1
00:00:02,141 --> 00:00:02,441
morning.

2
00:00:03,221 --> 00:00:06,041
I'm Belinda and this is Keon and we're from Anthropic.

3
00:00:06,421 --> 00:00:08,601
We work with financial services firms.

4
00:00:10,181 --> 00:00:13,781
As Keon gets set up here, we'll start with a little bit of background.

5
00:00:13,781 --> 00:00:17,721
My name is Belinda Neal, and I'm the Managing Director of Financial Services at Anthropic.

6
00:00:18,341 --> 00:00:24,521
Previously to joining Anthropic this year, I worked at Goldman Sachs for 19 years across engineering, global markets, and equity research.

7
00:00:24,881 --> 00:00:31,161
My role at Anthropic sits at the confluence of our industry, our product, our research, and our go-to-market teams,

8
00:00:31,521 --> 00:00:34,541
really in accelerating the adoption of AI within financial services.

9
00:00:35,601 --> 00:00:35,901
Keon.

10
00:00:36,341 --> 00:00:38,061
Good morning, everyone. Thanks for being here.

11
00:00:38,181 --> 00:00:41,321
My name is Keon Shahab. I'm an Applied AI Architect at Anthropic.

12
00:00:41,321 --> 00:00:44,141
Prior to Anthropic, I spent five years at Databricks

13
00:00:44,141 --> 00:00:45,821
helping financial services companies

14
00:00:45,821 --> 00:00:48,101
modernize their data and AI architecture.

15
00:00:49,701 --> 00:00:51,401
So we're excited to speak to you this morning,

16
00:00:51,401 --> 00:00:53,481
and our talk is going to cover three ideas.

17
00:00:54,401 --> 00:00:56,921
Firstly, the frontier of intelligence.

18
00:00:57,241 --> 00:00:59,341
How is this room taking the frontier models

19
00:00:59,341 --> 00:01:02,021
and applying it in ways that create alpha or edge

20
00:01:02,021 --> 00:01:02,761
within your firm?

21
00:01:03,381 --> 00:01:05,161
The second is the frontier of adoption.

22
00:01:05,741 --> 00:01:07,541
How are you scaling what you've built

23
00:01:07,541 --> 00:01:09,961
based on the foundation of a few to many?

24
00:01:09,961 --> 00:01:13,341
And then last, how are we applying it to financial services?

25
00:01:13,601 --> 00:01:17,301
We're going to bring an example in here about how we've been innovating in the wealth industry

26
00:01:17,301 --> 00:01:18,661
in Claude for financial advisors.

27
00:01:18,941 --> 00:01:23,181
And that paradigm compounds potentially to other areas of financial services as well.

28
00:01:24,081 --> 00:01:27,081
So let's get started and talk about the frontier of intelligence.

29
00:01:28,221 --> 00:01:30,881
So every so often, a model crosses a threshold.

30
00:01:31,161 --> 00:01:33,561
And when it does, a way of working changes.

31
00:01:34,001 --> 00:01:39,261
We first saw this encoding last year with Opus 4.5 and Claude code.

32
00:01:39,761 --> 00:01:41,921
Many of you in the room are using these tools,

33
00:01:41,981 --> 00:01:43,841
and you've gone from autocomplete encoding

34
00:01:43,841 --> 00:01:45,581
to managing fleets of agents

35
00:01:45,581 --> 00:01:47,101
where you're really driving and accelerating

36
00:01:47,101 --> 00:01:49,681
a multiple effect of how you're doing coding.

37
00:01:50,221 --> 00:01:52,741
We also saw this threshold cross this year

38
00:01:52,741 --> 00:01:54,841
in cyber with Claude Mythos Preview,

39
00:01:54,881 --> 00:01:56,861
and many of you are also spending time

40
00:01:56,861 --> 00:01:58,781
on the cyber programs in your firms.

41
00:01:59,541 --> 00:02:00,381
What's next?

42
00:02:00,881 --> 00:02:01,441
Finance.

43
00:02:02,061 --> 00:02:04,641
We've started now in finance with early agents

44
00:02:04,641 --> 00:02:06,841
and bringing Claude into the products and services

45
00:02:06,841 --> 00:02:08,161
that you use day to day.

46
00:02:08,801 --> 00:02:21,801
One example here is Citadel told us at our briefing event in May that their investment professionals started to use Claude for Excel, and it gives them a step change function in how they can test, build, and trust their work.

47
00:02:21,941 --> 00:02:26,941
We're really excited for the potential of Claude within the tools and the use cases that this room is building.

48
00:02:28,321 --> 00:02:29,461
But here's the challenge.

49
00:02:31,021 --> 00:02:32,621
Intelligence is ahead of adoption.

50
00:02:33,001 --> 00:02:36,901
What a model can do on its own is doubling every four months.

51
00:02:37,881 --> 00:02:39,481
PwC published early this year

52
00:02:39,481 --> 00:02:41,281
that the capabilities of AI

53
00:02:41,281 --> 00:02:44,661
are going to three quarters of companies.

54
00:02:45,121 --> 00:02:47,201
Sorry, three quarters of the economic gains of AI

55
00:02:47,201 --> 00:02:48,461
are going to a fifth of companies.

56
00:02:48,961 --> 00:02:50,761
This is what we call the diffusion gap.

57
00:02:50,961 --> 00:02:53,021
If you're not building close enough to the curve,

58
00:02:53,241 --> 00:02:55,501
there's opportunity for you to take the frontier

59
00:02:55,501 --> 00:02:57,521
and be able to apply it in your workflows.

60
00:02:58,821 --> 00:03:01,081
We work with firms across financial services,

61
00:03:01,381 --> 00:03:02,301
across our model family,

62
00:03:02,481 --> 00:03:04,661
most of which hopefully this room is familiar with.

63
00:03:05,101 --> 00:03:06,141
We have Haiku.

64
00:03:06,141 --> 00:03:07,921
for speed and volume.

65
00:03:08,421 --> 00:03:09,881
Sonnet for everyday work.

66
00:03:10,581 --> 00:03:12,961
Opus as the daily driver for financial services,

67
00:03:13,421 --> 00:03:15,181
complex tasks, and agentic coding.

68
00:03:15,441 --> 00:03:17,481
And Fable, our frontier model

69
00:03:17,481 --> 00:03:19,161
that can effectively be used

70
00:03:19,161 --> 00:03:20,521
for the hardest work in your firms.

71
00:03:21,261 --> 00:03:22,901
One example here from Millennium

72
00:03:22,901 --> 00:03:25,061
on our most recent model blog

73
00:03:25,061 --> 00:03:27,321
was a PM used Fable

74
00:03:27,321 --> 00:03:28,981
to find a piece of code

75
00:03:28,981 --> 00:03:30,201
that had a bug,

76
00:03:30,241 --> 00:03:32,341
and it was unexplained in five years.

77
00:03:32,601 --> 00:03:33,961
Fable was able to find it.

78
00:03:33,961 --> 00:03:38,141
We're starting to see our models get used in this industry in new and creative ways,

79
00:03:38,281 --> 00:03:40,961
and we believe the frontier of intelligence is just getting started.

80
00:03:42,601 --> 00:03:46,541
Every model release gets smarter, cheaper and safer, so it's really important to take

81
00:03:46,541 --> 00:03:49,981
a minute to call out how the model frontier is continuing to develop here.

82
00:03:50,701 --> 00:03:52,581
Firstly, the intelligence is going up.

83
00:03:52,901 --> 00:03:56,061
Now a model can work for 16 plus hours on its own.

84
00:03:56,341 --> 00:03:58,501
This is up from just minutes in 2023.

85
00:04:00,161 --> 00:04:02,481
In terms of cost per task, it's going down.

86
00:04:02,481 --> 00:04:08,881
The most recent Opus 5.5 model was 40% more cost-efficient than Opus 5.

87
00:04:08,941 --> 00:04:12,781
You can also see those impact metrics for Haiku, Sonnet, and Fable.

88
00:04:13,661 --> 00:04:14,501
And lastly, safety.

89
00:04:14,701 --> 00:04:16,401
This is something we care a lot about at Anthropic.

90
00:04:16,741 --> 00:04:22,781
In the most recent model of Fable 5.1, only 1% of propped injection attacks were successful.

91
00:04:23,021 --> 00:04:26,641
This is relative to other frontier models in the region of 9% to 53%.

92
00:04:26,641 --> 00:04:29,861
So we're really investing in safety alongside capability at Anthropic.

93
00:04:31,181 --> 00:04:32,381
And the banks know this.

94
00:04:32,481 --> 00:04:41,061
The last most recent Evident AI 2026 report that came out last week showed us that banks are 26% ahead of their AI maturity scores.

95
00:04:41,261 --> 00:04:44,961
We're really investing in scaling in applying this technology to our industry.

96
00:04:45,281 --> 00:04:52,641
A few use cases to call out from this report, which are publicly citable use cases that have been reported, is firstly on areas like fraud.

97
00:04:53,101 --> 00:05:00,301
One bank found that with auto loans, they were able to reduce $120 million of potential fraud in 12 months.

98
00:05:00,781 --> 00:05:03,461
In another area, in post-earnings research,

99
00:05:03,761 --> 00:05:08,061
a draft of a report was able to be generated in 15 minutes relative to 45 minutes.

100
00:05:08,381 --> 00:05:11,541
And you can see other use cases here and more broadly across the industry

101
00:05:11,541 --> 00:05:15,481
as we start to talk about the value of AI and the impact we're generating on our business.

102
00:05:17,101 --> 00:05:18,021
So where are we?

103
00:05:18,621 --> 00:05:20,821
In 2023, we built chatbots.

104
00:05:21,021 --> 00:05:24,121
Everyone was asking a question and going backwards and forwards with the model.

105
00:05:24,741 --> 00:05:27,701
In 2024, we brought pilots to firms

106
00:05:27,701 --> 00:05:31,401
and we started to get more knowledge bases and data into these AI systems.

107
00:05:32,101 --> 00:05:34,921
From 2025 to now, we're now building agents.

108
00:05:35,141 --> 00:05:39,361
And in the most case in this room, we're building them hand by hand, one by one.

109
00:05:39,521 --> 00:05:42,681
And our opportunity here to scale is not the model.

110
00:05:42,941 --> 00:05:47,361
Our opportunity here to scale is building the foundation for agents that can go from one to many.

111
00:05:47,681 --> 00:05:52,301
So the next part of the talk, we wanted to spend time focusing on how you can build the agent foundation

112
00:05:52,301 --> 00:05:54,941
that will allow you to scale the frontier of adoption.

113
00:05:55,221 --> 00:05:56,661
So over to you, Keon.

114
00:05:57,161 --> 00:05:57,601
Amazing.

115
00:05:57,701 --> 00:06:02,601
Thank you, Belinda. So, Belinda, as you've just said, there's no better time than now to start

116
00:06:02,601 --> 00:06:06,981
experimenting and building with agents. For this portion of the talk, we want to talk a little bit

117
00:06:06,981 --> 00:06:12,821
about how you can go from that experimentation phase towards productionizing and scaling agents.

118
00:06:13,761 --> 00:06:18,821
But first, some observations that we've seen across the industry. A year ago, we could think

119
00:06:18,821 --> 00:06:24,121
about this as the year of exploration, where engineers were working on a handful of use cases.

120
00:06:24,121 --> 00:06:26,481
They were in POC and pilot mode.

121
00:06:26,861 --> 00:06:29,721
And they took their learnings and transitioned that into this year,

122
00:06:29,741 --> 00:06:32,721
where we've seen this be the age of production.

123
00:06:33,021 --> 00:06:35,541
So we've seen these handful of agents go to production.

124
00:06:36,041 --> 00:06:37,841
A lot of them have been built by hand.

125
00:06:38,101 --> 00:06:41,641
What I think will happen next year is we'll go into the age of scale.

126
00:06:41,841 --> 00:06:46,021
We'll have many agents built on one solidified foundation.

127
00:06:49,341 --> 00:06:51,061
And as Belinda just mentioned,

128
00:06:51,581 --> 00:06:54,201
the models are now the easiest part of the equation here.

129
00:06:54,421 --> 00:06:57,081
We know that the models can handle high-quality tasks,

130
00:06:57,161 --> 00:06:58,601
and we're confident about that.

131
00:06:58,961 --> 00:07:00,481
But what we want to talk about here

132
00:07:00,481 --> 00:07:02,441
is building this five-layer framework

133
00:07:02,441 --> 00:07:05,041
for scaling and productionizing your agents.

134
00:07:05,301 --> 00:07:06,241
Starting from the bottom,

135
00:07:06,621 --> 00:07:09,481
we first want to look at what the agent can do.

136
00:07:09,801 --> 00:07:13,121
So we can give it context for our business use cases

137
00:07:13,121 --> 00:07:16,101
via connectors so that it can reach our systems and data.

138
00:07:16,461 --> 00:07:17,901
Then we want to give it context

139
00:07:17,901 --> 00:07:20,721
for our actual business rules and processes via skills.

140
00:07:21,521 --> 00:07:25,541
Once that's done, we want to think about building the trust layer for the agents.

141
00:07:25,921 --> 00:07:28,341
We can think about that in three general steps.

142
00:07:28,861 --> 00:07:34,781
Who the agent is acting as with an identity and permissions layer, how we can sandbox and

143
00:07:34,781 --> 00:07:38,641
determine the approval layer of the agent so that someone is signing off before it acts,

144
00:07:39,181 --> 00:07:44,901
and lastly, how we know the agent will still work once we adjust any variables over time.

145
00:07:45,601 --> 00:07:47,621
So let's look at each of these five.

146
00:07:49,101 --> 00:07:49,621
Perfect.

147
00:07:49,621 --> 00:08:01,221
So if you've started building agents already, you may think of giving those agents access to your underlying systems and data, manually wiring them by hand to a handful of different source systems.

148
00:08:01,441 --> 00:08:04,901
And that could work for maybe five agents and five systems.

149
00:08:04,901 --> 00:08:08,061
Then you'll have 25 systems to maintain and manage.

150
00:08:08,061 --> 00:08:14,401
But as you think about scaling up your agents into new specialized various tasks,

151
00:08:14,681 --> 00:08:15,021
you'll

152
00:08:15,221 --> 00:08:40,001
want to have those agents work off of one secure connectivity layer, which is MCP. That'll make it a lot easier for you to build at scale with one repeatable framework. Next, let's say now we have the agent have context to our underlying business data and systems. Now we want to give it more context to

153
00:08:40,021 --> 00:08:42,441
into our actual business rules and processes.

154
00:08:42,781 --> 00:08:44,361
And that's done through skills.

155
00:08:44,661 --> 00:08:46,901
And this is where the SMEs that really do the work

156
00:08:46,901 --> 00:08:49,741
and understand the process and the work come into the picture.

157
00:08:50,341 --> 00:08:54,081
And from here on, we're going to look at an agent use case

158
00:08:54,081 --> 00:08:56,941
for a trade break investigation agent.

159
00:08:57,141 --> 00:09:00,481
And we're going to follow this across for the next steps.

160
00:09:00,721 --> 00:09:04,861
In this case, a trade break agent is meant to look at two records

161
00:09:04,861 --> 00:09:06,721
of a single trade that don't match.

162
00:09:07,161 --> 00:09:09,661
And here you can see steps one through three.

163
00:09:09,661 --> 00:09:11,141
a little bit of a simplified version,

164
00:09:11,321 --> 00:09:14,081
but this is the business context that an SME provides.

165
00:09:14,241 --> 00:09:20,301
And we think about skills as living, breathing, versioned code, essentially,

166
00:09:20,521 --> 00:09:21,801
even though they're just plain text.

167
00:09:21,801 --> 00:09:23,181
They're meant to be dynamic.

168
00:09:23,381 --> 00:09:24,841
They're meant to be updated frequently.

169
00:09:27,521 --> 00:09:28,141
All right.

170
00:09:28,321 --> 00:09:31,241
So now we have our context layer for the agent.

171
00:09:31,481 --> 00:09:35,741
We have the agent having access to our source systems and data.

172
00:09:35,741 --> 00:09:42,381
We have the agent that understands how to go from a generalized model to a specialized agent with skills.

173
00:09:43,001 --> 00:09:45,981
Now we want to talk about building the trust layer with that agent.

174
00:09:46,481 --> 00:09:49,021
And first, let's start with the identity and permissions layer.

175
00:09:49,501 --> 00:09:52,241
So here you can see the identity of the agent.

176
00:09:52,441 --> 00:09:53,281
We're going to assign it.

177
00:09:53,461 --> 00:09:55,061
It's going to be the trade break investigator.

178
00:09:55,261 --> 00:09:58,141
We're going to give it a role with permissions called ops analyst.

179
00:09:58,441 --> 00:10:02,221
And you can see here, once that agent needs to go and reach our source systems,

180
00:10:02,401 --> 00:10:05,641
like our order data and our market data, it'll have permission to read.

181
00:10:05,741 --> 00:10:09,101
But if it tries to go reach for our CRM data, it'll be denied.

182
00:10:09,321 --> 00:10:14,321
And we will do that on purpose so that it won't overreach for sources of data that it shouldn't have access to.

183
00:10:15,141 --> 00:10:19,661
And lastly, we don't necessarily want this agent to perform on behalf of us.

184
00:10:19,701 --> 00:10:24,541
We want it to just propose new amendments to our orders data, not actually make those amendments.

185
00:10:24,841 --> 00:10:29,301
And that's where we'll look at on the next slide how we can keep the human in the loop here.

186
00:10:30,801 --> 00:10:36,541
All right. So this is kind of the layer of who signs off before the agent acts, right?

187
00:10:36,861 --> 00:10:40,681
Many of you have heard the term sandbox. We can use that terminology here as well

188
00:10:40,681 --> 00:10:47,501
for putting an approval layer around the agent so that you can see here the four-step process.

189
00:10:47,761 --> 00:10:53,001
The agent will actually go and propose a booking amendment. It'll actually try to

190
00:10:53,001 --> 00:10:57,941
proceed through the process with the context that it has and the source systems and data

191
00:10:57,941 --> 00:11:02,501
that it has access to, but then it'll look for a human, an SME, someone in the loop,

192
00:11:02,621 --> 00:11:04,781
in the process, to approve.

193
00:11:05,681 --> 00:11:09,861
Once they approve, then that amendment will be made and the change will be applied.

194
00:11:10,261 --> 00:11:16,341
This is the layer where you can also put deterministic guardrails called hooks around your agent

195
00:11:16,341 --> 00:11:20,521
so that it's never overreaching beyond what you would like it to do within its confines.

196
00:11:23,381 --> 00:11:24,021
All right.

197
00:11:24,341 --> 00:11:26,541
So now we have a full agent.

198
00:11:26,721 --> 00:11:27,821
Let's say it's in production.

199
00:11:27,941 --> 00:11:28,861
It's working great.

200
00:11:29,301 --> 00:11:32,401
And let's say that we want to make an adjustment to this agent.

201
00:11:32,601 --> 00:11:35,601
Let's say that a new model is shipped,

202
00:11:35,641 --> 00:11:39,821
or we want to make an adjustment to the business context layer with the skills.

203
00:11:40,301 --> 00:11:44,881
How do we actually know if these adjustments are improving the agent

204
00:11:44,881 --> 00:11:47,321
or if they're causing a regression?

205
00:11:47,641 --> 00:11:49,221
So that's our evals layer.

206
00:11:49,461 --> 00:11:53,861
This is where your SMEs, your business users who are in the loop,

207
00:11:53,861 --> 00:11:58,101
who understand the process help determine a set of test cases

208
00:11:58,101 --> 00:12:02,061
to test the agent against with a benchmark or a baseline

209
00:12:02,061 --> 00:12:05,321
so that once you apply any change to any variable,

210
00:12:05,661 --> 00:12:08,581
you'll be able to determine whether that variable change

211
00:12:08,581 --> 00:12:11,561
actually improved the behavior of the agent

212
00:12:11,561 --> 00:12:14,801
or if it caused it to regress, as you'll see here in some of the examples.

213
00:12:17,881 --> 00:12:18,621
All right.

214
00:12:19,001 --> 00:12:22,581
So hopefully you've seen, by building once,

215
00:12:22,581 --> 00:12:25,681
with a foundational layer that you trust,

216
00:12:25,681 --> 00:12:30,821
you can work at speed to deploy many agents at scale.

217
00:12:31,101 --> 00:12:34,481
For us, financial services and regulated industries

218
00:12:34,481 --> 00:12:35,381
doesn't mean slow.

219
00:12:35,501 --> 00:12:37,741
It means that we're seeing a lot of innovation

220
00:12:37,741 --> 00:12:38,841
from the financial sector,

221
00:12:38,881 --> 00:12:40,421
and we're seeing it go hand-in-hand

222
00:12:40,421 --> 00:12:43,701
with security and safety as well.

223
00:12:43,901 --> 00:12:46,461
And those three layers that we just walked through,

224
00:12:46,681 --> 00:12:49,641
who the agent is acting as, who signs off,

225
00:12:49,741 --> 00:12:51,941
and how you still know that agent works,

226
00:12:51,941 --> 00:12:55,681
are the key components of building a safe and secure agent framework.

227
00:12:58,601 --> 00:13:01,141
And once you have this framework that you've built,

228
00:13:01,341 --> 00:13:04,721
you can use it to apply to any new agent that you want to build

229
00:13:04,721 --> 00:13:06,161
for any specialty use case.

230
00:13:06,341 --> 00:13:08,481
So you can see at the top, you'll have your foundation.

231
00:13:09,401 --> 00:13:11,901
And then at the bottom, you'll see where you can actually

232
00:13:11,901 --> 00:13:16,041
bring in more specialty along the way as you continue to scale.

233
00:13:16,381 --> 00:13:20,841
That specialty is worked in combination with the business and engineering

234
00:13:20,841 --> 00:13:25,001
to make sure that the proper business rules and context

235
00:13:25,001 --> 00:13:27,601
are being integrated into the agent workflow.

236
00:13:30,581 --> 00:13:32,161
And you might be wondering,

237
00:13:32,181 --> 00:13:34,381
we haven't mentioned anything like Claude Code yet.

238
00:13:34,801 --> 00:13:37,161
Well, everything that we've talked about so far

239
00:13:37,161 --> 00:13:39,441
has, let's say, four different doors

240
00:13:39,441 --> 00:13:40,641
that you can access it from.

241
00:13:40,961 --> 00:13:44,541
So a business person can access it from Claude Cowork,

242
00:13:44,741 --> 00:13:46,981
an engineer can access it from Claude Code,

243
00:13:47,041 --> 00:13:50,101
or you can take this to the next step

244
00:13:50,101 --> 00:13:54,801
and give the agent more work to do in your own UI environment

245
00:13:54,801 --> 00:13:58,281
via things like managed agents or agent SDK.

246
00:14:01,921 --> 00:14:06,401
Now, so far we've talked about building a secure,

247
00:14:06,721 --> 00:14:09,041
trustable agent foundation framework.

248
00:14:09,341 --> 00:14:11,501
Now let's talk a little bit about the operating model.

249
00:14:11,941 --> 00:14:13,841
Some observations that we've seen,

250
00:14:14,081 --> 00:14:15,661
especially in financial services,

251
00:14:16,041 --> 00:14:19,481
is that the business is now becoming closer to engineering.

252
00:14:20,101 --> 00:14:25,681
And the way that we're observing this is that the business will directly work with the agent that engineering has built.

253
00:14:26,081 --> 00:14:32,401
They'll have direct capabilities to correct that agent or amend it, propose adjustments,

254
00:14:32,441 --> 00:14:35,141
and that's where engineering is in the loop as well.

255
00:14:35,541 --> 00:14:42,341
Once you see the list of proposed adjustments, you can solidify that as a scale change

256
00:14:42,341 --> 00:14:46,021
after you've reviewed and tested all that within your eval framework.

257
00:14:46,021 --> 00:14:54,781
And then anything new that you ship goes right back to the business, and they'll go through the same loop again to continuously improve the overall agent and the process.

258
00:14:58,121 --> 00:14:58,801
All right.

259
00:14:59,261 --> 00:15:05,741
So far we've talked about building the agent framework in a way that you can build once for scale.

260
00:15:06,041 --> 00:15:12,281
And we've talked about how you can turn a generalized, highly intelligent model into a specialized agent.

261
00:15:13,021 --> 00:15:17,221
Now I want to hand it off to Belinda to talk a little bit more about what we're seeing in the industry

262
00:15:17,221 --> 00:15:23,061
by building some of the foundations ourselves for the next step here, which is Claude for Financial Advisors.

263
00:15:25,241 --> 00:15:26,021
Thank you, Keon.

264
00:15:26,321 --> 00:15:29,801
So what you've just seen is us building a foundation once in a firm.

265
00:15:30,121 --> 00:15:33,941
Now I want to talk about how we could take that same approach to our industry.

266
00:15:34,861 --> 00:15:37,701
This foundation we've built in Claude for Financial Advisors,

267
00:15:37,861 --> 00:15:41,261
and whilst you might not all in the room sit in the wealth or the advisor space,

268
00:15:41,261 --> 00:15:47,201
I want you to take this as a pattern and an opportunity to start to think about how we can actually coordinate at an industry level.

269
00:15:48,121 --> 00:15:53,361
So in September, we launched Claude for Financial Advisors, the orchestrator across the stack of an advisor.

270
00:15:53,861 --> 00:15:59,581
This was built on the foundation that advisors only spend approximately 20% of their time with clients.

271
00:15:59,881 --> 00:16:05,161
We wanted to find a way where advisors could spend more time with clients and more time growing their franchise.

272
00:16:05,721 --> 00:16:06,781
So what did we do?

273
00:16:06,961 --> 00:16:09,921
We went through the agent foundation and found ways to scale.

274
00:16:09,921 --> 00:16:13,021
At layer one, we brought connectors in.

275
00:16:13,301 --> 00:16:16,061
The industry system, the deterministic numbers,

276
00:16:16,081 --> 00:16:18,581
and the scaled system that advisors trust,

277
00:16:18,581 --> 00:16:20,021
we brought into the ecosystem.

278
00:16:20,701 --> 00:16:22,661
Layer two, we built skills.

279
00:16:22,921 --> 00:16:25,081
We built a plugin that is available on GitHub

280
00:16:25,081 --> 00:16:27,761
that builds eight skills and workflows

281
00:16:27,761 --> 00:16:29,641
that an advisor can take out of the box,

282
00:16:30,061 --> 00:16:31,581
use today, and amend for tomorrow.

283
00:16:32,181 --> 00:16:33,781
Layer four, approval.

284
00:16:34,341 --> 00:16:37,241
The advisor approves everything that comes out of Claude.

285
00:16:37,461 --> 00:16:39,701
Nothing is saved or sent without the advisor.

286
00:16:39,921 --> 00:16:41,961
and that trust principle holds here too.

287
00:16:42,401 --> 00:16:44,601
And so you can see here on the diagram on the left,

288
00:16:45,321 --> 00:16:47,281
CRM, market data, custodians,

289
00:16:47,421 --> 00:16:49,381
all the tools and foundations and advisor needs

290
00:16:49,381 --> 00:16:50,301
brought into Claude.

291
00:16:50,961 --> 00:16:53,301
So let's go through layer one, the connectors.

292
00:16:53,821 --> 00:16:57,521
We brought in 25 platforms across custody,

293
00:16:57,781 --> 00:17:00,041
portfolio, planning, data, and CRM,

294
00:17:00,361 --> 00:17:04,241
names such as Schwab, Adapar, Salesforce, and more.

295
00:17:04,461 --> 00:17:06,221
And there's more connectors that you can find

296
00:17:06,221 --> 00:17:07,201
on the Claude marketplace.

297
00:17:07,701 --> 00:17:09,781
This is where the industry builds once

298
00:17:09,781 --> 00:17:13,421
the advisor can integrate once, and this is what compounding looks like.

299
00:17:14,361 --> 00:17:15,461
Layer 2 skills.

300
00:17:15,801 --> 00:17:18,221
These are the workflows we built for the advisor

301
00:17:18,221 --> 00:17:21,241
that they can start with and adjust and amend to their own firm.

302
00:17:21,681 --> 00:17:24,581
For example, pre-meeting prep allows an advisor

303
00:17:24,581 --> 00:17:28,641
to bring a full family plan into one brief across multiple systems.

304
00:17:28,981 --> 00:17:32,521
The follow-up allows the advisor to bring those meeting notes,

305
00:17:32,701 --> 00:17:34,381
enter them into the CRM system,

306
00:17:34,581 --> 00:17:37,541
and actually follow through the existing systems and tools.

307
00:17:38,181 --> 00:17:40,901
Nothing goes out until the advisor approves it.

308
00:17:41,081 --> 00:17:42,861
And this skill foundation is a baseline

309
00:17:42,861 --> 00:17:45,201
of which we imagine advisors will start to innovate

310
00:17:45,201 --> 00:17:47,521
and bring Claude into more and more of their workflows,

311
00:17:47,761 --> 00:17:49,941
allowing them to drive impact in their business.

312
00:17:51,301 --> 00:17:54,621
So, in closing, with 28 seconds on the clock,

313
00:17:55,501 --> 00:17:57,441
three questions to take to your firm today.

314
00:17:58,061 --> 00:18:00,121
The first one is the workflow.

315
00:18:00,381 --> 00:18:03,381
How are you applying the frontier intelligence to your firm?

316
00:18:03,501 --> 00:18:06,821
How are you gaining edge and alpha from the frontier models today?

317
00:18:07,661 --> 00:18:09,221
The second is the foundation.

318
00:18:09,621 --> 00:18:12,281
Do you all have a foundation that you trust

319
00:18:12,281 --> 00:18:14,961
for building and scaling agents for the future

320
00:18:15,021 --> 00:18:16,681
as things become more and more specialized?

321
00:18:17,841 --> 00:18:26,101
One connector. How are you actually building the foundations that others can use? Are you building connectors and skills for your users, for your firm, and for the industry?

322
00:18:27,101 --> 00:18:30,101
With that, thank you so much for having us. Enjoy the rest of the conference.