Who's Watching the Agent? | Rob Viglione | Horizen Labs [EP. 310]

1
00:00:00,040 --> 00:00:02,520
Who's watching the agent?
We're entering a new world in a

2
00:00:02,520 --> 00:00:06,040
gentle commerce where we do not
have guardrails yet and we have

3
00:00:06,040 --> 00:00:08,200
to race to build them.
Every asset class will be

4
00:00:08,200 --> 00:00:12,040
tokenized one day soon, and
agents will be operating in this

5
00:00:12,040 --> 00:00:15,520
entire domain autonomously.
They'll be operating in systems

6
00:00:15,520 --> 00:00:18,560
that don't even exist today, and
now your data is completely in

7
00:00:18,560 --> 00:00:22,120
the public domain all the time.
This is Money Never Sleeps.

8
00:00:22,160 --> 00:00:24,800
Sharp riffs, big ideas and real
insights from smart people.

9
00:00:25,000 --> 00:00:26,440
I'm Pete Townsend.
Let's go money.

10
00:00:29,120 --> 00:00:32,720
Never sleeps.
My guest today spent his career

11
00:00:32,720 --> 00:00:35,640
figuring out how machines make
decisions in high stakes

12
00:00:35,640 --> 00:00:39,240
environments and what happens
when those decisions go wrong.

13
00:00:39,880 --> 00:00:42,880
Now he's building the trust
infrastructure for the agentic

14
00:00:42,880 --> 00:00:46,320
economy.
Rob Viglione, Co founder and CEO

15
00:00:46,320 --> 00:00:48,920
at Horizon Labs.
Welcome to Money Never Sleeps.

16
00:00:49,760 --> 00:00:51,600
Happy to be here, Pete.
Thank you for having me.

17
00:00:52,280 --> 00:00:54,920
Awesome.
Just to set the scene, Rob, your

18
00:00:54,920 --> 00:00:58,720
Schwab account is already making
automated investment decisions

19
00:00:58,720 --> 00:01:02,480
for you.
Now imagine that at 100 XS speed

20
00:01:02,640 --> 00:01:05,920
24/7 across every asset class
globally.

21
00:01:06,600 --> 00:01:09,760
Who's watching the agent?
Yeah.

22
00:01:09,760 --> 00:01:12,680
Right now we just count on
Schwab and the other brokerage

23
00:01:12,680 --> 00:01:15,280
firms to handle the algorithmic
trading properly.

24
00:01:15,280 --> 00:01:18,040
But the reality is as these
things become agentic, where we

25
00:01:18,040 --> 00:01:21,960
have algorithms themselves that
have authority, spending power,

26
00:01:21,960 --> 00:01:25,280
trading authority, that are
making decisions on our behalf.

27
00:01:25,680 --> 00:01:28,040
This is where right now we're
entering a new world in a gentle

28
00:01:28,040 --> 00:01:31,360
commerce where we do not have
guardrails yet and we have to

29
00:01:31,360 --> 00:01:35,120
race to build them.
When it's 24/7, this isn't for

30
00:01:35,120 --> 00:01:37,920
me and you, this is for the
agents.

31
00:01:38,080 --> 00:01:40,600
In that absence of human
supervision, what are the stakes

32
00:01:40,600 --> 00:01:42,280
here?
Everyone knows what the stakes

33
00:01:42,280 --> 00:01:44,040
are.
The stakes are existential for

34
00:01:44,040 --> 00:01:46,040
your business, for society, some
could say.

35
00:01:46,360 --> 00:01:49,520
But we're seeing enterprises
spend billions of dollars on

36
00:01:49,520 --> 00:01:52,640
trying to put up guardrails and
really try to do monitoring and

37
00:01:52,640 --> 00:01:55,400
compliance.
Ford agents internally are doing

38
00:01:55,400 --> 00:01:57,360
already.
What we're trying to do, you

39
00:01:57,360 --> 00:01:59,720
know, like in our segment of
this industry is what we're

40
00:01:59,720 --> 00:02:02,160
building cryptographic
guardrails so that we can

41
00:02:02,160 --> 00:02:05,160
actually control and you know
it's not completely controlled

42
00:02:05,400 --> 00:02:08,680
monitor in real time exactly
what agents are doing on our

43
00:02:08,680 --> 00:02:10,560
behalf.
But what actually breaks when

44
00:02:10,560 --> 00:02:13,640
you put AI agents on financial
rails that weren't designed for

45
00:02:13,640 --> 00:02:14,800
them?
A lot, right.

46
00:02:14,800 --> 00:02:17,600
So the scale is just orders of
magnitude.

47
00:02:17,600 --> 00:02:20,000
We could talk 100 X and some
people are talking about a

48
00:02:20,000 --> 00:02:23,000
million times, kind of like
payments volume could be

49
00:02:23,000 --> 00:02:25,760
occurring on these rails that
just aren't designed for them.

50
00:02:26,160 --> 00:02:28,520
These rails are designed for
humans and human systems and

51
00:02:28,520 --> 00:02:31,560
institutions that have their own
internal checks and guardrails.

52
00:02:31,800 --> 00:02:34,080
So you don't have the rogue
trader, the rogue kind of

53
00:02:34,080 --> 00:02:36,960
invoice paid, you know, like
someone picking an invoice and a

54
00:02:36,960 --> 00:02:38,800
company for a billion dollars
when it was supposed to be

55
00:02:38,800 --> 00:02:42,400
$100,000.
We have already a labyrinth of

56
00:02:42,400 --> 00:02:45,960
systems and checks in place that
you are no longer valid.

57
00:02:45,960 --> 00:02:48,680
When we have agents that are
doing things at scale, they're

58
00:02:48,680 --> 00:02:51,680
just unimaginable right now.
And do you see more kind of

59
00:02:51,680 --> 00:02:54,280
instead of having additional
intermediaries that you will

60
00:02:54,280 --> 00:02:57,280
have more agents that are
performing the tasks that

61
00:02:57,280 --> 00:03:02,480
previously had been done by
intermediaries A. 100% So like

62
00:03:02,480 --> 00:03:05,960
my company comes at this from
what we call the Web three-part

63
00:03:05,960 --> 00:03:08,400
of the industry where we're
dealing with agents that are

64
00:03:08,400 --> 00:03:11,120
designed to be in the public
sphere and operating on chain

65
00:03:11,120 --> 00:03:12,840
with literally their own
wallets.

66
00:03:13,080 --> 00:03:16,600
But we also talked to like major
institutions, enterprises and

67
00:03:16,600 --> 00:03:20,240
systems integrators that are
building at a frantic pace,

68
00:03:20,240 --> 00:03:24,480
agents to kind of already
turbocharge internal operations

69
00:03:24,480 --> 00:03:26,400
for companies.
Like whether you're talking

70
00:03:26,400 --> 00:03:29,200
about Volkswagen or, you know, a
variety of other enterprises,

71
00:03:29,200 --> 00:03:32,760
they're building part of agentic
behavior into their business

72
00:03:32,760 --> 00:03:35,480
model so that they can collapse
their operating costs, which

73
00:03:35,480 --> 00:03:38,560
ultimately means probably
replacing human functions in the

74
00:03:38,560 --> 00:03:40,240
front.
You don't want to give them the

75
00:03:40,240 --> 00:03:41,880
keys to the Kingdom, right?
You don't want them to be

76
00:03:41,880 --> 00:03:44,760
completely unconstrained and
just operating completely on the

77
00:03:44,760 --> 00:03:47,520
without guardrails.
So like right now, this is a

78
00:03:47,520 --> 00:03:49,640
problem that companies are
attacking in different ways.

79
00:03:49,640 --> 00:03:52,600
You have to see So's or security
operations of firms that are

80
00:03:52,600 --> 00:03:55,280
just really getting in there and
sandboxing things.

81
00:03:55,280 --> 00:03:58,960
We do that internally at Horizon
Labs and finance Ages that could

82
00:03:58,960 --> 00:04:03,480
be trading 24/7, 365 on any type
of asset class that could be

83
00:04:03,480 --> 00:04:05,520
tokenized.
And the hint on that is every

84
00:04:05,520 --> 00:04:09,200
asset class will be tokenized
one day soon and Ages will be

85
00:04:09,440 --> 00:04:12,200
operating in this entire domain.
They'll be operating

86
00:04:12,200 --> 00:04:14,920
autonomously and they'll be
operating in systems that don't

87
00:04:14,920 --> 00:04:17,519
even exist today.
You then turn agents on to these

88
00:04:17,519 --> 00:04:19,880
things where previously was
humans pressing the trading

89
00:04:19,880 --> 00:04:23,200
button and now agents had the
full authority with a budget and

90
00:04:23,200 --> 00:04:26,920
a wallet with funds that could
disappear in an instant if you

91
00:04:26,920 --> 00:04:28,320
don't put in the proper
guardrails.

92
00:04:28,360 --> 00:04:31,560
It's a human assist AI operated
model that will probably work

93
00:04:31,560 --> 00:04:33,760
best.
Yeah, absolutely.

94
00:04:33,840 --> 00:04:36,760
The agents need to have that
accountability, and that

95
00:04:36,760 --> 00:04:39,600
accountability will come from
these human guardrails.

96
00:04:40,200 --> 00:04:44,520
One of the other sides of this
is exposure and the world being

97
00:04:44,520 --> 00:04:46,960
able to see what you're doing.
Because this is operating on

98
00:04:46,960 --> 00:04:50,720
blockchain rails now, there are
vault based solutions for this.

99
00:04:50,960 --> 00:04:53,280
There are a number of other
startups that are working on

100
00:04:53,280 --> 00:04:55,600
this.
For the person whose agent is

101
00:04:55,800 --> 00:04:58,800
out there acting on their
behalf, or for the organization

102
00:04:58,800 --> 00:05:02,520
whose agent is out there acting
on their behalf, how do you

103
00:05:02,520 --> 00:05:06,920
avoid that portfolio strategy
from becoming known and publicly

104
00:05:06,920 --> 00:05:09,640
exposed?
Yeah, I think this is the, the,

105
00:05:09,640 --> 00:05:12,400
because this is the difference
from, you know, the tribe flight

106
00:05:12,400 --> 00:05:16,120
to the future 5 stuff with D5
kind of as an intermediate stage

107
00:05:16,280 --> 00:05:19,400
in traditional finance, the
firms keep their proprietary

108
00:05:19,400 --> 00:05:22,280
trading secrets, their secret
sauce to their business and to

109
00:05:22,280 --> 00:05:24,600
their profitability secret.
It's on the servers.

110
00:05:24,600 --> 00:05:27,520
It's literally part of human
capital and IT and

111
00:05:27,520 --> 00:05:29,920
infrastructure that resides
within the firm.

112
00:05:30,280 --> 00:05:32,320
Now we're talking about
operating on rails that

113
00:05:32,400 --> 00:05:36,280
basically are global computers,
effectively, right, that are

114
00:05:36,280 --> 00:05:38,920
completely persistent, and now
your data is completely in the

115
00:05:38,920 --> 00:05:42,160
public domain all the time.
So now we need different

116
00:05:42,160 --> 00:05:45,000
technologies that can actually
operate in that public sphere

117
00:05:45,320 --> 00:05:48,680
while retaining privacy and
affording accountability.

118
00:05:48,880 --> 00:05:51,040
Like what?
One major class of technology

119
00:05:51,040 --> 00:05:53,800
here that we specialize in is
called 0 Knowledge Cryptography.

120
00:05:54,320 --> 00:05:56,480
And it's that class of
cryptography that allows you to

121
00:05:56,480 --> 00:06:00,200
basically execute transactions
or computation without revealing

122
00:06:00,200 --> 00:06:02,760
those parameters or the inner
workings of it directly.

123
00:06:02,920 --> 00:06:05,120
Now this is a key breakthrough
technology.

124
00:06:05,480 --> 00:06:08,320
You combine this with other
privacy Brazil technologies like

125
00:06:08,320 --> 00:06:11,240
trust execution environments,
even fully homomorphic

126
00:06:11,240 --> 00:06:14,720
encryption, FHE, this stack,
this sort of privacy and

127
00:06:14,720 --> 00:06:17,920
accountability stack, I think is
absolutely going to be at the

128
00:06:17,920 --> 00:06:20,280
heart of this future Phi
concept.

129
00:06:20,480 --> 00:06:23,160
I spoke with my pal Alejandro
Gutierrez about this and the

130
00:06:23,160 --> 00:06:26,800
question I had for him when he
started talking about privacy

131
00:06:27,200 --> 00:06:32,720
and how the institutional
adoption of on chain rails is

132
00:06:32,720 --> 00:06:35,000
driving this demand for more
privacy.

133
00:06:35,440 --> 00:06:41,320
I said is triad Phi integrating
with D Phi or is Tri fi

134
00:06:41,560 --> 00:06:43,520
infiltrating D fi?
What do you think?

135
00:06:43,520 --> 00:06:47,600
And additionally infiltrating
and as a long term central point

136
00:06:47,600 --> 00:06:49,640
and decrypted guy, it feels that
way.

137
00:06:49,640 --> 00:06:51,880
But I think the reality is it's
going to be an integration.

138
00:06:51,880 --> 00:06:53,760
And this is what I'm talking
about.

139
00:06:53,760 --> 00:06:56,160
These worlds are going to
combine because they just make

140
00:06:56,160 --> 00:06:58,760
sense.
We had some period of time

141
00:06:58,760 --> 00:07:01,800
called 10 years from, you know,
Satoshi not coming from a

142
00:07:01,800 --> 00:07:05,240
washing Bitcoin to in the advent
of D5 where it could scale

143
00:07:05,240 --> 00:07:07,080
sufficiently.
We can start talking about it

144
00:07:07,320 --> 00:07:09,120
and the Financial Times for
instance.

145
00:07:09,360 --> 00:07:12,400
But we're friendly at a point
where you add agentic behavior

146
00:07:12,400 --> 00:07:14,840
and AI in general.
And when you're talking about

147
00:07:14,840 --> 00:07:17,640
persistent trading and
operations on a world computer,

148
00:07:17,960 --> 00:07:20,360
it just makes sense.
So the world is converging.

149
00:07:20,360 --> 00:07:23,080
I think that this infiltration
is going to turn into a full

150
00:07:23,080 --> 00:07:25,960
blown integration.
That's exactly where you know,

151
00:07:26,000 --> 00:07:29,800
the minimum requirement for
traditional institutions to come

152
00:07:29,800 --> 00:07:32,240
on chain and start dealing with
these tokenized assets, which

153
00:07:32,240 --> 00:07:35,200
they absolutely have to.
They know it to be competitive,

154
00:07:35,200 --> 00:07:37,960
they have to and there's so much
opportunity for them to do it.

155
00:07:38,320 --> 00:07:41,560
They need strong privacy.
Now at the same time then very

156
00:07:41,560 --> 00:07:43,480
strong compliance.
And this is what is kind of

157
00:07:43,840 --> 00:07:46,680
shaped our industry over the
last 100 years is this kind of

158
00:07:47,040 --> 00:07:50,080
ever evolving system of
compliance and kind of oversight

159
00:07:50,080 --> 00:07:54,480
from our regulators is
absolutely necessary and must as

160
00:07:54,480 --> 00:07:56,680
a hard requirement be baked into
the system.

161
00:07:56,720 --> 00:07:58,640
Yeah, you.
Just use the term baked in.

162
00:07:58,680 --> 00:08:01,800
And I think that's the key is
that you have the privacy baked

163
00:08:01,800 --> 00:08:05,360
in at the protocol level rather
than these additional utilities

164
00:08:05,360 --> 00:08:10,600
that sit on top of the whole
stack that then are acting as a

165
00:08:10,600 --> 00:08:15,560
gatekeeper rather than protocol
baked in privacy.

166
00:08:15,880 --> 00:08:19,200
When we talked about kind of
protocol level privacy, I think

167
00:08:19,480 --> 00:08:22,800
like this industry, Bitcoin,
Ethereum and the major rails of

168
00:08:22,800 --> 00:08:24,880
web three that I think are going
to be this future financial

169
00:08:24,880 --> 00:08:27,080
rails.
They evolved in this permission

170
00:08:27,080 --> 00:08:29,480
less environment where anyone in
the world can run a node.

171
00:08:29,480 --> 00:08:31,280
You can verify every
transaction, you can see every

172
00:08:31,280 --> 00:08:34,600
transaction, you can broadcast a
transaction without anyone sense

173
00:08:34,600 --> 00:08:36,400
from you a blocking you from
doing so.

174
00:08:36,440 --> 00:08:39,400
When that intersection bumps up
against the regulated world

175
00:08:39,400 --> 00:08:41,320
where you have legal
jurisdictions like the United

176
00:08:41,320 --> 00:08:44,240
States that have things like
Olfact restrictiveness and

177
00:08:44,240 --> 00:08:47,440
there's participants or
countries and actors that are

178
00:08:47,440 --> 00:08:49,920
not allowed to participate in
the US financial system.

179
00:08:50,280 --> 00:08:53,200
You know, this is the concept of
like, well, can we actually bake

180
00:08:53,200 --> 00:08:56,120
that into the protocol itself or
you do or do you need these

181
00:08:56,120 --> 00:08:59,720
overlays now as of right now,
like while the debate is still

182
00:08:59,720 --> 00:09:01,960
happening and he'll probably
still happen for some years.

183
00:09:02,200 --> 00:09:04,360
Like what we're doing as an
example of Horizon Labs is we're

184
00:09:04,360 --> 00:09:07,840
actually baking in like an
overlay to the base blockchain.

185
00:09:07,840 --> 00:09:10,240
Like this is coin basis
blockchain ecosystem.

186
00:09:10,240 --> 00:09:12,680
We're baking in a privacy layer
there.

187
00:09:12,840 --> 00:09:14,120
And we're doing in a couple
different ways.

188
00:09:14,120 --> 00:09:15,640
Like our approach is just
pragmatic.

189
00:09:15,640 --> 00:09:17,960
We want to get out there to
market like a good old tech

190
00:09:17,960 --> 00:09:21,000
entrepreneurs see how things
work and generate demand and

191
00:09:21,000 --> 00:09:22,960
customers and everything kind of
integrate as we go.

192
00:09:23,320 --> 00:09:25,360
But I think the real interesting
question is what you hit on,

193
00:09:25,360 --> 00:09:27,880
which is like, can we actually
at some point take these

194
00:09:27,920 --> 00:09:31,240
overlays to ecosystems and bake
them into the cores of protocol?

195
00:09:31,280 --> 00:09:34,680
I think we're in this bridge
moment of, hey, let's just get

196
00:09:34,680 --> 00:09:39,520
trad Fi on chain and we're going
to need these layers, these

197
00:09:39,520 --> 00:09:43,880
utilities as we do that.
Once we have everyone on chains,

198
00:09:44,120 --> 00:09:48,720
that's where we can then enable
the protocol level to have these

199
00:09:48,720 --> 00:09:53,800
native inbuilt privacy
considerations 100.

200
00:09:54,320 --> 00:09:58,400
Percent for us is the business
in the space like AUS company we

201
00:09:58,400 --> 00:10:01,200
know we have to just comply with
a certain regulations, but not

202
00:10:01,200 --> 00:10:03,120
ever in the world is U.S.
company.

203
00:10:03,120 --> 00:10:06,840
And like reality is like I don't
know if we could bake in like an

204
00:10:06,840 --> 00:10:09,160
OAFAC restricted list into
Ethereum's core protocol.

205
00:10:09,160 --> 00:10:10,960
Like I just don't know if that's
ever possible.

206
00:10:11,600 --> 00:10:14,560
I doubt it.
Overall, Rob, now you've seen

207
00:10:14,560 --> 00:10:18,600
autonomous systems in war zones,
now you're building the trust

208
00:10:18,600 --> 00:10:21,200
infrastructure for the agentic
economy, right?

209
00:10:21,920 --> 00:10:25,880
Is this the moment that zero
knowledge has been waiting for,

210
00:10:25,880 --> 00:10:28,280
or is this the moment that zero
knowledge proofs have been

211
00:10:28,280 --> 00:10:31,600
waiting for?
Or is the market still not ready

212
00:10:31,600 --> 00:10:34,800
to have this conversation?
I've been in ZK saying ZK was

213
00:10:34,800 --> 00:10:38,440
the thing in crypto when Z Cash
launched and then we had a quick

214
00:10:38,440 --> 00:10:42,000
follow like other privacy coin
project on Zen Cash back in

215
00:10:42,000 --> 00:10:44,000
2017.
So I've seen it all.

216
00:10:44,000 --> 00:10:46,400
And the reality is the
technology was extremely clunky

217
00:10:46,400 --> 00:10:48,160
back then.
And our vision was always to

218
00:10:48,160 --> 00:10:50,720
bring ZK to the application
domain, not just have it for

219
00:10:50,720 --> 00:10:54,120
private money transfer.
But we were early and we could

220
00:10:54,120 --> 00:10:57,160
feel it and we saw the industry
evolve from like ZK roll ups to

221
00:10:57,160 --> 00:10:59,960
then you have the DZK virtual
machines that are out there now

222
00:10:59,960 --> 00:11:03,280
for general purpose computation.
You have like is it get proof

223
00:11:03,280 --> 00:11:05,640
generation networks, you have
the proof verification network

224
00:11:05,640 --> 00:11:08,120
or ZK verify.
We now have a stack, like a

225
00:11:08,120 --> 00:11:11,760
modular stack for ZK.
And now like even that one can

226
00:11:11,760 --> 00:11:13,320
say, OK, that's on the supply
side.

227
00:11:13,320 --> 00:11:15,920
But now I think AI is driving
the demand side of that

228
00:11:15,920 --> 00:11:17,440
equation.
And this is the year where we're

229
00:11:17,440 --> 00:11:19,840
actually seeing the supply side
and the demand side of the

230
00:11:19,840 --> 00:11:22,800
equation for ZK really take off
exponentially.

231
00:11:23,240 --> 00:11:26,880
That's fantastic, Rob, what's
the best way for people to get

232
00:11:26,880 --> 00:11:29,120
in touch and learn more about
what you're building at Horizon

233
00:11:29,120 --> 00:11:32,040
Labs?
Yeah, I mean Horizon Labs dot IO

234
00:11:32,040 --> 00:11:34,720
and you know the key there is
this Horizon like Zen, the

235
00:11:34,720 --> 00:11:37,320
cryptocurrency, like the our
first products that we lost in

236
00:11:37,320 --> 00:11:39,120
2017.
But that's the best way.

237
00:11:39,120 --> 00:11:41,440
You can also follow me on Acts
Ed Rob Vigliani.

238
00:11:42,120 --> 00:11:46,080
Fantastic.
Thank you, Rob and folks, if you

239
00:11:46,080 --> 00:11:49,160
enjoyed this episode, please
like subscribe, leave us a

240
00:11:49,160 --> 00:11:50,400
review.
It really does help.

241
00:11:51,160 --> 00:11:53,320
And again, appreciate you coming
out to the show.

242
00:11:53,320 --> 00:11:55,160
Rob, it's been great to talk to
you about this.

243
00:11:55,840 --> 00:11:56,720
Thanks for.
Having me, Pete.