1
00:00:00,340 --> 00:00:01,540
Who's excited for a great day?

2
00:00:02,980 --> 00:00:03,380
Woohoo.

3
00:00:03,500 --> 00:00:04,660
Let's get some energy going.

4
00:00:04,740 --> 00:00:05,540
It's an early morning.

5
00:00:07,620 --> 00:00:08,100
There you go.

6
00:00:08,420 --> 00:00:08,980
I love it.

7
00:00:10,680 --> 00:00:15,600
So special call out to all of those that are putting on this fabulous summit.

8
00:00:15,760 --> 00:00:20,640
My name is Casey Webster, and I'm co-founder of the PrecisionX system.

9
00:00:20,640 --> 00:00:26,400
I also am CEO and founder of Profit Quiver, which is an AI business transformation company.

10
00:00:26,720 --> 00:00:35,120
And I'm joined by Dave Mahofsky, who is my co-founder again for PrecisionX, and he is the CEO and founder of Mindset Innovations Consulting.

11
00:00:35,520 --> 00:00:37,440
So many of you have stopped by our booth.

12
00:00:37,440 --> 00:00:42,240
If you have not done that, please do put that on your list and in your app.

13
00:00:43,040 --> 00:00:45,680
Many people want to know what is the Precision X system.

14
00:00:45,680 --> 00:00:48,480
You're going to kind of notice any archery lovers in the room.

15
00:00:49,480 --> 00:00:51,200
Come on, where are my archery people?

16
00:00:51,600 --> 00:00:54,080
So I have this fascination for archery.

17
00:00:54,120 --> 00:01:16,240
And so you're going to notice that my business is named around archery, but the Precision X system metaphor that we often use is that no matter who is using AI, if it's done appropriately and correctly, and that we have prepared organizations from a human-centered perspective, that everyone can hit the bullseye and that we can create velocity and results within our organizations.

18
00:01:16,640 --> 00:01:19,840
So our PrecisionX system is focused just on that.

19
00:01:20,240 --> 00:01:31,280
Not only do we go in and evaluate tech stacks and look at what are the pain points and problems that we need to solve for businesses, but we start with the 12 domains of human-centered adoption.

20
00:01:31,280 --> 00:01:33,520
And that means psychological safety.

21
00:01:34,000 --> 00:01:42,960
It means leadership evolution, infinite mindset, and using diagnostics and assessments to ensure that we are being intentional.

22
00:01:43,520 --> 00:01:51,200
and that we are going to create human AI human experiences, and that we're going to amplify those human skills that are so critically important.

23
00:01:51,760 --> 00:01:53,400
So we have some activities.

24
00:01:53,400 --> 00:01:54,560
Stop by and see us.

25
00:01:54,560 --> 00:01:55,600
Excited to be here.

26
00:01:55,600 --> 00:01:58,560
Special thanks to everyone for letting us be your premier sponsor.

27
00:01:58,560 --> 00:01:59,400
We're honored to do that.

28
00:01:59,400 --> 00:02:02,320
And I'll turn it over to Dave for the keynote introduction.

29
00:02:11,120 --> 00:02:11,800
Thank you, Casey.

30
00:02:11,800 --> 00:02:13,360
I guess I don't have to add anything else to that.

31
00:02:13,360 --> 00:02:14,160
She did pretty good.

32
00:02:14,800 --> 00:02:15,920
So, okay.

33
00:02:15,920 --> 00:02:20,800
So next up, we have JC Hoyer from Pella Corporation.

34
00:02:21,280 --> 00:02:28,320
And so JC is really the leader of AI, data science, and advanced analytics at Pella Corporation.

35
00:02:29,280 --> 00:02:34,400
It's one of the North America's largest manufacturers of premium windows and doors.

36
00:02:35,440 --> 00:02:51,200
With experience spanning manufacturing, e-commerce, energy, financial services, and real estate, JC brings a cross-industry perspective on how organizations can move from experimentation to real-world AI impact.

37
00:02:51,200 --> 00:02:54,800
He holds master's degrees from Iowa State University.

38
00:02:55,360 --> 00:02:56,320
Round of applause, maybe.

39
00:02:56,880 --> 00:02:57,360
There we go.

40
00:02:58,320 --> 00:02:58,880
There we go.

41
00:02:58,880 --> 00:02:59,520
Go State.

42
00:03:00,280 --> 00:03:00,640
Okay.

43
00:03:01,920 --> 00:03:04,079
Both in business and data science.

44
00:03:04,640 --> 00:03:10,240
He is known for translating complex capabilities into practical, actionable strategies.

45
00:03:11,280 --> 00:03:16,680
In his keynote, JC will challenge a core assumption many organizations still hold.

46
00:03:17,680 --> 00:03:24,720
As AI capabilities accelerate and costs decline, intelligence itself is no longer scarce.

47
00:03:25,359 --> 00:03:33,280
The question is no longer how to access powerful tools, but how to build organizations that can effectively work alongside of them.

48
00:03:34,440 --> 00:03:44,400
Drawing on production grade examples from Pella and beyond, he will explore what it means to operate in a world where AI is not just a tool, but a teammate.

49
00:03:44,720 --> 00:03:58,080
From computer vision at scale to agentic systems that can plan and act, this session will examine how leadership, governance, and organizational design must evolve to keep pace.

50
00:03:58,720 --> 00:04:00,480
So on behalf of PrecisionX System,

51
00:04:01,200 --> 00:04:04,720
Please join me in welcoming JC Hoyer.

52
00:04:09,400 --> 00:04:09,920
Excuse me.

53
00:04:10,600 --> 00:04:11,000
Excuse me.

54
00:04:11,000 --> 00:04:11,360
Excuse me.

55
00:04:12,640 --> 00:04:14,560
That is a maze to get up here.

56
00:04:14,560 --> 00:04:16,960
So if I bumped anybody, forgive me.

57
00:04:17,360 --> 00:04:19,120
Good morning, everyone.

58
00:04:19,280 --> 00:04:21,920
Thank you for the introduction, Paul and team.

59
00:04:22,960 --> 00:04:23,840
Excited to be here.

60
00:04:24,280 --> 00:04:25,040
It's a great day.

61
00:04:25,040 --> 00:04:25,680
Fantastic day.

62
00:04:25,680 --> 00:04:27,200
We're going to learn about, talk about AI.

63
00:04:29,520 --> 00:04:33,840
The first place I want to start is with a question for you all.

64
00:04:33,840 --> 00:04:39,600
Now, I'm going to preface this question, though, emphasizing that you don't have to answer this out loud.

65
00:04:40,480 --> 00:04:45,480
Some of you may be more confident, and some may be more humble than others, but somewhere in between, and that's OK.

66
00:04:45,480 --> 00:04:58,080
But the question I have for you is I ask you to look around the room and consider for yourself whether or not you are the smartest thing in this room.

67
00:04:59,160 --> 00:05:02,480
So glance around, don't answer out loud, but just think on that for a moment.

68
00:05:07,600 --> 00:05:20,160
Now, to frame that a little bit more, it's probably fair that for any one of us, we are the smartest in a certain domain or certain expertise, right?

69
00:05:21,040 --> 00:05:27,680
But it's unlikely that across every single domain and area of expertise, we are the smartest thing in this room.

70
00:05:28,400 --> 00:05:29,600
and beyond, right?

71
00:05:31,680 --> 00:05:32,880
We go further on that.

72
00:05:33,040 --> 00:05:48,640
We know that intelligence, smartness, comes in different forms, from the carbon-based biological form that we are, to the silicon-based form, that is the machines that we're building, to what I believe is the likelihood.

73
00:05:48,640 --> 00:05:54,560
There's probably many other forms of intelligence across the vastness of the cosmos as you go out into the universe, right?

74
00:05:54,960 --> 00:05:56,600
Maybe we discover, maybe we don't.

75
00:05:59,920 --> 00:06:12,640
We know that intelligence also comes in different capabilities, from intelligence and science and mathematics, governance and politics, religion and art and music.

76
00:06:14,400 --> 00:06:16,880
And it's in music that I really want to start this morning.

77
00:06:17,680 --> 00:06:24,240
Music fascinates me as a domain of intelligence, because it combines both science and mathematics

78
00:06:24,840 --> 00:06:25,920
as well as creativity.

79
00:06:26,080 --> 00:06:30,640
And creativity from an intelligence standpoint is hard to validate.

80
00:06:30,640 --> 00:06:31,680
It's hard to verify, right?

81
00:06:32,160 --> 00:06:39,360
In music, if I put something out into the world, I don't know beforehand if it's going to be accepted as good.

82
00:06:39,920 --> 00:06:41,480
There's no clean answer to that.

83
00:06:41,480 --> 00:06:42,520
It may be a small group.

84
00:06:42,520 --> 00:06:44,040
It may go viral.

85
00:06:44,040 --> 00:06:47,160
It may be the next Taylor Swift or whatever it is, but it's hard to judge that.

86
00:06:47,200 --> 00:06:49,480
I think it's a fascinating space for intelligence because of that.

87
00:06:49,480 --> 00:06:52,720
And so we're going to explore this space of music

88
00:06:53,480 --> 00:06:55,440
for just a couple moments here to start off with.

89
00:06:55,440 --> 00:07:02,400
So I'm going to play three different songs, just short bits of three different songs, by three different artists.

90
00:07:03,120 --> 00:07:10,320
The challenge for you all is to decide whether or not that artist is human or AI.

91
00:07:11,200 --> 00:07:11,760
Sound fair?

92
00:07:12,840 --> 00:07:13,800
All right, let's try this.

93
00:07:13,800 --> 00:07:18,000
So when I tested it this morning, my caveat is that the sound system works.

94
00:07:18,000 --> 00:07:19,040
I think it's going to be good.

95
00:07:19,120 --> 00:07:20,200
So we'll see how this goes.

96
00:07:20,200 --> 00:07:22,000
So the first one,

97
00:07:23,600 --> 00:07:24,240
We're going to try.

98
00:07:24,560 --> 00:07:24,880
Ready?

99
00:07:27,080 --> 00:07:50,720
When I look at my life And all that I've been Every scar, every fall Every place I've been in I'm still here breathing Still finding my way It was a good song, right?

100
00:07:50,720 --> 00:07:51,840
Setting aside

101
00:07:52,240 --> 00:07:57,120
It may not be your genre of preference, but I think it was a good quality song for a few seconds there.

102
00:07:57,120 --> 00:07:59,960
You could go look it up on Spotify, I'm sure, and find it out.

103
00:08:01,280 --> 00:08:27,360
So the answer to this, was this human or AI? This song was written by an artist and performed by an artist named Inga Rose. And Inga is an AI. So this song, as we celebrate, is the name of this song. It made it to the top of the iTunes charts recently, and it's had some good popularity over the last couple of months as well. But it's entirely an AI song, right? The next one we've got for you.

104
00:08:30,160 --> 00:08:48,720
Ready for it? Oh, it's hearts on fire and crazy dreams Oh, the nights ignite like gasoline

105
00:08:51,520 --> 00:09:17,840
And light up those streets and never sleep when the sky goes dark. Now we're the wild things on. Good song, right? Maybe a little more popular in the song, but different beat, more country in its vibe. The song is from Luke Combs. So Luke Combs is a Grammy award-winning country artist. He actually had a concert just down the street here a few weeks ago, which

106
00:09:18,240 --> 00:09:46,200
was a great concert if you're able to go to it. Crazy amount of people were there, but it was a good concert, right? One more for us. Last one. Every scar's a story that I survived. I've been through hell, but I'm still alive. They said, slow down, boy. Don't go too fast. But I ain't never been one to live in the past. I keep

107
00:09:46,880 --> 00:10:15,120
A lot of emotion in that one, different style of song, a little rattle in it as well. So this song was by an artist named Breaking Rust, which the name might give this away, because I think it's the most country-sounding name I've ever come across, right? Breaking Rust is an AI. So this is a song of Breaking Rust that went viral last fall and made it to the top of a few different charts as well, but entirely AI.

108
00:10:16,560 --> 00:10:44,960
So that's three different artists, three different songs, two of them AI, one of them human. I think across all three of them, what's interesting is that you can start to make the claim that AI is at least, in a sample set, equivalent to human. I wouldn't say between the two AI and the one human that there was a significant difference. So in this domain of intelligence, there's at least an equivalence.

109
00:10:46,560 --> 00:11:14,560
I think you start to see as well that you could argue that AI is beginning to exceed human capability here too, right? There's A proliferation of more AI songs being put out on the Spotify and different streaming services as well, and they're gaining in popularity over time as well. Now across the human and the AI in this, in the space of intelligence, there's one common

110
00:11:15,880 --> 00:11:44,560
tie in and a nuance of this I want to step into a little bit. And that's whether it's human, whether it's non-human, the intelligence is born out of what is the most abundant resource in existence. And this is not, this is going to be a science joke, stay with me. It's not hydrogen. Anybody gets that? Sun, stars are made out of hydrogen, helium, they fuse together, it's a giant nuclear reactor, right? But the most abundant resource in existence

111
00:11:45,120 --> 00:12:13,200
From the smallest of the small, the Planck length, the largest of the large, super clusters of galaxies, and the universe, the multiverse, and beyond, the most abundant resource is information. Intelligence is emergent out of information. There's many, I think it's a great framing of this, that humans, we're actually nothing more, this is very reductionist to me, but nothing more than

112
00:12:13,680 --> 00:12:40,000
very capable information processing systems, right? Our senses take in any form of information and we put out what usually is intelligence out in the world, right? There's a great quote from Demis Hassabis on this as well that I think is just beautiful and elegant and captures this well. Demis is one of the co-founders of DeepMind. It's now Google DeepMind, Google bought him a few years ago.

113
00:12:41,200 --> 00:13:09,120
This is the group that's building Gemini. They built AlphaFold. AlphaFold won a, helped Demis and some others win a Nobel Prize for unlocking A decades-old challenge in biology related to protein structures. But this statement, a machine that can navigate an infinity of data will be infinite in its reach. I think it's just beautiful, telegant, captures it so well.

114
00:13:09,680 --> 00:13:37,120
the notion of how important information is to intelligence emerging. And so it's this concept of information, it's going to be the through line through the rest of the time this morning. Information, it's availability, our capability to process it, it's what drives the exponential growth of AI that we're witnessing and it's going to continue on its exponential curve. So we'll start there and we'll take the through line of that to

115
00:13:37,840 --> 00:14:05,760
frameworks for embracing intelligence that we have now, knowing that intelligence is only going to improve over time as more information, more computing capacity continues to become available to process that information. We'll wrap up here this morning with some of my thoughts on probabilities of the future. We know the probability that the future is nothing more than just some probability density function with any number of paths behind it.

116
00:14:06,760 --> 00:14:34,160
This is maybe too statistical for anybody here, but there's any number of things that can unfold in the future is the concept. I'm going to share a couple of thoughts on what I believe two of those might be meaningful for you all to take away for the rest of the day. So the capability of AI and its exponential is where we start. Now, there's one core takeaway that I want you all to grasp from this.

117
00:14:34,640 --> 00:14:58,320
I'll go into more detail over this in the next few minutes. The takeaway is that the capability of AI is doubling every four months. And that four-month window is continuing to compress shorter and shorter. And this is what I believe is the root of why there's such a whirlwind sensation of every day there's some new

118
00:14:58,720 --> 00:15:25,760
advancement or new feature or new breakthrough out in the world. If you're staying up to date with it, you have the sensation that just yesterday I learned how to do something. Today I have to learn something different, and tomorrow it's going to be something brand new again. A new model emerges or whatever that might be. From an enterprise standpoint, this drives a lot of challenge in trying to plan, to strategize, to allocate resources, and

119
00:15:26,000 --> 00:15:55,040
I start here because this is truly one of the areas, one of the topics I think I have conversations around the most often, at least weekly at Pella, discussing between the groups that believe the capability is real and the groups that are still doubters, right? And then the vast space of in-between of what's out there. All trying to work out how fast and far we actually need to integrate, should we integrate these capabilities of intelligence into the organization.

120
00:15:56,160 --> 00:16:22,640
The tool I use to help frame this and understand this is a forecast. Forecast is nothing more than three parts. You have some expectation of the future based upon some understanding of today, informed by the past. Nothing novel in that. It's A forecast. The forecast that I found to be most useful for this comes from an AI nonprofit research group called the AI Futures Project. This group, what I think is great about them, is

121
00:16:24,000 --> 00:16:52,800
Their forecast inputs and assumptions are very robust and it's very transparent. So some point today or when you leave here, using your AI tool of choice, go out and research this, look into it. The transparency of what they put in the forecast is great, from considerations for allocation to efforts to continue growing the grid capacity to advancements in compute,

122
00:16:53,280 --> 00:17:22,480
to advancements in foundational models and talent coming in the space and considerations for national security concerns and geopolitical concerns and so on. It's very broad, very deep in its inputs. Specifically what this group seeks to forecast is their call on when artificial general intelligence and super intelligence will be achieved. Now we'll give more definition to that in a moment.

123
00:17:24,480 --> 00:17:49,920
General intelligence, superintelligence, they put out a forecast on this last year. The first forecast came out in April 2025. The title of it was AI 2027. So the title gives us the way. The initial call was we'd achieve general intelligence in 2027. They made an update earlier this year. They bumped it out to 2028 as factors in their forecast have evolved. 2028 is now their mean of that call.

124
00:17:50,480 --> 00:18:19,200
The midpoint is early 2030s, and there's a long tail in this distribution that goes out in the 30s and 40s, all focused on making that call of when AGI and superintelligence will be achieved. There's more voices out that they've come across. They're beginning to converge on 2030 as being that kind of point in time when general intelligence is achieved. And again, I'll give more definition to that in a moment. But there's two parts then of this forecast that I think are meaningful

125
00:18:19,760 --> 00:18:48,480
and illustrate this quite well. And the first part we start on is this plot on your right, which is sourced from a AI benchmarking platform called Meter. It's M-E-T-R. Meter seeks to evaluate the task horizon capability of a given AI model. In other words, what they're looking for is that for any given task, can that AI do in a moment's time what it would take a human

126
00:18:48,960 --> 00:18:59,760
To do in some equivalent amount of time, right, or framing differently again, if you had a task that takes a human one week to accomplish at some level of proficiency, say 80% accuracy.

127
00:19:00,240 --> 00:19:03,520
Can an AI do that same task at the same level of proficiency?

128
00:19:03,520 --> 00:19:05,360
That's what they're looking to evaluate.

129
00:19:06,240 --> 00:19:11,120
Further, what they evaluate is the autonomous capability of any given AI.

130
00:19:12,080 --> 00:19:16,720
This begins to reveal agentic AI, the concept of agentic AI.

131
00:19:17,280 --> 00:19:23,040
Now, hopefully some of you, many of you have at least started to read about, use agents.

132
00:19:23,360 --> 00:19:25,440
But agents are the vehicle that are going to

133
00:19:26,000 --> 00:19:29,520
drive this transformation, deliver this transformation to an organization, right?

134
00:19:29,760 --> 00:19:32,000
The agentic enterprise is a term that's out there.

135
00:19:32,800 --> 00:19:37,360
We'll use agents more and more throughout our daily lives and everywhere in between.

136
00:19:37,360 --> 00:19:47,760
But an agent by itself is just, it's an entity that is given a task, a target, it's given tools to use, some outcome, and it's set for you to go execute.

137
00:19:48,280 --> 00:19:53,840
It goes and does it, comes back, might be a day, might be a week, might be months eventually.

138
00:19:54,640 --> 00:19:55,840
no different than a human.

139
00:19:56,360 --> 00:20:03,640
A human you give some target, some task, some tools, goes and executes, it comes back, and you've got some outcome, right?

140
00:20:03,640 --> 00:20:04,720
That's what the agent is doing.

141
00:20:05,440 --> 00:20:12,800
So a plot on the right is evaluating that exponential curve and these models that continue to be on that curve.

142
00:20:16,280 --> 00:20:19,600
There's one point of significance that the authors call out then.

143
00:20:19,600 --> 00:20:21,440
It's the one-year time horizon.

144
00:20:22,239 --> 00:20:23,679
Now, one-year time horizon

145
00:20:24,239 --> 00:20:32,719
is in their forecast, where they make the call that this is when we will achieve a true automated coder capability.

146
00:20:32,719 --> 00:20:35,760
It takes us to the left plot, right, the plot on the left-hand side.

147
00:20:36,320 --> 00:20:40,600
Automated coder, I wrestle with the right framing or term for this.

148
00:20:40,600 --> 00:20:44,000
The other term I'd give to it, it's a true agentic software developer.

149
00:20:45,120 --> 00:20:51,520
The definition they give on this left-hand plot is we'll achieve that by June of 2028.

150
00:20:52,000 --> 00:21:00,080
This is the idea that you'll have an agentic AI that is able to do the entirety of a software development function, an organization.

151
00:21:00,480 --> 00:21:13,520
Everything from planning to designing to building to deploying to debugging to supporting to maintaining, everything in that space will be able to be done by an agentic AI at that point in time.

152
00:21:13,520 --> 00:21:14,720
That's the call being made.

153
00:21:15,000 --> 00:21:20,080
Now, if you're using agentic tools right now, agentic software tools, programming tools,

154
00:21:21,280 --> 00:21:23,360
Maybe we're ahead of pace on that.

155
00:21:23,360 --> 00:21:24,800
There's some good evidence out there.

156
00:21:24,800 --> 00:21:31,680
You can make arguments kind of on either side, but it's seeming like it's probably going to be at least close to being that June of 2028 time frame.

157
00:21:33,040 --> 00:21:37,600
So once we hit that agentic coder, then you move all the way up to the right-hand side of this curve.

158
00:21:38,720 --> 00:21:43,280
By May of 2029, the call is that is when we achieve artificial superintelligence.

159
00:21:43,680 --> 00:21:48,880
Superintelligence broadly is the idea that it's AI, whatever form it is,

160
00:21:49,440 --> 00:21:52,800
that exceeds human capability across all domains.

161
00:21:53,360 --> 00:22:02,960
The definition they use in this forecast specifically is that it's AI that is two times better than the world's best human relative to the average human in that domain.

162
00:22:02,960 --> 00:22:05,840
I'm going to use my finger puppets to reframe that a little bit.

163
00:22:05,840 --> 00:22:12,640
So if you can see this, if you have the world's best human is this one here, you've got my thumb is the average human.

164
00:22:12,640 --> 00:22:14,240
There's a gap between those two.

165
00:22:14,720 --> 00:22:16,480
then you've got AI that's way up here.

166
00:22:16,720 --> 00:22:20,200
There's an exponential difference is what they're illustrating in this forecast, right?

167
00:22:22,600 --> 00:22:25,760
There's one other milestone of significance I want to call out on this.

168
00:22:26,080 --> 00:22:33,760
You step back one level from superintelligence, it's the superintelligent AI researcher that they call out on here.

169
00:22:34,560 --> 00:22:42,400
This is significant because the idea is this milestone is when we achieve what's termed the takeoff.

170
00:22:43,040 --> 00:22:44,120
The takeoff is the

171
00:22:44,800 --> 00:22:47,520
the idea that AI is now able to self-improve.

172
00:22:48,080 --> 00:22:59,960
So it's AI that's researching new methods, new algorithms, training on those new methods and algorithms, deploying the improved version of itself, and then this massive flywheel begins to take off.

173
00:22:59,960 --> 00:23:11,040
And once you achieve that, then it's thought to be a quick jump to superintelligence, because you've got AI that can now improve itself for whatever challenge that it has and get to superintelligence, right?

174
00:23:12,560 --> 00:23:14,240
So 2 plots illustrating that capability.

175
00:23:14,240 --> 00:23:24,040
I've got one more plot that I want to touch on to build this sentiment of the realness of the capability that is at least my perspective on it, right?

176
00:23:26,080 --> 00:23:35,360
This plot, I think, brings it to more of an immediate kind of perspective because it's focused on economics, on dollars.

177
00:23:36,000 --> 00:23:39,680
This is depicting the annualized revenue run rate

178
00:23:40,080 --> 00:23:43,600
of a number of leading frontier model providers.

179
00:23:44,480 --> 00:23:47,440
The 2 at the top are the specific ones to call out.

180
00:23:47,440 --> 00:23:54,960
You've got OpenAI at the very top, GPT models, and then Anthropic is the one right below it, which are the Claude models.

181
00:23:54,960 --> 00:24:06,080
And for the narrative here, focusing on Anthropic, you can see that at the top right of their line, the curve, that curve begins to steepen.

182
00:24:06,080 --> 00:24:07,600
The slope goes up more.

183
00:24:08,360 --> 00:24:21,840
And this illustrates well what has been the rapid pace of growth in terms of people, enterprises, voluntarily saying, I'm going to give you my dollars to get more of your capability.

184
00:24:22,240 --> 00:24:24,960
I think it's as telling as anything of how real that is.

185
00:24:24,960 --> 00:24:29,840
It's that true desire emerging, I need more, I need more of this, right?

186
00:24:30,560 --> 00:24:37,120
Specifically for Anthropic, so they're founded 2020 or 2021, so they're almost five years old.

187
00:24:38,000 --> 00:24:41,480
From 2023, the annualized revenue is $100 million.

188
00:24:42,000 --> 00:24:44,600
2024, it was $1 billion.

189
00:24:45,040 --> 00:24:47,040
2025, it was $10 billion.

190
00:24:47,040 --> 00:24:50,520
As of April of this year, it's $30 billion.

191
00:24:50,520 --> 00:24:55,760
And they're on pace that by the end of this year, it'll hit $100 billion in annualized revenue.

192
00:24:55,760 --> 00:25:05,920
That's 10xing each of those years, which is just unfathomable growth, but reflects significantly how real the capability is and how much demand there is for this capability.

193
00:25:07,440 --> 00:25:25,040
So I land on this because I think it's as telling as anything of it's real, and sorting out how to begin to embrace this technology, and the term is diffuse, diffuse it in the organization becomes the real challenge for all of us.

194
00:25:27,200 --> 00:25:34,400
The framework I then have for this, we've had our capability, now how do we walk through embracing this?

195
00:25:35,400 --> 00:25:40,160
The framework I use for this, and candidly, I'm going to admit, I don't have all the answers for it.

196
00:25:40,680 --> 00:25:42,800
My team at Pella doesn't have all the answers for it.

197
00:25:42,800 --> 00:25:50,240
I don't think I've chatted with anybody yet that truly has an answer for how to actually embrace and embed this technology through the organization.

198
00:25:50,240 --> 00:25:52,480
But the framework I use is one of the factory.

199
00:25:53,600 --> 00:25:58,400
So you've got in the physical world, Pella manufactures windows and doors.

200
00:25:58,400 --> 00:26:03,040
And then that factory, we've got saws and nailers and

201
00:26:03,520 --> 00:26:06,880
glass cutters and computer vision and robotics.

202
00:26:06,880 --> 00:26:15,760
And you've got people that are operating that equipment, ultimately trying to build quality, good windows and doors products for our customers, right?

203
00:26:16,080 --> 00:26:17,920
So the agentic factory is the same concept.

204
00:26:18,400 --> 00:26:26,640
So in the core of it, you've got the factory that's trying to provide safe, secure, trusted agents and capability to the organization.

205
00:26:27,120 --> 00:26:29,000
We've got agentic operators.

206
00:26:29,000 --> 00:26:29,360
So

207
00:26:29,680 --> 00:26:37,160
The novel ideas, you've got the technology itself, the AI itself running that factory, and you've got those that are consuming it, right?

208
00:26:38,000 --> 00:26:44,240
We're going to give more detail to this in a moment, but there's one other aspect of this I want to illustrate.

209
00:26:44,240 --> 00:26:49,960
I think it's foundationally important to how we bring this capability to life in the organization.

210
00:26:49,960 --> 00:26:55,200
It's one of culture, specifically the culture of experimentation, right?

211
00:26:55,200 --> 00:26:59,280
So not every organization, not every person natively has this

212
00:26:59,840 --> 00:27:04,320
culture of experimentation, but it's critical in the space of AI.

213
00:27:04,480 --> 00:27:09,840
My argument for this is because AI fundamentally is a probabilistic technology.

214
00:27:09,920 --> 00:27:14,600
It can use deterministic tools, but at its core, it's probability.

215
00:27:14,600 --> 00:27:21,200
And with probability, you don't really know how it's going to act until you deploy it, see what happens, and then refine around that, right?

216
00:27:21,600 --> 00:27:24,880
So having in place the right structure to be able to do that rapidly becomes critical.

217
00:27:25,760 --> 00:27:26,720
SpaceX,

218
00:27:27,440 --> 00:27:30,080
It's a fantastic example of this culture.

219
00:27:30,400 --> 00:27:34,600
This culture of rapid experimentation is at the core of what SpaceX is.

220
00:27:34,600 --> 00:27:40,480
And there's probably, I'm sure, some other SpaceX nerds out here in some form or some space, right?

221
00:27:40,480 --> 00:27:45,840
So forgive me if I don't mean to be offensive calling anybody a nerd, but I'll admit to be one of them, right?

222
00:27:47,480 --> 00:27:52,880
And so what you see on the screen are three Raptor engines from SpaceX, Raptor 1, 2, and 3.

223
00:27:53,560 --> 00:27:56,320
I think what's most immediately evident is how beautiful

224
00:27:56,800 --> 00:27:58,720
simple and elegant Raptor 3 is.

225
00:27:59,600 --> 00:28:03,360
Raptor 3 is the most powerful rocket engine in the world.

226
00:28:04,840 --> 00:28:06,960
Of the three on the screen, it's the most performant.

227
00:28:07,640 --> 00:28:11,440
Of the three on the screen, it costs the least to produce and costs the least to operate.

228
00:28:13,600 --> 00:28:23,840
A little side tangent within this, if you may have seen this, last week, SpaceX launched the 10th experimental flight of their Starship rocket, which is their super large rocket.

229
00:28:24,160 --> 00:28:30,320
They strap 33 of those rocket engines together, and that's what powers Starship, right?

230
00:28:30,400 --> 00:28:32,160
It shakes the earth when it launches.

231
00:28:32,320 --> 00:28:34,400
If you can go watch it, it's super incredible to see.

232
00:28:36,000 --> 00:28:45,760
The Raptor 3, SpaceX got to this not through careful planning and trying to design out all the risk, but they got there through rapid iteration, right?

233
00:28:45,920 --> 00:28:51,360
We're going to build it, we're going to launch it, and we're going to blow it up over and over and over and over again, right?

234
00:28:51,360 --> 00:28:52,720
We're going to feel the ephoria of it,

235
00:28:53,280 --> 00:28:54,640
We're going to fail, we're going to get up again.

236
00:28:54,640 --> 00:29:02,720
They buy in wholly to push the boundaries of engineering and physics to fail to learn as much as we can as quickly as we can.

237
00:29:03,240 --> 00:29:05,520
I think this concept becomes vital to how...

238
00:29:05,520 --> 00:29:09,680
I think I need to watch that again because it's just entertaining, right?

239
00:29:11,440 --> 00:29:11,920
That's great.

240
00:29:13,840 --> 00:29:16,680
That concept of rapid failure in a

241
00:29:17,200 --> 00:29:18,800
safe, secure kind of structure.

242
00:29:18,800 --> 00:29:28,600
I don't think it's anything new, but it's critical to AI more so than ever, because there's just an unknown nature of it all, and you don't know until you actually try and actually deploy it, right?

243
00:29:29,640 --> 00:29:38,800
And it is not always easy to illustrate that story to different leaders and different groups that shareholders and so on, but it's critical.

244
00:29:41,120 --> 00:29:43,760
Okay, so let's step into each of these.

245
00:29:44,800 --> 00:29:46,000
The first is the factory.

246
00:29:47,040 --> 00:29:50,080
So this is a quick snapshot, quick illustration of this concept.

247
00:29:50,080 --> 00:29:59,960
At the core of the factory, the Agenta factory, are all these components that allow for observing what an agent is doing, observing how and why an agent is making a decision.

248
00:30:00,040 --> 00:30:02,720
which in itself is still an unsolved space of research.

249
00:30:02,720 --> 00:30:10,280
It gets into what's called mechanistic interpretability, which is opening up the neuron kind of pathway of how a decision is being made.

250
00:30:10,280 --> 00:30:15,920
This is components for an agent registry, so you can see which agents are available in an organization.

251
00:30:16,280 --> 00:30:21,920
You've got management of safe, secure access to systems and data and tools.

252
00:30:22,320 --> 00:30:25,120
protocols built into this around MCP and A2A.

253
00:30:25,120 --> 00:30:33,280
These are all things that you're probably going to get some flavor of throughout the rest of the day and through the different talks and topics that you sit on as well.

254
00:30:34,240 --> 00:30:44,960
But it all gets to, in this framing, creating a governance and guardrail system that allows for safe, secure, trusted output of your intelligence.

255
00:30:45,120 --> 00:30:48,160
There's 2 components of this that I'm going to touch on with a little more in depth.

256
00:30:48,160 --> 00:30:49,280
The first is context.

257
00:30:50,160 --> 00:30:53,360
So context is another framing on information.

258
00:30:54,320 --> 00:31:05,120
Context specifically here, though, is within your organization, the enterprise, it's giving whatever AI you bring in the ability to gain expertise of that domain.

259
00:31:05,760 --> 00:31:17,120
Generally, I believe that as we go forward, models will continue to improve broadly, but there's always going to be some nuance of your organization that probably makes your organization special.

260
00:31:17,680 --> 00:31:20,000
that you need to be able to give context to that.

261
00:31:20,480 --> 00:31:25,440
Example I have of this, maybe somewhat crude, is surgery.

262
00:31:26,000 --> 00:31:28,360
I'm not a trained surgeon by any means.

263
00:31:28,360 --> 00:31:36,320
But if you give me a knife, a scalpel, needle, and thread, I can do surgery on you, move some stuff around, cut you open, sew you back up again.

264
00:31:36,880 --> 00:31:38,760
You probably won't be happy with the outcome.

265
00:31:38,760 --> 00:31:40,480
The quality won't be great, but I can do it.

266
00:31:41,040 --> 00:31:46,960
Versus if I spent the years learning the human body, learning how healing happens, learning about you,

267
00:31:47,800 --> 00:31:52,000
And then did the surgery, will I be a trained surgeon with context and I could do it with quality, right?

268
00:31:52,800 --> 00:31:57,600
The other example here I think drives us home well is NFL Thursday Night Football.

269
00:31:58,560 --> 00:32:00,560
So this is hosted by Amazon.

270
00:32:00,560 --> 00:32:01,760
Some of you may have seen this.

271
00:32:02,160 --> 00:32:05,280
They have a version of their telecast that is the Next Gen.

272
00:32:05,280 --> 00:32:06,560
Stats telecast.

273
00:32:07,120 --> 00:32:08,960
They do this for a few different sports now.

274
00:32:09,600 --> 00:32:17,040
But in real time, what they're showing is they'll predict what the next play is likely to be, everything from the kind of simple, benign,

275
00:32:17,680 --> 00:32:22,720
percentage of time they go left, right, or down the middle, to the more challenging of predicting a blitz.

276
00:32:22,720 --> 00:32:24,240
That's what they're doing on the screen here.

277
00:32:24,240 --> 00:32:27,680
And blitz in football is relatively a surprise.

278
00:32:27,680 --> 00:32:28,680
It is a surprise, right?

279
00:32:28,680 --> 00:32:30,800
It's an unexpected action by the defense.

280
00:32:31,280 --> 00:32:34,600
Catch the offense off guard to try to make a big play.

281
00:32:34,600 --> 00:32:36,480
And you see this on the screen.

282
00:32:36,720 --> 00:32:43,360
Right before the ball is snapped, a little red circle pops up on that Denver Broncos player on the top of the screen, the player in the white uniform.

283
00:32:43,960 --> 00:32:45,600
They're predicting that player is going to blitz.

284
00:32:45,680 --> 00:32:46,560
Circle pops up.

285
00:32:46,560 --> 00:32:49,920
He rushes across the line, makes a tackle for a loss.

286
00:32:50,000 --> 00:32:51,000
It's a good thing in football.

287
00:32:51,000 --> 00:32:56,320
And they're doing this by providing the AI with vast amounts of context.

288
00:32:56,560 --> 00:33:01,600
So the players all have IOT sensors, so devices on their bodies.

289
00:33:01,840 --> 00:33:08,280
They're tracking location on the field, acceleration, orientation, all this kind of information coming out of it.

290
00:33:08,280 --> 00:33:10,400
They've got the field itself.

291
00:33:10,960 --> 00:33:16,000
It's constantly gathering information on the environment, the current situation.

292
00:33:16,800 --> 00:33:24,680
The color commentators are commenting on the video, the game that's annotating video for them.

293
00:33:24,680 --> 00:33:38,880
And what might be the most useful of all of this is they've got a roster of Hall of Fame coaches and players that they have watched game film and then just stream their consciousness.

294
00:33:39,120 --> 00:33:40,000
Just talk about

295
00:33:40,720 --> 00:33:42,600
In this situation, what's this player doing?

296
00:33:42,600 --> 00:33:44,160
What's the coach thinking about?

297
00:33:44,600 --> 00:33:55,840
And it's that knowledge, that tribal knowledge, that tribal expertise, that fills in all these cracks at the hard data, the sensors, the video, doesn't capture.

298
00:33:56,160 --> 00:34:00,480
But it's probably the most important knowledge to give context to your AI.

299
00:34:00,480 --> 00:34:02,560
So Pella, we're trying to do this on a shop floor.

300
00:34:02,800 --> 00:34:08,000
We've got decades of knowledge in different areas of the shop floor, as an example.

301
00:34:08,640 --> 00:34:09,840
Framing I have for it is

302
00:34:10,320 --> 00:34:13,680
There's someone that might be working on a machine for 20 years.

303
00:34:13,680 --> 00:34:16,080
The machine itself is 50 years old.

304
00:34:16,320 --> 00:34:19,840
They just know how to smack the machine on the side of it to make it work.

305
00:34:20,400 --> 00:34:26,800
Well, that context becomes so important to an AI, being able to understand, to be able to be effective in that environment.

306
00:34:27,040 --> 00:34:31,920
It's more likely that 50-year-old machine is going to have another 50 years of life.

307
00:34:32,400 --> 00:34:34,720
It probably is that we're going to invest in a new one as well.

308
00:34:35,200 --> 00:34:44,960
So figuring out how to extract that context and give it to the AI becomes a critical component of enterprise AI, agentic AI capabilities emerging in the organization context, right?

309
00:34:49,280 --> 00:35:03,520
The other core element in this factor I'm going to touch on, and this kind of like spiritual visual that's out there, is what's called constitutional AI, or the soul of your AI.

310
00:35:04,040 --> 00:35:14,360
And so in this factory, working to create a core soul in which all other agents emerge from and have a link into becomes important.

311
00:35:14,360 --> 00:35:25,360
This is where ethical alignment, cultural alignment, practices of how you want the organization to function and act all tie back to what is the soul.

312
00:35:25,360 --> 00:35:29,440
Anthropic goes as far as to have they have an individual on their team

313
00:35:30,480 --> 00:35:34,800
whose sole responsibility is to craft the soul of Claude.

314
00:35:34,800 --> 00:35:42,000
They actually go further and they train that soul intrinsically into the core of Claude itself.

315
00:35:42,560 --> 00:35:45,600
And it's their attempt to try to solve what's called the alignment problem.

316
00:35:45,920 --> 00:35:54,480
So there's an alignment between where AI is emerging into and whether or not it aligns to human values.

317
00:35:54,880 --> 00:35:57,160
It's a real concern that's out there and they're trying to solve it with that.

318
00:35:57,160 --> 00:36:09,280
But from an enterprise standpoint, we can attach our own soul to the agents that are merging and again, drive that consistency or that alignment to our business as well out of that.

319
00:36:12,560 --> 00:36:13,760
So that's a factory.

320
00:36:14,480 --> 00:36:16,880
Now we need something to run this factory.

321
00:36:16,920 --> 00:36:19,200
This is the agentic operators.

322
00:36:20,240 --> 00:36:27,120
And so the idea behind this is that you've got a core group of agents that are tasked with running this factory.

323
00:36:27,280 --> 00:36:39,680
These are agents that are orchestrating interaction with the factory, that are helping to navigate intent, helping to understand tools and how they can be used, helping to manage security.

324
00:36:40,320 --> 00:36:44,040
This visual, I think, or animation helps illustrate this further.

325
00:36:44,040 --> 00:36:44,800
It comes from

326
00:36:46,040 --> 00:36:47,440
A developer built this.

327
00:36:47,480 --> 00:36:48,640
It's called Pixel Agent.

328
00:36:48,720 --> 00:36:50,400
You can go out and use this yourself.

329
00:36:50,800 --> 00:36:52,560
But I think it captures the sentiment well.

330
00:36:52,560 --> 00:36:59,920
Each of these little pixel characters is an agent in itself that you can see working with other agents, interacting with other agents.

331
00:37:00,240 --> 00:37:07,360
So this is exactly what this concept, this idea is, and that we're trying to build at Palo is this set of agentic operators that are in the midst of all this.

332
00:37:10,760 --> 00:37:14,320
And there's one specific part of this

333
00:37:14,840 --> 00:37:18,640
what these operators are tasked with doing that I want to step into for a moment.

334
00:37:19,400 --> 00:37:23,360
And it's the operator that manages compute, memory, and cost.

335
00:37:23,960 --> 00:37:37,280
And this is significant right now because there is, and some of you may, I'm sure, have seen some of this, an imbalance in the supply and demand of available compute resources and demand on top of trying to access all of that.

336
00:37:37,960 --> 00:37:42,240
And supply and demand mechanics would say if there's an imbalance, demand is above supply.

337
00:37:42,560 --> 00:37:44,000
What's going to change is cost.

338
00:37:44,280 --> 00:37:48,800
And so cost is beginning to be experimented with in this space.

339
00:37:49,120 --> 00:37:53,760
The framing I've got on this is, so AI usage is measured in energy.

340
00:37:54,400 --> 00:37:59,760
There's globally right now 20 gigawatts of AI compute available.

341
00:38:00,120 --> 00:38:02,640
Anthropic has claimed the 2 1/2 gigawatts.

342
00:38:02,960 --> 00:38:04,640
OpenAI has claimed the roughly 2.

343
00:38:05,600 --> 00:38:09,480
They're working to try to secure deals to gain more access to this.

344
00:38:09,480 --> 00:38:11,280
I just read about a company last night that is

345
00:38:13,200 --> 00:38:21,600
They are working to put, they're working with Pulte Group, which is a home builder, for homeowners to have many data centers on the side of their house.

346
00:38:22,200 --> 00:38:30,880
And then as you build a network of all these many data centers, they can link together and help try to resolve some of this capacity constraint that's out there as an example.

347
00:38:31,680 --> 00:38:33,280
Well, there's 20 gigawatts available now.

348
00:38:34,160 --> 00:38:41,880
Given the rate of manufacturing of new chips and capability to manufacture new chips by 2030,

349
00:38:42,560 --> 00:38:55,000
We likely get to 200 gigawatts, which will vastly still, vastly will be much smaller than what the demand still is in 2030.

350
00:38:55,000 --> 00:38:56,960
I also struggle to get that one out.

351
00:38:58,960 --> 00:39:00,640
And so what begins to change is cost.

352
00:39:01,240 --> 00:39:08,640
Anthropic is beginning to experiment with moving from their subscription-based buffet-style pricing to true usage-based pricing.

353
00:39:09,360 --> 00:39:12,320
So you can no longer consume all you want at a flat rate.

354
00:39:12,320 --> 00:39:15,600
And they're even beginning to experiment with outcome-based pricing.

355
00:39:16,080 --> 00:39:28,480
So pricing that varies based upon how the token, the output token, so AI is based upon tokens input and output, how that token drives value for you.

356
00:39:28,480 --> 00:39:35,680
So token that might be used for customer service is probably going to be a relatively low value token because customer service is

357
00:39:36,840 --> 00:39:47,840
broadly available, broadly similar, versus the token that's used for inventing the new molecule or the new drug that's going to be a billion dollar thing by itself.

358
00:39:48,000 --> 00:39:53,840
That new invention is going to be priced significantly different than the token that's helping with customer service.

359
00:39:54,240 --> 00:39:58,240
So I share this, I touch on this because it's significant from a planning standpoint.

360
00:39:58,720 --> 00:40:08,040
In an organization, it influences how fast and far do you start to ingrain these capabilities into business-critical functions and workflows is the key concept.

361
00:40:10,080 --> 00:40:14,800
So we've got our factory, we've got our operators, and we need the consumption of it.

362
00:40:15,160 --> 00:40:17,600
And so there's two parts to this I'm going to touch on.

363
00:40:18,080 --> 00:40:21,520
The first is now you've got safe, trusted, quality output.

364
00:40:22,040 --> 00:40:27,200
And so at this point, as a consumer, I can start to order or

365
00:40:27,600 --> 00:40:28,960
use these agents.

366
00:40:29,400 --> 00:40:32,720
And this is where the importance of what's called an agent registry comes to life.

367
00:40:33,280 --> 00:40:47,120
I frame it as a draft board, so you can go out and see for any given agent, what's that agent capable to do, what task can it accomplish, what tools and services can it connect to, how does the community feel about that agent?

368
00:40:47,360 --> 00:40:48,720
All this becomes available.

369
00:40:49,680 --> 00:40:57,200
As a project manager or someone in the business, I can decide upon, hey, I've got this project I need to go execute on.

370
00:40:57,520 --> 00:40:59,960
I can go grab these three agents, bring them into the business, and...

371
00:41:00,040 --> 00:41:06,080
give them a target, some reward that they're going to get, and keep them accountable, right?

372
00:41:06,080 --> 00:41:08,320
Not any different than how you would treat a human.

373
00:41:08,400 --> 00:41:15,600
You have a human on the team, I'm going to give them a target, I'm going to give them tools, give them a reward, and they're going to go to work, right?

374
00:41:15,600 --> 00:41:25,600
So the concept here is really pushing into these agents becoming more and more embedded into project teams and into different functions and becoming more managed like a human would be managed.

375
00:41:26,760 --> 00:41:28,880
And that takes me to the role of humans.

376
00:41:29,680 --> 00:41:33,680
Now, candidly, I don't know the answer for this.

377
00:41:33,680 --> 00:41:41,920
And I've touched on this because it's probably the question I get the most often is what does this all mean for my team, for my individuals, for humans, right?

378
00:41:43,040 --> 00:41:47,040
No one knows the answer to this is my position on this.

379
00:41:47,280 --> 00:41:48,480
Specifically, no one knows.

380
00:41:48,480 --> 00:41:55,280
You think about, you know, I started this off by illustrating artificial superintelligence being here in 2 1/2 years, whatever it is.

381
00:41:55,440 --> 00:41:57,360
What does that actually mean for humans at that point?

382
00:41:57,800 --> 00:41:58,840
No one's going to know, right?

383
00:41:58,840 --> 00:41:59,400
But at least

384
00:41:59,680 --> 00:42:06,480
Right now, I think it's still certainly the case that the role of human is AI and humans building together.

385
00:42:07,000 --> 00:42:10,080
And the human is more and more focused on three things.

386
00:42:10,080 --> 00:42:20,720
It's judgment, accountability, and decision making, judgment of what's right and wrong, accountability to the outcomes, decision making on we're going to go this way, we're going to go that way.

387
00:42:22,000 --> 00:42:23,360
That's my frame for it.

388
00:42:23,960 --> 00:42:27,040
I wish I could be more specific or more

389
00:42:28,200 --> 00:42:31,680
forthcoming and what the future's going to hold, but it'd be a guess, right?

390
00:42:33,280 --> 00:42:35,200
Okay, so we've got our capability.

391
00:42:35,880 --> 00:42:37,440
We've got our factory framework.

392
00:42:37,560 --> 00:42:40,880
You're probably going to get components of this in different forms throughout the rest of the day.

393
00:42:41,440 --> 00:42:45,120
The last bit here I'm going to touch on is let's fly into the future.

394
00:42:45,760 --> 00:42:47,440
So I've got two thoughts on this.

395
00:42:47,520 --> 00:42:50,320
One part of this is near term.

396
00:42:50,840 --> 00:42:59,040
And the other thought in this is a little bit more further out, but keep in mind this is an exponential, so further out is maybe 5 to 10 years or so is the framing for this.

397
00:43:00,240 --> 00:43:06,560
First one is the convergence of digital AI and physical AI becoming the same, right?

398
00:43:06,560 --> 00:43:12,240
Everything up to now that I've been thinking of or sharing with you all is really on the digital AI side.

399
00:43:12,240 --> 00:43:15,600
Most of how we consume AI right now is very much in the digital space.

400
00:43:15,600 --> 00:43:19,360
It's AI that has an indirect impact on the physical world.

401
00:43:19,720 --> 00:43:23,040
An example of this would be algorithmic high-frequency trading.

402
00:43:23,360 --> 00:43:30,880
So AIs that buy and sell shares, and something like 80 or 90% of all share transactions are AI-driven right now.

403
00:43:31,520 --> 00:43:42,840
Well, that buying and selling of shares influences capital allocation, which influences investment, influences building of buildings, manufacturing of goods, providing of services comes out of that as well.

404
00:43:42,840 --> 00:43:44,000
It's an indirect impact.

405
00:43:44,720 --> 00:43:48,400
The physical AI side is its direct impact on the physical world.

406
00:43:49,040 --> 00:43:50,560
This is the humanoid robots.

407
00:43:50,960 --> 00:43:53,120
This is the autonomous vehicle.

408
00:43:53,120 --> 00:44:00,560
So last year, Tesla had their first car roll off the end of the assembly line and autonomously deliver itself to the customer.

409
00:44:01,760 --> 00:44:03,920
This is digital and physical AI working together.

410
00:44:03,920 --> 00:44:06,320
The digital AI is evaluating the environment.

411
00:44:06,920 --> 00:44:13,920
Physical AI is in taking that evaluation decision from the digital and acting on it, steering the wheel, accelerating, braking, all that kind of stuff.

412
00:44:14,480 --> 00:44:17,840
The humanoid side of this, last year there was the first humanoid Olympics.

413
00:44:17,840 --> 00:44:19,120
There'll be another one this year.

414
00:44:19,440 --> 00:44:26,160
Most recently there was a humanoid first human marathon in China where the humanoids blew away the human records.

415
00:44:26,320 --> 00:44:27,120
That's another example.

416
00:44:27,120 --> 00:44:36,080
And there's many in this space that believe there's really two last sort of big milestones or discoveries to be made.

417
00:44:36,080 --> 00:44:38,880
One is solving for the information scarcity.

418
00:44:39,280 --> 00:44:47,120
So context, again, giving enough context to whatever form of physical AI there is to understand how to operate in the world.

419
00:44:47,600 --> 00:44:50,080
Getting that data is a challenge, but it's being resolved.

420
00:44:50,760 --> 00:44:53,840
An engineering challenge, specifically around the human hand.

421
00:44:54,400 --> 00:45:01,520
The human hand is a marvel of engineering, so figuring out how to create the humanoid version of that is another big challenge.

422
00:45:02,360 --> 00:45:03,520
I'd argue as well that

423
00:45:04,080 --> 00:45:09,200
Physical AI is going to take the form of whatever the application desires or needs, right?

424
00:45:09,200 --> 00:45:18,360
So a roofing robot or a robot that's installing Windows may not be a humanoid form, but it's going to be here soon, right?

425
00:45:18,400 --> 00:45:26,000
Pella, as an example, is starting to experiment with some of our own versions of multi-form physical AI in our factories.

426
00:45:29,520 --> 00:45:31,600
The last thought I have for you all,

427
00:45:33,200 --> 00:45:39,000
touches on, at least my perspective, of how we start to solve this capacity constraint back to cost.

428
00:45:39,000 --> 00:45:43,400
And frankly, it's just a fun kind of thought exercise to walk through.

429
00:45:43,400 --> 00:45:48,080
It has to do with us going to the moon and harvesting the moon to build chips, right?

430
00:45:48,560 --> 00:45:49,400
So stay with me on this.

431
00:45:49,400 --> 00:45:50,560
I'm going to go fast through this.

432
00:45:52,080 --> 00:45:53,760
So focus is on SpaceX.

433
00:45:54,400 --> 00:45:58,640
SpaceX is merging with xAI to go public later this year.

434
00:45:58,920 --> 00:46:01,280
It'll give them greater access to public markets.

435
00:46:01,440 --> 00:46:03,360
Public markets are much larger than private markets.

436
00:46:03,600 --> 00:46:05,920
In doing so, they're going to build a TeraFab.

437
00:46:06,320 --> 00:46:10,880
So TeraFab will be producing 1 terawatt of AI energy and one structure.

438
00:46:10,880 --> 00:46:12,040
They're already starting to build this thing.

439
00:46:12,040 --> 00:46:12,720
It's outside Austin.

440
00:46:12,840 --> 00:46:13,600
That's where they're at.

441
00:46:14,000 --> 00:46:23,440
So you go from 20 gigawatts of AI energy available globally to one vertically integrated manufacturing structure that has a terawatt available.

442
00:46:23,960 --> 00:46:27,800
They're going to 50 times that terrestrially, and they're going to go to space.

443
00:46:27,800 --> 00:46:32,640
So SpaceX is right now launching one rocket every two days.

444
00:46:33,600 --> 00:46:35,040
I saw one this spring.

445
00:46:35,040 --> 00:46:40,280
I took my family on a spring break trip down to Cape Canaveral and forced us to go watch a SpaceX for rocket launch, right?

446
00:46:40,280 --> 00:46:45,120
But they're pacing from one rocket every two days to one every 5 minutes.

447
00:46:45,520 --> 00:46:48,240
The target is that they'll start to build what's called the Dyson swarm.

448
00:46:48,600 --> 00:46:54,080
Dyson swarm is basically this giant array that is trying to consume as much of the sun's energy as possible.

449
00:46:54,680 --> 00:46:58,400
And that sun's energy is going to power AI data centers in space.

450
00:46:58,840 --> 00:47:11,760
And as we make our way to the moon, whether it's NASA and or SpaceX or others, and set up the infrastructure to start manufacturing and mining on the moon, we'll mine the moon and then using

451
00:47:12,320 --> 00:47:30,640
mass drivers, which are depicted here, start launching chips out into space to join that Dyson swarm, eventually get to harvesting one one-millionth of the sun's energy, which is, another context on that would be, that's roughly 2,000 Earth's worth of energy consumption.

452
00:47:31,040 --> 00:47:34,560
All of that to go to power our growing need for AI.

453
00:47:35,240 --> 00:47:35,920
over and over again.

454
00:47:36,040 --> 00:47:40,720
And I'll end on what I think is probably the most fascinating fact of all this, not even tied to AI.

455
00:47:40,720 --> 00:47:54,960
But once we're able to manufacture on the moon, it will be cheaper to build something on the moon and launch it to the earth and land it somewhere on the earth than it is to manufacture on the earth and ship it around the earth.

456
00:47:54,960 --> 00:47:58,520
I think it's just such a fascinating thought to me to think about that.

457
00:47:58,520 --> 00:48:00,040
We're shooting things back to the earth.

458
00:48:00,080 --> 00:48:03,360
So that's what I have for you all.

459
00:48:04,160 --> 00:48:04,960
Thank you all.

460
00:48:05,120 --> 00:48:06,240
Go and have a great day.

461
00:48:14,700 --> 00:48:15,020
Sure.

462
00:48:15,980 --> 00:48:20,220
Let's see if we can get time.

463
00:48:21,500 --> 00:48:22,140
All right.

464
00:48:22,540 --> 00:48:27,860
So does anyone have any questions for JC or are you afraid?

465
00:48:27,860 --> 00:48:29,220
I have one.

466
00:48:29,220 --> 00:48:30,300
Okay, we have one.

467
00:48:30,300 --> 00:48:31,220
Excellent.

468
00:48:31,220 --> 00:48:32,860
And let's get the microphone.

469
00:48:32,860 --> 00:48:34,540
Let's be patient with the microphone.

470
00:48:35,180 --> 00:48:35,540
Here we go.

471
00:48:39,120 --> 00:48:50,160
So I was wondering if you, what are your thoughts on China's court decision that AI cannot be used to replace human employees strictly for monetary purposes?

472
00:48:50,640 --> 00:48:52,880
Yeah, that is a very good question.

473
00:48:53,840 --> 00:49:03,040
Candidly, I don't know if I have a good thought on that yet, because I quickly get into ethical considerations, and it's a dicey space, right?

474
00:49:03,080 --> 00:49:03,440
I think

475
00:49:05,360 --> 00:49:07,840
economics, right or wrong, are going to win out.

476
00:49:08,280 --> 00:49:17,840
And so the pressure's always going to be on how do you produce goods and services at either a lower cost, greater profit margin, I think is going to hold, at least in the near term.

477
00:49:18,240 --> 00:49:25,840
And so whether or not that decision in China sets off a trend globally, we'll have to wait and see.

478
00:49:26,160 --> 00:49:31,440
I wish I had a better answer for that, but it's a dicey sort of area with no clean answer behind it.

479
00:49:31,440 --> 00:49:33,280
What happens to economics, though, when

480
00:49:34,400 --> 00:49:36,880
unemployment rates are crazy because of AI.

481
00:49:37,200 --> 00:49:40,760
So you're not producing something that nobody can buy.

482
00:49:41,120 --> 00:49:48,720
Yeah, So the promise or the, my framing for this, I go back to why is AI even a thing from its start.

483
00:49:48,760 --> 00:49:55,280
What's target is to at some point push the cost of goods and services to 0, as close to 0 as we can get.

484
00:49:55,640 --> 00:50:00,280
And you quickly get into this notion of some future where there's infinite abundance.

485
00:50:00,280 --> 00:50:03,360
So the scarcity that drives all of our economics

486
00:50:03,680 --> 00:50:09,800
goes away and I can go and build something, consume something, get it, whatever I need in any form.

487
00:50:09,800 --> 00:50:13,960
And there's going to be some interim period between now and getting to that.

488
00:50:13,960 --> 00:50:16,240
It's going to be very interesting.

489
00:50:16,560 --> 00:50:20,240
And I don't have the answer for what it's going to look like by any means.

490
00:50:20,480 --> 00:50:21,800
So it's...

491
00:50:21,800 --> 00:50:23,280
We've got another question here.

492
00:50:23,600 --> 00:50:30,960
I'm just going to suggest that we could make a conference for AI in society.

493
00:50:32,000 --> 00:50:35,680
That's not what this particular conference is, but I totally understand your issue.

494
00:50:44,320 --> 00:50:52,760
Mark Adressing had an interview three months ago, and he talked about that AI with the boom just started, right?

495
00:50:52,760 --> 00:50:53,920
We haven't even started.

496
00:50:53,920 --> 00:50:58,080
And he focuses on productivity, and he mentions something that

497
00:50:58,640 --> 00:51:04,560
We don't think productivity is the way it was like in the 1800s and the possibilities of growth.

498
00:51:04,880 --> 00:51:13,920
The way society was in the late 1800s to what it is in the early 2000s, it's really different, right?

499
00:51:14,480 --> 00:51:16,240
Everybody was expecting growth.

500
00:51:16,400 --> 00:51:26,960
And what he mentions is there's the way we're building things is actually, since everybody can do it, might reduce the cost

501
00:51:27,360 --> 00:51:28,960
of getting things like you mentioned.

502
00:51:29,440 --> 00:51:39,120
But one of the things, there's a discussion between him and Peter Thiel about there's not going to be new stuff, new technology being built.

503
00:51:39,520 --> 00:51:41,120
But what's your take on that?

504
00:51:41,440 --> 00:51:44,800
Because I find it kind of weird those two guys are fighting over.

505
00:51:45,040 --> 00:51:47,320
Like we've done everything under the sun and the other guys.

506
00:51:47,320 --> 00:51:50,120
No, there's like, there's more stuff coming down the road.

507
00:51:50,800 --> 00:51:52,800
And what's your take on that?

508
00:51:53,160 --> 00:51:53,920
Great question.

509
00:51:54,400 --> 00:51:59,920
I land on the side there's more, there's more stuff coming, and the simple framing I have on.

510
00:52:00,040 --> 00:52:01,840
This is where I ended with space, right?

511
00:52:01,920 --> 00:52:09,200
Space is truly the next frontier that we're starting to step into and embrace, and I was chatting with a group last night as an example.

512
00:52:09,200 --> 00:52:14,480
There's a whole other mirror economy to emerge in space, so you've got the notion of...

513
00:52:15,440 --> 00:52:16,800
take how we produce goods right now.

514
00:52:16,800 --> 00:52:19,200
How do you get a spatula into your kitchen?

515
00:52:19,320 --> 00:52:25,840
Well, there's, it's produced in China, gets on a barge, goes across the ocean, gets on a truck, gets to the store, and so on.

516
00:52:26,000 --> 00:52:30,720
That same supply chain has a place in space as well.

517
00:52:31,120 --> 00:52:34,000
SpaceX will be that barge that gets stuff in the low orbit.

518
00:52:34,240 --> 00:52:35,040
There'll be another

519
00:52:35,640 --> 00:52:46,720
economy, companies that emerge that then take that off the barge and they'll orbit and move it to the moon, to Mars, and so on, all that's going to require new invention, new technology to continue to emerge, right?

520
00:52:46,720 --> 00:52:57,120
So I'm very much in the camp of there's more technology, more invention, more creation to emerge, and AI is going to enable that, is my perspective.

521
00:52:57,120 --> 00:52:58,160
Productivity, right?

522
00:52:59,280 --> 00:53:03,920
The change we see that, some of us have seen in productivity, that is opening.

523
00:53:04,760 --> 00:53:08,440
other doors that some people even imagine that will be doing that stuff.

524
00:53:08,800 --> 00:53:09,240
Correct.

525
00:53:09,440 --> 00:53:12,280
Productivity is, I think, the easy place to start, right?

526
00:53:12,280 --> 00:53:20,160
Because it reflects on there's a fundamental inefficiency to how we run our economies or organizations, whatever it is.

527
00:53:20,160 --> 00:53:26,280
And so there's a gap there that can be closed with AI, which comes to life in productivity, right?

528
00:53:26,280 --> 00:53:27,440
And so it's a great starting point.

529
00:53:28,120 --> 00:53:28,640
Good question.

530
00:53:29,360 --> 00:53:33,680
My question comes from your initial bit about your music, your examples.

531
00:53:34,200 --> 00:53:37,440
And I'm thinking about the inspiration, right?

532
00:53:37,440 --> 00:53:43,920
The idea to take something old and make it new in a new way, like that new idea that didn't exist, like that inspiration part.

533
00:53:44,240 --> 00:53:47,160
I suspect the AI generated songs, a human still did that.

534
00:53:47,160 --> 00:53:51,760
Like it would still generate human inspiration, like combine these different things in this way with this outcome.

535
00:53:53,040 --> 00:53:56,480
How does the inspiration or that part become AI?

536
00:53:57,040 --> 00:53:59,040
That's what I'm trying to understand and think through.

537
00:53:59,120 --> 00:54:00,080
That's a good question.

538
00:54:00,080 --> 00:54:01,520
I mean, so

539
00:54:04,520 --> 00:54:08,160
Where I go with that is I get to what is the root of intelligence?

540
00:54:08,240 --> 00:54:14,080
And so is there something unique about humans that is beyond just finding patterns and information?

541
00:54:15,920 --> 00:54:19,600
I would argue that humans maybe aren't really that unique in that sense, right?

542
00:54:19,760 --> 00:54:22,960
It's all just we're consuming information and building patterns out of it.

543
00:54:22,960 --> 00:54:24,240
So from an AI standpoint,

544
00:54:25,040 --> 00:54:36,880
Our form of AI right now is really based upon the transformer architecture, which may not be the form that gets us to super intelligence it may change, it may come in a different architecture, right?

545
00:54:37,279 --> 00:54:40,240
But it's still baked into patterns, and so...

546
00:54:41,000 --> 00:54:55,359
There's no reason that an AI can't form that inspiration and create some new pattern that might seem like a net new-- it's never existed before, but everything is born out of something from the past, right?

547
00:54:55,439 --> 00:54:56,960
So that's my framing on it.

548
00:54:57,120 --> 00:55:01,359
So my probability is somewhere in there, maybe, right?

549
00:55:01,439 --> 00:55:07,880
So you mentioned the importance of context.

550
00:55:08,800 --> 00:55:10,000
How do you envision

551
00:55:11,120 --> 00:55:20,160
and the tribal knowledge, how do you envision the extraction of that from a long-tenured employee without disrupting that work?

552
00:55:20,640 --> 00:55:22,240
Yeah, it's a good question.

553
00:55:22,479 --> 00:55:32,960
Culture drives this, so being diligent and trying to historically in the manufacturing environment, there's a concept of lean manufacturing, which

554
00:55:33,960 --> 00:55:37,560
In simple terms, it's basically like any extra action was waste.

555
00:55:37,560 --> 00:55:40,560
So it was all focused on how do you take out unneeded actions.

556
00:55:41,120 --> 00:55:51,680
Well, a shift that has to be made is that action to get tribal knowledge out of the individuals is now of value to the organization, to the AI, and so on.

557
00:55:52,120 --> 00:55:59,040
And so that's the first step in this, is now building that back in to the process and trying to build it in as passive as you can.

558
00:55:59,240 --> 00:56:01,600
And so in our state, it's either

559
00:56:02,080 --> 00:56:18,320
allocating some amount of time, or it's just in their day to go and have an interview, an interaction with an AI that can help them work through some question and answer to extract that, to putting a device on them where they're just wearing this and speaking into a recorder throughout the day or for some period of time.

560
00:56:18,880 --> 00:56:20,400
There's no easy answer to that.

561
00:56:20,400 --> 00:56:27,520
It's going to vary as you go across enterprises and industries and so on to try to get that knowledge out of the individuals.

562
00:56:28,640 --> 00:56:30,800
Some areas it might be easier if

563
00:56:31,280 --> 00:56:34,160
The typical, the amount of work you do is focused on a computer.

564
00:56:34,160 --> 00:56:35,600
It's easier to track that.

565
00:56:35,600 --> 00:56:37,000
Meta's doing this as an example.

566
00:56:37,000 --> 00:56:44,880
It came out last week that they're forcing all their employees to have tracking software on their computers.

567
00:56:45,120 --> 00:56:48,960
So all their mouse clicks, everything is now going to be tracked and they don't have an option out of it.

568
00:56:49,320 --> 00:56:52,360
It's easier to enforce it there than it is on the factory floor.

569
00:56:52,360 --> 00:56:57,840
And it's going to be a challenge for any organization to have to wrestle with is what's the cultural impact

570
00:56:58,360 --> 00:57:04,240
of me trying to get information out of my workforce because it's going to benefit my AI to now be performant.

571
00:57:04,680 --> 00:57:05,800
All right.

572
00:57:06,280 --> 00:57:06,560
And