1
00:00:00,060 --> 00:00:03,807
Rita: People want, like, very clear, sometimes rigid rules about,

2
00:00:04,008 --> 00:00:08,466
like, tell me I'm allowed to bl- deploy X but not Y. And it's

3
00:00:08,466 --> 00:00:11,940
like, well, if it was that simple, then an agent could do it.

4
00:00:11,960 --> 00:00:15,833
The thing that I'm asking you to do is exercise human judgment and come back

5
00:00:15,833 --> 00:00:21,343
to me and tell me like, "Okay, if we released this thing on a Friday afternoon,

6
00:00:21,945 --> 00:00:25,995
could I explain to a customer w- what happened if it caused an incident?"

7
00:00:30,898 --> 00:00:34,184
Corey: Ladies and gentlemen, please welcome to the Frequency Podcast

8
00:00:34,195 --> 00:00:39,920
stage, Dane Connect and Rita Kozlov, joined by your host, Corey Quinn.

9
00:00:40,405 --> 00:00:42,315
No, thank you both for agreeing to do this.

10
00:00:42,346 --> 00:00:45,086
I'm pleasantly surprised that it required

11
00:00:45,086 --> 00:00:46,967
less arm twisting than I would have thought.

12
00:00:47,044 --> 00:00:50,580
Um, a question I love to ask people at conferences every chance

13
00:00:50,581 --> 00:00:53,623
I get, it- it's perfectly calibrated for just how much of a jerk

14
00:00:53,638 --> 00:00:56,250
I want to be to someone based upon the scale of their company.

15
00:00:56,421 --> 00:00:59,518
You ask someone where they work, and they give you an answer, and what-

16
00:00:59,518 --> 00:01:02,855
no matter what that answer is, I like to respond with, "Oh, great, what

17
00:01:02,855 --> 00:01:06,261
do you folks do?" So small, scrappy startup, they're thrilled to tell you.

18
00:01:06,270 --> 00:01:08,994
Large, giant bank, they get offended by this.

19
00:01:08,994 --> 00:01:10,719
And, like, the final form is some of the modern

20
00:01:10,731 --> 00:01:12,404
tech companies 'cause they don't know either.

21
00:01:12,596 --> 00:01:14,426
So we're gonna see where you all fall on that.

22
00:01:14,628 --> 00:01:16,169
Cloudflare, what do you folks do?

23
00:01:16,794 --> 00:01:19,351
Dane: We, we help, uh, protect the internet at, at, at, at our core,

24
00:01:19,526 --> 00:01:22,073
build a better internet and, you know, we move packets faster than

25
00:01:22,076 --> 00:01:25,579
anyone else around the world and, uh, uh, sprinkle some compute on that

26
00:01:25,580 --> 00:01:29,096
and, uh, some inference and, uh, uh, let people build amazing things.

27
00:01:29,672 --> 00:01:31,973
Corey: Would you agree, or are we gonna start a fistfight here?

28
00:01:32,270 --> 00:01:35,261
That's my boss, so… Oh, I- you kind of have to at that point.

29
00:01:35,279 --> 00:01:35,905
No, I get it.

30
00:01:35,917 --> 00:01:36,425
I get it.

31
00:01:36,985 --> 00:01:40,509
This episode is sponsored in part by my day job, Duckbill.

32
00:01:40,667 --> 00:01:43,286
Do you have a horrifying AWS bill?

33
00:01:43,550 --> 00:01:47,515
That can mean a lot of things, predicting what it's going to be,

34
00:01:47,715 --> 00:01:51,777
determining what it should be, negotiating your next long-term

35
00:01:51,780 --> 00:01:56,185
contract with AWS, or just figuring out why it increasingly

36
00:01:56,185 --> 00:02:00,719
resembles a phone number, but nobody seems to quite know why that is.

37
00:02:01,020 --> 00:02:04,595
To learn more, visit duckbillhq.com.

38
00:02:04,887 --> 00:02:08,906
Remember, you can't duck the Duckbill bill, which my

39
00:02:08,912 --> 00:02:13,185
CEO reliably informs me is absolutely not our slogan.

40
00:02:13,668 --> 00:02:18,970
I had Matthew Prince on this show twice in, uh, previous years, and at

41
00:02:18,971 --> 00:02:23,765
the time it really did feel like Cloudflare was a CDN, but no one wants

42
00:02:23,765 --> 00:02:27,293
to talk about a CDN because that's only interesting to networking nerds.

43
00:02:27,442 --> 00:02:27,823
Guilty.

44
00:02:28,319 --> 00:02:31,008
Uh, but then there was the other stuff you did sort of sprinkled on top,

45
00:02:31,170 --> 00:02:34,737
and, and I'm not sure when that happened, but these days talking about

46
00:02:34,737 --> 00:02:38,833
you as a cloud provider, specifically, I would argue, the premier cloud

47
00:02:38,834 --> 00:02:43,343
provider that is legible to agents, stopped being something, I guess,

48
00:02:43,366 --> 00:02:47,179
aspirational and started being an accurate description of the reality.

49
00:02:47,660 --> 00:02:49,260
It, it felt like it was a long time coming,

50
00:02:49,260 --> 00:02:51,047
but almost without noticing, you got there.

51
00:02:51,537 --> 00:02:53,117
Rita: I think the fists would've come out

52
00:02:53,126 --> 00:02:55,843
even 10 years ago if you called this a CDN.

53
00:02:55,966 --> 00:02:59,730
Um, that, that was a term that we never really embraced.

54
00:02:59,875 --> 00:03:02,430
Dane: 10 years ago is actually when the developer platform started.

55
00:03:03,144 --> 00:03:05,953
It really started as an internal platform, uh, for, uh,

56
00:03:06,106 --> 00:03:09,052
how, how we could accelerate our own, uh, development.

57
00:03:09,475 --> 00:03:12,266
Uh, we like to ship a lot of things, as I'm sure most people know.

58
00:03:12,673 --> 00:03:15,232
Uh, and, you know, in, in order to make it, uh, cost-effective,

59
00:03:15,232 --> 00:03:17,566
you had to have a way of, uh, uh, letting engineers

60
00:03:17,599 --> 00:03:20,512
iterate quickly and reduce, reduce that, that cost.

61
00:03:20,864 --> 00:03:22,341
Um, you know, it started out with just the

62
00:03:22,341 --> 00:03:24,959
one, one compute, uh, s- servers primitive.

63
00:03:25,130 --> 00:03:28,264
And actually at the time, Matthew kind of told us, uh, he said, you know, he

64
00:03:28,266 --> 00:03:32,899
said, "It takes 10 years for a platform. Uh, you know, maybe you get lucky and

65
00:03:32,899 --> 00:03:36,332
you do it in eight, but, you know, n- no one really becomes a, a platform in

66
00:03:36,332 --> 00:03:39,470
less than 10 years." And, uh, maybe that's changed with AI, but, you know, it,

67
00:03:39,470 --> 00:03:42,904
it took us the full 10 years, I think, uh, to really kind of reach that level

68
00:03:43,616 --> 00:03:48,364
Corey: It's strange in that the legacy cloud providers, for lack of

69
00:03:48,364 --> 00:03:50,572
a better term, and I'm, I'm increasingly starting to think of them

70
00:03:50,572 --> 00:03:54,462
that way, uh, everyone is saying that they are building for agents.

71
00:03:54,496 --> 00:03:56,049
That's the way of the future.

72
00:03:56,051 --> 00:03:57,685
That's the current state of the world.

73
00:03:57,938 --> 00:04:01,294
And you give an agent, a cloud code, whatever it is you're

74
00:04:01,295 --> 00:04:04,727
using, access to credentials in that environment and tell it to

75
00:04:04,729 --> 00:04:07,636
go and build a thing, and it gives the polite corporate robotic

76
00:04:07,637 --> 00:04:11,167
version of, "What the hell is this?" And trips all over itself.

77
00:04:11,179 --> 00:04:12,544
Computers, they're just like us.

78
00:04:13,116 --> 00:04:17,400
Increasingly, I have found that that is not the case in my

79
00:04:17,447 --> 00:04:21,279
weird, ridiculous experiments on top of Cloudflare specifically.

80
00:04:21,660 --> 00:04:26,386
What have you found changes when the agent is now the primary

81
00:04:26,393 --> 00:04:30,339
consumer/implementer of your platform instead of a human being?

82
00:04:30,904 --> 00:04:33,778
Rita: I mean, the legacy providers, they're

83
00:04:33,778 --> 00:04:35,735
still built on all the same primitives, right?

84
00:04:35,761 --> 00:04:38,697
And so then they just kind of change the label.

85
00:04:38,789 --> 00:04:41,443
They, they stock agent in front of it.

86
00:04:41,718 --> 00:04:43,294
Corey: But there's still, like, an instance, and then there's

87
00:04:43,294 --> 00:04:46,237
the block storage, and the key, the key defining characteristic

88
00:04:46,237 --> 00:04:48,251
is those teams are not allowed to talk to each other.

89
00:04:48,335 --> 00:04:48,979
Rita: Exactly.

90
00:04:48,979 --> 00:04:49,009
And it shows.

91
00:04:49,134 --> 00:04:52,347
At Cloudflare, you know, I, I wish I could, I wish I could

92
00:04:52,350 --> 00:04:55,668
say we 10 years ago predicted that agents would come,

93
00:04:55,676 --> 00:04:58,707
and so we built all of it f- for, for, for this moment.

94
00:04:59,085 --> 00:05:03,076
Um, but we did start out with fundamentally, first of all, very different

95
00:05:03,086 --> 00:05:06,131
architecture, which it's really hard for the clouds to back into, and

96
00:05:06,505 --> 00:05:10,193
that, that's kind of Matthew and Michelle are students of Clay Christensen.

97
00:05:10,218 --> 00:05:13,279
That's kind of the whole innovator's dilemma, right, is once you

98
00:05:13,288 --> 00:05:16,587
have this machine that's really working, that's been worked out

99
00:05:16,588 --> 00:05:19,778
around a single primitive, it's really, really hard to roll out

100
00:05:19,780 --> 00:05:22,980
something that's going to disrupt it and be 100 times more efficient.

101
00:05:23,422 --> 00:05:26,891
The foundation that we were starting from when agents were becoming

102
00:05:26,892 --> 00:05:30,622
a thing three years ago and, like, as they've evolved, was just

103
00:05:30,624 --> 00:05:33,789
very different, and that's allowed us to… I mean, one thing that

104
00:05:33,790 --> 00:05:38,455
we do very much have in our DNA is innovation, having retaining

105
00:05:38,455 --> 00:05:41,569
an ability to have agility and flexibility when things come up.

106
00:05:41,641 --> 00:05:44,739
And so I, I think it was a combination of we have the right

107
00:05:44,741 --> 00:05:50,209
foundation, but also once, you know, new, uh, protocols like MCP

108
00:05:50,217 --> 00:05:54,803
came out or we started paying attention to harnesses, like, we're

109
00:05:54,804 --> 00:05:58,445
able to adapt very quickly to whatever is happening in that moment.

110
00:05:59,080 --> 00:06:01,915
Dane: I mean, if you think about it, the whole cloud transition that

111
00:06:01,915 --> 00:06:05,705
everybody has been told to do for the last 20 years, uh, it really

112
00:06:05,706 --> 00:06:08,726
was just shifting from a data center to another data center you know,

113
00:06:08,729 --> 00:06:12,270
data center you own to a data center owned by one of a couple people.

114
00:06:12,708 --> 00:06:12,935
Um- Yeah,

115
00:06:12,936 --> 00:06:14,272
Corey: that's the lift portion of the lift and shift.

116
00:06:14,277 --> 00:06:14,421
Yeah, yeah.

117
00:06:14,421 --> 00:06:16,128
The shift, eh, we'll get to that later, as

118
00:06:16,130 --> 00:06:18,010
soon as we finish migrating off the mainframe.

119
00:06:18,298 --> 00:06:21,167
Dane: I think @S3 was a, was, was a new kind of primitive,

120
00:06:21,195 --> 00:06:23,501
uh, uh, but then it kind of regressed from there.

121
00:06:23,503 --> 00:06:25,806
Then it went back to just like, "Oh yeah, let's just give you a full machine

122
00:06:25,806 --> 00:06:29,861
and, uh," basically everything that you had, uh, in a data center before.

123
00:06:30,015 --> 00:06:32,671
Uh, never kind of going back to, uh, you know, the core

124
00:06:32,671 --> 00:06:34,590
principles of like what are you trying to accomplish?

125
00:06:34,625 --> 00:06:35,796
You're trying to accomplish efficiency.

126
00:06:36,184 --> 00:06:38,152
You know, how can you make sure that every machine that you have

127
00:06:38,152 --> 00:06:41,421
is running, you know, as close to capacity at, at a given point?

128
00:06:41,876 --> 00:06:45,457
Corey: It- it's weird in that things are speeding up incredibly, but also

129
00:06:45,460 --> 00:06:48,873
at the same time, it feels like there's a, there's a drag effect on it too.

130
00:06:48,873 --> 00:06:52,366
I remember a few weeks ago during your birthday celebration, uh, which

131
00:06:52,366 --> 00:06:55,995
is an interesting thing for a company to say, you released this CF

132
00:06:56,015 --> 00:06:59,061
command line tool, which, uh, solved between Wrangler and your other one.

133
00:06:59,063 --> 00:06:59,943
I forget what it was.

134
00:06:59,952 --> 00:07:01,385
They were constantly at odds with each other.

135
00:07:01,469 --> 00:07:02,815
Hey, a unified experience.

136
00:07:03,039 --> 00:07:05,787
And I remember giving that to my agent the first day it came out.

137
00:07:05,787 --> 00:07:07,515
Like, "Hey, use this instead.

138
00:07:07,676 --> 00:07:10,652
Cloudflare just released it." And like watching its thinking traces where, "I

139
00:07:10,655 --> 00:07:14,469
don't know, sounds like horse shit to me, but I'll give it a shot." Uh, and it

140
00:07:14,469 --> 00:07:18,500
also was trying to deploy things on top of Cloudflare pages, which is largely

141
00:07:18,501 --> 00:07:22,370
been deprecated and not the way that you folks envision these things going on.

142
00:07:22,640 --> 00:07:26,160
But getting the new way of doing things into the, into the

143
00:07:26,160 --> 00:07:28,525
weights and getting the old way, the old ways of doing them

144
00:07:28,590 --> 00:07:32,255
out when there's 10 years of historical blog posts on it being

145
00:07:32,269 --> 00:07:34,984
baked in feels like it's an incredibly difficult challenge.

146
00:07:35,859 --> 00:07:36,929
Dane: There were some arguments there.

147
00:07:36,962 --> 00:07:38,712
Uh, I, I was like, "No, it needs to be called

148
00:07:38,712 --> 00:07:40,721
Wrangler 'cause that's what the other ones know.

149
00:07:41,038 --> 00:07:44,729
Uh, uh, it needs to just be Wrangler 3." Uh, and, and they're like,

150
00:07:44,729 --> 00:07:47,515
"No, no," like, you know, s- "We should have called it CF all along.

151
00:07:47,520 --> 00:07:48,112
That's better.

152
00:07:48,433 --> 00:07:50,496
Uh, and we're gonna do it in a way that, you

153
00:07:50,496 --> 00:07:52,258
know, the agents will still just prefer it.

154
00:07:52,486 --> 00:07:55,121
It'll just be such a better experience for the agents." And

155
00:07:55,169 --> 00:07:57,592
I, I think, you know, been, it, they've been proven right.

156
00:07:57,650 --> 00:08:00,200
Like, uh, you know, first of all, CF's a lot easier to type, but, uh,

157
00:08:00,549 --> 00:08:03,655
the agents seem to naturally be able to understand it and, uh, it was

158
00:08:03,655 --> 00:08:06,842
kind of built actually more for an agent to use than a human to use.

159
00:08:07,232 --> 00:08:09,823
Corey: Oh, the, the search operator if I… Where you explain it in plain

160
00:08:09,823 --> 00:08:11,843
English the thing- Yeah … you're trying to do and it spits it out?

161
00:08:11,957 --> 00:08:12,357
Awesome.

162
00:08:12,399 --> 00:08:12,832
I love it.

163
00:08:12,840 --> 00:08:14,756
I'm still trying to figure out how I can exploit that.

164
00:08:14,758 --> 00:08:16,705
You know, write Python code for me because why not?

165
00:08:16,705 --> 00:08:18,260
Everything's a database if you hold it wrong.

166
00:08:18,931 --> 00:08:21,144
The other side of it though, I, I have to say Wrangler was

167
00:08:21,145 --> 00:08:23,601
a bad name in that you go six months without touching it

168
00:08:23,602 --> 00:08:26,383
and, "What is this Wrangler thing installed on my system?"

169
00:08:26,548 --> 00:08:28,876
CF at least you can trace down and sort of

170
00:08:28,876 --> 00:08:30,713
start to understand a bit more effectively.

171
00:08:30,917 --> 00:08:34,269
Legible to humans and legible to agents does seem to be converging.

172
00:08:34,683 --> 00:08:38,578
Rita: We hired an engineer who had just moved to Austin, and they were like,

173
00:08:39,001 --> 00:08:44,711
"Yee-haw, Wrangler." Um, and, you know, you You find these things endearing and

174
00:08:44,711 --> 00:08:49,314
then like seven, eight, nine years later, you're like, "So-called Wrangler."

175
00:08:49,844 --> 00:08:50,542
Corey: It, it's weird.

176
00:08:50,542 --> 00:08:53,265
The, the casual decisions we make, like, "Eh, I'll go back and fix

177
00:08:53,265 --> 00:08:55,833
that later," and you realize inadvertently it's like, "Yeah, I'm just

178
00:08:55,834 --> 00:08:58,915
gonna replace the entire foundation in that building," and oh no,

179
00:08:58,938 --> 00:09:01,366
suddenly, well, I guess that's what we're calling it forevermore.

180
00:09:01,501 --> 00:09:05,522
Rita: But it's interesting y- there's, uh, there's both sides to it.

181
00:09:05,522 --> 00:09:07,910
On one hand, there's everything that LLMs

182
00:09:07,923 --> 00:09:09,868
know, and obviously it's hard to displace that.

183
00:09:09,893 --> 00:09:14,094
On the other hand, it's so much easier in some ways to get LLMs

184
00:09:14,128 --> 00:09:16,901
to adopt something new if you can be a bit intentional about it

185
00:09:16,901 --> 00:09:19,318
because you can just give it a prompt and it will follow that

186
00:09:19,319 --> 00:09:22,482
prompt to the T, whereas humans are like very bad at following

187
00:09:22,482 --> 00:09:26,258
instructions, and we do fall back to old habits much more easily.

188
00:09:26,258 --> 00:09:30,044
So I think that that was something that the team put a lot of

189
00:09:30,053 --> 00:09:34,067
thought and effort into that helped make the transition to CF so

190
00:09:34,068 --> 00:09:36,566
much easier, like the fact that you can give it a single prompt.

191
00:09:36,567 --> 00:09:40,352
And, um, before even migrating between versions of Wrangler,

192
00:09:40,388 --> 00:09:44,962
that was something that I I would very painstakingly come

193
00:09:44,963 --> 00:09:47,347
to a customer and be like, "You guys really should do this."

194
00:09:47,386 --> 00:09:51,152
And they would be like, "I don't wanna fiddle around with this for a month.

195
00:09:51,152 --> 00:09:53,727
I wanna do other stuff that's actually important to my roadmap."

196
00:09:53,730 --> 00:09:55,825
And now it's like, "Just give it this prompt." And they're like,

197
00:09:55,826 --> 00:09:58,585
"Yep, I'm done." And we did this with a bunch of teams internally.

198
00:09:59,190 --> 00:10:01,952
Corey: It, it's a near certainty that there was an internal debate

199
00:10:01,972 --> 00:10:04,701
around this, so I wanna make sure I call out something that I like.

200
00:10:04,735 --> 00:10:08,692
Um, every provider now has skills you can give your agent that winds up

201
00:10:08,692 --> 00:10:11,944
explaining how to interface with the various things that we're using, how to,

202
00:10:12,011 --> 00:10:15,821
how do you contextualize this platform that you've suddenly gotten exposure to.

203
00:10:16,236 --> 00:10:22,204
And a, without naming names, a, a lot of them are basically SEO for agents

204
00:10:22,205 --> 00:10:25,528
of, "Oh, any time someone wants to deploy it, make sure you use this."

205
00:10:25,699 --> 00:10:29,764
So I went through yours, and it is remarkably even-handed across the board.

206
00:10:30,027 --> 00:10:32,990
It is, "If you are doing this on Cloudflare, then go ahead

207
00:10:32,991 --> 00:10:36,184
and do this other thing." It didn't turn into, "And remember,

208
00:10:36,214 --> 00:10:40,290
Cloudflare is the best," which some other skill packs do.

209
00:10:40,375 --> 00:10:44,011
And that has been wildly annoying to me when you, especially,

210
00:10:44,059 --> 00:10:45,924
God forbid, you have two of those things installed.

211
00:10:46,230 --> 00:10:49,775
So as fun as it is to wind up yelling at companies when they

212
00:10:49,775 --> 00:10:52,244
get things wrong, I do like to take a point every now and

213
00:10:52,244 --> 00:10:54,733
then of m- making it obvious when you've done something right.

214
00:10:55,052 --> 00:10:57,788
I'm sure someone was like, "Oh, think of the growth hacking we

215
00:10:57,788 --> 00:11:00,662
could achieve if we just basically sell our customers for glue."

216
00:11:01,090 --> 00:11:03,783
Dane: Well, I mean, I, I, I think that's pre-baked into our DNA.

217
00:11:03,981 --> 00:11:08,512
Uh, you know, things like even just from privacy to, uh, transparency

218
00:11:08,746 --> 00:11:11,771
to, uh, the customers are not our, our, our data and, and, you

219
00:11:11,771 --> 00:11:15,117
know, so, uh, that is not like a tops-down thing that, you know,

220
00:11:15,123 --> 00:11:18,491
that is kind of baked into, uh, uh, the people that we hire.

221
00:11:18,571 --> 00:11:20,552
Uh, there, there was no like conscious decision.

222
00:11:20,554 --> 00:11:21,222
I'm glad to hear that.

223
00:11:21,222 --> 00:11:22,653
I'm actually excited to hear that you, you say

224
00:11:22,655 --> 00:11:24,608
this, but, uh, uh, I actually didn't even know that.

225
00:11:25,102 --> 00:11:28,733
Corey: I was expecting that some providers would absolutely

226
00:11:28,759 --> 00:11:31,408
let marketing run loose inside of those skills, and-

227
00:11:31,408 --> 00:11:31,890
Rita: And some do

228
00:11:32,412 --> 00:11:33,801
… 
Corey: less than I was worried about

229
00:11:33,803 --> 00:11:35,228
originally, though, which is kind of awesome.

230
00:11:35,230 --> 00:11:37,685
Like, one other thing I'm realizing now is that we've been begging as humans

231
00:11:37,686 --> 00:11:41,307
for the internet to be better in various ways for 20 years, and only now

232
00:11:41,309 --> 00:11:44,378
that agents are a thing are companies actually getting off their corporate

233
00:11:44,386 --> 00:11:48,265
asses and doing something about it, where, oh yeah, turns out that we, if

234
00:11:48,266 --> 00:11:52,586
we stuff like eight megabytes of JavaScript into a two-kilobyte instruction

235
00:11:52,586 --> 00:11:55,759
file, agents are just not gonna click on the ads or read it anyway,

236
00:11:55,761 --> 00:11:59,526
so we should just put the text on a text file and make that available.

237
00:11:59,526 --> 00:12:01,846
It's like the second greatest thing to happen to the internet.

238
00:12:01,911 --> 00:12:05,903
Like, it's, it's reminiscent of GDPR day, when a bunch of companies were

239
00:12:05,903 --> 00:12:09,299
suddenly unprepared, didn't know how enforcement was gonna work, and for a few

240
00:12:09,299 --> 00:12:13,089
days they turned off the JavaScript and the entire internet was blazingly fast.

241
00:12:13,090 --> 00:12:16,351
It's like, ha, you coulda had this, but telemetry is important.

242
00:12:16,411 --> 00:12:20,372
Rita: So you know what's funny about that is on GDPR day was the first time

243
00:12:20,415 --> 00:12:23,723
I was- Initially the original Workers PM and it was the first time that

244
00:12:23,723 --> 00:12:28,746
we saw a jump in Workers adoption because so many European customers just

245
00:12:28,746 --> 00:12:34,484
deployed a worker that said, "If EU just don't serve that website," because US

246
00:12:34,505 --> 00:12:37,589
companies just didn't even wanna deal with it and didn't, didn't know how to.

247
00:12:37,929 --> 00:12:38,147
Corey: Yeah.

248
00:12:38,208 --> 00:12:39,443
I, I have thoughts on that.

249
00:12:39,443 --> 00:12:42,570
None of them are acceptable for, uh, public distribution at

250
00:12:42,570 --> 00:12:44,835
this point because I can't quite tamp down the blistering

251
00:12:44,836 --> 00:12:47,707
stream of profanity that winds up, uh, playing out.

252
00:12:48,062 --> 00:12:50,501
Uh, so a few weeks before this conference was a bunch

253
00:12:50,501 --> 00:12:52,829
of your releases, uh, during your birthday week.

254
00:12:52,865 --> 00:12:55,039
Folks have had time to play with them, and now basically

255
00:12:55,092 --> 00:12:58,609
talk to a number of the folks here about those enhancements.

256
00:12:58,638 --> 00:13:01,487
I think only a small subset of them started with, "You know what

257
00:13:01,487 --> 00:13:03,723
your problem is?" And then devolved into violence from there.

258
00:13:04,339 --> 00:13:06,032
So the question that has come up, though, do you

259
00:13:06,032 --> 00:13:09,337
consider Connect to be a developer conference?

260
00:13:09,726 --> 00:13:11,918
Dane: Well, I mean, I absolutely consider it to be a developer conference.

261
00:13:11,938 --> 00:13:14,983
We don't consider it a place to go announce things.

262
00:13:15,368 --> 00:13:18,374
We consider it a place to a conven- convenient, uh, experience,

263
00:13:18,375 --> 00:13:20,828
where you get to go meet peers, you get to go meet fellow

264
00:13:20,839 --> 00:13:23,282
Cloudflare employees, uh, you get to learn something.

265
00:13:23,461 --> 00:13:24,657
At least that's our, that's our hope.

266
00:13:24,862 --> 00:13:26,847
Like, that's why, you know, no one stood up on stage and

267
00:13:26,848 --> 00:13:29,198
said, "Guess what? You know, these three big announcements

268
00:13:29,198 --> 00:13:30,930
that we're doing," uh, that, that was very intentional

269
00:13:31,584 --> 00:13:34,623
Corey: So I wanna shift a little bit because something that, uh, you've

270
00:13:34,625 --> 00:13:37,266
talked about a number of sessions that I, uh, caught the abstracts

271
00:13:37,282 --> 00:13:41,347
of and, uh, poked my head in to listen to, uh, how have you folks

272
00:13:41,348 --> 00:13:45,016
changed how you're building and deploying things via the use of AI?

273
00:13:45,075 --> 00:13:49,126
Because I can talk about the insane nonsense that I build primarily in service

274
00:13:49,126 --> 00:13:52,839
of shitposting, but there's a difference between, "Hee-hee, if this goes

275
00:13:52,849 --> 00:13:56,614
really well, thousands of people will see it," whereas stuff that you do is

276
00:13:56,622 --> 00:14:01,019
now a statistically significant portion of the internet as a blast radius.

277
00:14:01,242 --> 00:14:04,198
I don't think you get to YOLO slam it into,

278
00:14:04,198 --> 00:14:06,406
uh, into creation quite the way that I do.

279
00:14:06,586 --> 00:14:08,597
Dane: About a, a little over a year and a half ago, we kinda wrote down

280
00:14:08,597 --> 00:14:11,824
what are, like, our, our AI principles were, and like, you know, number one

281
00:14:11,824 --> 00:14:14,795
and number two was no vibe coding, at least on, on the engineering side.

282
00:14:15,178 --> 00:14:18,217
Uh, and, and that's because, you know… And we also said, you know,

283
00:14:18,217 --> 00:14:20,836
if something gets deployed to production, there's no blaming AI either.

284
00:14:21,284 --> 00:14:23,492
So, uh, we still very much, you know, humans

285
00:14:23,493 --> 00:14:26,461
write code, uh, humans plan, humans do that.

286
00:14:26,522 --> 00:14:29,315
You know, they might use AI to assist them, but they

287
00:14:29,318 --> 00:14:31,704
own, uh, what they're building, what they're shipping.

288
00:14:32,174 --> 00:14:37,003
Uh, and, and we use AI as a tool to, uh, help accelerate that, uh, but not

289
00:14:37,003 --> 00:14:40,923
to replace hu- human judgment, uh, human thought, um, human creativity.

290
00:14:41,391 --> 00:14:46,179
Rita: That said, AI is really good at things that humans are bad at, right?

291
00:14:46,186 --> 00:14:49,250
So, um, you can- a human can review code

292
00:14:49,252 --> 00:14:52,206
and miss stuff that AI is going to catch.

293
00:14:52,246 --> 00:14:57,214
And so AI has been really instrumental to our code review process, and I think

294
00:14:57,214 --> 00:15:01,056
has actually preve- helped prevent incidents that otherwise would've happened.

295
00:15:01,583 --> 00:15:05,339
And similarly, yeah, I, I think kind of coming from a culture of

296
00:15:05,371 --> 00:15:10,849
ownership, like no vibe coding doesn't mean, uh, yeah, don't use AI.

297
00:15:10,983 --> 00:15:14,891
I think that one of the challenges that we're all kind of seeing now

298
00:15:14,891 --> 00:15:18,833
too is, um, there are things that vibe coding is really great at, like

299
00:15:19,185 --> 00:15:24,422
putting together a prototype, or we have a lot of non-engineering teams

300
00:15:24,422 --> 00:15:29,081
that really embrace vibe coding because it helps them build personal

301
00:15:29,088 --> 00:15:33,022
software that otherwise, you know, we, we'd have to go get a SaaS

302
00:15:33,032 --> 00:15:35,992
license for something that serves, like, maybe, like, 10% of their need.

303
00:15:35,992 --> 00:15:36,002
Yeah.

304
00:15:36,435 --> 00:15:39,577
Um, so, so in some ways it's been very empowering, but I think being

305
00:15:39,582 --> 00:15:43,979
very clear about what's been reviewed by a person and what's going into

306
00:15:43,981 --> 00:15:48,601
production, uh, having human judgment also for, you know, which code

307
00:15:48,611 --> 00:15:52,235
does need to be thoroughly reviewed, um, because it's gonna be deployed

308
00:15:52,259 --> 00:15:56,271
to the edge, and it's part of, like, a very, very core service versus,

309
00:15:56,276 --> 00:15:59,534
like, maybe this is something completely new and experimental and

310
00:15:59,544 --> 00:16:03,808
that's, you know, only a package that we're giving to people to try out.

311
00:16:03,831 --> 00:16:06,758
Like Those are, those things are very, very different

312
00:16:06,953 --> 00:16:09,432
Dane: Or, or, or things that are just, uh, annoying.

313
00:16:09,491 --> 00:16:13,904
A, a hu- a human's not gonna find excitement over, uh, rewriting Next to,

314
00:16:13,904 --> 00:16:18,018
uh, be, uh, faster and safer to run, and also be able to run on Cloudflare.

315
00:16:18,345 --> 00:16:18,780
Uh- There are few

316
00:16:18,790 --> 00:16:19,761
Corey: humans, but I hear you.

317
00:16:19,761 --> 00:16:19,771
Yeah.

318
00:16:19,832 --> 00:16:21,329
Dane: But, uh, yeah, the, I mean, there was, there

319
00:16:21,329 --> 00:16:24,380
was one human originally, a s- a slop fork Steve.

320
00:16:24,741 --> 00:16:26,914
You know, he gave it the plan, he gave it the guidelines.

321
00:16:27,667 --> 00:16:30,453
I've been asking for that for years, and they were like, "This,

322
00:16:30,456 --> 00:16:33,351
this project, uh, this, this is horrible." Uh, um, and then

323
00:16:33,354 --> 00:16:35,882
finally, like the AI was like… Well, the AI didn't have a choice.

324
00:16:36,080 --> 00:16:36,294
Yeah.

325
00:16:36,422 --> 00:16:38,000
Rita: Well, the, the interesting thing about

326
00:16:38,000 --> 00:16:39,993
that is he tried that experiment twice.

327
00:16:39,993 --> 00:16:43,678
Yeah Once in October of last year, and he kind of kicked the tires with

328
00:16:43,678 --> 00:16:46,375
it, and also got bored with it- Yeah … because the AI wasn't good enough.

329
00:16:46,717 --> 00:16:50,391
And then fast-forward to, um, you know, the models that came out in

330
00:16:50,391 --> 00:16:54,055
like November, December, and all of a sudden they're cooking with fire.

331
00:16:54,286 --> 00:16:54,429
Corey: Yeah.

332
00:16:54,659 --> 00:16:55,236
It's odd.

333
00:16:55,307 --> 00:16:57,615
There- there's a weird judgment aspect of it.

334
00:16:57,615 --> 00:16:59,178
People using these things love it.

335
00:16:59,244 --> 00:17:01,201
Uh, people have to consume the thing that

336
00:17:01,201 --> 00:17:02,712
gets put out, and the other end is awful.

337
00:17:02,712 --> 00:17:06,287
Like, uh, my, uh, AI factory, your slop cannon,

338
00:17:06,289 --> 00:17:07,931
and it, it becomes a different challenge.

339
00:17:08,134 --> 00:17:10,885
Something we've learned pretty clearly over the past year has been

340
00:17:10,885 --> 00:17:14,967
that vibe coding is awesome, and vibe maintaining is terrible,

341
00:17:14,986 --> 00:17:18,298
and vibe maintaining someone else's AI slop is like they're

342
00:17:18,299 --> 00:17:21,521
building an annex onto the last circle of hell just for that.

343
00:17:21,720 --> 00:17:22,873
It's terrible.

344
00:17:23,115 --> 00:17:27,054
So on some level you almost start to rebuild things five or six times.

345
00:17:27,054 --> 00:17:31,099
Where a rewrite used to be a complete anti-pattern variety of ways, and

346
00:17:31,153 --> 00:17:33,961
now it's starting to feel like for some things it's a best practice.

347
00:17:34,276 --> 00:17:36,581
Dane: I mean, we always thought of, uh, Workers as a platform

348
00:17:36,583 --> 00:17:39,387
where people build, uh, uh, greenfield, uh, applications.

349
00:17:39,755 --> 00:17:41,610
Uh, 'cause it's just a very opinionated architecture that

350
00:17:41,610 --> 00:17:43,203
you can get a lot of great things, uh, when you do it.

351
00:17:43,203 --> 00:17:46,337
But you're not gonna go port an, you know, a- an application

352
00:17:46,337 --> 00:17:48,988
that's been running, you know, for 10 years and move it over to us.

353
00:17:48,988 --> 00:17:50,861
And we were, like, kind of fine with that, 'cause

354
00:17:50,862 --> 00:17:52,943
still there's a lot more greenfield stuff being built.

355
00:17:53,293 --> 00:17:54,891
But all of a sudden, uh, we're seeing people that have

356
00:17:54,891 --> 00:17:57,550
these prompts that are literally- rewrite my application

357
00:17:57,552 --> 00:17:59,911
to run on Cloudflare Workers and, and migrate it for me.

358
00:18:00,365 --> 00:18:03,661
And, and all of a sudden, like, that, that process even becomes cheap.

359
00:18:04,144 --> 00:18:07,265
Corey: For me, the wild part that, of everything you've released

360
00:18:07,265 --> 00:18:10,437
this year, it is the ability to have an agent spin something up

361
00:18:10,534 --> 00:18:14,291
on Cloudflare without even having an account, and then okay, now

362
00:18:14,291 --> 00:18:16,497
you can go ahead and claim it if you decide you want to keep it.

363
00:18:16,806 --> 00:18:20,329
Uh, just the abuse concerns on something like that, the scale concerns

364
00:18:20,329 --> 00:18:23,960
you have to be dealing with, too, make that… It, it's a, it's miracle

365
00:18:23,960 --> 00:18:26,641
from my perspective it even got shipped, let alone stuck around.

366
00:18:26,782 --> 00:18:31,738
It, it's one of the, the, the fastest mean time to dopamine, uh, moments

367
00:18:31,739 --> 00:18:34,688
on deploying something for the world to see that I've ever heard of.

368
00:18:35,258 --> 00:18:36,763
Dane: Well, well, so two parts to that.

369
00:18:36,786 --> 00:18:37,985
One is, uh, the abuse part.

370
00:18:38,017 --> 00:18:39,810
We, we have a, an amazing trust and safety

371
00:18:39,811 --> 00:18:42,070
team, which we've, you know, had 15 years-

372
00:18:42,070 --> 00:18:42,751
Corey: You basically have to

373
00:18:42,838 --> 00:18:46,073
… 
Dane: Yeah, which we've had 15 years to, uh, uh, hone that skill.

374
00:18:46,429 --> 00:18:49,255
Uh, so, uh, you know, there's still obviously a lot of abuse,

375
00:18:49,263 --> 00:18:52,605
but we are, we're, we're pretty good at, uh, uh, uh, handling it.

376
00:18:52,959 --> 00:18:55,372
And then, you know, honestly just the, the cost, uh, to

377
00:18:55,372 --> 00:18:56,848
deploy something like that on our platform is really just

378
00:18:56,848 --> 00:18:59,214
minimal, and then that's just 'cause how we designed it.

379
00:18:59,620 --> 00:19:03,048
Um, and so, you know, uh, uh, the, you know, that, that,

380
00:19:03,072 --> 00:19:05,310
that, uh, the abuse part is probably the bigger issue,

381
00:19:05,340 --> 00:19:06,972
that was the bigger concern, but the cost is not really.

382
00:19:07,425 --> 00:19:11,308
Rita: The interesting thing is, w- we've over the past year really invested

383
00:19:11,309 --> 00:19:15,550
in our, uh, dashboard experience, which as someone that still uses it,

384
00:19:15,580 --> 00:19:20,232
has gotten so much better, but now we're seeing so many people on X being

385
00:19:20,233 --> 00:19:24,642
like, "Oh, I use Cloudflare without ever having touched a dashboard.

386
00:19:24,643 --> 00:19:26,719
I just told my agent to go X, Y, and Z.

387
00:19:26,756 --> 00:19:29,555
They created the account for me and everything." And that, that's the

388
00:19:29,556 --> 00:19:32,615
new era that we live in, and I think that that's a requirement now.

389
00:19:32,615 --> 00:19:32,788
Dane: It's the worst.

390
00:19:32,788 --> 00:19:34,297
It's not like I started getting good at it.

391
00:19:34,474 --> 00:19:34,624
Rita: Yeah.

392
00:19:35,024 --> 00:19:35,921
Dane: And then no one cares anymore.

393
00:19:36,051 --> 00:19:38,580
Corey: You have to have a dashboard because a demo on

394
00:19:38,584 --> 00:19:40,766
stage of, "And here it's working with Claude Code," like,

395
00:19:40,799 --> 00:19:43,201
what, what product are we demoing here doesn't work.

396
00:19:43,504 --> 00:19:47,191
But the thing that works well for customers who are using

397
00:19:47,192 --> 00:19:50,456
it is not necessarily the thing that looks great on a stage.

398
00:19:50,800 --> 00:19:53,011
Those are two very different use cases from a UX

399
00:19:53,021 --> 00:19:56,721
perspective as it turns out, and you have to have it.

400
00:19:56,763 --> 00:19:59,766
It is a baseline requirement, but your common

401
00:19:59,768 --> 00:20:01,524
case increasingly is customers never see it.

402
00:20:01,738 --> 00:20:05,699
Rita: I think an interesting take on it maybe is that, um, it, it allows

403
00:20:05,699 --> 00:20:10,155
us to focus on our efforts on things where you do kind of have to have

404
00:20:10,155 --> 00:20:14,763
it maybe to explain a unique concept or to really visualize something.

405
00:20:15,117 --> 00:20:17,858
And like, for example, CRUD dashboards should have

406
00:20:17,859 --> 00:20:19,928
never been complicated in the first place, right?

407
00:20:20,263 --> 00:20:25,059
Um, uh, if you have a decent enough design system in place, and that

408
00:20:25,060 --> 00:20:28,352
part's been accelerated for our internal teams, too, through having a

409
00:20:28,352 --> 00:20:33,072
design system where, um, we actually increasingly even have PMs that ship

410
00:20:33,079 --> 00:20:37,134
parts of the dashboard because- We already have the predefined criteria

411
00:20:37,607 --> 00:20:39,202
Corey: Well, that's the whole cloud company problem.

412
00:20:39,667 --> 00:20:42,074
To create a design system, ha, you only have one,

413
00:20:42,074 --> 00:20:45,178
we've got eight, and there's where the problem starts.

414
00:20:45,190 --> 00:20:47,050
It's… You can, you can tell when you're leaving

415
00:20:47,050 --> 00:20:49,451
one team's remit and heading into another.

416
00:20:49,465 --> 00:20:50,500
Consistency is horrible.

417
00:20:50,783 --> 00:20:52,173
Dane: Well, we like to pave our own way.

418
00:20:52,173 --> 00:20:53,504
We went 15 years without one.

419
00:20:53,656 --> 00:20:53,954
Yeah.

420
00:20:54,197 --> 00:20:55,659
Uh, and so we only have one.

421
00:20:56,036 --> 00:20:58,191
Corey: What I found interesting, some of this even exposed itself

422
00:20:58,193 --> 00:21:02,671
through CF, where there is still some inconsistency between namespaces,

423
00:21:02,703 --> 00:21:06,164
um, how things work with various services, and I, I still maintain fix

424
00:21:06,164 --> 00:21:09,406
it now because this problem only gets exponentially worse with time.

425
00:21:09,822 --> 00:21:15,097
But it's a… It is one of those areas that's very hard to wind up getting

426
00:21:15,098 --> 00:21:18,981
deployed it- because that requires centralized coordination in a company

427
00:21:18,991 --> 00:21:22,249
that benefits by launching things without that centralized coordination.

428
00:21:23,053 --> 00:21:26,595
This episode is sponsored by my own company, Duckbill.

429
00:21:26,886 --> 00:21:29,039
Having trouble with your AWS bill?

430
00:21:29,170 --> 00:21:32,524
Perhaps it's time to renegotiate a contract with them.

431
00:21:32,871 --> 00:21:34,943
Maybe you're just wondering how to predict

432
00:21:34,960 --> 00:21:38,227
what's going on in the wide world of AWS.

433
00:21:38,280 --> 00:21:40,938
Well, that's where Duckbill comes in to help.

434
00:21:41,147 --> 00:21:45,697
Remember, you can't duck the Duckbill bill, which I am reliably

435
00:21:45,698 --> 00:21:50,117
informed by my business partner is absolutely not our motto.

436
00:21:50,335 --> 00:21:53,589
To learn more, visit duckbillhq.com.

437
00:21:54,521 --> 00:21:55,889
Rita: It's definitely a spectrum, right?

438
00:21:55,933 --> 00:21:59,851
If you want innovation to go fast, then making it all go through a

439
00:21:59,856 --> 00:22:02,318
bottleneck is really, really hard, and so you kind of have to make

440
00:22:02,319 --> 00:22:07,259
the trade-off of, um, letting teams sometimes define their own.

441
00:22:07,261 --> 00:22:11,136
And, uh, a, a big part of our culture is ship but fix

442
00:22:11,136 --> 00:22:14,274
quickly, and don't let issues sit around and linger because

443
00:22:14,274 --> 00:22:16,987
the more they collect dust, the harder they become to fix.

444
00:22:17,180 --> 00:22:18,935
Dane: A- AI's kind of helped improve that somewhat.

445
00:22:18,970 --> 00:22:21,609
Uh, I mean, we built this thing called, uh, uh, our engineering

446
00:22:21,627 --> 00:22:25,771
codex, uh, which, uh, is, uh, you know, a set of RFCs on how we build.

447
00:22:26,177 --> 00:22:29,225
Uh, and it turns out agents are really good at following… You know, we should

448
00:22:29,225 --> 00:22:32,139
have had those rules already written down for humans, but we didn't really.

449
00:22:32,473 --> 00:22:32,872
Uh, but, uh- It's on

450
00:22:32,872 --> 00:22:33,368
Corey: the backlog.

451
00:22:33,368 --> 00:22:33,471
Yeah.

452
00:22:33,471 --> 00:22:33,911
We'll get to it

453
00:22:33,912 --> 00:22:34,269
Dane: next- Yeah, yeah

454
00:22:34,269 --> 00:22:34,844
… 
Corey: sprint for sure,

455
00:22:34,849 --> 00:22:34,927
Dane: right?

456
00:22:34,927 --> 00:22:36,887
But it, it turns out agents are even better at, uh,

457
00:22:36,998 --> 00:22:38,458
following instructions when you write them down.

458
00:22:38,827 --> 00:22:41,424
Uh, and but then, you know, that we pair that with another kind of

459
00:22:41,434 --> 00:22:44,781
Cloudflare principle that we like, uh, like guidelines and not rules.

460
00:22:45,206 --> 00:22:48,263
Uh, so you know, uh, uh, you know, 'cause ru- rules were

461
00:22:48,264 --> 00:22:50,346
made to be broken, so you're allowed to ask for exceptions.

462
00:22:50,627 --> 00:22:52,833
But, you know, you need to have a good reason, it needs to be thought out.

463
00:22:53,102 --> 00:22:55,496
But, uh, uh, at least we're all starting from a common place

464
00:22:55,497 --> 00:22:58,649
of, uh, here- here's how you build on Cloudflare successfully.

465
00:22:58,926 --> 00:23:03,171
Corey: Internally, how do you handle the approach of autonomy with AI?

466
00:23:03,245 --> 00:23:05,539
Because on, on the one hand, just go ahead

467
00:23:05,539 --> 00:23:07,469
and do whatever your little heart desires.

468
00:23:07,508 --> 00:23:09,645
Oh, no, the payments database is not great.

469
00:23:10,086 --> 00:23:12,991
Despite all the beautiful lies we tell ourselves about Agile, when

470
00:23:12,994 --> 00:23:16,108
it comes to payments, we all do Waterfall for very good reason.

471
00:23:16,391 --> 00:23:19,192
Uh, on the other, you wanna go full-on cautious.

472
00:23:19,385 --> 00:23:21,548
Great, now I'm just hitting approve on everything.

473
00:23:21,549 --> 00:23:23,249
I'm getting decision fatigue, and I feel like a

474
00:23:23,249 --> 00:23:26,168
Claude condom, which isn't a great place to be in.

475
00:23:26,196 --> 00:23:28,076
Like, that, that feels like my job of hitting the Enter

476
00:23:28,076 --> 00:23:30,336
button could be replaced by the Dickie Bird at some point.

477
00:23:31,239 --> 00:23:33,448
How do you find the right place on that continuum?

478
00:23:33,736 --> 00:23:34,250
How do you think about it?

479
00:23:34,665 --> 00:23:37,342
Dane: Well, I mean, there are one-way doors and, uh, not.

480
00:23:37,373 --> 00:23:40,549
I mean, if you're deploying something, uh, uh, to the core runtime

481
00:23:40,550 --> 00:23:43,822
of, uh, workers, that needs to go through a lot of reviews.

482
00:23:43,822 --> 00:23:47,842
And, uh, uh, but if you're deploying a worker on that, which, uh, is

483
00:23:47,844 --> 00:23:51,484
gonna have a much smaller blast radius, and then you can kind of, uh, uh,

484
00:23:51,501 --> 00:23:54,046
manage the surface area, then you might have a different set of criteria.

485
00:23:54,248 --> 00:23:56,017
Corey: The problem I found is that it's very easy to

486
00:23:56,017 --> 00:23:58,676
identify a one-way door usually after you've gone through it.

487
00:23:58,711 --> 00:24:01,035
It, it… I found interesting novel failure

488
00:24:01,036 --> 00:24:04,604
cases of AI doing- Call it curious things.

489
00:24:04,909 --> 00:24:07,151
Uh, uh, no one is as zealous about backups,

490
00:24:07,153 --> 00:24:08,939
for example, as someone who's just lost data.

491
00:24:09,231 --> 00:24:10,729
Like, it feels like it's one of those

492
00:24:10,731 --> 00:24:12,748
everything's easy to predict after it happens.

493
00:24:12,992 --> 00:24:14,974
H- h- what was the process of getting there like?

494
00:24:15,052 --> 00:24:17,233
Dane: So we have a, uh, a, a pretty, uh,

495
00:24:17,249 --> 00:24:19,891
regimented incident process, uh, where, uh-

496
00:24:19,898 --> 00:24:21,699
Corey: Your write-ups are gold, absolutely gold,

497
00:24:21,817 --> 00:24:23,019
Dane: and public is- Uh, and that's, that's

498
00:24:23,029 --> 00:24:24,438
the kind of the, the final outcome of it.

499
00:24:24,440 --> 00:24:27,583
But, but, you know, it starts from when we declare it, uh, to how we document

500
00:24:27,583 --> 00:24:31,043
it throughout it, uh, the follow-ups, the prevents that come out of that.

501
00:24:31,439 --> 00:24:32,999
Uh, and that's also what led to writing

502
00:24:32,999 --> 00:24:35,123
the codex of actually documenting, "Okay.

503
00:24:35,125 --> 00:24:37,553
Well, we know what works here and doesn't work here, and here's how

504
00:24:37,555 --> 00:24:40,334
to build if you want something to work." That's been, uh, uh, you

505
00:24:40,334 --> 00:24:42,601
know, very successful for us and… But, you know, I think it's a

506
00:24:42,602 --> 00:24:45,135
constant lea- I mean, we're, we're learning and documenting and,

507
00:24:45,190 --> 00:24:47,817
and hopefully, you know, it's, it's okay to make mistake once, most

508
00:24:47,818 --> 00:24:50,685
mistakes and, uh, you know, we don't, we don't punish anyone for that.

509
00:24:50,743 --> 00:24:52,642
Uh, i- if you make a mistake twice, then, uh-

510
00:24:52,642 --> 00:24:54,828
Corey: It's hard to fire an agent, too- Right … which makes it even harder.

511
00:24:54,902 --> 00:24:56,620
Rita: But, but that's why we have humans, right?

512
00:24:56,621 --> 00:25:00,331
Because some of that is very nuanced, and I

513
00:25:00,365 --> 00:25:01,983
think people are pretty uncomfortable with that.

514
00:25:02,011 --> 00:25:05,801
People want, like, very clear, sometimes rigid rules about,

515
00:25:06,050 --> 00:25:09,767
you know, like, "Tell me I'm allowed to b- deploy X but not Y."

516
00:25:10,107 --> 00:25:14,323
And it's like, well, if it was that simple, then an agent could do it.

517
00:25:14,342 --> 00:25:18,330
The thing that I'm asking you to do is exercise human judgment and come back to

518
00:25:18,332 --> 00:25:23,720
me and tell me, like, "Okay, if we released this thing on a Friday afternoon,

519
00:25:24,314 --> 00:25:28,543
could I explain to a customer w- what happened if it caused an incident, or do

520
00:25:28,543 --> 00:25:33,000
you feel so confident that, um, actually, you know what, ship it on a Friday.

521
00:25:33,011 --> 00:25:33,786
Why not," right?

522
00:25:34,032 --> 00:25:35,497
Dane: Or it's a brand-new product and no one's using it?

523
00:25:35,524 --> 00:25:35,841
Rita: Yeah.

524
00:25:35,978 --> 00:25:36,107
Dane: Sure.

525
00:25:36,107 --> 00:25:37,048
Ship it, ship it on a Friday.

526
00:25:37,209 --> 00:25:37,365
Corey: Yeah.

527
00:25:37,365 --> 00:25:39,583
There's almost like a tiering ring of at some

528
00:25:39,583 --> 00:25:41,635
point, oh, wow, this is, this runs the core network.

529
00:25:41,636 --> 00:25:44,039
We should probably have some diligence around

530
00:25:44,039 --> 00:25:45,831
this rather than, eh, it's probably fine.

531
00:25:45,997 --> 00:25:46,434
You'd hope.

532
00:25:46,631 --> 00:25:46,885
Mm-hmm.

533
00:25:46,885 --> 00:25:49,126
But then you see some of the tier one ISPs and

534
00:25:49,131 --> 00:25:51,274
you start to wonder how accurate that really is.

535
00:25:51,448 --> 00:25:54,799
So I'm curious at this point of it, it seems like every

536
00:25:54,799 --> 00:25:57,072
time I turn around and think, well, okay, this is what

537
00:25:57,072 --> 00:25:59,472
Cloudflare does, I draw a circle around it in my mind.

538
00:25:59,473 --> 00:26:02,130
And reasonable people can disagree, but for a long time,

539
00:26:02,131 --> 00:26:04,848
the primary value was back when it was hard to get a

540
00:26:04,880 --> 00:26:08,483
domain name with an SSL cert in front of a website, great.

541
00:26:08,498 --> 00:26:09,587
It was the thing that I went to.

542
00:26:09,596 --> 00:26:11,781
Y- yes, your website talked a lot about security, but

543
00:26:11,781 --> 00:26:14,285
people only care about that after they've just failed to.

544
00:26:14,749 --> 00:26:17,085
Fun day to say that with the entire internet melting down from a

545
00:26:17,087 --> 00:26:20,051
fundamental security layer, but which that's neither here nor there.

546
00:26:20,467 --> 00:26:25,200
Then it became this idea of, uh, of being

547
00:26:25,212 --> 00:26:27,956
increasingly interesting things, usually at the edge.

548
00:26:27,963 --> 00:26:29,182
It was where you were the strongest.

549
00:26:29,380 --> 00:26:34,333
And now it seems that with things like object store, things like D1, things

550
00:26:34,374 --> 00:26:40,546
like durable objects, things that are, that are turning into an absolute

551
00:26:40,834 --> 00:26:45,121
reimagining of what w- a cloud provider is, where it starts, where it

552
00:26:45,121 --> 00:26:49,586
stops, it feels like it's very hard to figure out what you do in some ways.

553
00:26:49,586 --> 00:26:53,045
I think probably for me, the straw that broke my back was the idea of

554
00:26:53,049 --> 00:26:57,417
using you as an AI gateway, where all inference can route through you.

555
00:26:58,396 --> 00:27:00,937
Now it feels like, okay, I start to need like a flowchart to

556
00:27:00,937 --> 00:27:04,336
figure out what aspects of this you are and what you aren't.

557
00:27:04,635 --> 00:27:06,300
Am I the only customer you have that's starting…

558
00:27:06,452 --> 00:27:09,899
I guess, struggling to keep abreast of what it is that you're putting out,

559
00:27:09,971 --> 00:27:13,832
or am I… I mean, again, I could easily be the dumbest person in this room.

560
00:27:13,832 --> 00:27:14,881
It would not be the first time.

561
00:27:15,112 --> 00:27:17,603
Dane: No, I'd definitely say, uh, uh, one, one of our Achilles

562
00:27:17,603 --> 00:27:20,606
heels is, uh, uh, customers tend to not know everything we do.

563
00:27:20,719 --> 00:27:25,204
And, uh, you know, uh, w- we- we tend to, uh, uh, you know, have, you

564
00:27:25,204 --> 00:27:27,586
know, a customer that comes from this, you know, one area and they know

565
00:27:27,587 --> 00:27:31,222
the workers very well, or, uh, they know our bot management very well.

566
00:27:31,540 --> 00:27:33,158
Uh, and that's probably how we were telling the story.

567
00:27:33,466 --> 00:27:38,105
Um, uh, you know, uh, but I think, you know, over the past year and, uh, uh,

568
00:27:38,222 --> 00:27:41,605
uh, hopefully at the, uh, our, our keynote here at the, the trade show, uh,

569
00:27:41,661 --> 00:27:45,461
uh, you know, people really got to see, uh, how we can explain the platform in

570
00:27:45,462 --> 00:27:49,988
a way, uh, that is a little more coherent and that kind of explains, uh, what

571
00:27:49,988 --> 00:27:53,560
we do and everything it- possible to do on it that you can kind of imagine.

572
00:27:53,908 --> 00:27:56,703
Corey: Part of the challenge too is that what is possible

573
00:27:56,705 --> 00:27:58,841
on the platform today and what is possible three months

574
00:27:58,841 --> 00:28:02,456
from now are going to not be completely congruent.

575
00:28:02,456 --> 00:28:04,439
I mean, one of the hardest problems in this era that we're

576
00:28:04,440 --> 00:28:09,165
in is building any plan for a release or a talk in advance.

577
00:28:09,334 --> 00:28:13,916
I gave a s- a, uh, talk earlier this spring about how AI speak

578
00:28:13,967 --> 00:28:17,591
Terraform like a tourist, and that thing aged like fine milk.

579
00:28:17,626 --> 00:28:20,958
It's, it's one of those areas of, yeah, it's really good at that

580
00:28:20,958 --> 00:28:24,067
now, which is helpful because Terraform clearly hates its user.

581
00:28:24,079 --> 00:28:26,395
That's kind of its feature, but we all put up with it.

582
00:28:26,416 --> 00:28:27,562
Now we don't have to.

583
00:28:27,982 --> 00:28:30,913
The other piece of it though is that I still maintain that The

584
00:28:30,913 --> 00:28:34,243
areas that are not gonna be industrialized are wisdom and judgment.

585
00:28:34,562 --> 00:28:37,826
And I'm hoping that that holds true at some point, but even if

586
00:28:37,826 --> 00:28:40,322
it do- if it doesn't, I'm not quite sure where that leaves us.

587
00:28:40,337 --> 00:28:42,585
Like, at some point, like, "Oh, yeah, your job is to sit

588
00:28:42,585 --> 00:28:44,934
here in case the agent screws up 'cause we need somebody

589
00:28:44,935 --> 00:28:47,308
to sue." Uh, uh, that's not a great seat to be in.

590
00:28:47,639 --> 00:28:49,966
Rita: First of all, I'm glad you're talking about how hard it is for

591
00:28:49,968 --> 00:28:54,259
people that present and speak, that you can't reuse talks anymore.

592
00:28:54,784 --> 00:28:56,659
The, the, that thing becomes out of date.

593
00:28:56,693 --> 00:28:58,982
Like, I used to be able to reuse things, like, six months

594
00:28:58,982 --> 00:29:01,234
apart, a year apart, but that's no longer the case.

595
00:29:01,345 --> 00:29:02,227
Corey: No, they, they, they melt.

596
00:29:02,228 --> 00:29:04,451
They are absolutely unusable.

597
00:29:04,458 --> 00:29:06,187
You can reuse a few of the jokes, a few of the

598
00:29:06,187 --> 00:29:08,504
general themes, but well, get out, get out, get out.

599
00:29:08,534 --> 00:29:08,795
I start…

600
00:29:08,815 --> 00:29:11,458
Even when I give talks now, I feel obligated to call out the

601
00:29:11,458 --> 00:29:14,314
date when I say it because otherwise it's gonna turn into

602
00:29:14,317 --> 00:29:17,909
a, "Ha, look at this bad take you had." Yeah, what, 2004?

603
00:29:17,927 --> 00:29:18,608
Imagine that.

604
00:29:18,956 --> 00:29:22,063
Rita: It, there's kind of two things to what you were saying before, too, right?

605
00:29:22,081 --> 00:29:27,694
One, that's what Cloudflare has been doing has evolved and continues to evolve.

606
00:29:28,033 --> 00:29:32,501
And sometimes, um, yeah, the, the release cycles are, our customer…

607
00:29:32,655 --> 00:29:36,127
W- we've always prided ourselves in innovation, which means that we

608
00:29:36,127 --> 00:29:38,851
release a lot, and our customers have always struggled to keep up.

609
00:29:39,288 --> 00:29:41,664
Um, but I, I think that, uh, yeah.

610
00:29:41,666 --> 00:29:45,273
And then there's the second part of it, which is how do you communicate it?

611
00:29:45,399 --> 00:29:48,233
Um, which, which I, I think we're getting better and better at.

612
00:29:48,656 --> 00:29:52,330
But I, I think an interesting thing, too, is y- you know, you said another

613
00:29:52,330 --> 00:29:56,514
word that we would generally shy away from, whi- which is edge, which

614
00:29:56,514 --> 00:29:59,129
is where I think people kind of get confused about what we do, right?

615
00:29:59,130 --> 00:30:02,485
Because they associate Cloudflare very closely with, like, "Oh, I

616
00:30:02,487 --> 00:30:05,600
would wanna use you for things that are very close to my users."

617
00:30:05,600 --> 00:30:08,103
And as you were saying, we're increasingly doing things like

618
00:30:08,463 --> 00:30:11,971
durable objects or object storage, and I think what that plays

619
00:30:12,000 --> 00:30:16,596
more to is we never thought of ourselves as an edge for that reason

620
00:30:16,596 --> 00:30:20,617
of we think of ourselves as a smart network where we can route

621
00:30:20,632 --> 00:30:23,532
things to the place where it makes the most sense for it to run.

622
00:30:23,775 --> 00:30:25,818
It doesn't necessarily always have to mean,

623
00:30:25,826 --> 00:30:27,989
like, everything everywhere all the time.

624
00:30:27,989 --> 00:30:31,471
It can also mean In one place and at one point in

625
00:30:31,483 --> 00:30:33,449
time, in another place and another point in time.

626
00:30:33,491 --> 00:30:34,809
That's what makes sense

627
00:30:35,248 --> 00:30:37,393
Corey: Now, the, the question of course is how you make that legible to

628
00:30:37,393 --> 00:30:40,745
people, where, huh, I have a bunch of users in one part of the world, and

629
00:30:40,767 --> 00:30:45,228
over time they find the experience gets better as the, the smart network,

630
00:30:45,238 --> 00:30:48,414
for lack of a better term, winds up relocating things closer to them.

631
00:30:48,761 --> 00:30:52,358
It does lead to challenges with things like data residency, which is

632
00:30:52,468 --> 00:30:55,247
becoming an increasing problem as various governments realize that they could

633
00:30:55,252 --> 00:30:58,329
legislate themselves a data center investment if they say the right things.

634
00:30:58,705 --> 00:31:01,290
It is a brave new world out there in a lot of

635
00:31:01,291 --> 00:31:03,250
respects, and it's certainly not getting simpler.

636
00:31:03,451 --> 00:31:06,512
I think the idea of being able to hold the entire stack in one

637
00:31:06,513 --> 00:31:10,995
person's head has long since ceased to be an accurate representation

638
00:31:10,995 --> 00:31:14,807
of reality, but I think we've held onto that fantasy for a while.

639
00:31:15,384 --> 00:31:18,300
It's getting harder and harder to credibly cling to that belief.

640
00:31:18,359 --> 00:31:21,635
Rita: I mean, I wonder if you fast-forward how many people will even have

641
00:31:21,635 --> 00:31:24,743
the stack in their heads, and how much of the stack is in their agents.

642
00:31:25,329 --> 00:31:29,080
Corey: A, a question that I'm having right now, and, and as a platform

643
00:31:29,081 --> 00:31:31,922
provider with a bunch of partners with booths and whatnot, many of

644
00:31:31,926 --> 00:31:35,967
whom fall squarely into this, what is the defining characteristic to

645
00:31:35,967 --> 00:31:40,777
your view of where someone should build their own implementation of

646
00:31:40,778 --> 00:31:45,471
a thing versus going and paying for a product that does that thing?

647
00:31:45,951 --> 00:31:49,092
Um, an easy example here is, great, I'm not gonna be able to

648
00:31:49,092 --> 00:31:53,802
go out and build a global network for, uh, for reasonable use,

649
00:31:53,805 --> 00:31:56,070
especially given the current price of components for things,

650
00:31:56,306 --> 00:31:58,929
and I don't have eight years to spend playing that game.

651
00:31:59,115 --> 00:32:01,864
Clearly I'll go with a provider that it's out there.

652
00:32:02,513 --> 00:32:06,523
But if I wanna have a, a service on a website that does a thing that's

653
00:32:06,523 --> 00:32:09,868
basically a, a digitized version where its biggest competitor is Microsoft

654
00:32:09,868 --> 00:32:13,195
Excel, well suddenly now I'm debating do I build it or do I buy it?

655
00:32:13,531 --> 00:32:14,533
How do you see that unfolding?

656
00:32:15,009 --> 00:32:18,601
Dane: So, uh, at least for Cloudflare we, we kind of think of, uh, uh, what are

657
00:32:18,601 --> 00:32:21,930
kind of the primary systems of record, and where is that kind of data gravity?

658
00:32:22,326 --> 00:32:24,754
And we wanna reduce that, but we wanna put, put it in places that

659
00:32:24,754 --> 00:32:28,238
have, uh, you know, that actually are pretty trustworthy software.

660
00:32:28,258 --> 00:32:30,667
Like, you know, uh, for HR systems, we wanna

661
00:32:30,668 --> 00:32:32,137
use something that un- understands compliance.

662
00:32:32,139 --> 00:32:33,627
We don't wanna vibe code compliance.

663
00:32:33,850 --> 00:32:36,951
Corey: All the Frontier labs are outsourcing payroll, for very good reason.

664
00:32:36,952 --> 00:32:37,179
Dane: Yeah.

665
00:32:37,299 --> 00:32:41,339
Things like, uh, payments or finance, uh, uh, tooling, uh, you know.

666
00:32:41,737 --> 00:32:45,155
Uh, I don't see, uh, any- anyone, uh, vibe coding an ERP anytime soon.

667
00:32:45,371 --> 00:32:47,153
That's just also because the risk, right?

668
00:32:47,153 --> 00:32:49,014
Like, y- you know, the CFO doesn't wanna go to jail

669
00:32:49,036 --> 00:32:51,831
'cause someone, you know, decided to use a, a bad model.

670
00:32:52,232 --> 00:32:55,155
Um, and so, so, you know, for us th- those are the types of

671
00:32:55,155 --> 00:32:57,212
things that we, we've said, "Okay, well, th- those are gonna

672
00:32:57,213 --> 00:32:59,336
stick around 'cause risk- that, that's, there's a, kind of a

673
00:32:59,337 --> 00:33:02,440
risk," but we'll give people a playground, uh, kind of above that.

674
00:33:02,801 --> 00:33:05,243
We'll make sure that that data's accessible, easy for them to use,

675
00:33:05,279 --> 00:33:07,344
put kind of governance around it to make sure it's not leaking.

676
00:33:07,744 --> 00:33:10,273
Um, and, and let them build their own interfaces on top of

677
00:33:10,273 --> 00:33:14,043
it, but, uh, keep the workflow and the, uh, um, you know,

678
00:33:14,045 --> 00:33:16,072
kind of lo- lower level engine inside of those places,

679
00:33:16,149 --> 00:33:16,370
Corey: you know?

680
00:33:17,339 --> 00:33:18,098
Dane: How do you think about it?

681
00:33:19,040 --> 00:33:20,091
Rita: You know, I agree with Dane.

682
00:33:20,104 --> 00:33:23,293
I, you know, I think conventional wisdom has always

683
00:33:23,294 --> 00:33:26,208
been th- does it make your beer taste better, right?

684
00:33:26,233 --> 00:33:28,934
So if it doesn't actually help your customers, then,

685
00:33:29,393 --> 00:33:32,187
um, you probably should buy it rather than build it.

686
00:33:32,549 --> 00:33:36,704
Uh, I, I think that that's no longer entirely true because we do spend so much

687
00:33:36,705 --> 00:33:42,942
time using tools that meet, like, 30% of the thing that we needed them for,

688
00:33:43,250 --> 00:33:47,463
and the cost for building, yeah, especially an interface on top of a system of

689
00:33:47,464 --> 00:33:52,405
record is so, so low that there's no reason not to just kind of build your own.

690
00:33:52,677 --> 00:33:54,170
Corey: It, it feels like the open source

691
00:33:54,172 --> 00:33:55,543
world figured this out a generation ago.

692
00:33:55,547 --> 00:33:58,248
Software could be free as in puppy, where the maintenance becomes

693
00:33:58,249 --> 00:34:00,426
a nightmare because it's every… Or you build a thing that

694
00:34:00,427 --> 00:34:02,639
service- that perfectly services your needs, and as soon as

695
00:34:02,640 --> 00:34:05,941
you get a second user, oh, hell is other people's requirements.

696
00:34:06,186 --> 00:34:09,643
Now it seems like that is less of a barrier than it once was, but there

697
00:34:09,643 --> 00:34:12,926
is still value in folks that think very deeply about a problem set.

698
00:34:12,931 --> 00:34:14,062
Rita: Yeah, I mean, I think it's okay for

699
00:34:14,062 --> 00:34:16,629
things to just be single user now, too, right?

700
00:34:16,703 --> 00:34:20,467
Like, uh, I, I think that that's where the maintenance burden

701
00:34:20,467 --> 00:34:24,476
becomes expensive is once you get the second, third, hundredth user.

702
00:34:24,476 --> 00:34:27,402
But there's, there's certain software that's like, this is for me.

703
00:34:27,433 --> 00:34:32,440
I want a view of my team so that I can, uh, you know, do a s-

704
00:34:32,440 --> 00:34:35,994
a particular activity maybe at a team onsite, and I… No one

705
00:34:35,995 --> 00:34:38,215
else is gonna use it, and it took me five minutes to build.

706
00:34:38,215 --> 00:34:41,739
I didn't need to go out and, um, actually sign off all of

707
00:34:41,764 --> 00:34:45,801
Cloudflare's, um, you know, data f- for me to have access to a tool.

708
00:34:46,208 --> 00:34:49,652
Dane: Or, I mean, we're seeing, you know, every, every person in sales building

709
00:34:49,654 --> 00:34:53,380
their own dashboards and their own, uh, tooling on top of Salesforce, right?

710
00:34:53,627 --> 00:34:55,161
Which would just be unthinkable in the past.

711
00:34:55,161 --> 00:34:56,656
But so that mean, can come as personal software.

712
00:34:56,656 --> 00:34:59,107
They don't have to worry about it scaling out to the rest of the company.

713
00:34:59,110 --> 00:35:01,786
They just have to worry about it doing what they need it to do that day.

714
00:35:01,786 --> 00:35:05,882
And, uh, you know, uh, I'm pretty bullish on personal software, uh, which, you

715
00:35:05,882 --> 00:35:08,529
know, I think fits very well into Cloudflare 'cause you can deploy it for one

716
00:35:08,532 --> 00:35:11,517
person and, or deploy it for 100 people, and it kind of scales the same way.

717
00:35:11,969 --> 00:35:14,913
Um, and, uh, you know, and, and every once in a while one of those

718
00:35:15,272 --> 00:35:18,343
pieces of personal software might end up being something that- Uh,

719
00:35:18,383 --> 00:35:21,613
you know, it, it, it is a super hit, and it does need to scale up

720
00:35:22,062 --> 00:35:23,921
Corey: I, I'm also a believer increasingly in the idea of

721
00:35:23,925 --> 00:35:26,302
killing your pets, where all right, I've built some personal

722
00:35:26,302 --> 00:35:28,563
software that I've been iterating on, and it's gotten…

723
00:35:28,589 --> 00:35:30,674
It's drifted a lot for its use case.

724
00:35:30,903 --> 00:35:31,222
Great.

725
00:35:31,587 --> 00:35:32,709
Throw it away and start over.

726
00:35:32,797 --> 00:35:34,202
Uh, here are the interface points.

727
00:35:34,219 --> 00:35:36,659
Maybe plan for data migration, backups, again,

728
00:35:36,665 --> 00:35:38,577
your friend, but build a second one in parallel.

729
00:35:38,639 --> 00:35:40,696
See if that winds up solving the need better.

730
00:35:40,731 --> 00:35:43,046
It almost becomes a schedulable tradition.

731
00:35:43,141 --> 00:35:43,390
Mm-hmm.

732
00:35:43,678 --> 00:35:46,464
Rita: I think that's where Cloudflare is not only uniquely suitable

733
00:35:46,464 --> 00:35:50,789
for scaling up but also for just stay- staying scaled at zero, right?

734
00:35:50,789 --> 00:35:54,220
It costs you nothing if you're getting no

735
00:35:54,220 --> 00:35:55,974
requests, and you're not doing anything with it.

736
00:35:56,410 --> 00:35:58,838
Corey: Uh, I guess the last big topic I want to get into with you

737
00:35:58,839 --> 00:36:02,517
folks before we call it an evening here is for a long time folks

738
00:36:02,518 --> 00:36:04,800
were talking about dead Internet theory, the sort of conspiracy

739
00:36:04,801 --> 00:36:07,669
theory that the Internet was largely bots talking to each other,

740
00:36:07,737 --> 00:36:11,731
and now that is very clearly a credible way of thinking about this.

741
00:36:11,954 --> 00:36:14,879
As the, as an increased percentage of traffic, I believe might

742
00:36:14,914 --> 00:36:19,568
be a majority now, uh, becomes, uh, automated rather than human

743
00:36:19,578 --> 00:36:24,872
driven or human directed, that changes the Internet foundationally.

744
00:36:25,418 --> 00:36:26,466
How do you think about that?

745
00:36:26,568 --> 00:36:29,535
Because, again, robots do many things, uh, more efficiently than humans.

746
00:36:29,786 --> 00:36:32,568
One thing they don't do is click on the ads, which is

747
00:36:32,570 --> 00:36:34,405
gonna be a problem for an awful lot of business models.

748
00:36:35,094 --> 00:36:38,914
Dane: Yeah, I mean, uh, you know, right now humans and, uh, agents

749
00:36:38,917 --> 00:36:42,102
are kind of surfing on the same, uh, Internet, and, uh, it's kind

750
00:36:42,102 --> 00:36:45,640
of this messy period where, uh, that, that needs to work itself out.

751
00:36:45,642 --> 00:36:47,530
Like, they probably should not be on the same Internet.

752
00:36:47,530 --> 00:36:50,301
They, you know, there probably are more efficient ways of the, each of them

753
00:36:50,302 --> 00:36:54,037
interfacing with it and as well as more efficient, uh, ways of monetizing.

754
00:36:54,485 --> 00:36:58,293
Um, you know, I, I think, uh, uh, you know, when people have

755
00:36:58,295 --> 00:37:00,885
ads on a page, they're, they're, they know what their CPM is.

756
00:37:00,885 --> 00:37:03,265
They know about how much they're gonna make, uh, you know,

757
00:37:03,312 --> 00:37:06,264
cost per million on, on an ad. There's no reason why an agent

758
00:37:06,264 --> 00:37:08,572
can't just, you know, with a micro transaction just pay that.

759
00:37:09,148 --> 00:37:10,526
Um, and, and that's kind of like what we're,

760
00:37:11,017 --> 00:37:12,623
we've been building with our monetization gateway.

761
00:37:13,096 --> 00:37:16,305
Uh, you know, put sites behind us, uh, let people decide what

762
00:37:16,330 --> 00:37:20,051
their CPM is, and, uh, uh, you know, give a, give an agent a

763
00:37:20,053 --> 00:37:23,462
wallet, and, you know, let them go, uh, go wild, uh, uh, using it.

764
00:37:23,904 --> 00:37:26,421
Uh, but, uh, we are trying to give, uh, you know, the, the

765
00:37:26,573 --> 00:37:29,574
site owners and creators, uh, those type of controls, whether,

766
00:37:29,574 --> 00:37:32,767
uh, uh, let them decide, uh, who gets in and who doesn't.

767
00:37:33,271 --> 00:37:34,769
Corey: Of course, defaults matter because that's

768
00:37:34,770 --> 00:37:36,982
going to be what most people will wind up taking.

769
00:37:37,001 --> 00:37:39,110
But it's a- It's weird.

770
00:37:39,121 --> 00:37:41,048
Some folks absolutely do not want, uh,

771
00:37:41,175 --> 00:37:43,754
agents interacting at all with their site.

772
00:37:43,754 --> 00:37:47,032
And one of the more inscrutable ones to my mind is Amazon's retail store,

773
00:37:47,047 --> 00:37:50,513
where they will actively block anything that looks to be bot-driven.

774
00:37:50,543 --> 00:37:53,136
Mm. It's like it- the bot is coming to your website

775
00:37:53,162 --> 00:37:57,191
to buy a thing and give you money in return for goods.

776
00:37:57,745 --> 00:37:59,982
Well, we're gonna build our own, because we're Amazon.

777
00:37:59,982 --> 00:38:02,828
It's gonna have a terrible UX, it's gonna ship three years late after

778
00:38:02,828 --> 00:38:06,526
everyone else wants it, and you are just gonna be surprised when it

779
00:38:06,540 --> 00:38:09,949
finds lukewarm success in the market, if I can Nostradamus that a second.

780
00:38:10,460 --> 00:38:13,105
But it feels like it's emblematic of a larger trend,

781
00:38:13,119 --> 00:38:16,657
where folks are clinging to the way things used to be,

782
00:38:16,885 --> 00:38:19,611
almost in a denial of the fact that the sea is changing.

783
00:38:19,935 --> 00:38:20,867
Rita: I think yes and no.

784
00:38:20,868 --> 00:38:24,176
If, uh, we've also seen customers shift their perspectives,

785
00:38:24,194 --> 00:38:28,718
and I- I think a lot of people are evolving on their thinking

786
00:38:28,719 --> 00:38:32,773
about, yeah, a few years ago, the world was, you know, bot versus

787
00:38:32,773 --> 00:38:37,129
human, and now it can increasingly be bot on behalf of human.

788
00:38:37,468 --> 00:38:40,032
And so, um, you- you wanna allow that traffic.

789
00:38:40,032 --> 00:38:43,016
And we had a whole, you know, part of our conversation that was about,

790
00:38:43,352 --> 00:38:48,597
um, AEO and agents buying things because there's information available.

791
00:38:48,614 --> 00:38:53,252
So we're also seeing businesses that are actually leaning in and trying to

792
00:38:53,442 --> 00:38:57,281
inversely give agents as much access as possible to their content, because

793
00:38:57,759 --> 00:39:01,094
they- they- they want agents to recommend them and things like that.

794
00:39:01,687 --> 00:39:02,708
Dane: I mean, we're early on.

795
00:39:02,709 --> 00:39:05,077
We saw a lot of sites that just did not want their traffic to be used

796
00:39:05,079 --> 00:39:09,296
for training, the personal agents coming, Dots, Muse, uh, Grokbot.

797
00:39:09,357 --> 00:39:11,024
Uh, you know, that's actually changing.

798
00:39:11,024 --> 00:39:13,090
We're actually having people that say, "Hey, how do I make sure

799
00:39:13,090 --> 00:39:15,723
I let this in?" Uh, 'cause they actually do want that traffic.

800
00:39:16,114 --> 00:39:18,849
Um, and you know, so you know, what we were trying to work with all

801
00:39:18,849 --> 00:39:21,135
those providers are to make sure they sign the requests, that they

802
00:39:21,136 --> 00:39:24,238
clearly state what their intentions are, and then the site editor

803
00:39:24,238 --> 00:39:27,525
can decide, uh, you know, th- I want this in or I don't want this in.

804
00:39:27,832 --> 00:39:30,308
You know, for the most part, you know, uh, their incentives

805
00:39:30,308 --> 00:39:32,624
are aligned that, you know, they want, they want them in.

806
00:39:32,839 --> 00:39:34,603
Corey: Yeah, it- it's a hard problem to solve for.

807
00:39:34,605 --> 00:39:36,890
I don't, I don't think there's any clear right answer,

808
00:39:36,890 --> 00:39:38,472
no matter what you do, especially at your scale.

809
00:39:38,570 --> 00:39:40,071
People are gonna be mad, mad about it.

810
00:39:40,120 --> 00:39:41,276
That, that is what the internet does.

811
00:39:41,281 --> 00:39:44,731
It's very good at, uh, not accepting the two very complex,

812
00:39:44,745 --> 00:39:47,676
complex things can both be true at the same time, and that

813
00:39:47,676 --> 00:39:50,336
doesn't fit in a tweet, so we're just gonna get angry about it.

814
00:39:50,842 --> 00:39:51,960
It's not ideal.

815
00:39:52,066 --> 00:39:55,177
Um, I think that we're gonna be seeing some interesting changes.

816
00:39:55,223 --> 00:39:57,942
So okay, Connect 2027.

817
00:39:57,945 --> 00:39:59,321
Let's assume it's a year from now.

818
00:39:59,934 --> 00:40:00,604
What's changed?

819
00:40:01,411 --> 00:40:03,801
Rita: Oh boy, I, even if you said connect,

820
00:40:03,843 --> 00:40:07,316
I don't know, January 15th, 2027 Yeah.

821
00:40:07,576 --> 00:40:10,243
Corey: It used to be what's gonna be different in 25 years?

822
00:40:10,258 --> 00:40:12,067
And like, oh, by then we're definitely gonna have quantum

823
00:40:12,069 --> 00:40:15,317
computing powered by cold fusion, and now 25 years later

824
00:40:15,319 --> 00:40:17,937
you can say the exact same things, but here we are.

825
00:40:18,194 --> 00:40:20,095
Some things are very clearly changing too.

826
00:40:20,103 --> 00:40:22,826
So I, I don't know that I would wanna put, uh, a prediction

827
00:40:22,826 --> 00:40:24,781
out there, which is why I'm making it your problem instead.

828
00:40:25,406 --> 00:40:26,307
Dane, what do you think?

829
00:40:26,598 --> 00:40:29,734
Dane: A, a higher majority of people will have personal agents that are, you

830
00:40:29,734 --> 00:40:34,009
know, are running, uh, for them, uh, both at work and, uh, uh, at, at home.

831
00:40:34,009 --> 00:40:36,339
Uh, you know, I think, uh, today, uh, you

832
00:40:36,339 --> 00:40:38,029
know, we hear a lot about the GPU crunch.

833
00:40:38,373 --> 00:40:39,817
Uh, I think you'll hear more and more about the CPU

834
00:40:39,823 --> 00:40:42,975
crunch because harnesses and things like that need CPU.

835
00:40:43,438 --> 00:40:46,462
Um, you know, uh, hopefully Cloudflare's the default place

836
00:40:46,466 --> 00:40:49,911
where you, uh, deploy all your durable agents, um, and everyone

837
00:40:49,920 --> 00:40:52,674
has, uh, you know, hundreds of them making their life easier

838
00:40:52,702 --> 00:40:55,952
and lets them go spend, uh, uh, time on higher value things.

839
00:40:56,569 --> 00:40:58,509
Corey: If I had to guess, I, I think a lot of the breakthrough might

840
00:40:58,511 --> 00:41:01,763
very well come from folks one layer up the stack who are building on

841
00:41:01,785 --> 00:41:04,884
top of you folks, where it becomes like that is the thing that becomes

842
00:41:04,884 --> 00:41:08,475
the household brand, the recognized name in non-technical circles.

843
00:41:08,480 --> 00:41:11,361
And I, I'm not sure that you'd want it to be the other

844
00:41:11,363 --> 00:41:13,596
way of, yes, we're now going into consumer products.

845
00:41:13,633 --> 00:41:18,525
Uh, feels like it's a bit of a pivot there, but being able to be the, the, the

846
00:41:18,654 --> 00:41:22,827
underlying infrastructure providence that everyone else starts to build upon.

847
00:41:23,188 --> 00:41:26,834
Uh, increasingly I am seeing at least on the, uh, at a lot of the

848
00:41:26,834 --> 00:41:29,995
transport layer that Cloudflare has well gotten into that point.

849
00:41:30,283 --> 00:41:34,176
Rita: Yeah, I mean, we were just, uh, just before this I was in a session where,

850
00:41:34,609 --> 00:41:40,824
um, John from OpenAI was showing how, uh, GPT Sites is built on Cloudflare.

851
00:41:41,116 --> 00:41:44,471
We had, uh, Fabian on stage with me yesterday

852
00:41:44,473 --> 00:41:46,567
who was showing how Lovable builds on Cloudflare.

853
00:41:46,571 --> 00:41:49,569
So, um, in that sense, we've always viewed ourselves as the

854
00:41:49,569 --> 00:41:54,373
infrastructure layer, and, um, our, our goal is to give people that

855
00:41:54,386 --> 00:41:57,894
are very good at building these consumer experiences, and that's

856
00:41:57,894 --> 00:42:00,908
why I think there's a lot that's been left to imagination still.

857
00:42:00,908 --> 00:42:02,848
There's so much that's been unexplored.

858
00:42:03,223 --> 00:42:06,730
Um, one thing that I'm actually very bullish on is, uh,

859
00:42:06,750 --> 00:42:11,234
real time and, and voice feels very early on, and we're,

860
00:42:11,259 --> 00:42:13,596
we're seeing more and more folks ex- experiment with that.

861
00:42:13,603 --> 00:42:16,210
So I, I think you're right that we're, we're trying

862
00:42:16,214 --> 00:42:19,033
to play at kind of the, the layer slightly below.

863
00:42:19,881 --> 00:42:21,800
Dane: W- I mean, at, at our heart, we're an infrastructure company.

864
00:42:22,404 --> 00:42:26,840
Uh, and so, uh, you know, uh, while, uh, there was a lot of, uh, you know, a

865
00:42:26,840 --> 00:42:29,412
lot of c- like, a lot of questions early on, "Well, why, why don't you build

866
00:42:29,414 --> 00:42:33,208
a, build a, you know, vibe coding, uh, website builder?" Uh, we're like,

867
00:42:33,244 --> 00:42:35,702
"Well, 'cause we wanna build the platform, and we wanna build the primitives

868
00:42:35,705 --> 00:42:39,932
that, uh, let, uh, there be 10 of them on us, um, or hundreds of them on us."

869
00:42:40,243 --> 00:42:42,194
Um, and I, and I think largely that's kind of played out.

870
00:42:42,741 --> 00:42:44,057
Corey: I, there, I feel like there are two kinds of

871
00:42:44,057 --> 00:42:46,652
infrastructure company in the world, the kind that are,

872
00:42:46,777 --> 00:42:48,914
have not… "We're an infrastructure company." Great.

873
00:42:48,932 --> 00:42:50,753
That empowers an awful lot of things.

874
00:42:50,761 --> 00:42:51,872
Please build on top of us.

875
00:42:52,079 --> 00:42:55,381
And infrastructure companies that are mad about it and like to delude

876
00:42:55,381 --> 00:42:57,397
themselves into thinking, "Nope, nope, we're gonna move up the stack

877
00:42:57,398 --> 00:43:00,527
now and compete with all of the people that build on top of us."

878
00:43:00,842 --> 00:43:03,832
I almost wonder at some level if that's an inevitable consequence

879
00:43:03,833 --> 00:43:06,836
of getting to a certain point of scale, but it's refreshing to

880
00:43:06,842 --> 00:43:09,989
hear that you folks are not inflict- that affected by that yet.

881
00:43:10,231 --> 00:43:11,950
That's the one thing I hope doesn't change.

882
00:43:12,135 --> 00:43:15,114
Rita: We have a lot of conversations about it internally every once in a

883
00:43:15,115 --> 00:43:18,889
while, and, you know, what, what that layer on top keep, keeps changing.

884
00:43:19,119 --> 00:43:22,825
And, and one thing that we will do sometimes, um, mostly just so

885
00:43:22,825 --> 00:43:26,981
that we have experience in-house in understanding our customer's

886
00:43:26,984 --> 00:43:30,205
perspective is we will build, like, an internal ins- instance,

887
00:43:30,577 --> 00:43:33,991
and, um, sometimes we'll open source it as an example, right?

888
00:43:33,995 --> 00:43:37,869
To even show people what putting together the pieces would look like.

889
00:43:37,886 --> 00:43:42,039
Because also just being infrastructure that doesn't show you, you

890
00:43:42,039 --> 00:43:45,634
know, it's like how to draw a horse, and step one is very simple.

891
00:43:45,634 --> 00:43:48,709
And then, no, I s- I swear, you, you could get to a horse.

892
00:43:48,744 --> 00:43:53,454
Um, so, so that's one direction that we've found ourselves leaning in, but

893
00:43:53,533 --> 00:43:56,455
it's not to say that I, I think as a company, you can definitely build new

894
00:43:56,457 --> 00:43:59,652
muscles, and there are muscles that I'm very proud that we've built over time.

895
00:43:59,652 --> 00:44:03,377
But also understanding where your strengths are and where you're better off,

896
00:44:03,407 --> 00:44:07,673
um, you know, either letting people build on top of you or partnering, um,

897
00:44:08,049 --> 00:44:11,532
allows you to focus your energy on the things… Like, building infrastructure

898
00:44:11,533 --> 00:44:15,153
is really hard, and we have no shortage of problems to solve over there.

899
00:44:15,517 --> 00:44:18,662
Corey: You, you would think that would be more broadly accepted than it

900
00:44:18,664 --> 00:44:21,591
is, but now we're into the realm of people promising data centers in three

901
00:44:21,593 --> 00:44:24,095
months, and they think it's basically, uh, some tents and a generator.

902
00:44:24,150 --> 00:44:25,653
Um, good luck.

903
00:44:25,653 --> 00:44:26,604
Have fun with that.

904
00:44:27,120 --> 00:44:29,777
It, it takes work, legitimate hard work, to get these

905
00:44:29,783 --> 00:44:32,078
things built in a way that's sustainable and scalable, and

906
00:44:32,079 --> 00:44:34,412
it's nice to see that folks are sweating those details.

907
00:44:34,831 --> 00:44:35,291
Rita: That's our job.

908
00:44:35,318 --> 00:44:35,522
Yeah.

909
00:44:36,183 --> 00:44:38,232
Corey: I wanna thank you both for taking the time to speak with me.

910
00:44:38,278 --> 00:44:40,440
If people want to learn more, probably not the people

911
00:44:40,440 --> 00:44:42,599
here in this room, uh, you, you probably should know

912
00:44:42,601 --> 00:44:45,060
where to go to figure out more what Cloudflare is up to.

913
00:44:45,088 --> 00:44:47,565
But for folks listening, where's the best way

914
00:44:47,565 --> 00:44:49,246
to get started with Cloudflare these days?

915
00:44:50,043 --> 00:44:53,056
Dane: Go to cloudflare.com or point your agent at it and- There we go … and

916
00:44:53,068 --> 00:44:55,928
ask, uh, what, what, you know, what, what can I, what can I do with Cloudflare?

917
00:44:55,946 --> 00:44:56,530
How can I use it?

918
00:44:56,582 --> 00:44:57,764
How can I make my apps better?

919
00:44:58,063 --> 00:44:59,572
Corey: How can I make it hysterical and shit

920
00:44:59,583 --> 00:45:01,484
post-y is usually my prompt, but it works.

921
00:45:01,512 --> 00:45:01,714
Dane: Yeah.

922
00:45:01,970 --> 00:45:03,756
Rita: Someone had a great suggestion the other day.

923
00:45:03,798 --> 00:45:08,050
We, uh, published a blog post about what, what we use for our own

924
00:45:08,069 --> 00:45:12,107
internal AI review tool, and they were like, "I just started pointing

925
00:45:12,120 --> 00:45:15,285
my agent at Cloudflare blog posts and being like, 'Extract the parts

926
00:45:15,287 --> 00:45:18,213
that I could reuse from this,' and, like, start implementing it."

927
00:45:18,238 --> 00:45:18,969
So yeah.

928
00:45:19,327 --> 00:45:21,857
Corey: I still have this weird behavior every once in a while where an agent

929
00:45:21,857 --> 00:45:24,380
will look at this, re- well, I'll point it to one of your blog posts, and it'll

930
00:45:24,387 --> 00:45:28,625
be skeptical and try to fact check the things that you say in the blog post.

931
00:45:28,625 --> 00:45:31,477
Like I, like I made it up or something, like I'm trying to trick it.

932
00:45:32,185 --> 00:45:35,230
I'm sure someone out there has a good reason for it, but yeah,

933
00:45:35,336 --> 00:45:37,371
Rita: it- I promise we do a lot of fact-checking

934
00:45:37,387 --> 00:45:39,250
before any of our blog posts go live.

935
00:45:39,587 --> 00:45:40,384
Dane: A lot of lawyers.

936
00:45:40,500 --> 00:45:40,822
Corey: Yeah.

937
00:45:40,972 --> 00:45:43,420
Yeah, it turns out that they have opinions

938
00:45:43,429 --> 00:45:45,489
that you, you don't heed at your own peril.

939
00:45:45,563 --> 00:45:46,968
I've learned that over the years.

940
00:45:47,504 --> 00:45:50,383
Thank you both so much for taking the time out of your day to speak with me.

941
00:45:50,819 --> 00:45:51,165
Thanks again.

942
00:45:51,230 --> 00:45:51,484
Thanks.

943
00:45:51,484 --> 00:45:51,573
Dane: Have a great one.

944
00:45:51,634 --> 00:45:51,912
Thank you.

945
00:45:52,114 --> 00:45:52,403
Thank you.