1
00:00:00,440 --> 00:00:04,717
I think a decade looking back,
it'll it'll be obvious that

2
00:00:04,800 --> 00:00:09,437
intelligence became abundant and that
agency became abundant. Like the current

3
00:00:09,520 --> 00:00:10,637
trends we're on are just going to keep

4
00:00:10,720 --> 00:00:13,117
continuing. And um this will be very

5
00:00:13,200 --> 00:00:16,477
strange. I mean, I think for the history

6
00:00:16,560 --> 00:00:19,317
of humanity, um you know, groups of

7
00:00:19,400 --> 00:00:23,637
smart people getting together towards
a shared goal was was the bottleneck of

8
00:00:23,720 --> 00:00:27,437
progress. You know, US The United States
of America in some sense was an example.

9
00:00:27,520 --> 00:00:30,757
It's like the United States of America
is formed from a smart group of very

10
00:00:30,840 --> 00:00:33,917
smart people getting together and having
a vision for the future that they wanted

11
00:00:34,000 --> 00:00:38,077
to enact. And that's the story of nearly
every company um in America and the

12
00:00:38,160 --> 00:00:41,637
story of every YC company. Um and

13
00:00:41,720 --> 00:00:43,957
that's going to change. Like all of a

14
00:00:44,040 --> 00:00:48,957
sudden the scarce resource isn't going
to be intelligence or agency. I really

15
00:00:49,040 --> 00:00:52,917
think it's going to be vision and
ambition. It's like, do you have a clear

16
00:00:53,000 --> 00:00:56,197
view of what you want
the world to look like in

17
00:00:56,280 --> 00:00:59,797
the future? What is the like one way in
which you want to put your finger on the

18
00:00:59,880 --> 00:01:04,117
scale for how the future of the world
will develop and how the how the world

19
00:01:04,200 --> 00:01:07,517
will look like in 5 to 10 years that it
does not look like today? And do you

20
00:01:07,600 --> 00:01:10,077
have the ambition and drive to like go

21
00:01:10,160 --> 00:01:11,637
through all the crap to make that

22
00:01:11,720 --> 00:01:14,597
happen? And AI will make that easier.

23
00:01:14,680 --> 00:01:19,717
Like agents and AI makes that maybe 10
times or 100 times easier than it was a

24
00:01:19,800 --> 00:01:22,637
decade ago. But the flip side of that
is then you all of a sudden you can dream

25
00:01:22,720 --> 00:01:26,117
bigger. Like I think And the world is like

26
00:01:26,200 --> 00:01:28,317
um you know, there's so many things that

27
00:01:28,400 --> 00:01:31,557
need to evolve for us to be able to

28
00:01:31,640 --> 00:01:36,077
fully embrace this technology. Um you
know, the world is like really just, you

29
00:01:36,160 --> 00:01:39,597
know, barely even ready for this
technology today. And I think, you know,

30
00:01:39,680 --> 00:01:44,277
as a builder, we have a responsibility
to prepare the world, right? Like we

31
00:01:44,360 --> 00:01:48,517
have to help enterprises and governments,
you know, to adapt to this

32
00:01:48,600 --> 00:01:52,037
new technology. We have to help figure
out how we secure the world from a

33
00:01:52,120 --> 00:01:54,997
biosecurity perspective or a
cybersecurity perspective. We have to

34
00:01:55,080 --> 00:01:59,877
figure out how we um how we're going
to to manage all these risks that we see

35
00:01:59,960 --> 00:02:03,997
with this new technology. But on the
flip side, it's also the time of like,

36
00:02:04,080 --> 00:02:07,397
you know, unprecedented opportunity
for humans. Like we can develop new

37
00:02:07,480 --> 00:02:10,677
sciences. We can solve problems
in health and biology that have been

38
00:02:10,760 --> 00:02:13,357
forever unsolved. We can build new
businesses that you couldn't have even

39
00:02:13,440 --> 00:02:16,277
imagined before. There's like new
creative opportunities that couldn't

40
00:02:16,360 --> 00:02:18,837
have existed before. So it's like

41
00:02:18,920 --> 00:02:21,757
it's just this incredible cradle of of

42
00:02:21,840 --> 00:02:24,317
opportunity and risks that

43
00:02:24,400 --> 00:02:29,557
I think makes it like no better time
to be someone who's a builder and

44
00:02:29,640 --> 00:02:32,797
has a strong view of how
the world should change.

45
00:02:32,880 --> 00:02:35,877
>> Do you think the path has changed?
I mean, one of the things I saw, I think

46
00:02:35,960 --> 00:02:39,837
Stanford, the amount of computer
science majors actually dropped by some

47
00:02:39,920 --> 00:02:44,277
double-digit percentage. People sort of
worried like, which is sort of insane to

48
00:02:44,360 --> 00:02:48,277
me. Like you still sort of need those
skills to even create agents that are

49
00:02:48,360 --> 00:02:51,797
that good. Maybe that won't
be true. I'm not really sure.

50
00:02:51,880 --> 00:02:55,077
How you know, have you changed you know,
what do you what would you say to people

51
00:02:55,160 --> 00:02:59,077
in this audience right now? Like this is
sort of a real question that people are

52
00:02:59,160 --> 00:03:03,077
sort of facing. Like should they become
more word cell and less shape rotator?

53
00:03:03,160 --> 00:03:08,717
Like what you know, what's the move?
And you has that changed um

54
00:03:08,800 --> 00:03:11,717
the kind of people you're looking
to hire and you know, how you manage your

55
00:03:11,800 --> 00:03:13,877
teams right now at Meta?

56
00:03:13,960 --> 00:03:16,757
>> I think systematic
and rigorous thinking

57
00:03:16,840 --> 00:03:20,917
are still incredibly important because
you know, the abstraction layer, I mean,

58
00:03:21,000 --> 00:03:23,237
I didn't used to believe that this is
how this is going to play out, but it

59
00:03:23,320 --> 00:03:27,557
really has. Like the abstraction layer
just keeps changing. So, you know, when

60
00:03:27,640 --> 00:03:32,437
I started a company, back in my day,
we wrote code. Um >> [laughter]

61
00:03:32,520 --> 00:03:35,837
>> And now, you know, I'm sure nobody here
writes code anymore. That's ridiculous.

62
00:03:35,920 --> 00:03:39,597
But um but now it's about how do you
orchestrate the agents together? And

63
00:03:39,680 --> 00:03:43,117
then it's like, how do you develop these
organizations of agents? Like how do you

64
00:03:43,200 --> 00:03:45,997
get like a million agents to work
together well? And then it'll be, how do

65
00:03:46,080 --> 00:03:49,397
you get like a trillion agents to work
together well? Like I think that there's

66
00:03:49,480 --> 00:03:51,957
going to be this continued um

67
00:03:52,040 --> 00:03:55,597
uh need to figure out how you structure,

68
00:03:55,680 --> 00:03:59,477
uh, workflows at the abstraction layer
that we're going to be operating at. And

69
00:03:59,560 --> 00:04:02,157
that form of like rigorous systematic

70
00:04:02,240 --> 00:04:06,597
thinking, I mean, traditionally the way
this would work, like in my era of

71
00:04:06,680 --> 00:04:10,117
starting companies, is you would start
by writing code, and then you would have

72
00:04:10,200 --> 00:04:12,637
organizations of humans, and you'd
figure out how you how to organize those

73
00:04:12,720 --> 00:04:16,477
humans. Um, and that requires system
thinking. And now, maybe it's like much

74
00:04:16,560 --> 00:04:20,037
more much closer to first you you work
with the agent, then you figure out how

75
00:04:20,120 --> 00:04:23,037
to organize like these agents, but, um,

76
00:04:23,120 --> 00:04:27,597
but I think systems thinking is never
going to go out of style. So, I think

77
00:04:27,680 --> 00:04:31,197
it's definitely a mistake to go all
in on Word cell. Like, I think you need to

78
00:04:31,280 --> 00:04:33,797
you need to shape rotate. Um, but then I

79
00:04:33,880 --> 00:04:36,197
think the sort of like, um,

80
00:04:36,280 --> 00:04:38,357
much more of I think what's necessary

81
00:04:38,440 --> 00:04:41,277
going in the future is having, um,

82
00:04:41,360 --> 00:04:46,357
a deeper sort of compass
and philosophical view on how the world

83
00:04:46,440 --> 00:04:48,037
should develop. Because I think there

84
00:04:48,120 --> 00:04:50,677
are there are many many lessons, um,

85
00:04:50,760 --> 00:04:53,597
from human history around, um,

86
00:04:53,680 --> 00:04:55,917
how how we think civilization go through

87
00:04:56,000 --> 00:05:00,637
this period, and, um, so,
you know, humanity will change more

88
00:05:00,720 --> 00:05:02,197
in the next decade than it has in the

89
00:05:02,280 --> 00:05:05,397
past 100 years, probably. And, um,

90
00:05:05,480 --> 00:05:07,797
and so, I think like the imperative for

91
00:05:07,880 --> 00:05:10,077
us to have positive visions for that,

92
00:05:10,160 --> 00:05:14,957
and have coherent articulations
of how that should develop are are more

93
00:05:15,040 --> 00:05:19,357
important than ever. >> Um,
let's get a little more concrete. I

94
00:05:19,440 --> 00:05:21,957
mean, one of the things I'm curious
about is like, are there sort of

95
00:05:22,040 --> 00:05:26,757
applications of AI that you're seeing
among your friends or internal to Meta

96
00:05:26,840 --> 00:05:30,597
that you can talk about that are, you
know, they're sort of obvious near term,

97
00:05:30,680 --> 00:05:34,627
maybe people haven't figured out yet.
I mean, give us some alpha.

98
00:05:34,710 --> 00:05:37,597
>> [laughter] >> Um, I mean,

99
00:05:37,680 --> 00:05:42,117
I think there's still just like
astronomical opportunity in, uh,

100
00:05:42,200 --> 00:05:44,317
agentic looping, and and figuring out

101
00:05:44,400 --> 00:05:47,277
how you develop systems that enable you

102
00:05:47,360 --> 00:05:50,717
to spend like 1,000 x more or 1 million

103
00:05:50,800 --> 00:05:54,437
x more on tokens to drive an outcome in

104
00:05:54,520 --> 00:05:58,117
a in a continuous feedback loop.
Like if you think about most companies,

105
00:05:58,200 --> 00:06:02,037
companies are just these like
large-scale feedback loops where humans

106
00:06:02,120 --> 00:06:05,917
are operating each of the edges.
Like, you know, companies they

107
00:06:06,000 --> 00:06:08,717
they get customers and they figure out
to make those customers happier. And if

108
00:06:08,800 --> 00:06:11,477
customers are happier, then they spend
more. And if they spend more, then they

109
00:06:11,560 --> 00:06:13,957
can hire more people who can then go
figure out how to get more customers and

110
00:06:14,040 --> 00:06:17,477
make those customers happier. And that's
like this you know, that in some sense

111
00:06:17,560 --> 00:06:19,997
is the feedback loop of

112
00:06:20,080 --> 00:06:22,477
every startup or every business. And you

113
00:06:22,560 --> 00:06:25,117
know, these within that there are micro

114
00:06:25,200 --> 00:06:29,717
feedback loops that exist. And I think
developing agentic systems that can

115
00:06:29,800 --> 00:06:34,517
operate and optimize these feedback
loops is there's like just huge amounts

116
00:06:34,600 --> 00:06:38,317
of alpha there. Like I think
we've seen internally at Meta

117
00:06:38,400 --> 00:06:42,877
cases where if you can develop the right
agentic loop and you have the right eval

118
00:06:42,960 --> 00:06:44,917
or the right metric for the agents to

119
00:06:45,000 --> 00:06:48,757
optimize, you can have a swarm of agents

120
00:06:48,840 --> 00:06:54,077
accomplish more than like a team
of 100 engineers in you know,

121
00:06:54,160 --> 00:06:58,077
very very handily actually, very very
easily. And so, I think figuring out

122
00:06:58,160 --> 00:07:03,077
what the world looks like with lots of

123
00:07:03,160 --> 00:07:06,437
sort of this like these
agentic coordination problems. I

124
00:07:06,520 --> 00:07:08,757
think that is like one of the most
interesting problems today. So,

125
00:07:08,840 --> 00:07:10,837
mechanically speaking, I mean, markdown

126
00:07:10,920 --> 00:07:13,237
files, cron jobs. Is I mean, is it

127
00:07:13,320 --> 00:07:15,877
that's and then basically pointing the

128
00:07:15,960 --> 00:07:19,157
agent at enough data so that it can

129
00:07:19,240 --> 00:07:23,117
figure something out that you know,
maybe isn't in distribution. Yeah, I

130
00:07:23,200 --> 00:07:25,717
think figuring out yeah, yeah,
mechanically figuring out what the

131
00:07:25,800 --> 00:07:31,557
metric is. And yeah, it just comes down
to skills, markdown files, cron jobs.

132
00:07:31,640 --> 00:07:35,877
>> Slash goal. >> Yeah, slash goal.
Like I think I think it's always funny how

133
00:07:35,960 --> 00:07:38,557
mundane everything is once
you really dig into it. But um

134
00:07:38,640 --> 00:07:41,437
>> So, it's not magic. You know, I mean,
some some people put a lot of magic

135
00:07:41,520 --> 00:07:44,797
There's like some LinkedIn threads
out there about some magic stuff.

136
00:07:44,880 --> 00:07:47,077
>> advice, ignore LinkedIn. >> Hey.

137
00:07:47,160 --> 00:07:49,557
>> LinkedIn is where you get customers.

138
00:07:49,640 --> 00:07:55,277
>> [laughter] >> So, uh I like
to end on this, which is um

139
00:07:55,360 --> 00:07:59,157
you know, you get
a telegram to send to the

140
00:07:59,240 --> 00:08:03,397
18-year-old version of yourself.
You know, what do you say to that person

141
00:08:03,480 --> 00:08:07,437
right now, given all, you know, I mean
thank you for coming back and sharing

142
00:08:07,520 --> 00:08:10,917
your wisdom with this audience. I mean,
you know, what would you send in a a

143
00:08:11,000 --> 00:08:13,997
message in a bottle to the 18-year-old
version of yourself right now?

144
00:08:14,080 --> 00:08:16,557
>> Yeah, I think the

145
00:08:16,640 --> 00:08:18,917
I think it really boils down to

146
00:08:19,000 --> 00:08:21,437
develop your own internal compass for

147
00:08:21,520 --> 00:08:23,717
how you think the future will develop

148
00:08:23,800 --> 00:08:26,157
and have strong conviction in it

149
00:08:26,240 --> 00:08:29,477
because, you know, you will get so you

150
00:08:29,560 --> 00:08:33,877
will get inundated with noise and people
telling you and like you'll get be

151
00:08:33,960 --> 00:08:35,957
very confusing and it'll be very hard.

152
00:08:36,040 --> 00:08:40,237
And especially when you're young and you
don't have experience like it can feel

153
00:08:40,320 --> 00:08:45,077
very difficult to um have true
conviction in what you believe

154
00:08:45,160 --> 00:08:47,717
and and what you want to do. But,
I think that's the most important thing.

155
00:08:47,800 --> 00:08:50,517
Kind of as we talked about, you know, um

156
00:08:50,600 --> 00:08:53,397
it took a deep, deep conviction in what

157
00:08:53,480 --> 00:08:55,957
we were building to be able to weather

158
00:08:56,040 --> 00:08:58,957
the the sort of storms of many years of

159
00:08:59,040 --> 00:09:02,197
um of uh chaos in the market, in the

160
00:09:02,280 --> 00:09:06,557
industry, and the people around us. And
so, um and and the other piece of advice

161
00:09:06,640 --> 00:09:08,917
I would have is try to identify what is

162
00:09:09,000 --> 00:09:11,597
the what is the exponential in the world

163
00:09:11,680 --> 00:09:14,997
that has both the steepest curve and

164
00:09:15,080 --> 00:09:17,397
will go the longest. And, you know, many

165
00:09:17,480 --> 00:09:20,517
decades ago this curve was uh was

166
00:09:20,600 --> 00:09:24,557
Moore's law. And that probably was,
you know, that was yeah, at the time like

167
00:09:24,640 --> 00:09:27,637
clearly the right thing to invest on.
I think right now it's AI progress, but

168
00:09:27,720 --> 00:09:31,077
there will be more of these very steep
curves in the future. And it's fine if

169
00:09:31,160 --> 00:09:36,357
these curves start um you know, the
starting point is very boring or like it

170
00:09:36,440 --> 00:09:39,757
doesn't even seem that interesting.
Like it you know when we started

171
00:09:39,840 --> 00:09:44,037
when I started working on scale, um
you know we had cat detectors in YouTube

172
00:09:44,120 --> 00:09:48,677
videos and that felt, you know,
it's hard to say explain the story that

173
00:09:48,760 --> 00:09:51,277
that's like the most important
technology of our time, but it was on

174
00:09:51,360 --> 00:09:55,320
just this like unbelievable exponential.