1
00:00:00,461 --> 00:00:00,701
Okay.

2
00:00:01,121 --> 00:00:01,181
Yeah.

3
00:00:02,481 --> 00:00:03,801
Do we want to call it Jiro or no?

4
00:00:04,781 --> 00:00:05,601
Yeah, we could call it Jiro.

5
00:00:05,721 --> 00:00:07,001
We could just be like, this is our agent Jiro.

6
00:00:07,481 --> 00:00:08,841
I think it's also like a helpful potential cut,

7
00:00:08,841 --> 00:00:10,661
which is like if we ask Jiro,

8
00:00:10,741 --> 00:00:11,781
are you ready to get started?

9
00:00:12,301 --> 00:00:12,661
Yeah, sure.

10
00:00:12,821 --> 00:00:15,501
And then our video can move,

11
00:00:15,621 --> 00:00:17,181
Jiro can draw something,

12
00:00:17,301 --> 00:00:17,901
acknowledgement, whatever.

13
00:00:18,101 --> 00:00:18,441
Yeah, yeah, yeah.

14
00:00:18,481 --> 00:00:19,541
It can be like a transition pain,

15
00:00:20,261 --> 00:00:20,901
transition cut.

16
00:00:22,281 --> 00:00:23,821
All right, let's get started.

17
00:00:24,601 --> 00:00:26,541
Thanks, guys, for coming to listen to a talk

18
00:00:26,541 --> 00:00:28,341
on the AIE online track.

19
00:00:28,501 --> 00:00:30,421
I guess coming to listen is not quite right

20
00:00:30,421 --> 00:00:33,181
because obviously you guys are not in the room with us.

21
00:00:33,281 --> 00:00:34,161
Making the long, long trip.

22
00:00:34,401 --> 00:00:35,581
Yeah, making the long, long trip.

23
00:00:35,701 --> 00:00:36,081
Yeah, yeah, yeah.

24
00:00:36,221 --> 00:00:37,521
The arduous journey to YouTube.com.

25
00:00:37,521 --> 00:00:41,201
Whenever we do these kinds of talks, we always want to play around with the format a little bit.

26
00:00:41,301 --> 00:00:45,321
I'm not a huge fan of just standing in front of a loom and doing slides late at night in my room.

27
00:00:45,961 --> 00:00:48,601
So I dragged Cliff along to do something a little bit different.

28
00:00:48,721 --> 00:00:50,561
We wanted to do this in a little bit of a podcast format.

29
00:00:50,941 --> 00:00:55,561
And part of that is just because it's nice to have multiple people to riff with and bounce ideas off of.

30
00:00:55,701 --> 00:00:58,781
But also because we wanted to find a way to include Jiro, our agent.

31
00:00:59,481 --> 00:01:03,081
So obviously, on our side, Jiro is not quite in the room with us.

32
00:01:03,121 --> 00:01:04,261
He's kind of living on this computer.

33
00:01:04,801 --> 00:01:08,921
But when you guys are watching this on YouTube, hopefully you'll have all three of us present.

34
00:01:09,081 --> 00:01:14,981
You'll have me and Cliff talking and then Jiro will be sort of in the middle, drawing things out and describing, describing things.

35
00:01:15,301 --> 00:01:16,741
Jiro, please wave hi for the viewers.

36
00:01:17,081 --> 00:01:18,641
Yeah, Jiro, please wave hi for the viewers.

37
00:01:19,781 --> 00:01:23,301
So this talk is all about agentifying a SecMaster.

38
00:01:23,561 --> 00:01:30,481
This is part of AIE New York, which, of course, is very, very focused on the finance and like AI and finance in 2026.

39
00:01:31,701 --> 00:01:37,161
and so i think uh before we really dive in maybe it makes sense for just us to briefly introduce

40
00:01:37,161 --> 00:01:41,201
ourselves and then we probably want to just give some definitions even like what is a sec master

41
00:01:41,201 --> 00:01:45,961
what is a cloud agent how do you like use agents to do any of these things um so i'll go first

42
00:01:45,961 --> 00:01:51,821
my name is amol i am the ceo of a company called nori agentic and as you might have guessed we

43
00:01:51,821 --> 00:01:55,841
spend a lot of time thinking about cloud agents and how we can put agents in the cloud to solve

44
00:01:55,841 --> 00:02:01,021
problems, both in finance and kind of all in a bunch of different industries. And yeah, Cliff.

45
00:02:01,321 --> 00:02:05,941
I'm Cliff. I'm the CTO at Nori Agentic. We've been working on dev tooling, AI agents,

46
00:02:06,221 --> 00:02:10,921
this form factor for almost two years now. And before that, I was working on data engineering

47
00:02:10,921 --> 00:02:14,721
and building out a new data platform for a startup hedge fund. And before that,

48
00:02:14,881 --> 00:02:17,781
staff software engineering. And there's kind of a through line in all those things with data

49
00:02:17,781 --> 00:02:22,021
engineering, dev tools, some amount of cloud inference, some amount of web apps.

50
00:02:22,121 --> 00:02:24,241
Yeah, it's kind of funny, actually. This is almost like a perfect talk for you. Like,

51
00:02:24,241 --> 00:02:29,041
I'm sort of the odd person out. My background's in ML, so I've been doing the AI ML thing for

52
00:02:29,041 --> 00:02:32,581
12 years, but I don't know anything about finance. You know all the finance stuff.

53
00:02:33,461 --> 00:02:36,781
So it's kind of an interesting conversation, right? Because we have two potentially very

54
00:02:36,781 --> 00:02:40,401
different audiences in the room. You have the finance folks who maybe don't know anything

55
00:02:40,401 --> 00:02:44,061
about AI, and you have the AI engineer folks who maybe don't know anything about finance.

56
00:02:44,521 --> 00:02:47,221
So maybe it makes sense to even just get started with some definitions like,

57
00:02:47,361 --> 00:02:52,081
what is a cloud agent and what is a SEC master before we can even dive into how these things

58
00:02:52,081 --> 00:02:59,301
fit together. Yes. So cloud agent, the most succinct definition, I think, for just agent to

59
00:02:59,301 --> 00:03:05,301
start an LLM or a model that you run in a loop and give access to tools. And so it can take some

60
00:03:05,301 --> 00:03:09,561
action that you hopefully find useful. Specifically cloud agents, you take that whole form factor,

61
00:03:10,281 --> 00:03:14,441
usually the code that is calling the model in a loop and you put it in some cloud process and

62
00:03:14,441 --> 00:03:18,861
you give it access to tools that don't exist on your laptop or your phone or wherever else.

63
00:03:18,861 --> 00:03:30,441
And now that whole entity, the hopeful, useful primitive, runs in the cloud, hopefully taking action with either your company's data and processes or your own needs.

64
00:03:30,641 --> 00:03:33,141
And I mean, so much of the important data now lives in the cloud.

65
00:03:33,281 --> 00:03:37,161
So like in some sense, it actually is like a very natural home for all of that to even run.

66
00:03:37,421 --> 00:03:43,541
Yeah, I think it's there's a certain historicity of, I think, when we saw local agents take off.

67
00:03:43,641 --> 00:03:43,741
Right.

68
00:03:43,901 --> 00:03:45,541
With the Claude Code moment last year.

69
00:03:45,681 --> 00:03:45,761
Right.

70
00:03:45,921 --> 00:03:48,521
And before that, with earlier agent primitives.

71
00:03:48,681 --> 00:03:48,741
Right.

72
00:03:49,221 --> 00:03:52,801
I think it's almost like a quirk of history that it became local first.

73
00:03:53,201 --> 00:03:53,561
Yeah.

74
00:03:53,901 --> 00:03:54,241
Just to start.

75
00:03:54,481 --> 00:03:58,601
In the long run, we're going to see it take a lot longer for actual on-edge

76
00:03:59,361 --> 00:04:01,721
devices to deliver value.

77
00:04:01,921 --> 00:04:05,101
So that's cloud agents, agents that run in the cloud that you can basically do things

78
00:04:05,101 --> 00:04:07,921
with asynchronously and talk through across a bunch of different platforms.

79
00:04:08,301 --> 00:04:09,101
What about a Secmaster?

80
00:04:09,781 --> 00:04:10,141
Yes.

81
00:04:10,321 --> 00:04:13,681
Now, we probably want to spend more time here because I think that this is a lot harder.

82
00:04:13,921 --> 00:04:17,461
It's harder for the average, I think, person who is not in finance to understand what a

83
00:04:17,461 --> 00:04:20,721
a sec master is compared to the other way around. And I think that people in finance have heard a

84
00:04:20,721 --> 00:04:24,081
lot more about agents than the other way. Yeah. And software engineers have heard about a stack

85
00:04:24,081 --> 00:04:28,821
master. Yeah. I certainly felt that way several years ago, kind of jumping into that field to

86
00:04:29,501 --> 00:04:33,901
bring data engineering expertise to help a startup hedge fund begin trading. So I mean,

87
00:04:33,981 --> 00:04:39,641
training, I think is the place to start. So you have in theory, a tradable universe of assets that

88
00:04:39,641 --> 00:04:44,141
are core to your business as an asset manager, many different types of asset managers. We don't

89
00:04:44,141 --> 00:04:49,001
even need to go there right now. But for your business to operate, you need some master database

90
00:04:49,001 --> 00:04:53,941
of all the securities that are in that tradable universe. Yeah. And so it's basically just like

91
00:04:53,941 --> 00:04:57,341
a table. Like for the engineering side, you can kind of think of this as a table that lives in

92
00:04:57,341 --> 00:05:01,621
Snowflake or Databricks or something like that. But in practice, it's actually like a really,

93
00:05:01,761 --> 00:05:04,941
really important operational primitive. Yeah. And I think it's,

94
00:05:06,381 --> 00:05:10,601
if you just give the rote definition of identifiers and assets and instruments,

95
00:05:10,601 --> 00:05:12,821
and you start to explain the difference between,

96
00:05:13,161 --> 00:05:16,121
okay, you have instruments and some of them are securitized

97
00:05:16,121 --> 00:05:17,781
and there will be different asset classes.

98
00:05:18,081 --> 00:05:19,961
And okay, that seems like most of it.

99
00:05:20,161 --> 00:05:20,301
Yeah.

100
00:05:20,561 --> 00:05:21,221
And actually?

101
00:05:21,581 --> 00:05:25,501
Actually, building the thing, going inside the primitive,

102
00:05:25,961 --> 00:05:28,421
there's a lot of complexity, a lot of accidental complexity.

103
00:05:28,681 --> 00:05:30,261
And once you get a centralized representation,

104
00:05:30,561 --> 00:05:32,801
it makes all the downstream processes very, very easy.

105
00:05:33,001 --> 00:05:35,881
And that's why almost every asset manager has one.

106
00:05:35,961 --> 00:05:36,121
Yeah.

107
00:05:36,381 --> 00:05:38,801
But actually munging all that data together

108
00:05:38,801 --> 00:05:47,301
And every time you bump into a new asset class or a new ID format, a new upstream source of data, you actually have new needs that you need to abstract inside the security master.

109
00:05:47,541 --> 00:05:54,461
I mean, I tend to think of it almost like an operational primitive where you just have all of this data, all these different data sources and all these different data consumers.

110
00:05:54,461 --> 00:06:09,881
And if you don't have something in the middle that basically ensures that the right data is getting to the right places, it's impossible to otherwise guarantee in a distributed way that you're going to have clean data, that you're going to be able to report the right things, that your audits are correct, that your pricing models are correct.

111
00:06:10,001 --> 00:06:14,141
Like all of that stuff essentially is downstream of having a good, clean sec master.

112
00:06:14,681 --> 00:06:14,821
Yeah.

113
00:06:14,921 --> 00:06:20,921
So the sec master for most analysis that you're going to run, it's going to be the left side of almost every join.

114
00:06:21,101 --> 00:06:21,261
Yeah.

115
00:06:21,721 --> 00:06:23,961
It's going to be the starting point for a lot of analysis that you do.

116
00:06:24,101 --> 00:06:24,241
Yeah.

117
00:06:24,241 --> 00:06:28,141
And that drives some of the biggest processes at any asset manager.

118
00:06:28,401 --> 00:06:28,501
Right.

119
00:06:28,721 --> 00:06:36,101
And that could be everything from your reporting requirements to looking at historical performance to understanding your...

120
00:06:36,101 --> 00:06:37,121
I'm sorry, this is just regulatory.

121
00:06:37,321 --> 00:06:40,261
Like you literally have to have some sort of a record of what assets you have.

122
00:06:40,541 --> 00:06:41,141
Yeah, exactly.

123
00:06:41,461 --> 00:06:41,581
Yeah.

124
00:06:42,801 --> 00:06:50,521
I think it's very easy as software engineers for us to jump over some of the accidental complexity that gets encapsulated within an abstraction.

125
00:06:50,801 --> 00:06:50,981
Sure.

126
00:06:50,981 --> 00:06:55,821
and so the comparison i've been thinking about when we've explained this to other engineers on

127
00:06:55,821 --> 00:07:01,821
our team and talking with our customers it's like date time libraries yeah it's very it's very easy

128
00:07:01,821 --> 00:07:06,221
for engineers to think oh it's a daytime library yeah you're just tracking the time how hard could

129
00:07:06,221 --> 00:07:10,241
it be and then you realize like time zones are a thing and like people track things differently and

130
00:07:10,241 --> 00:07:14,241
it's all sorts of pain yeah everything from the localization to how people see the same

131
00:07:14,241 --> 00:07:18,761
representation to actually moving something between different time zones and uh daylight

132
00:07:18,761 --> 00:07:20,701
savings actually hits at different points in different countries.

133
00:07:22,581 --> 00:07:27,921
So just like the briefest snapshot of all the accidental complexity that you hide inside

134
00:07:27,921 --> 00:07:33,641
the perimeter of the Secmaster, I think the easiest place to start, just so people grasp

135
00:07:33,641 --> 00:07:35,261
this, is publicly traded equities.

136
00:07:35,441 --> 00:07:35,581
Yeah.

137
00:07:35,841 --> 00:07:37,861
Like a Google or an Apple stock or something like that.

138
00:07:37,961 --> 00:07:39,801
You go to the exchange, you buy Google or something like that.

139
00:07:39,841 --> 00:07:39,981
Yeah.

140
00:07:40,141 --> 00:07:43,541
And this is by far the easiest asset class to even handle.

141
00:07:43,861 --> 00:07:47,481
And already you just mentioned one of the first quirks, which is Google.

142
00:07:47,481 --> 00:07:50,321
that's not even one-to-one between tickers and company.

143
00:07:50,441 --> 00:07:51,281
Yeah, what do you mean by Google?

144
00:07:51,401 --> 00:07:52,601
Like, are you buying Alphabet stock?

145
00:07:52,741 --> 00:07:53,601
Are you buying Google stock?

146
00:07:53,761 --> 00:07:54,581
There's actually two completely different

147
00:07:54,581 --> 00:07:55,481
things. Yeah, and there's Google Class A and B.

148
00:07:55,681 --> 00:07:57,401
And they are different on different exchanges,

149
00:07:57,401 --> 00:07:59,021
and they're different in different countries.

150
00:07:59,421 --> 00:08:00,301
So tickers aren't unique.

151
00:08:00,481 --> 00:08:02,821
They're not one-to-one with the actual entities.

152
00:08:03,041 --> 00:08:03,161
Right.

153
00:08:03,481 --> 00:08:04,621
And then on different exchanges,

154
00:08:04,621 --> 00:08:07,301
you can actually have the same stock listed

155
00:08:07,301 --> 00:08:11,321
across potentially a different exchange in the EU.

156
00:08:11,481 --> 00:08:11,641
Right.

157
00:08:11,681 --> 00:08:13,061
You might have actually different listings

158
00:08:13,061 --> 00:08:15,561
of the same underlying shares.

159
00:08:15,561 --> 00:08:20,241
So you got this rat's nest of data, basically, and it's really, really hard to unentangle.

160
00:08:20,441 --> 00:08:22,021
And you have to basically have a team of people

161
00:08:22,041 --> 00:08:25,921
who are doing this pretty labor-intensive role of keeping this all clean.

162
00:08:26,341 --> 00:08:34,701
Yeah. And so a lot of it's kind of architecting how the SecMasters work for the area or particular asset classes that your business will be investing in.

163
00:08:35,701 --> 00:08:47,221
I think that, especially in the publicly traded world, oftentimes most of the structure to the data comes from the moments where that data encounters different centralizing forces.

164
00:08:47,601 --> 00:08:54,101
So the two biggest examples are the regulatory constraints and the actual exchanges where it's traded.

165
00:08:54,261 --> 00:08:54,421
Right.

166
00:08:54,421 --> 00:08:56,121
So kind of point of sale, point of exchange.

167
00:08:56,881 --> 00:09:01,701
And those two have the effect of creating most of the identifiers that you care about in the publicly traded world.

168
00:09:01,901 --> 00:09:01,961
Right.

169
00:09:01,961 --> 00:09:07,421
So if those are the assets that you're interested in, you have to make sure that you have those identifiers.

170
00:09:07,541 --> 00:09:08,341
You have to make sure they're aligned.

171
00:09:08,501 --> 00:09:15,241
You have to make sure you have them somewhere recorded in your company, because anytime anybody wants to do anything with those assets, they have to reference them back to those things.

172
00:09:15,501 --> 00:09:19,321
Yeah. And at first it can be relatively easy because you're dealing with a particular asset class.

173
00:09:19,401 --> 00:09:24,661
You're dealing with one intersection of different centralizing structures.

174
00:09:24,941 --> 00:09:28,301
You can just be like, I'm just going to use Bloomberg's IDs for everything and not worry about it.

175
00:09:28,301 --> 00:09:31,301
Yeah, so you have Bloomberg's IDBB Globals.

176
00:09:31,301 --> 00:09:33,321
It's like a column in the source data that you're going to get from them.

177
00:09:33,701 --> 00:09:37,421
And these are called FIGGYs sometimes because of a previous acronym.

178
00:09:37,661 --> 00:09:38,141
It's called FIGGY?

179
00:09:38,241 --> 00:09:39,001
F-I-G-G-Y?

180
00:09:39,141 --> 00:09:41,841
Yeah, it's like Financial Instrument Global Identifier.

181
00:09:42,041 --> 00:09:42,361
Okay.

182
00:09:42,721 --> 00:09:45,101
But I don't know if it's officially renamed.

183
00:09:45,181 --> 00:09:46,021
Sure, yeah, sure, whatever.

184
00:09:46,201 --> 00:09:47,941
It's a little bit easier to say than IDBB Global.

185
00:09:48,441 --> 00:09:51,341
And so from that, that's going to answer a lot of your problems.

186
00:09:51,601 --> 00:09:51,741
Yeah.

187
00:09:52,281 --> 00:09:54,781
At the moment you get a little bit bigger, then it starts to get a little bit messy.

188
00:09:54,901 --> 00:09:55,901
Once there's a little bit more complexity.

189
00:09:56,141 --> 00:09:56,301
Yeah.

190
00:09:56,301 --> 00:10:00,181
or you're not dealing with things that are understood in the Bloomberg universe of assets.

191
00:10:00,681 --> 00:10:04,241
And so actually one of the things I think is really interesting is the last three years,

192
00:10:04,241 --> 00:10:09,641
I think, has seen a massive growth for asset managers in assets that are not of this sort

193
00:10:09,641 --> 00:10:13,401
of like standardized, globalized public entities, right?

194
00:10:13,481 --> 00:10:17,081
Like Google and Apple in some sense are pretty well understood, but so many people are now

195
00:10:17,081 --> 00:10:20,501
investing in like data centers and like clusters of GPUs.

196
00:10:20,621 --> 00:10:24,041
And if you're just like buying a data center from somebody or buying like shares of a data

197
00:10:24,041 --> 00:10:28,001
center, that's not something that trades on a liquid market. And so you have to like track that

198
00:10:28,001 --> 00:10:33,681
in some other way. Yeah. So, I mean, this is not my area of expertise, but the data centers are

199
00:10:33,681 --> 00:10:39,281
a very interesting one because I think it touches on multiple different areas of private, privately

200
00:10:39,281 --> 00:10:44,141
traded instruments. Right. So you might have actual real estate that's involved in where do

201
00:10:44,141 --> 00:10:47,781
we build this data center? The actual land. Yeah. And then from there, there might be a

202
00:10:47,781 --> 00:10:54,521
privately owned company that does issue equity. Maybe it's held by a publicly traded company.

203
00:10:54,881 --> 00:10:59,881
Maybe there's some publicly traded bonds issued to fund some part of this, or maybe there's private

204
00:10:59,881 --> 00:11:04,221
credit that's taken on to fund the construction. I mean, increasingly the GPUs itself feel like

205
00:11:04,221 --> 00:11:09,161
an asset class. Yeah. And there actually are efforts right now to securitize GPUs.

206
00:11:09,181 --> 00:11:13,281
GPU clusters. Yeah, sure. Well, I mean, because another private asset that's a very interesting

207
00:11:13,981 --> 00:11:15,041
niche is airplanes.

208
00:11:15,421 --> 00:11:15,521
Sure.

209
00:11:15,781 --> 00:11:16,661
So one reason why

210
00:11:16,661 --> 00:11:17,601
the entry costs

211
00:11:17,601 --> 00:11:19,361
for the airline business

212
00:11:19,361 --> 00:11:20,501
are low is because...

213
00:11:20,501 --> 00:11:20,601
Oh, low.

214
00:11:20,821 --> 00:11:21,121
Low.

215
00:11:21,521 --> 00:11:22,441
The entry costs are so low

216
00:11:22,441 --> 00:11:24,301
is because it holds collateral

217
00:11:24,301 --> 00:11:24,841
against itself

218
00:11:24,841 --> 00:11:25,601
when you're losing these

219
00:11:25,601 --> 00:11:25,921
planes. Oh, interesting.

220
00:11:26,001 --> 00:11:26,781
You could lose the plane.

221
00:11:26,981 --> 00:11:27,141
Yeah.

222
00:11:27,301 --> 00:11:28,441
So it's going to get us to

223
00:11:29,521 --> 00:11:30,841
most private assets.

224
00:11:31,041 --> 00:11:31,341
This is actually

225
00:11:31,341 --> 00:11:32,101
a helpful distinction.

226
00:11:32,241 --> 00:11:32,921
Most private assets

227
00:11:32,921 --> 00:11:33,921
are not very fungible.

228
00:11:34,261 --> 00:11:34,621
Yeah, sure.

229
00:11:34,921 --> 00:11:35,261
Yeah, yeah,

230
00:11:35,401 --> 00:11:36,181
That's kind of a homework

231
00:11:36,181 --> 00:11:36,681
of the problem

232
00:11:36,681 --> 00:11:37,821
that we're going to come back to.

233
00:11:38,001 --> 00:11:38,821
There are some of these

234
00:11:38,821 --> 00:11:40,301
private assets that actually are...

235
00:11:40,301 --> 00:11:41,341
Are weirdly fungible.

236
00:11:41,421 --> 00:11:41,821
Weirdly fungible,

237
00:11:41,921 --> 00:11:43,261
like airplanes or GPUs.

238
00:11:43,281 --> 00:12:05,661
Yeah. So it might be worth even diving into then the specifics, right? So what does it mean to agentify a securities master? And I think it could be helpful to even ground this in a particular example, which is, you know, we're working very closely with one of these big asset managers in New York City, one of the benefits of being in New York, New York local. Like, let's talk through a little bit of like what problems that they are running into. And then how are they using agents to solve that problem?

239
00:12:05,661 --> 00:12:12,321
Yeah, so a lot of this is that last distinction we hit on for private assets.

240
00:12:14,141 --> 00:12:22,101
And the first important distinction here is that the amount of data and the amount of work related to each asset,

241
00:12:22,101 --> 00:12:25,061
it scales linearly with a number of assets because they're unique.

242
00:12:25,301 --> 00:12:33,201
So each private asset, it's not like you can easily write software that will handle a group of these private assets because each one is different.

243
00:12:33,381 --> 00:12:34,121
Yeah, oftentimes.

244
00:12:34,121 --> 00:12:38,101
There's no API that you can hit to be like, give me the price of this data center.

245
00:12:38,241 --> 00:12:39,801
The price lives in a contract somewhere.

246
00:12:40,101 --> 00:12:40,221
Yeah.

247
00:12:40,481 --> 00:12:49,161
And if you're part of a business like a VC where the number of these different assets already scales linear to the amount of time put in,

248
00:12:49,301 --> 00:12:57,301
so the number of different checks that you write is already very related to the number of different person hours spent with those potential investments.

249
00:12:57,581 --> 00:12:58,221
It's extremely manual.

250
00:12:58,441 --> 00:12:59,181
It's very manual.

251
00:12:59,181 --> 00:13:07,681
And it might be, you know, that might be scalable because it's already, you know, the source data is proportional to the person hours and then the downstream effects then cost proportional person hours.

252
00:13:08,021 --> 00:13:20,121
But then once you start looking at everything from real estate to private credit and the different ways that these assets can be packaged, that ends up being its own whole new way that the scale of the data increases.

253
00:13:21,001 --> 00:13:25,201
So you have this like, you have this like, you know, exponentially increasing complexity.

254
00:13:25,561 --> 00:13:26,981
How does that then impact?

255
00:13:27,161 --> 00:13:29,021
Like, let's dive into this kind of specific use case.

256
00:13:29,021 --> 00:13:30,181
So we're working with this asset manager.

257
00:13:30,441 --> 00:13:32,461
Like, what are the things that they are trying to do?

258
00:13:32,581 --> 00:13:33,961
And then where do agents come in?

259
00:13:34,441 --> 00:13:34,581
Yeah.

260
00:13:34,801 --> 00:13:38,361
So they're already seeing a process where they've created a new sec master.

261
00:13:38,581 --> 00:13:38,701
Yep.

262
00:13:39,141 --> 00:13:41,261
And they have both internal and external sources.

263
00:13:41,521 --> 00:13:41,641
Yep.

264
00:13:42,061 --> 00:13:47,421
Many of these are quantitative, but some of them are actually related to different documents

265
00:13:47,421 --> 00:13:49,901
and other unstructured sources of data.

266
00:13:50,681 --> 00:13:52,961
And they have data pipelines that are mapping that.

267
00:13:52,961 --> 00:14:03,081
And just the lowest barrier for adoption is accurately being able to tie out individual changes to upstream pieces of evidence.

268
00:14:03,181 --> 00:14:03,901
Yeah, yeah.

269
00:14:03,981 --> 00:14:11,361
And that, just for a couple hundred columns or a couple hundred rows with even fewer than that columns, is already a very labor-intensive process.

270
00:14:12,281 --> 00:14:15,181
And they manage a lot more than a couple hundred assets.

271
00:14:15,461 --> 00:14:16,101
Yeah, yeah, yeah, yeah.

272
00:14:16,421 --> 00:14:19,581
So just to give a kind of layman's view of this, right?

273
00:14:19,581 --> 00:14:30,261
Like anytime an asset, anytime one of these rows in this table changes, there's a guy, like somebody who gets called up, who basically has to then go and figure out why that change happened.

274
00:14:30,461 --> 00:14:30,801
Right.

275
00:14:30,861 --> 00:14:30,961
Yeah.

276
00:14:31,121 --> 00:14:36,341
So basically be able to look at any cell in the table and look at the lineage for what produced that.

277
00:14:36,361 --> 00:14:36,941
Not every row.

278
00:14:37,061 --> 00:14:37,821
It's every cell.

279
00:14:37,981 --> 00:14:38,501
Every cell.

280
00:14:38,641 --> 00:14:38,781
Yeah.

281
00:14:38,861 --> 00:14:39,001
Yeah.

282
00:14:39,061 --> 00:14:40,641
So every row slash column.

283
00:14:40,881 --> 00:14:41,161
Yeah.

284
00:14:41,341 --> 00:14:41,621
And.

285
00:14:42,681 --> 00:14:43,901
So extremely labor.

286
00:14:43,901 --> 00:14:50,401
Like anytime anything changes for any of these columns and rows, there's a person who basically has to go and find the provenance for that data.

287
00:14:50,641 --> 00:14:57,661
And yeah, I can't undersell the trickiness raised by the intersection of both software and finance here.

288
00:14:57,721 --> 00:14:57,841
Sure.

289
00:14:57,981 --> 00:15:09,121
Because that person is not going to both be able to perfectly understand all the Python data munching and data flows, whether that's like some notebooks that run in Snowflake all the way down to like the build your own.

290
00:15:09,241 --> 00:15:11,481
We have our own lake house and we're running our own Apache Spark jobs.

291
00:15:12,341 --> 00:15:18,821
If they are a finance expert, someone who's conducted a lot of operational work in back office for different managers...

292
00:15:18,821 --> 00:15:20,301
They probably don't have that software engineering system.

293
00:15:20,301 --> 00:15:27,761
A lot of those people are really savvy, actually, with how they can use Python to solve problems on their machine with different CSVs or tie-outs or exports from Excel.

294
00:15:28,341 --> 00:15:31,761
But being able to look at the repo and understand...

295
00:15:33,061 --> 00:15:39,021
It's interesting, spreadsheets and software are kind of opposite views of the same process.

296
00:15:39,021 --> 00:15:43,541
space. Yeah. Where it's like with a spreadsheet, you see the data and you don't see the logic.

297
00:15:43,881 --> 00:15:48,381
And then with a software, you know, all these data pipelines, you see the process and the logic,

298
00:15:48,521 --> 00:15:51,681
but you don't see the data. Yeah. And so there's completely different representations. And on the

299
00:15:51,681 --> 00:15:55,381
flip side of it, you have the software engineer who has implemented the data pipelines and they've,

300
00:15:55,461 --> 00:15:59,121
but they don't have the financial knowledge. They don't know what it looks like for something to be

301
00:15:59,121 --> 00:16:03,141
incorrect in this particular cell that's now causing problems for an investment professional

302
00:16:03,141 --> 00:16:07,441
who needs to look at data and have it be correct. Right. Right. Right. So, so you have this process

303
00:16:07,441 --> 00:16:09,361
that right now, anytime a cell changes,

304
00:16:09,941 --> 00:16:10,821
and for private data,

305
00:16:10,941 --> 00:16:12,381
it's not even like you can go hit an API.

306
00:16:12,561 --> 00:16:14,581
You literally have a guy going through a Google Drive

307
00:16:14,581 --> 00:16:15,861
or going through somebody's emails

308
00:16:15,861 --> 00:16:17,861
to find a PDF of a contract

309
00:16:17,861 --> 00:16:19,601
to make sure that the data is correct.

310
00:16:19,741 --> 00:16:19,961
Yeah.

311
00:16:20,161 --> 00:16:21,781
Is this the most up-to-date version?

312
00:16:21,941 --> 00:16:24,381
Is this the latest notarized version of the contract

313
00:16:24,381 --> 00:16:26,161
and not some draft that was sent back and forth?

314
00:16:27,141 --> 00:16:28,901
Is the actual information I need here

315
00:16:28,901 --> 00:16:30,581
or is there a more recent update?

316
00:16:31,861 --> 00:16:34,361
And so I just feel like in those terms, right,

317
00:16:34,501 --> 00:16:36,501
it's just so obvious that cloud agents

318
00:16:36,501 --> 00:16:37,321
would solve this problem.

319
00:16:37,441 --> 00:16:43,921
yeah yeah or maybe you know ameliorate the problem yeah solve is a tough word yeah so i think this

320
00:16:43,921 --> 00:16:50,661
kind of gets to like the most interesting part of the problem which is why is this hard yeah we as

321
00:16:50,661 --> 00:16:55,021
software engineers as software professionals we see agents solving accidental complexity every

322
00:16:55,021 --> 00:16:59,821
single day yes so i think it's helpful to just deep dive why is this hard both from a technical

323
00:16:59,821 --> 00:17:06,361
perspective and a business perspective yeah and to your point about there's someone who needs to go

324
00:17:06,361 --> 00:17:11,981
look for this, right? Being able to know what good looks like, what correct looks like,

325
00:17:11,981 --> 00:17:16,981
that's more of a business question. Or it can also be a search problem where,

326
00:17:17,161 --> 00:17:22,401
you know, when have you seen enough? When do you have all of the evidence? It's tough to prove.

327
00:17:22,561 --> 00:17:25,441
It's going to be different from different companies in different situations. Yeah.

328
00:17:25,501 --> 00:17:30,781
For different assets, even. Exactly. And yeah, I mean, one of the reasons why this particular

329
00:17:30,781 --> 00:17:35,481
instance, this particular project is a good fit is because they've already actually thought very

330
00:17:35,481 --> 00:17:41,081
hard about what does that workflow look like? What does it look like to reconcile inconsistencies or

331
00:17:41,081 --> 00:17:47,421
any changes with the data? What would they want that back office function to actually look at

332
00:17:47,421 --> 00:17:52,021
step by step? And when would they want them to escalate an inconsistency they're seeing up to

333
00:17:52,021 --> 00:17:56,341
a data owner or to a financial? Sec masters already have this sort of process in place.

334
00:17:56,681 --> 00:18:01,881
If there's an error somewhere, they already know, okay, there's an error in the cell. We know who

335
00:18:01,881 --> 00:18:05,821
is going to be owner of that cell? Yeah, it could be everything from there is an investment

336
00:18:05,821 --> 00:18:14,901
professional whose book or whose investments are related to this particular number. It could be

337
00:18:15,381 --> 00:18:20,561
actually, you know, when this particular data source doesn't look correct, we talked to Jerry.

338
00:18:20,801 --> 00:18:21,961
Sure. Because he

339
00:18:22,021 --> 00:18:25,361
was the one who talked to them on the initial extracts of data that they sent over.

340
00:18:25,641 --> 00:18:25,721
Yep.

341
00:18:26,201 --> 00:18:30,001
So this, I think, is like the first of the problems of why this is hard.

342
00:18:30,341 --> 00:18:30,401
It's like a type of knowledge, yeah.

343
00:18:30,521 --> 00:18:38,121
Yeah, there's knowledge that's not encoded in either the code or the spreadsheets or a lot of the purpose behind the tasks.

344
00:18:38,361 --> 00:18:43,081
The tasks are very encoded at your company, but the purpose behind those tasks is often not.

345
00:18:43,081 --> 00:18:47,041
And that lives with all of the people who are making up those different functions.

346
00:18:47,721 --> 00:18:50,601
Which I think also gets to the second part of why this is difficult.

347
00:18:51,641 --> 00:18:55,901
In order to properly assist them, you need to make assistive technologies.

348
00:18:56,181 --> 00:19:01,941
And this is a common failure pattern we've seen, especially on the software side of the market for the past year or

349
00:19:02,021 --> 00:19:05,441
There's so many people who want the agents to take everything to 100%.

350
00:19:05,441 --> 00:19:11,241
And so they create these software factories, quote unquote, that are really inflexible.

351
00:19:11,241 --> 00:19:15,641
you know they they spin up some process where it's like oh i tagged this agent and it'll like

352
00:19:15,641 --> 00:19:22,241
run in s in uh ec2 and then there's no thought put into well what if you need to intervene what

353
00:19:22,241 --> 00:19:25,761
if you need the agent to be able to do something else but if you need to respond there's got to be

354
00:19:25,761 --> 00:19:31,041
this like very rich human interaction pattern and if your thought process is i'm just going to have

355
00:19:31,041 --> 00:19:36,381
this thing run autonomously and try and do the whole thing to 100 well you're going to fail

356
00:19:36,381 --> 00:19:40,861
because there's actually like so much depth that like a human needs to even if it's just

357
00:19:40,861 --> 00:19:44,041
accountability. Sometimes you need a human just to be able to hit the merge button.

358
00:19:44,421 --> 00:19:50,621
Yeah. And it's kind of funny that people are shooting for that particular target when that's

359
00:19:50,621 --> 00:19:56,461
one of the biggest benefits of agents is their malleability and the ability to correct them.

360
00:19:56,941 --> 00:20:00,281
I mean, I think it's mid-task. I think a big part of it is, and this is kind of the thing about

361
00:20:00,281 --> 00:20:04,201
cloud agents. I think a lot of people think, oh, a cloud agent is just something that we can just

362
00:20:04,201 --> 00:20:09,001
have an endpoint and then kick off a task. But actually the real cloud agent infrastructure

363
00:20:09,001 --> 00:20:10,461
is when it feels seamless.

364
00:20:10,921 --> 00:20:13,421
It feels so seamless to be able to talk to it in the cloud

365
00:20:13,421 --> 00:20:16,501
as if it was running on your computer or on your phone

366
00:20:16,501 --> 00:20:19,001
or you were just talking to it over a voice call.

367
00:20:19,321 --> 00:20:22,541
And that's the level of depth and product area

368
00:20:22,541 --> 00:20:23,821
that most people just don't want to do

369
00:20:23,821 --> 00:20:24,341
because it's hard.

370
00:20:24,421 --> 00:20:25,121
It's hard to build that.

371
00:20:25,321 --> 00:20:27,021
Yeah, I know that's the second part that's difficult,

372
00:20:27,161 --> 00:20:29,081
which is from the technical side.