1
00:00:06,160 --> 00:00:08,317
Good morning.

2
00:00:08,400 --> 00:00:11,197
Right after World War II, there was a

3
00:00:11,280 --> 00:00:13,997
massive building boom, the largest in

4
00:00:14,080 --> 00:00:16,877
history. In the US, people were coming

5
00:00:16,960 --> 00:00:18,877
home from war. There were new

6
00:00:18,960 --> 00:00:21,597
technologies, new construction methods,

7
00:00:21,680 --> 00:00:24,157
and buildings were going up like gang

8
00:00:24,240 --> 00:00:29,197
busters. The builders of that day
drew on the style of modernism from the

9
00:00:29,280 --> 00:00:30,877
1920s.

10
00:00:30,960 --> 00:00:34,157
That's really simple geometries, clean,

11
00:00:34,240 --> 00:00:36,877
bare surfaces, monochromatic color

12
00:00:36,960 --> 00:00:39,757
palettes, and no adornment. Now, say

13
00:00:39,840 --> 00:00:43,997
what you will about modernism,
but the founders behind the movement had

14
00:00:44,080 --> 00:00:47,437
intentionality.
There was a point of view behind the

15
00:00:47,520 --> 00:00:50,317
Villa Savois. There were principles

16
00:00:50,400 --> 00:00:53,357
behind the Bow House. But after World

17
00:00:53,440 --> 00:00:56,557
War II, with urgency to build and new

18
00:00:56,640 --> 00:00:59,197
construction methods, the style was

19
00:00:59,280 --> 00:01:02,237
essentially copied again and again

20
00:01:02,320 --> 00:01:06,957
without intentionality.
The thinking got thinner and thinner,

21
00:01:07,040 --> 00:01:09,037
and all that was left were generic

22
00:01:09,120 --> 00:01:11,677
patterns ills suited to the context at

23
00:01:11,760 --> 00:01:15,037
hand. For example, we used to have banks

24
00:01:15,120 --> 00:01:17,757
that looked like this, signaling trust,

25
00:01:17,840 --> 00:01:20,637
reliability, and security.

26
00:01:20,720 --> 00:01:24,477
And then we got this.

27
00:01:24,560 --> 00:01:26,237
You're all too familiar with this

28
00:01:26,320 --> 00:01:28,237
reality because it carried on for

29
00:01:28,320 --> 00:01:32,157
decades until this day. Zombie buildings

30
00:01:32,240 --> 00:01:34,797
all over the place. Nothing

31
00:01:34,880 --> 00:01:37,677
differentiates them. Nothing says this

32
00:01:37,760 --> 00:01:40,957
is intentional. This is fit to purpose.

33
00:01:41,040 --> 00:01:43,037
They essentially show no care for the

34
00:01:43,120 --> 00:01:45,277
user and the habitants around it. No

35
00:01:45,360 --> 00:01:48,397
care for context or the brand. I think

36
00:01:48,480 --> 00:01:50,797
about this a lot. One, we're surrounded

37
00:01:50,880 --> 00:01:53,437
by them. And two, we're in another

38
00:01:53,520 --> 00:01:58,237
building boom now. The AI building boom.

39
00:01:58,320 --> 00:02:00,157
Everybody and their mother can build

40
00:02:00,240 --> 00:02:02,957
now. Teams of three can do what teams of

41
00:02:03,040 --> 00:02:05,757
30 used to be needed for. But there's

42
00:02:05,840 --> 00:02:08,317
echoes of the post-war building boom.

43
00:02:08,400 --> 00:02:12,637
Similarities that are actually watch
points for us. I want to tell you about

44
00:02:12,720 --> 00:02:14,717
these watch points and then we'll talk

45
00:02:14,800 --> 00:02:17,437
about how we can navigate them together.

46
00:02:17,520 --> 00:02:21,037
First, LLMs are really good at telling

47
00:02:21,120 --> 00:02:23,677
you the most probable answer, which

48
00:02:23,760 --> 00:02:25,677
essentially means that they're able to

49
00:02:25,760 --> 00:02:28,157
tell you what has been or is popular

50
00:02:28,240 --> 00:02:31,597
now, what has worked, what was in style.

51
00:02:31,680 --> 00:02:33,597
They're less good at telling you what's

52
00:02:33,680 --> 00:02:36,317
original or specific to you, your brand,

53
00:02:36,400 --> 00:02:41,037
and your users context. I'll give you
an example. Can you guess what this brand

54
00:02:41,120 --> 00:02:42,637
sells?

55
00:02:42,720 --> 00:02:47,837
Software? No. Korean barbecue.

56
00:02:47,920 --> 00:02:50,637
Now this is I'm sure a very fine website

57
00:02:50,720 --> 00:02:53,197
but it has no character no context for

58
00:02:53,280 --> 00:02:55,597
the user or the context of the use which

59
00:02:55,680 --> 00:02:57,677
brings me to my second point than

60
00:02:57,760 --> 00:03:01,117
temptation of done. AI makes things feel

61
00:03:01,200 --> 00:03:04,637
finished really really fast all too

62
00:03:04,720 --> 00:03:07,117
often prematurely. I'll give you an

63
00:03:07,200 --> 00:03:09,357
example that I probably shouldn't admit.

64
00:03:09,440 --> 00:03:12,157
I have a microwave burrito for lunch all

65
00:03:12,240 --> 00:03:16,477
too often. I take the cold, hard rock
out of the freezer, plop it on the

66
00:03:16,560 --> 00:03:19,037
plate, put it in the microwave. In 90

67
00:03:19,120 --> 00:03:21,757
seconds, I go from hungry to lunch. I'm

68
00:03:21,840 --> 00:03:24,077
willing to overlook some pretty serious

69
00:03:24,160 --> 00:03:27,997
flaws. It's nearly inedible, but I'm so

70
00:03:28,080 --> 00:03:31,037
enamored by the speed of execution.

71
00:03:31,120 --> 00:03:34,477
The same happens with AI. You type in a

72
00:03:34,560 --> 00:03:37,117
sentence and boom, you have an

73
00:03:37,200 --> 00:03:38,957
interface. There's some nice little

74
00:03:39,040 --> 00:03:41,837
corners and maybe some drop shadows and

75
00:03:41,920 --> 00:03:44,957
it's good to be done, but an apparently

76
00:03:45,040 --> 00:03:47,917
polished state can often be misleading.

77
00:03:48,000 --> 00:03:52,477
Does it really solve the problem?
Does it really differentiate? All the

78
00:03:52,560 --> 00:03:54,637
questions Claire Vu brought up earlier

79
00:03:54,720 --> 00:03:57,997
apply here. And the third watch point,

80
00:03:58,080 --> 00:04:00,877
this work is so easy to do, so quick

81
00:04:00,960 --> 00:04:03,997
that it often feels disposable and worst

82
00:04:04,080 --> 00:04:06,877
is treated that way. We sometimes let

83
00:04:06,960 --> 00:04:11,037
the responsibility of our decisions fall
by the wayside and don't really think

84
00:04:11,120 --> 00:04:15,437
about the long term, like who's
maintaining this. Anyways, all three of

85
00:04:15,520 --> 00:04:17,437
these watch points can come together to

86
00:04:17,520 --> 00:04:19,757
a world that looks a little like this,

87
00:04:19,840 --> 00:04:22,557
an actual place in Turkey. If we're not

88
00:04:22,640 --> 00:04:27,357
careful, our building boom could end
up a little too much like the post-war

89
00:04:27,440 --> 00:04:29,517
building boom. The proliferation of

90
00:04:29,600 --> 00:04:32,157
patterns ills suited to the context at

91
00:04:32,240 --> 00:04:36,877
hand. essentially zombie UI. Zombie UI

92
00:04:36,960 --> 00:04:39,917
that's monotonous, vacant, or uncared

93
00:04:40,000 --> 00:04:42,477
for. We spend half our waking lives

94
00:04:42,560 --> 00:04:45,437
looking at screens. We want software

95
00:04:45,520 --> 00:04:48,797
that feels actually cared for. And that

96
00:04:48,880 --> 00:04:53,677
means has a little personality
as something as simple as the Grockbot and

97
00:04:53,760 --> 00:04:55,357
the little animation and life that it

98
00:04:55,440 --> 00:04:58,477
brings. Or even these tiny details like

99
00:04:58,560 --> 00:05:02,477
actually having the date in the tab
on the calendar. You probably clicked by

100
00:05:02,560 --> 00:05:05,117
this all day long, but that shows the

101
00:05:05,200 --> 00:05:07,997
builder cared about you. Or the

102
00:05:08,080 --> 00:05:10,637
understanding of context like the link

103
00:05:10,720 --> 00:05:13,997
agent wallet. It anticipates the issues

104
00:05:14,080 --> 00:05:16,397
that the agent could have while buying

105
00:05:16,480 --> 00:05:18,797
online and has the troubleshooting built

106
00:05:18,880 --> 00:05:21,597
in because they anticipate the context

107
00:05:21,680 --> 00:05:24,397
of the user even when it's an agent.

108
00:05:24,480 --> 00:05:28,317
These examples show builders that are
care about their users and bring that

109
00:05:28,400 --> 00:05:30,317
care into the work. It shows the

110
00:05:30,400 --> 00:05:33,837
character of the brand and it has soul.

111
00:05:33,920 --> 00:05:36,317
We can do this with AI even though it

112
00:05:36,400 --> 00:05:38,797
can sometimes be the culprit. And I have

113
00:05:38,880 --> 00:05:41,357
four recommendations on how we can use

114
00:05:41,440 --> 00:05:44,877
AI to build products with care and soul.

115
00:05:44,960 --> 00:05:48,477
First one, you have to have a point of

116
00:05:48,560 --> 00:05:51,677
view. If you don't, AI will give it for

117
00:05:51,760 --> 00:05:56,077
you. And as we've already talked about,
that's very likely to be generic and

118
00:05:56,160 --> 00:05:58,157
looking at the past. You need to define

119
00:05:58,240 --> 00:06:00,957
your brand. What is it for you? What is

120
00:06:01,040 --> 00:06:03,117
it for your users? Who do you want to be

121
00:06:03,200 --> 00:06:05,677
to them? And what do they care about?

122
00:06:05,760 --> 00:06:08,477
This forms the basis of your standards.

123
00:06:08,560 --> 00:06:10,797
And I'd say this in any era, but it's

124
00:06:10,880 --> 00:06:13,357
all the more important right now. When

125
00:06:13,440 --> 00:06:18,237
building is distributed, ownership
is diffuse. We're all contributing to the

126
00:06:18,320 --> 00:06:19,917
products in more ways than we've been in

127
00:06:20,000 --> 00:06:22,797
the past. It's too easy to abdicate to

128
00:06:22,880 --> 00:06:26,077
AI. So, for example, at Stripe, we care

129
00:06:26,160 --> 00:06:29,037
a lot about optimism. So we put that

130
00:06:29,120 --> 00:06:32,397
into all the details big and small. The

131
00:06:32,480 --> 00:06:35,677
colors we choose, the way we write, what

132
00:06:35,760 --> 00:06:38,237
we write about, and the products that we

133
00:06:38,320 --> 00:06:41,117
ship to try to support entrepreneurs.

134
00:06:41,200 --> 00:06:43,757
Now, these details are so important to

135
00:06:43,840 --> 00:06:48,477
be aligned on because everybody
is contributing and so much is changing.

136
00:06:48,560 --> 00:06:52,237
And I'll give you an example of where
that friction can show up if there isn't

137
00:06:52,320 --> 00:06:56,477
alignment. We were recently in a design
crit looking at an advertisement. So,

138
00:06:56,560 --> 00:06:58,397
this is an ad we were working on that

139
00:06:58,480 --> 00:07:03,997
shows our brand, the parallelogram,
with our users visuals. And as we were

140
00:07:04,080 --> 00:07:06,637
looking at this, we're like, "All right,
cool. You kind of get what's happening,

141
00:07:06,720 --> 00:07:11,037
but some things are off." And in the room
was a cross functional partner that

142
00:07:11,120 --> 00:07:14,317
was looking to move fast and get
something shipped. Sure, you can't

143
00:07:14,400 --> 00:07:19,277
relate. And we had a discussion because
they brought up a really interesting

144
00:07:19,360 --> 00:07:24,877
question. What's the quality standard
for something made with AI? What a curious

145
00:07:24,960 --> 00:07:28,877
question. Why should
that matter how it was made?

146
00:07:28,960 --> 00:07:32,557
It doesn't matter to the users how it
was made. It matters to them if it's

147
00:07:32,640 --> 00:07:34,957
good or not. And so that is the basis of

148
00:07:35,040 --> 00:07:37,597
our standards. It's the output at the

149
00:07:37,680 --> 00:07:42,877
end of the day that matters and they
care if it's good and so should we. So

150
00:07:42,960 --> 00:07:45,837
through that discussion, we've got
alignment on what really matters here

151
00:07:45,920 --> 00:07:49,517
and what we're striving for and then
walked away with 17 bullet points of

152
00:07:49,600 --> 00:07:52,157
improvements to be made. the frost a

153
00:07:52,240 --> 00:07:54,397
little bit here and there, softer edges,

154
00:07:54,480 --> 00:07:57,837
a bit smaller bubblers. And now we use

155
00:07:57,920 --> 00:08:00,477
this term as a verb to mean meticulous

156
00:08:00,560 --> 00:08:04,957
craft around the office called Pepsi
bubbling. Now, we're not shooting for

157
00:08:05,040 --> 00:08:08,477
perfection. We're shooting to go one
level deeper than what your customer

158
00:08:08,560 --> 00:08:10,877
could see. That meticulous craft will

159
00:08:10,960 --> 00:08:14,077
show up for them. And this point of view

160
00:08:14,160 --> 00:08:17,917
really drives the standards and drives
the work, especially when there's agents

161
00:08:18,000 --> 00:08:19,917
involved. And so much more is happening

162
00:08:20,000 --> 00:08:22,717
on a daily basis. So how do you develop

163
00:08:22,800 --> 00:08:25,357
it? Frankly, it's all about getting

164
00:08:25,440 --> 00:08:28,237
really good at noticing. Notice what

165
00:08:28,320 --> 00:08:30,557
your users need and what they want, not

166
00:08:30,640 --> 00:08:33,677
just what they say. Notice what about

167
00:08:33,760 --> 00:08:36,397
the world around you signals good and

168
00:08:36,480 --> 00:08:39,917
great and meh. really take note of these

169
00:08:40,000 --> 00:08:41,917
things in the products that you use, but

170
00:08:42,000 --> 00:08:44,637
also in analogous situations so you can

171
00:08:44,720 --> 00:08:49,117
be inspired by art, science, and a
broader point of view in the work that

172
00:08:49,200 --> 00:08:52,397
you create. Building your sense of taste
and your understanding of the world

173
00:08:52,480 --> 00:08:54,477
around you and what's good and great can

174
00:08:54,560 --> 00:08:57,277
really improve your own standards that

175
00:08:57,360 --> 00:09:00,397
you will then need to scale. Which

176
00:09:00,480 --> 00:09:02,317
brings me to my second recommendation,

177
00:09:02,400 --> 00:09:05,357
which is encode your standards into the

178
00:09:05,440 --> 00:09:10,237
machine. It would be a wonderful
world if every human had the shared

179
00:09:10,320 --> 00:09:14,077
understandings and could make the same
decision. But the reality is is humans

180
00:09:14,160 --> 00:09:18,077
are not going to be a part of all
of these decisions going forward. We are

181
00:09:18,160 --> 00:09:22,557
seeing interfaces built without
a designer in the room. We're seeing

182
00:09:22,640 --> 00:09:24,797
agents finding problems and fixing them

183
00:09:24,880 --> 00:09:27,277
while we sleep. And we're certainly

184
00:09:27,360 --> 00:09:30,237
seeing generative UI built in real time

185
00:09:30,320 --> 00:09:33,997
for users. This is why design systems

186
00:09:34,080 --> 00:09:36,317
are having a moment again. And the

187
00:09:36,400 --> 00:09:38,557
system of yesterday is different than

188
00:09:38,640 --> 00:09:43,357
the system of today. In the past,
you didn't have to write everything down

189
00:09:43,440 --> 00:09:47,917
because there was always a designer
in the room to fill in the blanks. Now, we

190
00:09:48,000 --> 00:09:50,237
must enable distributed building and

191
00:09:50,320 --> 00:09:53,677
agentic construction and the decisions

192
00:09:53,760 --> 00:09:56,957
need to be easier to define. So, the

193
00:09:57,040 --> 00:09:59,277
object of design is no longer the

194
00:09:59,360 --> 00:10:02,637
screen. It is the system itself. Now,

195
00:10:02,720 --> 00:10:05,277
everybody knows Gutenberg created the

196
00:10:05,360 --> 00:10:07,997
first printing press. The lesserk known

197
00:10:08,080 --> 00:10:10,877
part of the story is the system he built

198
00:10:10,960 --> 00:10:14,717
around it. He didn't create just 26

199
00:10:14,800 --> 00:10:18,077
lowercase and 26 uppercase letters. He

200
00:10:18,160 --> 00:10:22,397
created 290 unique characters, different

201
00:10:22,480 --> 00:10:26,317
widths, abbreviations, liatures. And the

202
00:10:26,400 --> 00:10:30,397
reason that he did this is because when
he type set it and wanted it to be fully

203
00:10:30,480 --> 00:10:34,477
justified, he wanted to make sure that
there wouldn't be the weird rivers and

204
00:10:34,560 --> 00:10:38,877
lakes of whites space that erode
the beauty of the final product.

205
00:10:38,960 --> 00:10:43,117
Essentially, he was making it both
extensible and opinionated enough that

206
00:10:43,200 --> 00:10:45,517
it could feel as good as handmade,

207
00:10:45,600 --> 00:10:47,757
although it was machine-made. This is

208
00:10:47,840 --> 00:10:50,317
what we should be aiming for as well.

209
00:10:50,400 --> 00:10:54,157
Now, the old system scaled consistency,

210
00:10:54,240 --> 00:10:58,477
but the new system needs to scale intent.
We've been working on this at

211
00:10:58,560 --> 00:11:01,997
Stripe and, you know, have hit a few
snags along the way. And so, we're we're

212
00:11:02,080 --> 00:11:03,997
figuring it out. But one of the things

213
00:11:04,080 --> 00:11:05,837
we want to do is we want to empower

214
00:11:05,920 --> 00:11:09,517
builders to go from prompt to production

215
00:11:09,600 --> 00:11:12,077
nearly instantaneously. Like many, we

216
00:11:12,160 --> 00:11:16,797
started with an MCP that understood
our design documentation, but the results

217
00:11:16,880 --> 00:11:21,197
were not great. it wasn't specific
enough and it wasn't driving the right

218
00:11:21,280 --> 00:11:23,677
outcomes. Three different people could

219
00:11:23,760 --> 00:11:25,917
put in the same prompt and get three

220
00:11:26,000 --> 00:11:28,637
different results. So since then we've

221
00:11:28,720 --> 00:11:33,517
evolved and now we have created a CLI
built on our design system. It makes a

222
00:11:33,600 --> 00:11:37,597
harness that makes the AI far more
obedient. Now it's where the builders

223
00:11:37,680 --> 00:11:40,237
are building and it consumes the

224
00:11:40,320 --> 00:11:42,477
documentation at the right time and

225
00:11:42,560 --> 00:11:47,277
place to avoid context rot. Now,
importantly, the big difference between

226
00:11:47,360 --> 00:11:51,757
this version and a previous design
system is it's not just components and

227
00:11:51,840 --> 00:11:54,077
atomic parts, but it's actually full

228
00:11:54,160 --> 00:11:58,877
templates and flows. So, the system
actually knows what our product is

229
00:11:58,960 --> 00:12:03,437
supposed to behave like and how it all
comes together. Essentially, it's far

230
00:12:03,520 --> 00:12:07,837
more opinionated than it's been in the
past. We embed these standards into the

231
00:12:07,920 --> 00:12:10,317
means of production, which enables a

232
00:12:10,400 --> 00:12:13,597
more coherent product. But this is just

233
00:12:13,680 --> 00:12:16,397
the baseline. Christopher Alexander

234
00:12:16,480 --> 00:12:19,357
said, "A system can satisfy every rule

235
00:12:19,440 --> 00:12:22,557
and still be dead." This brings me to my

236
00:12:22,640 --> 00:12:25,597
third recommendation.

237
00:12:25,680 --> 00:12:29,917
Refuse to confuse done with good.

238
00:12:30,000 --> 00:12:33,277
The filter is gone. It used to be.

239
00:12:33,360 --> 00:12:36,717
Quality filter was essentially
built into every stage of the product

240
00:12:36,800 --> 00:12:39,437
development process, even before a

241
00:12:39,520 --> 00:12:42,877
project started. We've got 20 ideas and

242
00:12:42,960 --> 00:12:46,637
we can only staff one. And then we poked

243
00:12:46,720 --> 00:12:51,437
and we prodded and we pruned along
the way and products grew somewhat

244
00:12:51,520 --> 00:12:53,437
methodically.

245
00:12:53,520 --> 00:12:56,237
Well, now we can build 20 ideas in a

246
00:12:56,320 --> 00:12:59,677
week. This is awesome in many ways and

247
00:12:59,760 --> 00:13:03,997
we can finally skip theoretical meetings
talking about hypothetical products and

248
00:13:04,080 --> 00:13:06,477
react to the real thing in our hands.

249
00:13:06,560 --> 00:13:08,637
But the filtering that used to be

250
00:13:08,720 --> 00:13:10,797
throughout the process now needs to

251
00:13:10,880 --> 00:13:13,997
happen post build when it is a lot

252
00:13:14,080 --> 00:13:16,957
harder to say no. This is where the role

253
00:13:17,040 --> 00:13:19,437
of an editor comes in and it is ultra

254
00:13:19,520 --> 00:13:24,237
important in this day and age. Who
is doing this in your organization? Who's

255
00:13:24,320 --> 00:13:27,117
looking at the end to end
and understanding whether or not we're

256
00:13:27,200 --> 00:13:29,597
actually building a whole? It's a new

257
00:13:29,680 --> 00:13:33,357
behavior. We have to unlearn that built

258
00:13:33,440 --> 00:13:37,357
means done and that done means good. The

259
00:13:37,440 --> 00:13:41,197
most important thing I can say
to anybody working on their editing skills

260
00:13:41,280 --> 00:13:44,317
is you have to experience it like a user

261
00:13:44,400 --> 00:13:47,277
would. Does it actually solve the

262
00:13:47,360 --> 00:13:49,837
problem? Is it actually attuned to the

263
00:13:49,920 --> 00:13:52,637
way the user thinks about things? Is it

264
00:13:52,720 --> 00:13:56,957
actually coherent? A lot of things look
good in isolation, but then when you

265
00:13:57,040 --> 00:13:58,717
pull it to all together into the user

266
00:13:58,800 --> 00:14:01,917
journey, it feels a little disconnected

267
00:14:02,000 --> 00:14:06,717
and it's not just about saying yes or no,
this is good or it shouldn't go.

268
00:14:06,800 --> 00:14:08,797
It's actually about is it even fully

269
00:14:08,880 --> 00:14:10,717
formed? How can we push this to

270
00:14:10,800 --> 00:14:13,757
completion? We don't want done to be the

271
00:14:13,840 --> 00:14:16,637
enemy of good. A very insightful article

272
00:14:16,720 --> 00:14:19,757
by Nabil Koreshi talks about what makes

273
00:14:19,840 --> 00:14:24,957
art great. And he says it's about
the unexpected details, those little

274
00:14:25,040 --> 00:14:28,077
surprising things you're like, "Wow,
I can't believe they thought of that." And

275
00:14:28,160 --> 00:14:32,557
then the deeper meaning that sits behind
the surface or the continuous themes

276
00:14:32,640 --> 00:14:34,477
that tie it all together to a whole.

277
00:14:34,560 --> 00:14:36,397
These are the things that AI is not

278
00:14:36,480 --> 00:14:39,917
great at. But knowing what the gaps are

279
00:14:40,000 --> 00:14:43,037
in AI help us better navigate that and

280
00:14:43,120 --> 00:14:47,037
fill these gaps in. As Nibil says, one

281
00:14:47,120 --> 00:14:49,197
of the things that so offends us about

282
00:14:49,280 --> 00:14:51,997
AI slop is the sense that the details

283
00:14:52,080 --> 00:14:55,837
don't matter. The cup is green but may

284
00:14:55,920 --> 00:14:58,797
as well have been blue. An editor takes

285
00:14:58,880 --> 00:15:01,677
accountability for every decision, which

286
00:15:01,760 --> 00:15:05,277
is in many ways every pixel. I'll give

287
00:15:05,360 --> 00:15:08,237
you an example from this event. So my

288
00:15:08,320 --> 00:15:10,317
team had the pleasure of designing the

289
00:15:10,400 --> 00:15:12,877
event. Stefan worked on this fabulous

290
00:15:12,960 --> 00:15:15,357
opening animation.

291
00:15:15,440 --> 00:15:17,917
He and the team started with a 3D model

292
00:15:18,000 --> 00:15:22,717
of the scene and then they brought
it into AI to help with the different

293
00:15:22,800 --> 00:15:24,317
animations.

294
00:15:24,400 --> 00:15:27,597
Now here's the first version.

295
00:15:27,680 --> 00:15:31,357
It's cool. It's, you know, fun to see
the different little parts of it and

296
00:15:31,440 --> 00:15:35,837
it's nice the way it moves. But if
you scrutinate, scrutinize it pixel by

297
00:15:35,920 --> 00:15:37,837
pixel, something feels off, right?

298
00:15:37,920 --> 00:15:39,117
That's not quite the way it should move.

299
00:15:39,200 --> 00:15:40,717
It's a little jilted. You kind of want

300
00:15:40,800 --> 00:15:43,757
to get more of the scene, right? So he

301
00:15:43,840 --> 00:15:46,157
did it again

302
00:15:46,240 --> 00:15:48,317
and again

303
00:15:48,400 --> 00:15:52,557
and again. 56 times,

304
00:15:52,640 --> 00:15:55,597
56 iterations later, he got to something

305
00:15:55,680 --> 00:15:58,237
truly beautiful. Now, it's got more

306
00:15:58,320 --> 00:16:00,477
realism. It's got a little bit of life

307
00:16:00,560 --> 00:16:03,117
and it's subtle differences, but you can

308
00:16:03,200 --> 00:16:05,517
sense the care.

309
00:16:05,600 --> 00:16:07,997
Now, what one of the things that so uh

310
00:16:08,080 --> 00:16:10,877
impresses me about this is that if he

311
00:16:10,960 --> 00:16:13,357
didn't use AI, he may not have built

312
00:16:13,440 --> 00:16:18,397
such a complex scene or he may not
have tried so many different viewpoints.

313
00:16:18,480 --> 00:16:20,317
And while it wasn't one and done and

314
00:16:20,400 --> 00:16:22,077
certainly took a lot of wherewithal and

315
00:16:22,160 --> 00:16:25,437
scrutiny, AI opened the possibility

316
00:16:25,520 --> 00:16:28,557
space. Now, this brings me to my last

317
00:16:28,640 --> 00:16:30,797
and definitely my favorite fourth

318
00:16:30,880 --> 00:16:32,717
recommendation.

319
00:16:32,800 --> 00:16:36,957
Unleash creativity and artistry. Yes, AI

320
00:16:37,040 --> 00:16:39,837
can help us manufacture monotony, but it

321
00:16:39,920 --> 00:16:44,237
is also the greatest creative
catalyst we've ever had.

322
00:16:44,320 --> 00:16:47,997
This matters more now because when
everybody in their mother is building

323
00:16:48,080 --> 00:16:50,557
something, it is going to be harder and

324
00:16:50,640 --> 00:16:53,677
more important to differentiate.

325
00:16:53,760 --> 00:16:57,437
and the interfaces of today, chats,

326
00:16:57,520 --> 00:17:02,077
charts, CLIs. No way is that the epitome

327
00:17:02,160 --> 00:17:05,197
of great interactions in the modern era.

328
00:17:05,280 --> 00:17:07,037
There is so much more we can do and so

329
00:17:07,120 --> 00:17:09,437
much more we should do to make things

330
00:17:09,520 --> 00:17:13,357
feel generative, alive, responsive,

331
00:17:13,440 --> 00:17:16,157
dynamic. We can literally talk to the

332
00:17:16,240 --> 00:17:19,197
computers. So, let's branch out. It is

333
00:17:19,280 --> 00:17:22,237
time to invent new interfaces. It is

334
00:17:22,320 --> 00:17:25,117
time to invent new aesthetics. The best

335
00:17:25,200 --> 00:17:30,237
practices have not been written yet. We
get to do that and now we have the tools

336
00:17:30,320 --> 00:17:34,477
to do it. It's just like when
multi-touch made it possible to do

337
00:17:34,560 --> 00:17:36,397
wholly new interactions or the

338
00:17:36,480 --> 00:17:38,717
synthesizer allowed us to create totally

339
00:17:38,800 --> 00:17:41,917
new sounds. AI is allowing all sorts of

340
00:17:42,000 --> 00:17:44,237
new creativity. And we're seeing so much

341
00:17:44,320 --> 00:17:46,797
of this online. We're even seeing folks

342
00:17:46,880 --> 00:17:49,117
showing up the Stripe design team with

343
00:17:49,200 --> 00:17:51,597
way more interesting data viz. And we're

344
00:17:51,680 --> 00:17:53,917
seeing websites for restaurants with

345
00:17:54,000 --> 00:17:56,877
character and personality and people

346
00:17:56,960 --> 00:17:59,677
using AI to paint and create art

347
00:17:59,760 --> 00:18:02,717
themselves. It is a really interesting

348
00:18:02,800 --> 00:18:05,837
time. And at Stripe, we're using AI to

349
00:18:05,920 --> 00:18:08,717
essentially amplify the abilities of the

350
00:18:08,800 --> 00:18:11,517
creative team. Our more most recent

351
00:18:11,600 --> 00:18:14,557
cover for Built to Grow, it was created

352
00:18:14,640 --> 00:18:17,917
by a human marbler who worked on these

353
00:18:18,000 --> 00:18:21,597
stunning iterations. And then we used AI

354
00:18:21,680 --> 00:18:24,237
to fine-tune the details ever further to

355
00:18:24,320 --> 00:18:26,157
ensure we had the colors and the lines

356
00:18:26,240 --> 00:18:29,117
in all the right places.

357
00:18:29,200 --> 00:18:33,517
It was basically taking the good
judgment of the humans and helping us

358
00:18:33,600 --> 00:18:38,157
scale it to make something truly stunning.
Now, to make AI a stronger

359
00:18:38,240 --> 00:18:39,757
creative partner, there's a couple of

360
00:18:39,840 --> 00:18:43,197
things I recommend. One, improve your

361
00:18:43,280 --> 00:18:46,157
inputs. You want more specity going in

362
00:18:46,240 --> 00:18:49,037
the direction of your brand, your unique

363
00:18:49,120 --> 00:18:51,997
interests. So, put specific prompts.

364
00:18:52,080 --> 00:18:56,237
Don't just say, "Hey, I need a website
for my Korean barbecue, but this is what

365
00:18:56,320 --> 00:19:01,437
I believe in. This is what good is. This
is what we care about." Add your source

366
00:19:01,520 --> 00:19:03,117
material that you're using to build your

367
00:19:03,200 --> 00:19:05,837
own standards with. Help make it think

368
00:19:05,920 --> 00:19:09,357
in different ways. And then stress your

369
00:19:09,440 --> 00:19:11,757
outputs. Don't get tempted by the

370
00:19:11,840 --> 00:19:14,557
burrito dilemma. always push a step

371
00:19:14,640 --> 00:19:17,277
further and then of course use

372
00:19:17,360 --> 00:19:21,917
adversarial agents to help you critique
it and bang it up a little bit, but you

373
00:19:22,000 --> 00:19:26,717
yourself should always be pushing for
better. Now, these are just tactics. The

374
00:19:26,800 --> 00:19:30,237
much harder thing is definitely cultural.

375
00:19:30,320 --> 00:19:32,717
It is easy to follow cookie cutter

376
00:19:32,800 --> 00:19:36,797
patterns. It's safe and cozy, but it is

377
00:19:36,880 --> 00:19:39,757
much more impressive and much harder to

378
00:19:39,840 --> 00:19:42,477
find something unique that improves the

379
00:19:42,560 --> 00:19:45,117
status quo. So, if you're leading a

380
00:19:45,200 --> 00:19:47,917
team, don't just tell them to use AI,

381
00:19:48,000 --> 00:19:50,637
but give them room to explore.

382
00:19:50,720 --> 00:19:54,477
Protect the strange. AI lowers the cost

383
00:19:54,560 --> 00:19:58,957
to create. Let's spend some of that
savings on making something truly

384
00:19:59,040 --> 00:20:03,197
special. If we only use
AI to make the things

385
00:20:03,280 --> 00:20:07,597
that we already make just faster,
then we are definitely missing out on the

386
00:20:07,680 --> 00:20:10,157
most interesting part.

387
00:20:10,240 --> 00:20:13,117
AI can help us raise the ceiling, not

388
00:20:13,200 --> 00:20:16,077
just the floor. The most important thing

389
00:20:16,160 --> 00:20:19,597
is to be very intentional.

390
00:20:19,680 --> 00:20:22,957
It's your point of view. It's your

391
00:20:23,040 --> 00:20:27,437
system. It's your quality bar and your

392
00:20:27,520 --> 00:20:29,117
ambition

393
00:20:29,200 --> 00:20:30,957
that you will want to bring to life with

394
00:20:31,040 --> 00:20:34,477
AI. In total contrast to the post-war

395
00:20:34,560 --> 00:20:37,197
building boom and modernism design was

396
00:20:37,280 --> 00:20:39,517
Gothic architecture.

397
00:20:39,600 --> 00:20:43,117
In 1850, John Ruskin wrote a lot about

398
00:20:43,200 --> 00:20:45,997
quality and craft and highlighted Gothic

399
00:20:46,080 --> 00:20:50,077
architecture as the epitome of great. He

400
00:20:50,160 --> 00:20:52,237
noticed that no Gothic building was

401
00:20:52,320 --> 00:20:55,117
alike. Frankly, not even one column was

402
00:20:55,200 --> 00:20:58,237
alike the other. Each detail was

403
00:20:58,320 --> 00:21:00,717
uniquely crafted and showed the unique

404
00:21:00,800 --> 00:21:04,477
hand and mind of the maker behind it. It

405
00:21:04,560 --> 00:21:07,517
felt truly cared for.

406
00:21:07,600 --> 00:21:10,717
This is what our users want. They're not

407
00:21:10,800 --> 00:21:13,037
impressed if we animated something with

408
00:21:13,120 --> 00:21:16,397
3JS and Blender in 30 minutes.

409
00:21:16,480 --> 00:21:18,797
They are impressed by us solving their

410
00:21:18,880 --> 00:21:21,997
problems and clever touches in the

411
00:21:22,080 --> 00:21:26,637
details that show we anticipated
their needs and that our brand has some

412
00:21:26,720 --> 00:21:31,117
character behind it. We have the choice
in this building boom to not make the

413
00:21:31,200 --> 00:21:33,597
digital equivalent of zombie buildings.

414
00:21:33,680 --> 00:21:36,877
We can make this a creative renaissance.

415
00:21:36,960 --> 00:21:38,557
So, let's make some products that are

416
00:21:38,640 --> 00:21:41,677
more powerful and show the hand and care

417
00:21:41,760 --> 00:21:44,687
of the maker. Thanks everybody.

418
00:21:44,770 --> 00:21:46,790
[applause]