2026-02-16_brainforge_x_ey_ai_strategy_sync_cf07ade7

Meeting Title: Brainforge x EY AI Strategy Sync Date: 2026-02-16 Meeting participants: Uttam Kumaran, Michael Tran

---

WEBVTT

1 00:01:17.490 ⇒ 00:01:18.660 Uttam Kumaran: Hey, Michael.

2 00:01:20.970 ⇒ 00:01:21.900 Michael Tran: Hey!

3 00:01:22.380 ⇒ 00:01:23.630 Uttam Kumaran: Hey, how are you?

4 00:01:24.320 ⇒ 00:01:26.699 Michael Tran: Good. Give me one second. I’m trying to…

5 00:01:26.700 ⇒ 00:01:27.210 Uttam Kumaran: kid.

6 00:01:28.810 ⇒ 00:01:34.030 Michael Tran: get my video to work. I’m using my personal laptop, which I don’t really do video calls on, so…

7 00:01:34.270 ⇒ 00:01:36.299 Uttam Kumaran: Okay, I appreciate it.

8 00:01:42.720 ⇒ 00:01:43.830 Michael Tran: There you go.

9 00:01:44.650 ⇒ 00:01:45.729 Uttam Kumaran: How’s everything?

10 00:01:46.000 ⇒ 00:01:46.830 Michael Tran: Good.

11 00:01:48.870 ⇒ 00:01:49.330 Uttam Kumaran: Busy?

12 00:01:50.770 ⇒ 00:01:59.080 Michael Tran: Yeah, definitely busy, and yeah, busy, and we just,

13 00:01:59.470 ⇒ 00:02:03.020 Michael Tran: I just got access to, like, Factory AI droids.

14 00:02:03.260 ⇒ 00:02:04.000 Uttam Kumaran: Nice.

15 00:02:04.630 ⇒ 00:02:11.490 Michael Tran: And so, I’m like, Working within it now, to… to try to… Kind of gold.

16 00:02:11.490 ⇒ 00:02:12.169 Uttam Kumaran: Appreciate everything.

17 00:02:12.170 ⇒ 00:02:14.490 Michael Tran: And build, like, a strategy and stuff, and so…

18 00:02:15.230 ⇒ 00:02:24.710 Michael Tran: Dude, the pace of, like, working with AI and, like, the density and volume of information and decisions that I have to make is just, like, insane.

19 00:02:24.970 ⇒ 00:02:30.420 Uttam Kumaran: I feel like a lot has changed, also, because I’ve been so… I built this company using, like.

20 00:02:30.720 ⇒ 00:02:35.279 Uttam Kumaran: AI the whole way, like, I… we started this business, like, maybe two and a half years ago.

21 00:02:35.500 ⇒ 00:02:35.860 Michael Tran: Yeah.

22 00:02:35.860 ⇒ 00:02:46.189 Uttam Kumaran: And… but in the last 3 months, I’m starting to be unable to keep up. Like, I… I… it’s my job to sort of, like, really be on the nose on everything, and right.

23 00:02:46.810 ⇒ 00:03:05.939 Uttam Kumaran: it’s nice because I was doing that in my… I was doing that even before this, but I was always doing it in my free time. Now it’s like, oh great, like, I… in order to do well, I have to, like, do… basically be on Twitter all the time and, like, talk to everybody and understand and try things, but in the last 3 months, it’s really moved very fast.

24 00:03:06.250 ⇒ 00:03:07.570 Uttam Kumaran: And so…

25 00:03:07.860 ⇒ 00:03:17.570 Uttam Kumaran: It’s… it’s just super hard to… it’s also super hard to understand, like, what is just, like, a toy or a flash in the pan versus what is, like, an enterprise-grade solution.

26 00:03:17.570 ⇒ 00:03:17.950 Michael Tran: Right?

27 00:03:17.950 ⇒ 00:03:23.380 Uttam Kumaran: But there’s also things about AI that, like, haven’t changed, like, for example, just, like, having great context.

28 00:03:23.730 ⇒ 00:03:25.470 Uttam Kumaran: Like…

29 00:03:25.610 ⇒ 00:03:43.290 Uttam Kumaran: choosing, like, not choosing, like, a random startup, but choosing, like, some foundational tools that, like, have skin in the game. Right. It’s also, like, I feel like only in the last, like, six, seven months did, like, a lot of the older, like, I say Snowflake, like, an old platform, but, like, finally, they finally caught up with, like, Cortex.

30 00:03:43.630 ⇒ 00:03:44.060 Michael Tran: Yeah.

31 00:03:44.060 ⇒ 00:03:53.329 Uttam Kumaran: Right? And they waited for a while to release it. I don’t still think there’s still, like, a lot of hiccups, but, like, it’s helpful because now I can go to clients and I don’t have to pitch, like.

32 00:03:53.450 ⇒ 00:03:56.770 Uttam Kumaran: other tools, I can be like, okay, Snowflake’s gonna offer a lot of this, how’s…

33 00:03:57.240 ⇒ 00:03:57.600 Michael Tran: Yeah.

34 00:03:57.600 ⇒ 00:03:58.450 Uttam Kumaran: the box.

35 00:03:58.770 ⇒ 00:04:03.980 Uttam Kumaran: And… But ultimately, like, I still think that the fact that just, like, you’re…

36 00:04:04.840 ⇒ 00:04:19.029 Uttam Kumaran: your average teammate needs to be using AI, but beyond ChatGPT. I think everybody’s trying to use ChatGPT for, like, write me this email, or they throw CSV in, but, like, the step above that is what is, like, needs to happen now, where it’s, like, a shared context layer.

37 00:04:19.029 ⇒ 00:04:26.289 Uttam Kumaran: Like, at our company, everybody uses Cursor, from sales, marketing, like, everybody. The engineers, of course, like, they’re totally fine, but, like.

38 00:04:26.460 ⇒ 00:04:31.070 Uttam Kumaran: we’ve been using Cursor for engineering work for, like, more than a year now, so that’s, like, table stakes.

39 00:04:31.330 ⇒ 00:04:36.590 Uttam Kumaran: I’m more like, hey, the people that are, like, Like, thinking through marketing strategy.

40 00:04:36.890 ⇒ 00:04:43.470 Uttam Kumaran: what the hell? Like, why aren’t they on Cursor 2? Like, all of our context is there, and they’re gonna have to take a bigger jump.

41 00:04:43.740 ⇒ 00:04:48.339 Uttam Kumaran: Because they’ve never even used GitHub, they’ve never used an IDE in their life.

42 00:04:49.060 ⇒ 00:04:49.560 Uttam Kumaran: But…

43 00:04:49.560 ⇒ 00:04:49.980 Michael Tran: Yeah, no.

44 00:04:49.980 ⇒ 00:04:54.590 Uttam Kumaran: the cursor is the ideal… it’s like an ideal UX for a lot of the knowledge work that we do.

45 00:04:55.150 ⇒ 00:04:59.570 Uttam Kumaran: But you can slot in anything. You can slot Claude for work, you can slot in…

46 00:04:59.790 ⇒ 00:05:07.990 Uttam Kumaran: anything into that cursor, but, like, okay, so how does the Brainforge platform enable the average BrainForge employee to use AI?

47 00:05:08.100 ⇒ 00:05:20.780 Uttam Kumaran: okay, we need, like, a sick context layer. We need a lot of, like, SOPs for, like, how to use something like this. Yep. And I need to reinforce that, like, if you’re asking a question in Slack, did you try to ask Cursor first?

48 00:05:20.780 ⇒ 00:05:21.280 Michael Tran: Right.

49 00:05:21.280 ⇒ 00:05:25.540 Uttam Kumaran: like, and I have to be that sort of annoying person, but there’s all these, like, things that…

50 00:05:25.990 ⇒ 00:05:28.750 Michael Tran: To get back to your point, yes, it’s moving fast.

51 00:05:28.750 ⇒ 00:05:32.520 Uttam Kumaran: But I think, like, it’s working for us. It’s really, really working for us.

52 00:05:32.780 ⇒ 00:05:33.300 Uttam Kumaran: Yeah.

53 00:05:33.300 ⇒ 00:05:36.579 Michael Tran: Yeah, it is, it is a big challenge, and, like, I think…

54 00:05:37.120 ⇒ 00:05:45.930 Michael Tran: Well, one of the challenges that I have, that maybe you don’t… you wouldn’t really face as much is, like, we’re pretty locked down on what we can use.

55 00:05:46.290 ⇒ 00:05:54.549 Michael Tran: And so, like, I have access to Droid, and I’m using it, and I’m, like, getting my, like, immediate team access, but it’s still, like, not a…

56 00:05:54.750 ⇒ 00:05:59.199 Michael Tran: Well, deployed application across, like, the organization.

57 00:05:59.200 ⇒ 00:05:59.780 Uttam Kumaran: Yeah.

58 00:06:00.300 ⇒ 00:06:02.320 Michael Tran: And also, I’m like.

59 00:06:02.500 ⇒ 00:06:07.979 Michael Tran: terrified of giving access to Droid to, like, the average person on my, like, greater team.

60 00:06:08.160 ⇒ 00:06:13.319 Michael Tran: Right? Because they’re just gonna turn into, like, slop cannons, and it’s just like, you know…

61 00:06:13.320 ⇒ 00:06:15.110 Uttam Kumaran: Yeah, yeah, yeah.

62 00:06:15.110 ⇒ 00:06:19.280 Michael Tran: Because, like, oh my god, I can do all this stuff, and then you just, like, if you haven’t really thought about, like.

63 00:06:19.620 ⇒ 00:06:29.450 Michael Tran: context and thought about, you know, how to manage it, right? When you have the ability to produce so much content, it can get overwhelming, and so, like, I’m just, like, terrified of that.

64 00:06:29.620 ⇒ 00:06:33.239 Michael Tran: But I’m trying to, like, spread it out, at least within my, like, tight team.

65 00:06:33.350 ⇒ 00:06:35.770 Michael Tran: So we can kind of, like, figure out how to use it together.

66 00:06:35.940 ⇒ 00:06:39.529 Michael Tran: And, like, Constance is really important, and it’s crazy because, like.

67 00:06:39.970 ⇒ 00:06:45.820 Michael Tran: you know, I’m putting together the, the strategy of, like, What we’re going to

68 00:06:46.010 ⇒ 00:06:53.479 Michael Tran: go to market with? Like, what is it that is going to differentiate what we offer from, like, what exists today, and, you know…

69 00:06:53.620 ⇒ 00:06:58.330 Michael Tran: where can we make use of, like, our inherent advantages at EY, right?

70 00:06:58.860 ⇒ 00:06:59.970 Michael Tran: And so…

71 00:07:00.120 ⇒ 00:07:10.080 Michael Tran: you know, designing features and, like, a commercialization plan and all that stuff, and it’s just, like, a ton of context within, within, like, a… basically a Git repo, right? It’s just, like, a folder structure.

72 00:07:10.350 ⇒ 00:07:11.000 Uttam Kumaran: Yes.

73 00:07:11.000 ⇒ 00:07:17.029 Michael Tran: And then, what I’m thinking about doing is… and I’ve tested it a little bit, like, we got an RFP come in from a client.

74 00:07:17.240 ⇒ 00:07:27.590 Michael Tran: And I just, like, put it into my repo, and I was like, hey, let’s see where the gaps are and where we’re, like, really well positioned. Dude, phenomenal, right? Just, like…

75 00:07:27.590 ⇒ 00:07:34.540 Uttam Kumaran: No, you’re gonna… wait till you… if I tell you how… so, exactly the thing that you’re doing is what I decided to do, like.

76 00:07:34.690 ⇒ 00:07:35.850 Uttam Kumaran: 4 months ago, is…

77 00:07:35.850 ⇒ 00:07:36.170 Michael Tran: Yeah.

78 00:07:36.170 ⇒ 00:07:42.939 Uttam Kumaran: We have a fat monorepo structure for everything. And, I mean, I’ll show you,

79 00:07:43.080 ⇒ 00:07:52.200 Uttam Kumaran: You know, because it’ll hopefully give you some… some inspiration. But we have, here, let me just kind of show you, like, what it looks like.

80 00:07:52.390 ⇒ 00:07:57.070 Uttam Kumaran: So… This is our…

81 00:07:58.550 ⇒ 00:08:04.260 Uttam Kumaran: this is our, just, like, main platform. We have a couple, like, core folders.

82 00:08:04.380 ⇒ 00:08:15.890 Uttam Kumaran: we have, one, we have, like, we have an internal UI, so I don’t know if EY has, like, sort of an internal platform, but we have, like, helpful UIs, and, like, we’re building things into that, so that’s… all of that lives under, like.

83 00:08:16.060 ⇒ 00:08:35.120 Uttam Kumaran: a thing called, like, our apps, and we have, like, our platform, we have, like, Slack applications we’re building, MCP servers we’re building. Additionally, we have, like, what’s called, playbooks. So playbooks are sort of, like, I would say, the canonical SOWs for how to do things.

84 00:08:35.120 ⇒ 00:08:38.919 Michael Tran: It’s not… it’s not the materials need… you need to do them.

85 00:08:39.100 ⇒ 00:08:45.320 Uttam Kumaran: But it is, like, the, how do I write an SOW, right? So in here, we have, like, how do I write a PRD?

86 00:08:45.880 ⇒ 00:08:46.370 Michael Tran: Yeah.

87 00:08:46.370 ⇒ 00:09:00.539 Uttam Kumaran: Right? How do I… how do I set up a project? How do I write an SOW? And for example, this is where, like, my thing I’m gonna probably repeat 100 times is, like, I don’t need my team to…

88 00:09:00.980 ⇒ 00:09:05.539 Uttam Kumaran: I’m not trying to convince them to use Cursor. They need to use it.

89 00:09:06.000 ⇒ 00:09:24.679 Uttam Kumaran: But also, like, I don’t need them to understand how it works either, right? They just need to use it. I’ve, like, passed that point, because, like, by the time I could show everybody how, why monorepo structure, why Kersher’s good at indexing, tons of stuff, this agentic reasoning, no. They just need to use it.

90 00:09:24.920 ⇒ 00:09:26.899 Uttam Kumaran: So, for example.

91 00:09:27.230 ⇒ 00:09:39.540 Uttam Kumaran: you have a client, they said, hey, I would like you guys to help with this. You have that meeting transcript. You could literally put it in here and just say, like, hey, I need help writing an SOW,

92 00:09:39.840 ⇒ 00:09:41.220 Uttam Kumaran: Tell me how

93 00:09:41.640 ⇒ 00:09:57.969 Uttam Kumaran: And it’s… it’s literally gonna go through our thing. It’s gonna look through and say, what’s the playbook for SOW templates? And it’s gonna… in there, we have not only information on, like, how to write a really good one, we have examples of past really good ones.

94 00:09:57.970 ⇒ 00:09:58.360 Michael Tran: Yeah.

95 00:09:58.360 ⇒ 00:10:03.639 Uttam Kumaran: Of course, like, we… because all of our SNLWs don’t ever… we were previously freestyling, meaning, like.

96 00:10:04.080 ⇒ 00:10:23.369 Uttam Kumaran: one person would write a certain way, one person… we have guidelines, right? And you guys have guidelines. And, like, okay, so now I actually… all I need is my people to go from transcript to SOW fast, and get to, like, the B… the B-bindness version of it, and send it for review. Because then it’s really where all the action is. The action is not on this part.

97 00:10:23.490 ⇒ 00:10:38.600 Uttam Kumaran: Right. The action is on the refinements and the, like, well, strategically, like, should we consider layering on more? How do we loop this in with another thing? Like, that’s the thing that, like, we need to spend time on. And so we now drive from meeting to SOW is, like, shrunk to, like, nothing.

98 00:10:38.680 ⇒ 00:10:43.889 Uttam Kumaran: The other thing that’s happened is I can now go to my delivery engineers and say, hey.

99 00:10:44.060 ⇒ 00:10:53.159 Uttam Kumaran: I don’t need to hire account managers. You guys can now do SOWs. So I’ve increased the, like, the TAM of people that can do SOWs now, right? I’ve, like.

100 00:10:53.160 ⇒ 00:10:53.600 Michael Tran: Yeah.

101 00:10:53.600 ⇒ 00:11:06.720 Uttam Kumaran: it down a notch. And so, if you expand this to, like, all parts of, like, sort of client delivery work at our size, like, we’re starting to do. So SOPs, PRDs, and then also it’s just, like, helpful prompts for.

102 00:11:06.990 ⇒ 00:11:11.849 Uttam Kumaran: Meeting summarization, creating great linear tickets, like.

103 00:11:12.100 ⇒ 00:11:15.839 Uttam Kumaran: I almost wanted to be like, hey, I am so-and-so on a project.

104 00:11:16.390 ⇒ 00:11:26.159 Uttam Kumaran: like, I want to do X activity associated with a project, create tickets, create meeting summaries, prep an agenda, work on a Gantt chart.

105 00:11:26.370 ⇒ 00:11:31.759 Uttam Kumaran: help me do that, and we should be able to do that. So, playbooks is the guidelines. This is, like.

106 00:11:32.080 ⇒ 00:11:44.650 Uttam Kumaran: these are, like, the rules, right? These are, like, the cursor rules for, like, knowledge work, basically, if you think about it that way. And then we have all our vaults. So in our vault is really where it’s, like, all of the content.

107 00:11:44.700 ⇒ 00:11:56.840 Uttam Kumaran: Right? So, like, for example, like, who is… who is on our team? What are our OKRs? On the go-to-market side, we have all of our marketing assets are here. Like, our… how do we do…

108 00:11:57.000 ⇒ 00:12:02.699 Uttam Kumaran: How do we do pricing? Because again, I don’t want my team to think about how to do pricing, because it saves.

109 00:12:02.700 ⇒ 00:12:03.300 Michael Tran: Right.

110 00:12:03.300 ⇒ 00:12:12.899 Uttam Kumaran: sales has a scheme for pricing, right? Like, how do we do our… if it’s an hourly approach, how do we think about it? Is it a strategic client versus, like, an existing relationship?

111 00:12:13.030 ⇒ 00:12:32.989 Uttam Kumaran: But again, the… they… how would this have been done prior? Oh, we would have, like, a pricing policy guide, nobody would ever read that. SOW just gets sent to the pricing team, and then they… probably they don’t even… nobody there even read it, either. I don’t… I don’t care. I don’t want to eliminate that. The SOW people should spend their time, write the first version.

112 00:12:32.990 ⇒ 00:12:40.689 Uttam Kumaran: get the SOW reviewed. Then, once that’s reviewed, they can send it back into Cursor for pricing, send all of that to sales.

113 00:12:41.590 ⇒ 00:12:49.550 Uttam Kumaran: And then sales can sprinkle a little couple things on top, and we get it out the door. And that’s so much better than, like.

114 00:12:49.700 ⇒ 00:13:05.890 Uttam Kumaran: where this gets hung up on, it’s like an SOW takes… it’s like a couple weeks to write, gets reviewed, someone’s off, blah blah blah, it’s now a month later, the client is like, oh, we’re moved on, like, that’s not… you know, all that nonsense. And again, think about a company my size, like.

115 00:13:06.150 ⇒ 00:13:10.330 Michael Tran: we have nobody. We have, like, most of our team is delivery folks, so… Right.

116 00:13:10.660 ⇒ 00:13:19.129 Uttam Kumaran: this was birthed out of necessity. Like, I don’t… I can’t hire… I don’t want to hire more people that don’t work on clients, and so…

117 00:13:20.160 ⇒ 00:13:22.270 Uttam Kumaran: This is what we use, yeah.

118 00:13:22.270 ⇒ 00:13:24.800 Michael Tran: the pattern is great, and like, you know, I think…

119 00:13:24.900 ⇒ 00:13:44.429 Michael Tran: Because, like, for us, you know, we have a ton of SharePoint sites, right? Different team sites for different clients. And, like, my practice, which is, like, the tax technology transformation, we have, like, 350 people in the US. So it’s like a big practice, right? And so, I mean, everyone works on, like, different kinds of things, and there’s a whole gamut of things, but…

120 00:13:44.690 ⇒ 00:13:57.989 Michael Tran: you know, whenever we have a new client where we need to do some SOW, or put together, like, a response, you know, for, like, an RFP or something, you know, people are, like, the senior managers are, like, pinging each other, like, hey, have you had a client similar to this, or I know you.

121 00:13:57.990 ⇒ 00:13:58.539 Uttam Kumaran: What’s this movie?

122 00:13:58.540 ⇒ 00:14:02.570 Michael Tran: can I see your SOW, right? And so, like, we’re kind of doing this stuff in the background.

123 00:14:02.570 ⇒ 00:14:03.469 Uttam Kumaran: Yeah, yeah, yeah.

124 00:14:03.470 ⇒ 00:14:04.480 Michael Tran: And, like, all these…

125 00:14:04.480 ⇒ 00:14:05.579 Uttam Kumaran: All the DMs, yeah.

126 00:14:05.580 ⇒ 00:14:09.299 Michael Tran: Yeah, and all these SharePoint sites, and all the files that are on them.

127 00:14:09.530 ⇒ 00:14:11.369 Michael Tran: They basically go there to die.

128 00:14:11.550 ⇒ 00:14:17.380 Michael Tran: Until, like, someone remembers that, like, oh, you know, someone’s working on something, then you go back there and look at it, right? And so…

129 00:14:17.380 ⇒ 00:14:25.229 Uttam Kumaran: That’s exactly, it’s like, hey, we have this client, do they rhyme with anyone we work for? But, like, why isn’t that a… that should be a question to AI. Right.

130 00:14:25.560 ⇒ 00:14:38.540 Michael Tran: Yeah, but the thing is, when you have AI running on it, like, all those files are not dead, right? They get brought into context, like, whenever needed, if you have it set up well. Yeah. So now, it’s like, you’re improving everything as you go, right? And so…

131 00:14:38.540 ⇒ 00:14:39.820 Uttam Kumaran: Yes, yes.

132 00:14:39.820 ⇒ 00:14:41.540 Michael Tran: This is crazy, though, like, you have, you have…

133 00:14:41.540 ⇒ 00:14:57.290 Uttam Kumaran: So, another process, and this is why, like, our SOWs and our pricing, this is, like, sort of matured parts of our repo, but, like, for example, in the beginning, we had, like, some guidelines on SOWs, but it caused us to ask questions, like.

134 00:14:57.420 ⇒ 00:15:01.930 Uttam Kumaran: in this situation, what do we do? And, like, okay, we need to codify that back into the rule, so…

135 00:15:02.260 ⇒ 00:15:17.669 Uttam Kumaran: One thing that we often do is, like, we’ll be working on something, we’ll realize that there is not already a policy or a rule around it, and as part of the end of my thread, I will say, cool, now that we’ve finished this, any learnings from this process that we can go right back to playbooks, so that everybody can get this, right?

136 00:15:17.670 ⇒ 00:15:18.030 Michael Tran: Yeah.

137 00:15:18.180 ⇒ 00:15:23.850 Uttam Kumaran: And so that’s… but that’s not something that I asked… I need everybody on my team to do.

138 00:15:23.960 ⇒ 00:15:33.479 Uttam Kumaran: Right? That’s because other… it’s too complicated for them. Yeah. Again, I… this is where, like, you know, me and Clarence kind of argue, because I think he is a lot more of, like.

139 00:15:33.940 ⇒ 00:15:40.190 Uttam Kumaran: I think he has a lot of faith in people sometimes, and he’s like, they should all be writing back to the repo, and I’m like.

140 00:15:40.800 ⇒ 00:15:49.330 Uttam Kumaran: I… in order to win, I just need them to use it. Right. Just use it, and we will win. I don’t need them to think about, like.

141 00:15:49.820 ⇒ 00:15:57.460 Uttam Kumaran: cursor rules, or adding back contacts, I just need them, in the moment of doubt, on anything, they need to use this.

142 00:15:57.600 ⇒ 00:16:12.230 Uttam Kumaran: And, that’s sort of been our driving MO, and it’s working in our company, and we are biasing towards the least AI native, the least AI-conscious person needs to be able to use this. And, like, we are attacking that, like, every single day.

143 00:16:12.660 ⇒ 00:16:19.719 Uttam Kumaran: We’re going through activity by activity of a person on a client, and being like, is this possible using

144 00:16:19.920 ⇒ 00:16:22.429 Uttam Kumaran: using Cursor and using the platform.

145 00:16:23.090 ⇒ 00:16:24.380 Uttam Kumaran: If not, why not?

146 00:16:24.550 ⇒ 00:16:28.449 Uttam Kumaran: And how much time savings could be accomplished if they did, and like…

147 00:16:28.910 ⇒ 00:16:37.450 Uttam Kumaran: stack rank, and, like, let’s go. Like, is it one-on-one trainings? Is it, like, the fact that people are just lazy, and I need to go yell at somebody, like…

148 00:16:37.620 ⇒ 00:16:39.569 Michael Tran: Right? And, and .

149 00:16:40.240 ⇒ 00:16:44.879 Uttam Kumaran: Yeah, I mean, we’re finding that it’s working super, super, super, super well.

150 00:16:45.280 ⇒ 00:16:47.799 Michael Tran: That’s good to hear, because, like, I think, you know.

151 00:16:48.160 ⇒ 00:16:56.170 Michael Tran: how I’ve been thinking about this is… because basically, like I said, not everyone’s going to have a droid, and I don’t want them all to have droids. I mean.

152 00:16:56.580 ⇒ 00:17:03.810 Michael Tran: there might be a decision made, like, way above my pay grade to give our practice odd droids, but I don’t… I would… I would disagree with that.

153 00:17:03.940 ⇒ 00:17:09.569 Michael Tran: But everyone has access to, like, Copilot, right, and M365 Copilot, and so…

154 00:17:10.190 ⇒ 00:17:14.150 Michael Tran: what I’m gonna be doing is, you know, once we have this repo in a decent place.

155 00:17:14.280 ⇒ 00:17:22.529 Michael Tran: After every session ends, I have this skill that runs, and it just, like, does a bunch of, like, housekeeping things. So one of them is gonna be pushing this repo to, like, a SharePoint site.

156 00:17:22.740 ⇒ 00:17:26.410 Michael Tran: Right? And basically, everyone on my team is going to, like, basically have.

157 00:17:26.410 ⇒ 00:17:32.620 Uttam Kumaran: Are you guys doing any, like, RAG or anything on SharePoint stuff? Like, how are you,

158 00:17:32.750 ⇒ 00:17:36.309 Uttam Kumaran: Yeah, just talk to me about that, because that’s, I think.

159 00:17:37.140 ⇒ 00:17:42.910 Uttam Kumaran: Part of the… probably the most crucial thing is, like, that, plus, like, some type of, like, these playbooks.

160 00:17:42.910 ⇒ 00:17:45.239 Michael Tran: Yeah. How to do things, like, how to steer.

161 00:17:45.240 ⇒ 00:17:47.130 Uttam Kumaran: You’re a droid, basically.

162 00:17:47.130 ⇒ 00:18:02.419 Michael Tran: Well, yeah, and so, like, for the droid piece, I’m approaching it from… I have some playbooks that I’ve created. I also have skills that come with the repo. Like, factory AI, you can create skills and droids, which is, like, sub-agents that do specific things.

163 00:18:02.610 ⇒ 00:18:06.590 Michael Tran: It’s like, they come with the repo, so if you pull the repo down, you get those…

164 00:18:06.730 ⇒ 00:18:12.750 Michael Tran: those skills and those droids, if you’re using factory.ai, and you just spin it up and say, hey, you know, I want to get started.

165 00:18:13.350 ⇒ 00:18:19.090 Michael Tran: access to all those, you have access to, like, the session start, session end, like, skills and everything, and so, like, everyone kind of…

166 00:18:19.310 ⇒ 00:18:22.179 Michael Tran: like… the way I’m looking at it is, like.

167 00:18:22.340 ⇒ 00:18:33.570 Michael Tran: yes, I want the person who’s using this to kind of understand what’s happening, but I really want their agent, who is, like, their teammate, to, like, have a much deeper understanding of, like, how to use this repo and, like, what we’re trying to do.

168 00:18:33.950 ⇒ 00:18:34.440 Uttam Kumaran: Yes.

169 00:18:34.480 ⇒ 00:18:43.700 Michael Tran: A lot of stuff is, like, in these MD files that are really just designed for their droid to read through and get, like, full context of everything, and know how to, like, start working.

170 00:18:44.120 ⇒ 00:18:46.210 Uttam Kumaran: Yes, that’s exactly right.

171 00:18:46.580 ⇒ 00:18:52.790 Michael Tran: Yeah, but for, like, the rest of the people, right? Because, like, basically, you know, my little experiment with, like, just putting an RFP in, like.

172 00:18:52.980 ⇒ 00:18:58.949 Michael Tran: awesome, right? And so, I want that capability for, like, all the people that are going out to talk to clients. And we have…

173 00:18:58.950 ⇒ 00:18:59.360 Uttam Kumaran: Yes.

174 00:18:59.360 ⇒ 00:19:04.900 Michael Tran: and tons of people, so I want them to be able to, like, put an RFP into, like, into,

175 00:19:05.650 ⇒ 00:19:10.480 Michael Tran: co-pilot, And just have it, like… and they already have access to, like, the…

176 00:19:10.630 ⇒ 00:19:17.530 Michael Tran: You know, kind of the repo, or maybe we create, like, a sales repo or something like that with roadmap and features and things like that, and some playbooks.

177 00:19:17.730 ⇒ 00:19:31.409 Michael Tran: and have Copilot go in there and, like, understand it, and then tie it back to it, right? And we’ll have to figure out, like, you know, how to exactly make Copilot do that specifically, but the biggest thing I want to do is, like, I want to make it available.

178 00:19:31.980 ⇒ 00:19:32.400 Uttam Kumaran: Yeah.

179 00:19:32.400 ⇒ 00:19:37.480 Michael Tran: and meet the team where they are, which is, like, not gonna be working in a terminal on Droid.

180 00:19:37.480 ⇒ 00:19:37.850 Uttam Kumaran: Yes.

181 00:19:37.850 ⇒ 00:19:39.220 Michael Tran: Maybe it’s been up co-pilot.

182 00:19:39.220 ⇒ 00:19:40.690 Uttam Kumaran: No, it has to be co-pilot, yeah.

183 00:19:42.850 ⇒ 00:19:43.889 Michael Tran: So yeah, it’s a…

184 00:19:43.890 ⇒ 00:19:46.180 Uttam Kumaran: No problem.

185 00:19:47.660 ⇒ 00:19:51.440 Michael Tran: Yeah, but I mean, this is just, like, the research piece of it, though, right? I mean…

186 00:19:51.440 ⇒ 00:19:52.260 Uttam Kumaran: Yeah.

187 00:19:52.260 ⇒ 00:19:57.269 Michael Tran: like, what I’m building out is, you know, as we keep building, and you probably know this.

188 00:19:57.470 ⇒ 00:20:14.050 Michael Tran: like, it’s very… it’s very easy to, like, design this, like, amazing thing with… with Opus, right? Because, like, it’s… it’s just, like, unlimited what it can do, and it… you have an idea, like, makes it work, and all that stuff, and so, one of the challenges

189 00:20:14.290 ⇒ 00:20:18.790 Michael Tran: It’s gonna be, like, how do we move from that to, like, actually deployment and coding?

190 00:20:19.020 ⇒ 00:20:22.619 Michael Tran: And then also, how do I turn this whole team, like, not only from

191 00:20:22.740 ⇒ 00:20:25.940 Michael Tran: Strategy and design and, like, this phase of it.

192 00:20:26.170 ⇒ 00:20:35.720 Michael Tran: all the way through to, like, development and management of development, all that stuff, and make it, like, as AI-native as possible. And it’s not easy, because…

193 00:20:35.900 ⇒ 00:20:37.759 Michael Tran: Like, we went from basically, like.

194 00:20:38.580 ⇒ 00:20:45.140 Michael Tran: no AI at all. I mean, it’s not no AI at all, like, Teams… people were using, like, GitHub, like, you know, kind of autocorrect and stuff like that, or whatever.

195 00:20:45.250 ⇒ 00:20:48.769 Michael Tran: for a while, nobody had access to agents,

196 00:20:49.320 ⇒ 00:20:56.879 Michael Tran: they’re not using, probably, Opus until they get Droid. And so it’s just, like, it’s a huge shift, right?

197 00:20:57.140 ⇒ 00:20:57.740 Uttam Kumaran: Yeah.

198 00:20:57.970 ⇒ 00:21:03.469 Michael Tran: And the people who would be good at it Are not necessarily, like.

199 00:21:03.670 ⇒ 00:21:07.480 Michael Tran: The people who are working in development on it today.

200 00:21:07.770 ⇒ 00:21:08.260 Uttam Kumaran: Yes.

201 00:21:08.260 ⇒ 00:21:09.390 Michael Tran: You know what I mean? And so it’s just like.

202 00:21:09.390 ⇒ 00:21:10.020 Uttam Kumaran: Yeah.

203 00:21:11.230 ⇒ 00:21:22.529 Uttam Kumaran: So what’s, like, what’s the game plan? Like, how do you… how do you both, like… droids and stuff have only been here for, like, 6 months, so how do you, like, keep up with, like, this R&D, but also, like.

204 00:21:22.660 ⇒ 00:21:24.630 Uttam Kumaran: Actually start to go see, like.

205 00:21:25.920 ⇒ 00:21:29.660 Uttam Kumaran: benefits, you know, like, actually try to, like, hit that, too, while you’re, like.

206 00:21:30.240 ⇒ 00:21:33.179 Uttam Kumaran: Figuring out the… because again, like, at my company.

207 00:21:33.710 ⇒ 00:21:36.810 Uttam Kumaran: I’m like, I’m… I’m sold. I’m like, it’s working.

208 00:21:37.020 ⇒ 00:21:44.839 Uttam Kumaran: People just need to start using it today, even if there’s, like, small cracks. But you… but also, we have, like, you can’t ship… no one can ship PRs unless they’re reviewed.

209 00:21:44.960 ⇒ 00:21:47.870 Michael Tran: And there’s a limited amount of us that can affect the playbook.

210 00:21:47.870 ⇒ 00:21:49.720 Uttam Kumaran: A broader say it can affect the vault.

211 00:21:49.890 ⇒ 00:21:51.330 Uttam Kumaran: But ultimately, like.

212 00:21:52.070 ⇒ 00:21:58.349 Uttam Kumaran: like, we’re trying to be, like, yes, your AI may tell you, like, oh, I need to go delete all this and, like, do something.

213 00:21:58.400 ⇒ 00:22:14.510 Uttam Kumaran: it won’t, like, permeate through the entire system, right? So, like, we’re defending against some of that, but it’s commonly, like, we’re defending it kind of like traditional software engineering defends it, which is, like, PR reviews, scoped access to certain things.

214 00:22:14.610 ⇒ 00:22:17.570 Uttam Kumaran: like, You know, things like that.

215 00:22:18.230 ⇒ 00:22:21.830 Michael Tran: Yeah, I mean, like, part of the strategy and part of the design is, like.

216 00:22:22.070 ⇒ 00:22:26.570 Michael Tran: is how to make this AI, like, more AI native, right?

217 00:22:26.820 ⇒ 00:22:28.120 Michael Tran: And,

218 00:22:28.340 ⇒ 00:22:34.579 Michael Tran: And, you know, I have a couple research, like, droids that, you know, go out and do research for me.

219 00:22:34.840 ⇒ 00:22:38.209 Michael Tran: And so, they did some research on, like, kind of.

220 00:22:38.440 ⇒ 00:22:44.969 Michael Tran: what teams are doing, you know, like, I think Coda, OpenAI came out with, like, an article yesterday, two days ago, something…

221 00:22:45.080 ⇒ 00:22:49.079 Michael Tran: about, like, this project they did 100% with AI.

222 00:22:49.080 ⇒ 00:22:49.690 Uttam Kumaran: Hmm.

223 00:22:49.690 ⇒ 00:22:53.160 Michael Tran: Billion lines of code over a few months, like, what were their learnings from it, right?

224 00:22:53.310 ⇒ 00:23:04.900 Michael Tran: lots of companies are putting out stuff like that, and so I’m just trying to figure out, you know, like, what are the best practices, and then see how we can build them into, kind of our process, because, like, this would be…

225 00:23:05.300 ⇒ 00:23:08.160 Michael Tran: the first attempt I know of, where we’re like.

226 00:23:08.960 ⇒ 00:23:16.479 Michael Tran: you know, going AI native, right? And, like, having AI write more than, like, let’s say 60-70% of the code.

227 00:23:17.100 ⇒ 00:23:17.560 Uttam Kumaran: Yes.

228 00:23:17.560 ⇒ 00:23:18.360 Michael Tran: my goal.

229 00:23:18.580 ⇒ 00:23:22.660 Michael Tran: But there’s a… there’s a lot of structural stuff that…

230 00:23:22.790 ⇒ 00:23:24.569 Michael Tran: That we have to get over.

231 00:23:24.920 ⇒ 00:23:26.959 Michael Tran: Not the least of which I have to convince, like.

232 00:23:27.270 ⇒ 00:23:29.910 Michael Tran: The partners to let me try this.

233 00:23:30.440 ⇒ 00:23:31.350 Michael Tran: Let this team try.

234 00:23:31.350 ⇒ 00:23:31.700 Uttam Kumaran: Yes.

235 00:23:31.700 ⇒ 00:23:34.279 Michael Tran: Because, you know, this is a very big departure

236 00:23:34.830 ⇒ 00:23:39.530 Michael Tran: from, like, our traditional development processes, which is, like, you know, send everything to India, basically.

237 00:23:40.130 ⇒ 00:23:40.830 Uttam Kumaran: Absolutely.

238 00:23:41.680 ⇒ 00:23:49.610 Michael Tran: And I’m like, well, you know, some of these developers, they’re great at, like, coding, but they don’t understand, like, the big picture, and…

239 00:23:49.730 ⇒ 00:23:54.199 Michael Tran: We really need everyone who’s working in this to kind of, like, have that understanding, because that’s going to be more important.

240 00:23:54.360 ⇒ 00:23:56.799 Michael Tran: Then, the ability to write the code.

241 00:23:57.020 ⇒ 00:23:59.839 Uttam Kumaran: So, like, how are you… are you both managing, like.

242 00:24:00.050 ⇒ 00:24:04.150 Uttam Kumaran: I guess, like, what green… like, who is the gatekeeper between, like.

243 00:24:04.260 ⇒ 00:24:07.760 Uttam Kumaran: okay, droids can move past Mike’s team.

244 00:24:07.970 ⇒ 00:24:13.480 Uttam Kumaran: Or is it, like, your team is the POC? Yeah. Yeah, like, what is the…

245 00:24:14.470 ⇒ 00:24:19.109 Uttam Kumaran: Yeah, I mean, I’m roughly, I’m like, okay, how can I be helpful to Mike? I mean, one is, like.

246 00:24:19.390 ⇒ 00:24:24.770 Uttam Kumaran: in my company, I think a lot about, like, how do I measure both adoption and

247 00:24:25.150 ⇒ 00:24:29.499 Uttam Kumaran: like, efficacy. So, is it effective, right? And so, one is, like.

248 00:24:29.670 ⇒ 00:24:34.069 Uttam Kumaran: purely, like, I think half the battle is just, like, are people even just using…

249 00:24:35.110 ⇒ 00:24:50.409 Uttam Kumaran: Right? And I feel like the bigger the company, the more, and we consult with now several sort of multi-hundred million dollar companies, it’s just, like, the fact that people aren’t just using it. So that’s a training that is, like, a buy-in, executive buy-in, and sort of, like, push down on people.

250 00:24:50.880 ⇒ 00:24:58.050 Uttam Kumaran: Then it’s sort of like, okay, now are people using it, and are we seeing the benefit? And so part of this is, like.

251 00:24:58.580 ⇒ 00:25:14.039 Uttam Kumaran: okay, are we… are we able to do, like, time studies and show that it’s now, like, it’s possible? Are we seeing that people are using the things that we put out? Like, for example, I may be like, hey, go through our cursor logs, and how many times is the SOW generator thing being referenced at cursor logs, right?

252 00:25:14.380 ⇒ 00:25:21.989 Uttam Kumaran: You should… you could also see on, like, downstream metrics, right? Like, in your… in my sales, my HubSpot, I should see time to SOW go down, right?

253 00:25:21.990 ⇒ 00:25:22.340 Michael Tran: Yeah.

254 00:25:22.340 ⇒ 00:25:31.740 Uttam Kumaran: between the time that we get the R, the whatever, and then we generate, and we send out. So there’s some of these downstream metrics where you can prove that, hey, my team using it has now been able to affect

255 00:25:31.950 ⇒ 00:25:35.180 Uttam Kumaran: A couple of these, and so they’re… it’s very obvious.

256 00:25:35.460 ⇒ 00:25:39.420 Uttam Kumaran: Apart from that, though, because what you want to avoid is a sort of shiny demo of, like.

257 00:25:40.240 ⇒ 00:25:55.489 Uttam Kumaran: oh my god, this is working, versus, like, hey, like, this is reduced X KPI, and if you were to now push that across the org, like, here’s a savings, yo, green light, like, what else, what other bet do we have in the company that’s about to do that, right?

258 00:25:55.690 ⇒ 00:25:59.299 Michael Tran: Right, yeah, I mean, it’s definitely gonna be, like, the POC kind of thing.

259 00:25:59.650 ⇒ 00:26:09.189 Michael Tran: And I think that, one of the… so, like, there’s two sides to this, right? There’s one which is, like, what I’m working on right now, which is, like, the strategy and, like, the planning and all that stuff.

260 00:26:09.390 ⇒ 00:26:20.949 Michael Tran: And there is a potential benefit of this that is, like, you know, kind of bottom line benefit, and that’s if we get an RFP in, we run it through the system, we create, like, a response that’s really good, and we win the work.

261 00:26:21.060 ⇒ 00:26:29.380 Michael Tran: Right? That, in and of itself, would be like, okay, we need to… we need to make this available to, like, everyone else on the team, right? And so that’s the co-pilot SharePoint kind of flow.

262 00:26:29.880 ⇒ 00:26:40.550 Michael Tran: The other side of it is, okay, well, you know, we want to be able to deliver all these things that we’re planning on doing. They’re really, really ambitious, and, you know, we are looking at, like.

263 00:26:40.650 ⇒ 00:26:43.340 Michael Tran: a two-year window, right? Because, like, we are…

264 00:26:43.710 ⇒ 00:26:45.569 Michael Tran: You know, kind of fighting against

265 00:26:46.160 ⇒ 00:26:56.279 Michael Tran: other people developing the same thing, and everyone has access now to really powerful AI, and so, like, in my mind, I’m like, well, things are gonna get commoditized really quickly.

266 00:26:56.670 ⇒ 00:27:02.300 Michael Tran: And so, like, we need to start building, like, our moat. We need to start building, like, something that’s gonna be, you know, more lasting and harder.

267 00:27:02.300 ⇒ 00:27:13.379 Uttam Kumaran: But your moat is the SharePoint thing, and your moat is the playbooks, dude, because ultimately, it’s gonna get easier and easier to access it. Like, the fact that a company like mine is able to use Cursor to do this just means, like.

268 00:27:13.550 ⇒ 00:27:25.150 Uttam Kumaran: some… it’s gonna come out in Copilot, but again, what’s not gonna come out is the fact that your data is, like, in an organized fashion, that you’ve already talked about what makes a great SOW. Like, you can’t ask Cop… like.

269 00:27:25.440 ⇒ 00:27:28.559 Uttam Kumaran: you can’t offload that to the vendor, right?

270 00:27:28.560 ⇒ 00:27:32.449 Michael Tran: So that’s the alpha, is that, like, holy shit, we’ve written down what…

271 00:27:32.450 ⇒ 00:27:44.550 Uttam Kumaran: what good looks like for all of these different procedures. We’ve looked… we’ve written down how to get to good, and the materials needed to produce are also readily available. Like.

272 00:27:44.870 ⇒ 00:27:47.570 Uttam Kumaran: That is, I think, the thing that, like.

273 00:27:48.170 ⇒ 00:27:54.819 Uttam Kumaran: to focus on, because it’ll be droids today, you may shift to codecs, will be, like, the thing tomorrow, like…

274 00:27:54.820 ⇒ 00:27:55.560 Michael Tran: Yeah.

275 00:27:55.560 ⇒ 00:28:01.269 Uttam Kumaran: I’m like, whatever, whatever. Like, no matter what, my GitHub repo is fixed, I’ve written everything down.

276 00:28:01.700 ⇒ 00:28:04.999 Uttam Kumaran: No matter what it is that comes in, we’ll just turn that on.

277 00:28:05.000 ⇒ 00:28:05.830 Michael Tran: Right, yeah.

278 00:28:05.830 ⇒ 00:28:24.880 Uttam Kumaran: the adoption of that is totally dependent on this, like, raw, like, the setup is there. And again, dude, not every tool doesn’t have a CLI yet, every tool doesn’t even have an API yet, so those will come out, we will build some if they don’t exist, or we will… we’ll use whatever it takes to get the data.

279 00:28:24.910 ⇒ 00:28:33.070 Uttam Kumaran: But, like, that is actually what the limiting factor is. I don’t think it’s necessarily a tool.

280 00:28:33.250 ⇒ 00:28:39.750 Uttam Kumaran: But whatever tool you have access to now, just showing that it’s possible.

281 00:28:40.380 ⇒ 00:28:42.349 Uttam Kumaran: We’ll just get by for whatever the next…

282 00:28:42.350 ⇒ 00:28:42.730 Michael Tran: Yeah.

283 00:28:42.730 ⇒ 00:28:44.310 Uttam Kumaran: Sort of shift has to be.

284 00:28:44.710 ⇒ 00:28:50.270 Michael Tran: Right. The biggest thing that, like, Droid has allowed me to do is create context.

285 00:28:50.820 ⇒ 00:28:51.560 Uttam Kumaran: Yes.

286 00:28:51.920 ⇒ 00:28:54.829 Michael Tran: And then, you know, I have some… I have, like, some red team…

287 00:28:55.070 ⇒ 00:28:59.279 Michael Tran: processes that, like, you know, goes in and does some adversarial review as well. It’s like.

288 00:28:59.280 ⇒ 00:28:59.730 Uttam Kumaran: Okay.

289 00:28:59.730 ⇒ 00:29:06.439 Michael Tran: the content is turning pretty good, right? And it’s, like, pretty cohesive, and it’s getting better as we go, as I’m, like, learning.

290 00:29:06.570 ⇒ 00:29:09.959 Michael Tran: you know, what else I can, like, offload to AI.

291 00:29:10.420 ⇒ 00:29:25.380 Michael Tran: And so, like, and that’s my goal, right? My goal is, like, to create enough context so that I can use it for, like, things. So, for example, I, I was just playing around, like, on Saturday night, and I was like, alright, let’s, let’s see how I can make, like, a presentation, right?

292 00:29:25.490 ⇒ 00:29:33.370 Michael Tran: And so, like, I don’t want to do slides, slides are annoying. Let me just do, like, an HTML, like, full-page scroll presentation.

293 00:29:33.630 ⇒ 00:29:37.789 Michael Tran: And it created, like, an amazing presentation. It had, like, actual…

294 00:29:37.920 ⇒ 00:29:56.060 Michael Tran: demos, like, using, you know, D3JS for, like, some of the… some of the, charts and stuff, use… I mean, like, demos that work, inline, based on, like, what we have put into our features, like, what we’re planning to build. And it just, like, spit this thing out in, like, you know, 10 minutes, or 50 minutes, or whatever.

295 00:29:56.110 ⇒ 00:30:08.450 Michael Tran: And, like, this is amazing, because now we get, like, a client who has certain use cases, I’m gonna build up, like, our corpus of, like, demos that we can put in, and then just have, like, this totally, you know, customized.

296 00:30:08.450 ⇒ 00:30:21.360 Michael Tran: demo with, like, or presentation with all the demos in it, all the additional information they want, personalized to that client, and it just generates it in, like, 15 minutes, because of all the contacts that I have in the repo, all the research that we’ve done, right?

297 00:30:21.770 ⇒ 00:30:22.589 Michael Tran: And so it’s like…

298 00:30:22.590 ⇒ 00:30:36.129 Uttam Kumaran: Dude, I honestly wouldn’t even think much about how long it takes, more that it’s accurate… Yeah. …with, like, in the first one or two passes, because the time will go down. Like, time will go down, but for example, if someone says something, and then it’s like.

299 00:30:36.190 ⇒ 00:30:50.419 Uttam Kumaran: tons of paper cuts for them to get to something, that’s what’s gonna slow down. Because, like, me and you will be patient, we’ll tweak it, but then I’m like, okay, if it can’t, like, at least one-shot, like, most of it, with, like, very little context, because that’s what everybody’s… that’s what, like, the average

300 00:30:50.530 ⇒ 00:31:01.980 Uttam Kumaran: person is gonna turn on cursor and just be like, I need this. So one, it should be pushed back, like, it should be like, hey, I can’t do this until you give me these, right? There should be some input validation. But also, like, it…

301 00:31:02.170 ⇒ 00:31:10.829 Uttam Kumaran: Just the fact that, like, you’re able to do that, and it’s easy, and you don’t have to, like… it’s not missing, or it’s not hallucinating, that’s the stuff that, like, you have to kind of crap first.

302 00:31:11.010 ⇒ 00:31:11.910 Michael Tran: Yeah.

303 00:31:11.910 ⇒ 00:31:12.610 Uttam Kumaran: Yeah.

304 00:31:13.650 ⇒ 00:31:19.839 Michael Tran: Yeah, so that’s, it’s been, it’s been an interesting… like, I started, I started working on this, maybe 10 or 12 days ago.

305 00:31:20.200 ⇒ 00:31:20.690 Michael Tran: And it’s just.

306 00:31:20.690 ⇒ 00:31:22.700 Uttam Kumaran: Nice! Holy shit!

307 00:31:22.700 ⇒ 00:31:23.220 Michael Tran: Yeah.

308 00:31:24.860 ⇒ 00:31:29.940 Michael Tran: Sick. It’s been, like, crazy, because, yeah, it’s just so fast, right? It’s just, like.

309 00:31:29.940 ⇒ 00:31:30.570 Uttam Kumaran: Yes.

310 00:31:30.570 ⇒ 00:31:39.389 Michael Tran: it’s coming at me, like, information’s coming at me so fast, it’s all dense, and, like, I’m building, like, droids and skills to help me, like, dig through it and, like, all that stuff, right?

311 00:31:39.390 ⇒ 00:31:39.720 Uttam Kumaran: Yeah.

312 00:31:40.850 ⇒ 00:31:42.690 Michael Tran: It’s, it’s cool,

313 00:31:43.150 ⇒ 00:31:54.289 Michael Tran: I was playing around with Droid and, like, Cloud Code and stuff, personally, for, like, a year now, just, like, using it to develop things. I got access to Droid, I’m like, alright, I’m gonna try to spin up, like, a…

314 00:31:54.600 ⇒ 00:32:00.490 Michael Tran: a, kind of a POC app that I was thinking about doing, like, with graph databases and stuff.

315 00:32:00.640 ⇒ 00:32:05.020 Michael Tran: Yeah. It wasn’t until, like, I guess about almost 2 weeks ago now that I’m like, hey…

316 00:32:05.250 ⇒ 00:32:16.730 Michael Tran: you know, I’m seeing all this stuff online about, like, cloud code just being more of a general agent versus, like, a coding agent. I’m like, Droid is probably the same, even though Droid is more focused on coding, but let’s see if I can, like.

317 00:32:16.730 ⇒ 00:32:22.440 Uttam Kumaran: But see, that’s the thing, is the problem is the UI and the UX of these things are driven by the coding tools, because.

318 00:32:22.440 ⇒ 00:32:23.120 Michael Tran: permit.

319 00:32:23.340 ⇒ 00:32:29.980 Uttam Kumaran: The developers are the first to adopt it, and it’s so… it’s, like, a ton of structured language.

320 00:32:30.050 ⇒ 00:32:45.169 Uttam Kumaran: they will… once that’s done, which is… it seems like it’s almost done, it will move to, like, more knowledge work. Like, there will be codecs for… you already see Cloud for Work is probably the first foray into that, but I think it’s kind of… it’s kind of stupid, like, I’m like…

321 00:32:45.480 ⇒ 00:32:52.630 Uttam Kumaran: Yes, if you just want to open something and be like, edit my, like, desktop, it’s like, okay. But we’re talking about more sophisticated knowledge work, like.

322 00:32:52.960 ⇒ 00:32:55.840 Uttam Kumaran: Now you’re gonna start to see, like.

323 00:32:56.370 ⇒ 00:33:01.529 Uttam Kumaran: like, clog code for legal, clog code for sales. You will see that.

324 00:33:01.670 ⇒ 00:33:18.939 Uttam Kumaran: But the open AIs and these guys are not going to go after it first. They are going to want to power the people that build those. They’re going to release the most generic version. Like, you can code anything in Codex. Okay, like, that’s, like, literally the most generic, like…

325 00:33:19.490 ⇒ 00:33:38.650 Uttam Kumaran: thing ever, right? The data’s like, fine, we’ll release that application, because it’s gonna drive more usage of our LLMs. But, like, okay, I only want to focus on, like, mobile application. I want codecs for mobile. Like, they’re not gonna do that, right? So, part of this is, like, I see it as my company, I’m like.

326 00:33:38.830 ⇒ 00:33:41.080 Uttam Kumaran: I want to maintain our lead.

327 00:33:41.200 ⇒ 00:33:45.310 Uttam Kumaran: So we need to… we need to take these coding tools and use it for knowledge work.

328 00:33:45.430 ⇒ 00:33:45.800 Michael Tran: Yeah.

329 00:33:45.800 ⇒ 00:33:52.309 Uttam Kumaran: just deal with the fact that it’s ugly, and like, yes, it’s IDE, and it’s VS Code, like, who cares? Whatever.

330 00:33:52.490 ⇒ 00:33:53.459 Uttam Kumaran: everything marked up, I was working.

331 00:33:53.460 ⇒ 00:33:54.979 Michael Tran: No one really knows how to use…

332 00:33:54.980 ⇒ 00:34:08.259 Uttam Kumaran: Yeah, and I’m like, don’t worry about it, don’t worry about it. Here, just install this extension, it Printifies it, and like, whatever. But I’m like, we’re gonna use the coding tools for knowledge work, because eventually there will be… there’s gonna be a lot of vendors offering, like.

333 00:34:08.429 ⇒ 00:34:22.750 Uttam Kumaran: Hirscher-style editing for legal review, but that is gonna be probably in two years from now. Like, it’s not here right now, and I can’t wait. This… our advantage is the fact that we are capable of dealing with droids for knowledge work.

334 00:34:22.929 ⇒ 00:34:28.440 Uttam Kumaran: But in two years, it’ll be like, oh yeah, remember when we were using droids? Like, yes, but, like, you don’t want to have to wait.

335 00:34:28.719 ⇒ 00:34:29.239 Michael Tran: Right.

336 00:34:29.239 ⇒ 00:34:33.389 Uttam Kumaran: we have to build it now for ourselves. We have to take this thing and, like.

337 00:34:33.769 ⇒ 00:34:40.109 Uttam Kumaran: just, like, customize it for us, and, we’re seeing the same thing, exactly the same thing.

338 00:34:41.800 ⇒ 00:34:42.659 Michael Tran: Cool.

339 00:34:42.810 ⇒ 00:34:50.600 Michael Tran: So, switching gears just a little bit, so, like, you know, the tech stack that we’re…

340 00:34:51.000 ⇒ 00:34:58.800 Michael Tran: that we’re building, right? We have Snowflake as our, kind of main… Like, foundational technology.

341 00:34:58.980 ⇒ 00:35:10.870 Michael Tran: And, it’s… it basically functions like a data warehouse. We have some, like, kind of data quality stuff in there, automated quality checks. We have a medallion architecture, a common data model.

342 00:35:11.020 ⇒ 00:35:15.159 Michael Tran: And, you know, we’ve deployed that baseline, to a few clients.

343 00:35:15.220 ⇒ 00:35:16.450 Uttam Kumaran: Okay.

344 00:35:16.760 ⇒ 00:35:23.720 Michael Tran: What we wanna… extend that, you know, that kind of baseline and foundation with is, with Graph.

345 00:35:24.140 ⇒ 00:35:26.590 Michael Tran: And I think that I’m, like, kind of…

346 00:35:27.440 ⇒ 00:35:31.209 Michael Tran: Doing more research on, like, what all the potential, like.

347 00:35:31.600 ⇒ 00:35:41.110 Michael Tran: use cases and value from a graph could be. But, like, the core one, the initial one, is basically, like, you know, all of our… all of our clients, we work… work on financial services, so they’re all, like.

348 00:35:41.680 ⇒ 00:35:43.800 Michael Tran: Massive partnership structures, right?

349 00:35:44.040 ⇒ 00:35:45.630 Michael Tran: Most of them global.

350 00:35:45.940 ⇒ 00:35:58.370 Michael Tran: And, you know, it’s really hard to really get a handle on, like, what their entire structure is. All the way from, like, these entities at the very bottom that are, like, making the investments, all the way up through.

351 00:35:58.440 ⇒ 00:36:09.840 Michael Tran: you know, those deal structures into the funds, into the fund structure, and with all their investors in different vehicles, and then all up to the management company, and, like, all the owners, and everything like that, right? It’s just, like, from top to bottom.

352 00:36:09.920 ⇒ 00:36:10.739 Michael Tran: I don’t know.

353 00:36:11.040 ⇒ 00:36:15.799 Michael Tran: For a big client, could be, like, 10 to 15, 16 layers.

354 00:36:15.960 ⇒ 00:36:16.520 Uttam Kumaran: Yes.

355 00:36:16.520 ⇒ 00:36:29.849 Michael Tran: Right? So it’s, like, massive. And then the thing is, like, they’re all interconnected, and so, like, something that happens at the very bottom, you know, you have a transaction of some kind, it generates some kind of income, it can impact, you know, maybe 10 tiers away an investor, right?

356 00:36:30.010 ⇒ 00:36:30.670 Uttam Kumaran: Yes.

357 00:36:30.670 ⇒ 00:36:34.919 Michael Tran: Today, it’s, like, really hard to understand that in any kind of a…

358 00:36:35.350 ⇒ 00:36:43.389 Michael Tran: and then kind of scale. Like, if you kind of have an idea, because, like, you know, all the structures are designed to do something, so you kind of know, like, what should happen.

359 00:36:43.570 ⇒ 00:36:49.179 Michael Tran: And if there’s, like, a big transaction, then, you know, they’ll put people onto it and say, hey, let’s trace this, and, like, they’ll do this whole…

360 00:36:49.330 ⇒ 00:36:52.390 Michael Tran: You know, analytics project or whatever to see the impact.

361 00:36:52.740 ⇒ 00:36:56.490 Michael Tran: But, like, how can we make that, you know, kind of, one, just

362 00:36:56.940 ⇒ 00:37:03.520 Michael Tran: not automatic, but, like, very easy to do, and then get more value out of, like, that structure itself. There’s a lot of different use cases for.

363 00:37:03.520 ⇒ 00:37:03.870 Uttam Kumaran: Yeah.

364 00:37:03.910 ⇒ 00:37:05.630 Michael Tran: Being able to tier through.

365 00:37:06.020 ⇒ 00:37:07.749 Michael Tran: And then also…

366 00:37:07.750 ⇒ 00:37:12.360 Uttam Kumaran: Are you thinking about using, like, Snowflake Cortex for stuff? Are you thinking about layering something else on?

367 00:37:12.700 ⇒ 00:37:17.209 Michael Tran: Yeah, and so, the Snowflake Cortex capabilities are…

368 00:37:17.600 ⇒ 00:37:21.120 Uttam Kumaran: We just started doing a lot with it in, like, the last 3 weeks.

369 00:37:21.120 ⇒ 00:37:21.490 Michael Tran: Yeah.

370 00:37:21.490 ⇒ 00:37:25.550 Uttam Kumaran: We need… we need to do, like, our limited… we just have to add so much more

371 00:37:25.700 ⇒ 00:37:28.739 Uttam Kumaran: Context to the… like, semantic layer.

372 00:37:28.740 ⇒ 00:37:29.110 Michael Tran: Yeah.

373 00:37:29.110 ⇒ 00:37:32.279 Uttam Kumaran: So, like, we’re… We’re working on that right now.

374 00:37:32.510 ⇒ 00:37:35.740 Uttam Kumaran: I feel like, though, like, It’s gonna work.

375 00:37:35.850 ⇒ 00:37:42.850 Uttam Kumaran: If we just steer it, like, really appropriately, and are really narrow with the use cases we’re solving.

376 00:37:43.290 ⇒ 00:37:47.070 Uttam Kumaran: The other thing is the ability to create Streamlit apps way faster.

377 00:37:48.540 ⇒ 00:37:51.869 Uttam Kumaran: So you may… you may actually be able to, like, create

378 00:37:52.150 ⇒ 00:37:56.740 Uttam Kumaran: And you’ll be able to create, like, sophisticated apps that do actions.

379 00:37:57.100 ⇒ 00:38:10.750 Uttam Kumaran: and actually make that capability more open to people on the team, versus, like, okay, I need to vibe code something and host it somewhere else. Like, you can actually do all of that for simple, like, stream mode apps, all within Snowflake. It’s sort of, like, how we’re…

380 00:38:11.570 ⇒ 00:38:14.080 Uttam Kumaran: We’re starting to think about it, like, how this changes.

381 00:38:15.150 ⇒ 00:38:22.220 Michael Tran: Yeah, that’s interesting. And, I haven’t really looked into Cortex as much, yet.

382 00:38:22.450 ⇒ 00:38:25.840 Michael Tran: It is, like, a research bike that literally is running right now.

383 00:38:26.620 ⇒ 00:38:27.070 Uttam Kumaran: Okay.

384 00:38:27.070 ⇒ 00:38:31.189 Michael Tran: I have a researcher, DSPY versus Cortex.

385 00:38:31.190 ⇒ 00:38:32.050 Uttam Kumaran: Okay, great, great, great.

386 00:38:32.050 ⇒ 00:38:34.530 Michael Tran: hybrid, right? And so, like.

387 00:38:34.760 ⇒ 00:38:35.940 Uttam Kumaran: Yeah. Because, like…

388 00:38:35.950 ⇒ 00:38:36.670 Michael Tran: Yeah, exactly.

389 00:38:36.670 ⇒ 00:38:41.169 Uttam Kumaran: But, like, the thing is, with DSPY, I feel like you’re… it’s so… it’s… for me, it’s just, like, I don’t want to…

390 00:38:41.820 ⇒ 00:38:46.129 Uttam Kumaran: I’m also, again, I’m like, okay, if Cortex is like this, maybe in 6 months it’s better, like.

391 00:38:46.660 ⇒ 00:38:50.260 Uttam Kumaran: should I try to figure out my own thing with TSPY, or just use that?

392 00:38:51.010 ⇒ 00:38:54.720 Uttam Kumaran: Like, the other thing with… now they have Cortex CLI,

393 00:38:55.030 ⇒ 00:38:58.649 Uttam Kumaran: I can orchestrate Snowflake via a CLI interface.

394 00:38:58.650 ⇒ 00:38:59.410 Michael Tran: Yeah.

395 00:38:59.620 ⇒ 00:39:01.649 Uttam Kumaran: as a develop- as a Snowflake developer.

396 00:39:01.650 ⇒ 00:39:02.090 Michael Tran: Right.

397 00:39:02.090 ⇒ 00:39:07.670 Uttam Kumaran: Like, everything from, like, grants and groups to, like, actually creating assets, creating synthetic data sets.

398 00:39:08.110 ⇒ 00:39:11.040 Uttam Kumaran: So I’m almost like, okay.

399 00:39:11.250 ⇒ 00:39:13.250 Uttam Kumaran: I think there’s, like, 3 layers. There’s, like.

400 00:39:13.410 ⇒ 00:39:22.560 Uttam Kumaran: where I’m, like, more of the platform person, then there’s, like, an individual developer who, like, may be on a client, like, but who wants to create apps, and then there’s, like, the end…

401 00:39:22.640 ⇒ 00:39:41.690 Uttam Kumaran: the end user or end analyst, like, they’re maybe just, like, using natural language on the sidebar, or something like that, or using a streamlined app that was created for them. Those are, like, the kind of the personas I’m, like, kind of thinking about, most of which I think will be in, like, you have platform people, and then you have, like, the users, like, the people in between.

402 00:39:42.070 ⇒ 00:39:42.430 Michael Tran: Yeah.

403 00:39:42.430 ⇒ 00:39:52.080 Uttam Kumaran: It’s like, if you’re already… if you’re someone in between, you might as well join the platform, because there’s not many people that could do that right now. I’m like, if you’re already able to do it, just like, you should start building… help us build a platform.

404 00:39:52.080 ⇒ 00:39:52.630 Michael Tran: Yes.

405 00:39:52.630 ⇒ 00:39:54.490 Uttam Kumaran: Yeah.

406 00:39:55.530 ⇒ 00:40:00.200 Michael Tran: Yeah, that makes sense. And, you know, we’re…

407 00:40:00.310 ⇒ 00:40:02.909 Michael Tran: I’m thinking about it just from the platform, mostly.

408 00:40:03.320 ⇒ 00:40:06.960 Michael Tran: Platform, and also, like, the user, the end user, right?

409 00:40:07.210 ⇒ 00:40:07.780 Uttam Kumaran: Yes.

410 00:40:07.780 ⇒ 00:40:15.739 Michael Tran: And so, like, the… you know, one of the things about the natural language processing, or natural language querying that I want to be able to… to do is, like, bring back, like, a…

411 00:40:16.050 ⇒ 00:40:18.720 Michael Tran: a trace of, like, where the answer came from, right? Because…

412 00:40:18.920 ⇒ 00:40:23.530 Michael Tran: There’s just so many different things that you could ask, Tax-related?

413 00:40:23.680 ⇒ 00:40:28.939 Michael Tran: And, you know, it’s… We need to be able to provide some kind of a…

414 00:40:29.250 ⇒ 00:40:30.830 Michael Tran: Confidence to the user.

415 00:40:30.830 ⇒ 00:40:37.390 Uttam Kumaran: Yes. You know, the answer’s correct, and so, like, it’s not just giving them the answer, but it’s, like, being able to expose the entire trace of, like.

416 00:40:37.920 ⇒ 00:40:38.450 Uttam Kumaran: Yes.

417 00:40:38.450 ⇒ 00:40:42.099 Michael Tran: All the different decisions and kind of points that things have…

418 00:40:42.390 ⇒ 00:40:48.020 Michael Tran: These are changes, so, like, basically, I’m thinking about, like, another use for the graph itself, which is, like.

419 00:40:48.210 ⇒ 00:40:52.089 Michael Tran: kind of a… some kind of a process graph or something, where you have, like, codified

420 00:40:52.920 ⇒ 00:40:55.470 Michael Tran: stuff like that, right? Because there’s so many rules and regulations.

421 00:40:55.990 ⇒ 00:40:56.520 Uttam Kumaran: Yes.

422 00:40:56.520 ⇒ 00:41:02.200 Michael Tran: And, like, a lot of them are like, oh, if you have, like, over 15%, you know, consolidated ownership.

423 00:41:02.360 ⇒ 00:41:08.419 Michael Tran: from this country, then, like, you need to do this form, right? And it’s just, like, really complex, you know, kind of…

424 00:41:08.650 ⇒ 00:41:09.240 Uttam Kumaran: Yeah. It’s a success.

425 00:41:09.240 ⇒ 00:41:11.149 Michael Tran: simple calculation and whatever, and so I’m…

426 00:41:11.400 ⇒ 00:41:15.050 Michael Tran: That’s the kind of stuff that we have to expose, so that you say, hey, yes, you need to file this form.

427 00:41:15.230 ⇒ 00:41:17.119 Michael Tran: And then it just tells you why, right? It’s like.

428 00:41:17.360 ⇒ 00:41:17.870 Uttam Kumaran: Yeah.

429 00:41:17.870 ⇒ 00:41:18.820 Michael Tran: good reasoning.

430 00:41:19.430 ⇒ 00:41:27.029 Uttam Kumaran: Another thing, like, a client wanted us to think about is, like, how do we get, like, feedback from users of the AI? Like, thumbs up, thumbs down, but, like, if they say thumbs down.

431 00:41:27.130 ⇒ 00:41:32.309 Uttam Kumaran: Can they put a note in? Yes. So we built some other, like, internal-facing chatbots.

432 00:41:32.370 ⇒ 00:41:50.649 Uttam Kumaran: That we… we do that for, so that… and then we… we’re trying to kind of build something that’s a little bit closer to, like, self-healing, which is, like, on a weekly basis, we take all the bad feedback. Some of it is just, like, user error, some of it is, like, okay, actually, like, yes, this is, like, incorrect, and we classify them.

433 00:41:50.650 ⇒ 00:41:58.600 Uttam Kumaran: Is it context issues? Is it, like, okay, the thing was super slow? Is it just some other error? And then we can say, like, okay.

434 00:41:58.990 ⇒ 00:42:04.619 Uttam Kumaran: is that, for example, some of our clients are on the hook for making the context, like, clean. So we can, like, hey.

435 00:42:04.760 ⇒ 00:42:21.080 Uttam Kumaran: you had two pieces of information that just… it, like, disagreed with each other, and so it, like, it’s… it fucked that up. Also, there’s gotta be stuff on, like, on our side, where it’s like, okay, this question, like, it looped for way too long, and it took, like, a minute to answer, and then it timed out. Okay, that’s, like.

436 00:42:21.120 ⇒ 00:42:39.770 Uttam Kumaran: On us to figure out. So we kind of classify the… the negative feedback, and then ultimately, like, I think we want to get to a system that’s, like, more eval-based, like, overall, where we have a really robust golden data set, and so, like, we are able to, like, without waiting for the user to give feedback, we can score.

437 00:42:40.020 ⇒ 00:42:42.600 Uttam Kumaran: the kind of Q&A pairs, you know?

438 00:42:43.300 ⇒ 00:42:49.250 Michael Tran: Yeah, yeah, that’s… and that’s always the… the… a lot of work, right? Because you have… you have to create that, like, golden data.

439 00:42:49.250 ⇒ 00:42:49.920 Uttam Kumaran: Yeah.

440 00:42:50.090 ⇒ 00:42:54.920 Uttam Kumaran: It’s… it’s tough, it’s tough, and again, like, these are all these things where if you want to get it perfect.

441 00:42:55.670 ⇒ 00:43:00.770 Uttam Kumaran: Versus just, like, get something out there that’s, like, can start to get a little bit of feedback, like, I don’t know.

442 00:43:02.230 ⇒ 00:43:02.800 Michael Tran: Yes.

443 00:43:03.140 ⇒ 00:43:04.850 Michael Tran: Go?

444 00:43:06.910 ⇒ 00:43:08.179 Michael Tran: Alright, one second.

445 00:43:08.390 ⇒ 00:43:09.749 Uttam Kumaran: Okay, no worries, no worries.

446 00:43:27.100 ⇒ 00:43:30.560 Michael Tran: President’s Day today, so the kids are, like, the school’s closed.

447 00:43:30.810 ⇒ 00:43:31.979 Michael Tran: What’s the hell?

448 00:43:31.980 ⇒ 00:43:37.239 Uttam Kumaran: I know, we have… a lot of our clients are off, but that means I get… I get silence. Like, it’s.

449 00:43:37.240 ⇒ 00:43:37.570 Michael Tran: Perfect.

450 00:43:38.750 ⇒ 00:43:40.780 Uttam Kumaran: I get to call and talk about this stuff.

451 00:43:40.930 ⇒ 00:43:48.140 Uttam Kumaran: So, I mean, like, I don’t know, like, where… do you think, like, we could be helpful? Like, I mean, I think, like, both are… I mean, one…

452 00:43:48.210 ⇒ 00:43:51.550 Uttam Kumaran: I’m pumped to hear that someone’s trying to use droids in Enterprise, but, like.

453 00:43:51.580 ⇒ 00:44:09.549 Uttam Kumaran: I think I’m… as you can tell, we’re super passionate in thinking about this same problem. We’re also thinking very, very seriously about how to leverage Snowflake, and not only for structured and unstructured querying. We, like, my background, I’ve used Snowflake since, like, 2018. I’ve probably done, like, 40.

454 00:44:09.660 ⇒ 00:44:20.399 Uttam Kumaran: plus Snowflake implementations. So, like, just used it for most of my career. We probably have active, like, 10 active snowflake contracts right now through Brainforge.

455 00:44:20.410 ⇒ 00:44:38.370 Uttam Kumaran: And we’re really, really close with, like, that team, so I’ve asked them to give us, like, as much private preview access of all the AI stuff, so we can get ahead of it. We have a lot of clients right now that are interested in, like, hey, do we need a BI tool? Or, like, can we start to use Cortex for certain things? Yeah.

456 00:44:38.680 ⇒ 00:44:43.679 Uttam Kumaran: And so we’re going really, really deep. I think the ability to cons… I think at an enterprise scale.

457 00:44:43.820 ⇒ 00:44:48.510 Uttam Kumaran: The ability to go with one vendor that can do a lot of those things is such an easier sell.

458 00:44:48.510 ⇒ 00:44:51.049 Michael Tran: And thinking about y’all, where you’re thinking about, like.

459 00:44:51.410 ⇒ 00:45:08.930 Uttam Kumaran: doing something and spreading it to, like, multiple clients, like, via data share, or, like, streamlet share, or, like, marketplace private shares, like, I think it’s a great opportunity to, like, do that. I mean, the challenge for a lot of this is gonna come in on, like, can you have structured data streams that everybody can use?

460 00:45:09.050 ⇒ 00:45:17.060 Uttam Kumaran: do you have that common… do you have a common data model, basically? Yeah. And then also, can you deploy the, like, system prompts and, like, the SOWs?

461 00:45:17.360 ⇒ 00:45:24.719 Uttam Kumaran: To each client, but of course, understand that they may have… they may have to do some bespoke work based on, like, their… their data and the shape of it.

462 00:45:24.720 ⇒ 00:45:25.320 Michael Tran: Right.

463 00:45:25.860 ⇒ 00:45:30.340 Uttam Kumaran: You know, so, like, I don’t know, that’s sort of, like, what is immediate, like, come to mind for me.

464 00:45:30.340 ⇒ 00:45:32.559 Michael Tran: Yeah, I mean, I think that definitely…

465 00:45:32.740 ⇒ 00:45:39.930 Michael Tran: You know, kind of the nuts and bolts of, like, using AI in, like, the…

466 00:45:40.310 ⇒ 00:45:44.180 Michael Tran: Development and client service piece of it. So, like.

467 00:45:44.180 ⇒ 00:45:45.010 Uttam Kumaran: Yes.

468 00:45:45.010 ⇒ 00:45:57.199 Michael Tran: you know, how do you use AI to help with, say, like, data mapping, for example, right? Or, you know, like, what, what do you, like, what do you guys do for, like, the kind of engineering…

469 00:45:57.970 ⇒ 00:46:01.410 Michael Tran: You know, kind of… Automatic checks and, like, you know, what’s working.

470 00:46:01.410 ⇒ 00:46:01.750 Uttam Kumaran: Yeah.

471 00:46:01.750 ⇒ 00:46:03.760 Michael Tran: to kind of keep the…

472 00:46:04.330 ⇒ 00:46:07.030 Uttam Kumaran: Yes, I can… I’m happy to show you that, too, because…

473 00:46:07.340 ⇒ 00:46:12.650 Uttam Kumaran: this is another area where, like, a few weeks ago, I was like.

474 00:46:13.030 ⇒ 00:46:26.159 Uttam Kumaran: Sam on my team runs our, like, sort of AI… he’s, like, sort of our architect on our AI platform, but me and him are sort of, like, both that, but I told him, like, hey, consider me, like, our most enabled business user.

475 00:46:26.320 ⇒ 00:46:36.499 Uttam Kumaran: So my job is to break this so we figure out what it’s like if someone wakes up and ships… clicks… goes on linear, assigns 30 tickets to cursor, and then, like, goes to bed.

476 00:46:36.660 ⇒ 00:46:37.650 Michael Tran: So…

477 00:46:37.650 ⇒ 00:46:53.839 Uttam Kumaran: So I’m like, you’re gonna find out what it’s like if one person’s able to ship 30 PRs, what’s it gonna be like if the whole company is shipping very fast? Right. So we had to work on an auto-reviewer, we worked on, like, like, I’ll show you, like, sort of, like, what this,

478 00:46:54.230 ⇒ 00:46:56.920 Uttam Kumaran: Where we’re at right now, it’s, like, not perfect, so…

479 00:46:56.920 ⇒ 00:46:57.420 Michael Tran: Yeah.

480 00:46:57.420 ⇒ 00:47:02.660 Uttam Kumaran: But I… I kind of want to give you a sense of, like, how it looks like, so…

481 00:47:03.430 ⇒ 00:47:09.889 Uttam Kumaran: we have, like, for example, we tried to come up with, like, a labeling scheme, and really address a couple things. One is, like.

482 00:47:10.460 ⇒ 00:47:24.400 Uttam Kumaran: it’s quickly apparent that reviewing code is going to become the next bottleneck. And so, I want to start by just making it easy for anybody reviewing code, which right now is primarily me and Sam, to get a sense of, like, if a feature… if it’s a feature or bug fix.

483 00:47:24.620 ⇒ 00:47:26.290 Uttam Kumaran: What the priority is.

484 00:47:26.620 ⇒ 00:47:31.569 Uttam Kumaran: whether it’s, raw, half-baked, or fully baked, I like… I like cooking, so I like…

485 00:47:31.570 ⇒ 00:47:32.260 Michael Tran: sometimes.

486 00:47:32.260 ⇒ 00:47:36.769 Uttam Kumaran: using these, like, cooking buns, but is it, like, half-baked, or is it, like, fully baked? Meaning, like.

487 00:47:36.960 ⇒ 00:47:40.749 Uttam Kumaran: For example, this is a, this is a feature that’s, like.

488 00:47:41.320 ⇒ 00:47:48.240 Uttam Kumaran: This is like a security, like, feature, where we’re updating something. Pretty simple feature, fully baked.

489 00:47:48.350 ⇒ 00:48:04.209 Uttam Kumaran: probably should just get… just get pushed. What happens, though, is, like, when this PR comes in, there is a cursor bug bot review cycle that kind of goes in and just adds a quick note on, like, what it is that’s happening here that’s actually separate from what the user put in here, right?

490 00:48:04.400 ⇒ 00:48:10.380 Uttam Kumaran: The next thing is there’s actually… it gives you the results on, like, why it auto-labeled it the way it did.

491 00:48:11.250 ⇒ 00:48:16.979 Uttam Kumaran: And it sort of looks through, like, it gives you some insight into, like, what it… what it used, right, to, like, auto-label it.

492 00:48:17.260 ⇒ 00:48:23.180 Uttam Kumaran: And then this isn’t a great PR, because this is sort of like a security thing, but if I was to pull in something like…

493 00:48:23.560 ⇒ 00:48:33.309 Uttam Kumaran: like, this is something that I did, whereas I wanted to add Kimi K2 and 2.5 as an option for people in our platform to use for, like, certain applications.

494 00:48:33.460 ⇒ 00:48:39.050 Uttam Kumaran: So it will try… it’ll do multiple… it’ll give you, like, a PR review guide, which is, like.

495 00:48:39.870 ⇒ 00:48:49.899 Uttam Kumaran: here’s the estimated time it takes, there’s no relevant concerns, here’s, like, where you should focus your time. We also, like… I’m sort of trying, like, everything right now, so code ideas.

496 00:48:49.900 ⇒ 00:48:50.230 Michael Tran: Yeah.

497 00:48:50.230 ⇒ 00:49:05.550 Uttam Kumaran: their own review, BugBot does a review, so it’s, like, super, super noisy, but we’re throwing everything at the wall to, like, figure out what the best process is, and then what I do as a developer, I go back into Cursor, and I’m like, okay, look at the comments on the PR, and

498 00:49:05.550 ⇒ 00:49:13.289 Uttam Kumaran: let’s go… let’s go back and review them. And so, as a developer, I’m able to, like… the comment on the PR structure gives me a good loop.

499 00:49:13.400 ⇒ 00:49:33.230 Uttam Kumaran: And then eventually, I think we’ll have some type of autofix process. Like, we already use Cursor’s autofix, which it will make a PR… it’ll make a draft PR with the fix, if it thinks it can make it. I think we’ll get to something that’s a little bit more like, hey, if you’re, like, 95% confidence that you can fix it, go ahead and fix it, and…

500 00:49:33.380 ⇒ 00:49:44.020 Uttam Kumaran: like, either commit it back or create a PR, and then there’s gonna be some things that are, like, really tough. For example, I’m trying to do… add role-based access control to our platform, and

501 00:49:44.240 ⇒ 00:49:46.920 Uttam Kumaran: I, like, tried to one-shot it, and so I’m like.

502 00:49:47.510 ⇒ 00:49:56.249 Uttam Kumaran: I’m like, okay, this is, like, way… this is, like, this is way too complicated. And I was talking to Sam, I’m like, okay, so as a business user, what should I have done? Should I have spent more time

503 00:49:56.670 ⇒ 00:49:59.089 Uttam Kumaran: with Cursor, working on, like, the PRD.

504 00:49:59.290 ⇒ 00:50:01.959 Michael Tran: Yeah. And the TD, and then handed it to you?

505 00:50:02.150 ⇒ 00:50:20.439 Uttam Kumaran: Or… and he was like, yeah, I think maybe that’s how we should do it, versus, like, I have to do… I want to think about this big-ass thing, like, I want users at Brainforge to only have access to their meetings, and, like, think about these admins. It’s kind of, like, similar structure to, like, how you do Snowflake role-based access control. And I was like, I know how to… I know what I want, but, like.

506 00:50:20.530 ⇒ 00:50:22.610 Uttam Kumaran: I’m not an expert in auth, and like.

507 00:50:22.740 ⇒ 00:50:40.679 Uttam Kumaran: front-end auth, I’m like, you guys are, and I’m like, okay, so probably I should have, like, worked on, like, a comprehensive PRD, and then handed it to Sam to be like, okay, now you can logically break this into tickets, if you can one-shot it, you can one-shot it, and, like, that’s maybe what we should have done. So these are, like, the ergonomics we’re thinking through.

508 00:50:42.460 ⇒ 00:50:42.850 Michael Tran: Yeah.

509 00:50:42.850 ⇒ 00:50:44.360 Uttam Kumaran: Small team of ours, right?

510 00:50:44.790 ⇒ 00:50:53.459 Michael Tran: I mean, I think that’s… that’s gonna be, like, where I would find a lot of value, because, like, I… I’m not an engineer in… or a developer, I would say.

511 00:50:54.270 ⇒ 00:51:00.700 Uttam Kumaran: Dude, I don’t know, man, like, you’re, you’re basically there, dude. The lines are very blurry now.

512 00:51:01.330 ⇒ 00:51:08.990 Michael Tran: Yeah, like, if AI does all the development, then yeah, that’d be… then I can do some stuff, but, like, you know, like, my background is not…

513 00:51:09.130 ⇒ 00:51:11.729 Michael Tran: engineering or development, and I’ve just picked up

514 00:51:12.150 ⇒ 00:51:17.839 Michael Tran: technology over the past 10 years working at UI. And so, like, you know, there are…

515 00:51:18.560 ⇒ 00:51:29.479 Michael Tran: kind of, like, the engineering cycle and the kinds of things that… that, you know, should be happening. Like, I’m picking it up, because I’m reading a ton right now, and there’s, like, so many articles out about this stuff, so, like, I am picking a lot of that stuff up.

516 00:51:29.680 ⇒ 00:51:37.129 Uttam Kumaran: But you’re coming at it from the business side, dude. This is where it’s kind of interesting, because some… a lot of people are coming at it from the engineering side up, or from the business side down.

517 00:51:37.280 ⇒ 00:51:52.130 Uttam Kumaran: It’s… it’s sort of, like, one way or another. Like, for me, I’m, like, sort of a little bit in both. I’m, like, I learned how consulting works through… through running this business, but then I’m also, like, an engineer, so I’m sort of, like, trying to marry them both.

518 00:51:52.310 ⇒ 00:52:01.810 Uttam Kumaran: But, like, again, it’s… I think you’re… you’re not coming at it from a good angle, where you’re sort of, okay, there’s one LLM, one workflow. Okay, then there’s reasoning, and there’s, like, agentic steps. Now I’m, like.

519 00:52:02.000 ⇒ 00:52:10.679 Uttam Kumaran: okay, but I can kick off multiple things in parallel. Okay, now I need to think about, like, unified context between all of them. Okay, now I need to think about, like.

520 00:52:11.420 ⇒ 00:52:17.799 Uttam Kumaran: all the research I did to get to this point, is anyone at my company, like, smart enough or, like, has enough free time to, like.

521 00:52:17.920 ⇒ 00:52:24.550 Uttam Kumaran: also get there? Okay, so no? How do I, like, bring this to them, then? Right. You know?

522 00:52:24.550 ⇒ 00:52:28.829 Michael Tran: It’s very iterative, and, you know, these past two weeks has been very fast, and…

523 00:52:29.270 ⇒ 00:52:37.290 Michael Tran: And I’m like, I don’t fully trust the AI either. Not in the way that, like, some people don’t trust AI, and they just, like, don’t want to… I just know that…

524 00:52:37.290 ⇒ 00:52:37.840 Uttam Kumaran: Yeah, yeah, yeah.

525 00:52:37.840 ⇒ 00:52:39.880 Michael Tran: You know, it’s, like, different perspective, and so, like, one of the…

526 00:52:39.880 ⇒ 00:52:41.520 Uttam Kumaran: Garbage in, garbage out, and so.

527 00:52:42.200 ⇒ 00:52:54.210 Uttam Kumaran: if you do say, give me an SOW, it’s gonna be like, what? Like, but you needed to… you need to push back. Yeah. So that’s the system I… I needed to be like, you did not give me enough context, I’m not supposed to do this, basically.

528 00:52:54.210 ⇒ 00:52:57.050 Michael Tran: So that’s why I created the red team concept this morning.

529 00:52:57.050 ⇒ 00:52:58.120 Uttam Kumaran: Yeah, yeah, yeah, yeah, yeah.

530 00:52:58.120 ⇒ 00:53:14.400 Michael Tran: And I’m using a different model. I’m like, okay, I use Opus 4.6 for everything, I’m doing my Red Team with, like, GPT-5.3 codecs extra high, and, like, it’s great. And then I’m like, okay, this is great. Give me a full report. Like, basically, it runs Red Team on every, like, feature, every big thing that we do.

531 00:53:14.560 ⇒ 00:53:28.479 Michael Tran: And I’m like, I don’t want to, like, go through all this with you. And so, like, now there’s another loop where, like, Opus and Codex iterate on the red team report and fix everything, and then I don’t get it back until, like, they agree that it’s good.

532 00:53:28.590 ⇒ 00:53:30.540 Michael Tran: Yes. You know?

533 00:53:31.300 ⇒ 00:53:48.649 Michael Tran: And so I’m just slowly, as I’m working with it, I’m like, okay, well, you know, my mind is like, how can I use the AI that’s available to get me to that without me having to do it? And then, like, I only have to, like, look at it when it’s, like, really at a place, like, a good place, right? And then I can go in and add context or whatever if needed.

534 00:53:49.090 ⇒ 00:53:53.940 Michael Tran: And so it’s, it’s, it’s been, like, it’s been fun, and interesting.

535 00:53:54.520 ⇒ 00:53:56.430 Michael Tran: I just need to get to the next…

536 00:53:56.580 ⇒ 00:54:00.140 Michael Tran: The next thing is to make it valuable for someone else.

537 00:54:00.560 ⇒ 00:54:00.920 Uttam Kumaran: Yeah, yeah.

538 00:54:00.920 ⇒ 00:54:08.350 Michael Tran: It needs to push the project forward, it needs to, like, you know, kind of align on, like, a development. It needs to, like, bring in more…

539 00:54:08.690 ⇒ 00:54:13.900 Michael Tran: Investment or revenue, like, it needs to, like, have some real-world impact, and that’s, like, the next thing.

540 00:54:14.740 ⇒ 00:54:15.500 Uttam Kumaran: Yeah.

541 00:54:16.920 ⇒ 00:54:22.390 Michael Tran: But… Yeah, it’s yeah, I think,

542 00:54:22.660 ⇒ 00:54:25.930 Michael Tran: I don’t know. You know, if I think about how…

543 00:54:27.160 ⇒ 00:54:29.790 Michael Tran: My biggest gap, between, like.

544 00:54:30.190 ⇒ 00:54:34.950 Michael Tran: I think between what you know and, like, what I know is gonna be on the engineering side.

545 00:54:35.120 ⇒ 00:54:39.900 Michael Tran: Right? Like, I kind of have an idea of, like, what…

546 00:54:40.190 ⇒ 00:54:45.379 Michael Tran: things need to be in there, with, like, the automated reviews and, like, guardrails and stuff, and…

547 00:54:45.490 ⇒ 00:54:48.569 Michael Tran: And I’m able to put my research droid to go out and bring me that.

548 00:54:48.570 ⇒ 00:54:50.360 Uttam Kumaran: Yeah, yeah, yeah.

549 00:54:50.430 ⇒ 00:54:51.430 Michael Tran: Which is, like…

550 00:54:51.430 ⇒ 00:55:07.300 Uttam Kumaran: What you’re gonna find is just some of these are just battle, like, this is just how normal engineering works, and so we’ve taken a lot of, like… for example, there’s no need to rewrite what great software engineering looks like, it’s just great software engineering takes a shitload of time, so it doesn’t… it doesn’t now, right? But, like.

551 00:55:07.480 ⇒ 00:55:16.249 Uttam Kumaran: Unit tests, end-to-end tests, smoke tests, all of that is, like, well-written. It’s just that you can never do that in the time you have as, like, a normal company.

552 00:55:16.250 ⇒ 00:55:20.930 Michael Tran: Right. So that’s where I feel like we’re leveraging a lot of the best practices, just putting them all on rails.

553 00:55:20.930 ⇒ 00:55:23.149 Uttam Kumaran: like, as many on Rails, and…

554 00:55:23.270 ⇒ 00:55:29.790 Uttam Kumaran: just attacking the most immediate bottleneck, which, again, like, when I woke up one day and I shipped all those PRs, I was like, okay.

555 00:55:30.120 ⇒ 00:55:35.770 Uttam Kumaran: in 6 months, hopefully faster, when a lot of people at the company are shipping PRs, the bottleneck is gonna be reviews.

556 00:55:35.940 ⇒ 00:55:36.320 Michael Tran: Yeah.

557 00:55:36.480 ⇒ 00:55:52.369 Uttam Kumaran: So… but then I also said, hey guys, some of these things I tried to ship, our codecs environment wasn’t set up in a way it could do all the end-to-end testing, like, it could pull it up, it could take a screenshot, it could then understand. So, okay, we need to do that, right? And then…

558 00:55:52.420 ⇒ 00:55:57.649 Uttam Kumaran: now that we nailed that, okay, now let’s think about scaling the review process. Great, now that that’s done.

559 00:55:58.060 ⇒ 00:56:03.179 Uttam Kumaran: we sh… there should be no reason why I’m the only one shipping PRs. Okay, let’s go call people and ask.

560 00:56:03.290 ⇒ 00:56:20.830 Uttam Kumaran: what’s your hesitation? Okay, I never… I just never thought I could do this. Okay, let’s do it on the phone, together. Like, you wanna… you wanna edit something in the platform? You can change that. And, like, there’s an adoption piece. So, yeah, I mean, if… if, like, if we can be helpful as a thought partner, or even help engineer some of these, and help you scale.

561 00:56:20.830 ⇒ 00:56:21.380 Michael Tran: Yup.

562 00:56:21.520 ⇒ 00:56:22.370 Uttam Kumaran: steps.

563 00:56:22.660 ⇒ 00:56:24.380 Uttam Kumaran: And focus more on, like.

564 00:56:24.500 ⇒ 00:56:41.329 Uttam Kumaran: okay, how does this actually get into the team? And, like, how do I get the business buy-in? Like, that’s a great way to leverage us, and we can’t do… we… we would come in, and doing that would be, of course, way hard for us. Like, you know, that’s, like, that’s where your expertise and

565 00:56:41.460 ⇒ 00:56:49.909 Uttam Kumaran: and buy an inside UI is gonna allow that, but it’s almost like, okay, can we… can we help you skip a few steps on the engineering side to get to the output, you know?

566 00:56:50.610 ⇒ 00:56:56.409 Michael Tran: Yeah, I think that… that would be really helpful, and I was… I know you talked to Claire, too, and I was, like, talking to her about it, and…

567 00:56:56.580 ⇒ 00:57:06.410 Michael Tran: And, you know, I think initially she was thinking, you know, we’re just trying to figure out what resources we can bring in to help us deliver, right? And I was like, you know, like, and I know that you guys

568 00:57:06.640 ⇒ 00:57:25.339 Michael Tran: our… do Snowflake, and that’s, like, your specialty, and you have Neo4J, which is also really awesome, because we don’t have that many people that know anything about Neo4j. Yeah. But to me, I was like, that’s not why I want to bring you guys in, isn’t to, like, deliver faster, it’s like, I want you guys to teach our team, like, all those little…

569 00:57:25.970 ⇒ 00:57:26.360 Uttam Kumaran: Yeah.

570 00:57:26.360 ⇒ 00:57:29.290 Michael Tran: how to work with AI effectively, and… Yeah.

571 00:57:29.290 ⇒ 00:57:30.559 Uttam Kumaran: Like, put us in the middle.

572 00:57:30.560 ⇒ 00:57:32.459 Michael Tran: Right. Like, kind of sandwiched between…

573 00:57:32.510 ⇒ 00:57:35.160 Uttam Kumaran: Everyone on the ground, your vision.

574 00:57:35.160 ⇒ 00:57:40.260 Michael Tran: and just, like, kind of bobble in here. I actually… I was talking to Clarence about that, too. I was like…

575 00:57:41.450 ⇒ 00:57:42.760 Uttam Kumaran: I was like, okay.

576 00:57:42.940 ⇒ 00:57:48.890 Uttam Kumaran: yes, of course, like, we can try to build it, but EY has the resources, they don’t have the know-how.

577 00:57:49.280 ⇒ 00:58:02.660 Uttam Kumaran: But they also… but also, like, the problem with an EY is, like, there also has to be dedication going to get buy-in, and, like, so when it does work, to get adoption. So there is, like, almost, like, those three work streams, which is, like, can we get the thing to work?

578 00:58:03.190 ⇒ 00:58:12.989 Uttam Kumaran: can Michael go, like, make sure they can get adoption and buy-in? And then can we start to just train people so that they start using it? And that’s, like, part of the jump, like, between those three layers, you know?

579 00:58:13.110 ⇒ 00:58:20.160 Uttam Kumaran: Because you’re similar, like, our guys are… it’s a much better use of EY funds to pay us to go teach and, like, scale that.

580 00:58:20.160 ⇒ 00:58:20.730 Michael Tran: Right.

581 00:58:21.260 ⇒ 00:58:21.819 Uttam Kumaran: And scale that.

582 00:58:22.180 ⇒ 00:58:23.280 Uttam Kumaran: Exactly.

583 00:58:23.580 ⇒ 00:58:38.319 Michael Tran: And so, like, I guess right now, if I’m thinking about this, I want to make sure, like, the ecosystem is, like, ready for it, right? So, like, you know, all the people who are going to be doing the development, they all have droids, they at least know how to open it and use it, so they have access to AI.

584 00:58:38.630 ⇒ 00:58:40.380 Michael Tran: The, the…

585 00:58:40.580 ⇒ 00:58:47.110 Michael Tran: Strategy and, like, the framework needs to be, like, at least pretty well thought out, and, like, we have a plan for, like, what we need to do.

586 00:58:47.310 ⇒ 00:59:05.159 Michael Tran: And then, so when you guys come in, it’s like, alright, this is where you fit in, you can help us with both sides of it, you know, just kind of, like, polish them up, but you’re not coming in on, like, a blank slate, and then trying to, like, help us design the thing, which I think would not be… I mean, you guys would probably do a good job of it, but, like, I feel like we’d be wasting

587 00:59:05.930 ⇒ 00:59:12.590 Michael Tran: The real value that we need you for, which is, like, you know, kind of taking the dev team and the dev process to the next level.

588 00:59:12.810 ⇒ 00:59:14.070 Uttam Kumaran: Yes, yeah.

589 00:59:15.750 ⇒ 00:59:17.250 Michael Tran: Okay, I’m gonna…

590 00:59:17.600 ⇒ 00:59:22.449 Michael Tran: talk to Claire, and just kind of keep this on. We’re just kind of figuring out, like, resourcing and stuff, and…

591 00:59:22.820 ⇒ 00:59:27.019 Michael Tran: And what options we have available to us, but, yeah, this is.

592 00:59:27.020 ⇒ 00:59:33.650 Uttam Kumaran: Yeah, dude, and if I can even just, like… this is what I’m… my free time is figuring this out for my company, and so if I can…

593 00:59:33.770 ⇒ 00:59:39.389 Uttam Kumaran: Even just, like, kind of shoot the shit like this, or you want to text me, like, happy to, because, like, we are learning…

594 00:59:39.820 ⇒ 00:59:42.769 Uttam Kumaran: The hard way, like, every day, so…

595 00:59:42.950 ⇒ 00:59:52.750 Michael Tran: Yeah, man, absolutely. If you’re, like, kind of talking through something, or… I’m happy to jump on and talk. I know Clarence said that he and I have had many late night conversations about this stuff.

596 00:59:52.750 ⇒ 00:59:57.950 Uttam Kumaran: Cool, okay. He’s usually, like, my go-to text of, like.

597 00:59:58.590 ⇒ 00:59:58.950 Michael Tran: Yeah.

598 00:59:58.950 ⇒ 01:00:15.750 Uttam Kumaran: what do you think about this? But, like, again, like, I think my challenge, I’m thinking about, like, the business inertia that’s preventing me to do some of these. It’s like, how do we overcome that, you know? Because I think the tools are there for us to do it. Like, a year ago, it wasn’t… they weren’t there. Now that they’re there, I’m like, okay, what can we do? But…

599 01:00:15.970 ⇒ 01:00:20.559 Uttam Kumaran: Yeah, I’m, like, more than happy… are you in… are you in San Antonio, by the way?

600 01:00:20.940 ⇒ 01:00:21.930 Michael Tran: I’m in Dallas.

601 01:00:22.110 ⇒ 01:00:23.190 Uttam Kumaran: Okay, I’m in Austin.

602 01:00:23.530 ⇒ 01:00:24.650 Michael Tran: Oh, nice. Yeah.

603 01:00:24.650 ⇒ 01:00:25.210 Uttam Kumaran: Yeah.

604 01:00:25.390 ⇒ 01:00:36.530 Uttam Kumaran: Okay, so… yeah, we should all get together. I’ll tell… I think I may come to Dall sometime next 2 months, but you ever find yourself in Austin? Or I guess we could meet in the middle somewhere.

605 01:00:36.530 ⇒ 01:00:40.430 Michael Tran: I’m gonna be in Austin in… May, for sure.

606 01:00:41.160 ⇒ 01:00:46.109 Uttam Kumaran: Okay, I mean, let’s definitely… let’s definitely link up. The world will be way different.

607 01:00:46.110 ⇒ 01:00:47.830 Michael Tran: I know, right? It could be like a…

608 01:00:48.150 ⇒ 01:00:51.140 Uttam Kumaran: I don’t even know what life is gonna look like.

609 01:00:51.140 ⇒ 01:00:55.869 Michael Tran: Yeah, GPT-6? None of this is even gonna matter anymore.

610 01:00:55.870 ⇒ 01:01:00.689 Uttam Kumaran: No, my AI will be meeting with your… my AI will be meeting with your AI while we’re.

611 01:01:03.010 ⇒ 01:01:07.129 Michael Tran: Like, hey, we can all just talk… stuff to talk about, like, you and I would just get beers and just hang out in, like, our AR.

612 01:01:07.970 ⇒ 01:01:09.500 Michael Tran: You know, the business.

613 01:01:12.570 ⇒ 01:01:13.160 Uttam Kumaran: That’s funny.

614 01:01:13.160 ⇒ 01:01:17.359 Michael Tran: Cool. I’ll email you my number. Feel free to text me whenever.

615 01:01:17.360 ⇒ 01:01:17.930 Uttam Kumaran: Okay.

616 01:01:17.930 ⇒ 01:01:19.520 Michael Tran: I love talking about this stuff, obviously.

617 01:01:19.820 ⇒ 01:01:25.680 Michael Tran: And it’s rare to find people that, like, can get deep on it, right? Who thought about it a lot already, too, so…

618 01:01:26.420 ⇒ 01:01:31.049 Uttam Kumaran: Yeah, and again, we’re coming at it from the constraint-driven side, like.

619 01:01:31.050 ⇒ 01:01:31.500 Michael Tran: Yeah.

620 01:01:31.500 ⇒ 01:01:35.019 Uttam Kumaran: we’re a small business, and we’re growing, and I’m like, how do we grow efficiently? And so we’ve…

621 01:01:35.230 ⇒ 01:01:41.109 Uttam Kumaran: found this. I think EY is almost, like, an opposite problem, where it’s, like, scale, and there’s so much…

622 01:01:41.270 ⇒ 01:01:45.059 Uttam Kumaran: potentially pushing back on this, and it’s a lot, it’s like, but…

623 01:01:45.520 ⇒ 01:01:48.480 Uttam Kumaran: But again, it is just a challenge, and it’s sort of like…

624 01:01:48.630 ⇒ 01:01:53.850 Uttam Kumaran: it’s sort of figuring out, like, okay, what are the constraints of the system? Like, what do we need to optimize for? And, like.

625 01:01:54.330 ⇒ 01:02:00.789 Uttam Kumaran: I’m really, really excited that, like, hopefully some of the things that we’ve learned through doing this can be helpful, you know?

626 01:02:00.970 ⇒ 01:02:01.990 Michael Tran: Yeah, yeah.

627 01:02:02.400 ⇒ 01:02:03.350 Michael Tran: I think so.

628 01:02:03.800 ⇒ 01:02:04.670 Michael Tran: Cool.

629 01:02:04.900 ⇒ 01:02:07.480 Uttam Kumaran: Perfect. Okay, thank you, appreciate the time.

630 01:02:07.480 ⇒ 01:02:08.749 Michael Tran: Yeah, of course.

631 01:02:08.750 ⇒ 01:02:12.020 Uttam Kumaran: Yeah, and, like, text me your number, I’m glad to stay hi and stay in touch, so…

632 01:02:12.020 ⇒ 01:02:13.180 Michael Tran: Yeah, sounds good.

633 01:02:13.180 ⇒ 01:02:14.369 Uttam Kumaran: Cool, okay, thanks.