfect. I used to, like, spend myself, like, waste time trying to get a near-perfect V1 when the client would just be like, oh, this new thought I had in the shower last night means we’re changing this. I’m like, dude, we’ll fucking.
394 00:34:36.150 ⇒ 00:34:44.060 Uttam Kumaran: No, but it’s the same thing we’re dealing with, right? Where if you… the faster you can get them this version for them to give feedback on, maybe that’s the KPI.
395 00:34:44.060 ⇒ 00:34:49.409 Matthew Good: That… exactly. It’s like, if I can get the V1 that’s, like, a workable V1 that I’ve put thought into it, reviewed.
396 00:34:49.590 ⇒ 00:34:57.860 Matthew Good: and then that gets out the door more quickly, then we can, like, start that review process with them, because I’m… I’m becoming the bottleneck of doing that. …
397 00:34:58.570 ⇒ 00:35:01.109 Matthew Good: But yeah, okay, so in terms of next steps, like.
398 00:35:02.210 ⇒ 00:35:06.680 Matthew Good: Well, one, this was great, thank you guys for building this, it was super helpful. …
399 00:35:07.040 ⇒ 00:35:15.229 Matthew Good: two, would love to, like, keep doing stuff together, like I kind of talked about. There’s a couple ideas in my head, I need to just, like, I guess, write them down, like, in bullet points.
400 00:35:15.230 ⇒ 00:35:19.580 Uttam Kumaran: Well, like, what I I mean, I think we should just do, like, a session like this.
401 00:35:19.850 ⇒ 00:35:20.290 Matthew Good: Okay.
402 00:35:21.060 ⇒ 00:35:37.640 Uttam Kumaran: you know, I think you should just literally… we can record this, we’ll use a transcript, but, like, I think it’s clear that me asking you these, like, questions and getting deeper is, like, would get us there, so, like, and I would rather you not try to, like, go find, like, an hour of time and go do that. Like, I’ll literally just grill you.
403 00:35:37.760 ⇒ 00:35:43.259 Uttam Kumaran: on a call about this, we’ll find the answers. Like, if that’s helpful, then, like, that’s something that we should do.
404 00:35:43.260 ⇒ 00:35:45.550 Matthew Good: Yeah, that would be super helpful. …
405 00:35:45.850 ⇒ 00:35:48.439 Matthew Good: And just, yeah, cause I’ve… the way I’m, like…
406 00:35:48.870 ⇒ 00:35:56.330 Matthew Good: thinking about, like, I used to just be in, like, the Cloud desktop, now I’m like, okay, let me use Workbench, now let me get these, like, different agents. It’s, like, starting to evolve, and I’m just, like.
407 00:35:56.630 ⇒ 00:36:00.359 Matthew Good: I don’t really have a rhyme or reason for, like, doing it, I’m just like, well, this seems like a better output, so I’m like.
408 00:36:00.360 ⇒ 00:36:00.940 Uttam Kumaran: Yeah.
409 00:36:01.190 ⇒ 00:36:07.149 Matthew Good: But having someone that, like, has… like, that’s your domain of expertise much more than it is mine. …
410 00:36:07.260 ⇒ 00:36:14.200 Matthew Good: So yeah, I’m down to do that, and figure out kind of, like, the right surface area for what the next iteration of this would be based on.
411 00:36:14.200 ⇒ 00:36:15.700 Uttam Kumaran: than you have right now.
412 00:36:16.000 ⇒ 00:36:16.560 Uttam Kumaran: Okay.
413 00:36:16.970 ⇒ 00:36:29.489 Uttam Kumaran: Okay, cool. So then let me send you a text, I mean, let’s talk, if you want to run in the same, like, kind of structure we’ve been doing it, let’s do that, and then… Yeah. We can book, like, another meeting to.
414 00:36:29.490 ⇒ 00:36:29.880 Matthew Good: Yeah, yeah.
415 00:36:30.000 ⇒ 00:36:32.710 Uttam Kumaran: I think, like, you know, especially thinking about my time, and I…
416 00:36:32.940 ⇒ 00:36:51.389 Uttam Kumaran: we’ve done things. If I sit with you and sort of grill you, it’s gonna all come out, and then I will… I then, kind of, the role I’m playing is, like, as solutions, it’s like, we’ll go to the team, be like, okay, couple proof of concepts in this, this, and this we should just rip. We’ll at least have something demoable that you can see. My only other ask is.
417 00:36:51.490 ⇒ 00:37:01.880 Uttam Kumaran: grill this agent, like, try to use it this week, or next week, if you can, and see what breaks. Like, that’s the only way we’re gonna be able to see the edge cases.
418 00:37:02.670 ⇒ 00:37:11.990 Uttam Kumaran: we’ve got it to the point where, like, I feel kind of good, but again, like, I don’t know your flow as well as you do, so… Yeah. When a new guy starts, on day one, you should say, this is the way
419 00:37:12.110 ⇒ 00:37:17.990 Uttam Kumaran: We do things, like… I think his or her feedback will be best to, you know.
420 00:37:17.990 ⇒ 00:37:26.349 Matthew Good: Okay, got it. Yeah, yeah, absolutely. I’ll do that when he starts, it should be on Tuesday, because I’ve got the holiday on Monday, and then I’ll, you know, I’ll shoot you a text and find some time.
421 00:37:26.470 ⇒ 00:37:27.420 Matthew Good: Bye.
422 00:37:27.670 ⇒ 00:37:28.420 Matthew Good: Coming up.
423 00:37:28.910 ⇒ 00:37:29.550 Uttam Kumaran: Okay.
424 00:37:29.550 ⇒ 00:37:30.500 Matthew Good: Cool. Second.
425 00:37:30.500 ⇒ 00:37:33.809 Uttam Kumaran: Yeah, this is the project, this is great. I feel like the…
426 00:37:34.030 ⇒ 00:37:36.590 Uttam Kumaran: We got to do a lot more, like, writing-related
427 00:37:36.710 ⇒ 00:37:43.620 Uttam Kumaran: prop work than we usually do. I feel like most of our stuff is, like, more technical, like, so it was nice being able to build this system, so it’s cool.
428 00:37:43.620 ⇒ 00:37:51.289 Matthew Good: Yeah, man, it’s great. I’m excited for what we can, like, how we can… like, the next iteration of this, whatever it’ll look like. So, appreciate it. Thanks, guys.
429 00:37:51.410 ⇒ 00:37:54.499 Matthew Good: Thank you. Appreciate the time. Yeah, I’ll shoot you a text for Tom.
430 00:37:54.870 ⇒ 00:37:55.519 Mustafa Raja: Thanks, Q.