1
00:00:00,120 --> 00:00:04,000
every 3 months things have just kept

2
00:00:02,000 --> 00:00:05,240
getting progressively better and now

3
00:00:04,000 --> 00:00:07,480
we're at this point where we're talking

4
00:00:05,240 --> 00:00:09,519
about full-on vertical AI agents that

5
00:00:07,480 --> 00:00:11,519
are going to replace entire teams and

6
00:00:09,519 --> 00:00:13,559
functions and Enterprises that

7
00:00:11,519 --> 00:00:15,759
progression is still mind-blowing to me

8
00:00:13,558 --> 00:00:17,559
a lot of the foundation models are kind

9
00:00:15,759 --> 00:00:19,800
of coming head-to-head there used to be

10
00:00:17,559 --> 00:00:23,118
only one player in town with open AI but

11
00:00:19,800 --> 00:00:26,039
we've been seeing in the last batch this

12
00:00:23,118 --> 00:00:28,599
has been changing thank God it's like

13
00:00:26,039 --> 00:00:30,800
competition is you know the the soil for

14
00:00:28,599 --> 00:00:33,159
a very fertile Market Marketplace

15
00:00:30,800 --> 00:00:36,120
ecosystem uh for which consumers will

16
00:00:33,159 --> 00:00:39,159
have choice and uh Founders have a shot

17
00:00:36,119 --> 00:00:44,359
and that's the world I want to live

18
00:00:39,159 --> 00:00:47,599
[Music]

19
00:00:44,359 --> 00:00:50,558
in welcome to another episode of the

20
00:00:47,600 --> 00:00:52,920
light cone I'm Gary this is Jared Harge

21
00:00:50,558 --> 00:00:55,439
and Diana and collectively we funded

22
00:00:52,920 --> 00:00:57,399
hundreds of billions of dollars worth of

23
00:00:55,439 --> 00:01:01,599
startups right when they were just one

24
00:00:57,399 --> 00:01:04,879
or two people starting out and and today

25
00:01:01,600 --> 00:01:09,040
Jared is a man on fire and he's going to

26
00:01:04,879 --> 00:01:11,799
talk about vertical AI yes I am I am

27
00:01:09,040 --> 00:01:15,080
fired up about this because I think

28
00:01:11,799 --> 00:01:17,759
people especially startup Founders

29
00:01:15,079 --> 00:01:20,679
especially young ones are not fully

30
00:01:17,759 --> 00:01:22,478
appreciating just how big vertical AI

31
00:01:20,680 --> 00:01:23,680
agents are going to be it's not a new

32
00:01:22,478 --> 00:01:25,478
idea some people are talking about

33
00:01:23,680 --> 00:01:27,200
vertical AI agents we funded a bunch of

34
00:01:25,478 --> 00:01:28,879
them but I think the world has not

35
00:01:27,200 --> 00:01:31,000
caught on to just how big it's going to

36
00:01:28,879 --> 00:01:33,519
get and so I'm going going to make the

37
00:01:31,000 --> 00:01:34,599
case for why I think there are going to

38
00:01:33,519 --> 00:01:38,078
be

39
00:01:34,599 --> 00:01:40,759
$300 billion plus companies started just

40
00:01:38,078 --> 00:01:43,399
in this one category nice I'm going to

41
00:01:40,759 --> 00:01:46,718
do it by analogy with SAS and I

42
00:01:43,399 --> 00:01:49,439
think in a in a similar fashion people

43
00:01:46,718 --> 00:01:51,039
don't understand just how big SAS is

44
00:01:49,438 --> 00:01:53,639
because most startup Founders especially

45
00:01:51,040 --> 00:01:54,960
young ones tend to see the startup

46
00:01:53,640 --> 00:01:56,680
industry through the lens of the

47
00:01:54,959 --> 00:01:58,438
products that they use as a consumer and

48
00:01:56,680 --> 00:01:59,640
as a consumer you don't tend to use that

49
00:01:58,438 --> 00:02:02,639
many sass tools because they're mostly

50
00:01:59,640 --> 00:02:04,759
built for companies and so I think a lot

51
00:02:02,640 --> 00:02:06,399
of people have missed the basic point

52
00:02:04,759 --> 00:02:08,159
that if you just look at what Silicon

53
00:02:06,399 --> 00:02:10,439
Valley has been funding for the most for

54
00:02:08,159 --> 00:02:12,199
like for the last 20 years like we've

55
00:02:10,439 --> 00:02:14,120
mostly been producing SAS companies guys

56
00:02:12,199 --> 00:02:16,039
like that's literally been like most of

57
00:02:14,120 --> 00:02:18,840
what has been coming out of Silicon

58
00:02:16,039 --> 00:02:21,159
Valley it's over 40% of all venture

59
00:02:18,840 --> 00:02:23,200
capital dollars in that time period went

60
00:02:21,159 --> 00:02:26,318
to SAS companies and we produced over

61
00:02:23,199 --> 00:02:28,479
300 SAS unicorns in that 20-year time

62
00:02:26,318 --> 00:02:30,280
period which is way more than every

63
00:02:28,479 --> 00:02:32,560
other category software is pretty

64
00:02:30,280 --> 00:02:35,080
awesome software is pretty awesome I was

65
00:02:32,560 --> 00:02:37,199
thinking back to the history of this

66
00:02:35,080 --> 00:02:39,319
because we we always like to talk about

67
00:02:37,199 --> 00:02:42,878
the sort of how the how the history of

68
00:02:39,318 --> 00:02:45,359
Technology informs the future and um the

69
00:02:42,878 --> 00:02:48,639
the real Catalyst for for the SAS boom

70
00:02:45,360 --> 00:02:51,319
was a do you guys remember XML HTTP

71
00:02:48,639 --> 00:02:53,039
request oh my God like I I'd argue that

72
00:02:51,318 --> 00:02:57,119
that was quite literally the Catalyst

73
00:02:53,039 --> 00:02:59,479
for the S boom like uh Ajax Ajax yeah in

74
00:02:57,120 --> 00:03:01,840
2004 browsers added this JavaScript

75
00:02:59,479 --> 00:03:03,119
function XML HTTP request which was the

76
00:03:01,840 --> 00:03:04,840
missing piece that enabled you to build

77
00:03:03,120 --> 00:03:06,319
a rich internet application in a web

78
00:03:04,840 --> 00:03:08,200
browser so for the first time you could

79
00:03:06,318 --> 00:03:10,439
make things in websites that looked like

80
00:03:08,199 --> 00:03:12,798
desktop applications and then that

81
00:03:10,439 --> 00:03:15,919
created Google Maps and Gmail and set up

82
00:03:12,799 --> 00:03:18,680
this whole like SAS boom essentially the

83
00:03:15,919 --> 00:03:20,399
the key technology atlock was that

84
00:03:18,680 --> 00:03:22,360
software moved from being a thing that

85
00:03:20,400 --> 00:03:23,920
you got on a CD ROM and installed on

86
00:03:22,360 --> 00:03:25,680
your desktop to being something that you

87
00:03:23,919 --> 00:03:28,000
use through a website and on your phone

88
00:03:25,680 --> 00:03:29,400
yeah Paul Graham actually uh shares in

89
00:03:28,000 --> 00:03:31,759
that lineage in that he was one of the

90
00:03:29,400 --> 00:03:34,879
first people to realize that he could

91
00:03:31,759 --> 00:03:37,840
take the HTTP request and then actually

92
00:03:34,878 --> 00:03:40,598
hook it up to a Unix prompt and you

93
00:03:37,840 --> 00:03:43,239
didn't actually have to you know have a

94
00:03:40,598 --> 00:03:46,199
separate computer program that would

95
00:03:43,239 --> 00:03:48,680
change a website so via web was a online

96
00:03:46,199 --> 00:03:50,598
store kind of like Shopify but way back

97
00:03:48,680 --> 00:03:52,680
in the day yeah it was basically like

98
00:03:50,598 --> 00:03:55,878
the first SAS app ever like like PG

99
00:03:52,680 --> 00:03:57,480
actually invented SAS in like 1995 it's

100
00:03:55,878 --> 00:03:59,239
just that those first SAS apps kind of

101
00:03:57,479 --> 00:04:00,878
suck because they didn't have XML HTTP

102
00:03:59,239 --> 00:04:02,280
request and so every time you would like

103
00:04:00,878 --> 00:04:04,598
click a button you would have to reload

104
00:04:02,280 --> 00:04:05,840
the whole page and so it's just a shitty

105
00:04:04,598 --> 00:04:08,639
experience and so it didn't really catch

106
00:04:05,840 --> 00:04:11,319
on until 2005 when X XML HTP request

107
00:04:08,639 --> 00:04:14,119
white spread anyway I I see this llm

108
00:04:11,318 --> 00:04:16,079
thing as like actually very similar um

109
00:04:14,120 --> 00:04:17,918
it's like it's a new Computing Paradigm

110
00:04:16,079 --> 00:04:20,720
that makes it possible to just like do

111
00:04:17,918 --> 00:04:23,758
something fundamentally different and in

112
00:04:20,720 --> 00:04:25,720
2005 when cloud and mobile finally took

113
00:04:23,759 --> 00:04:27,680
off there is this sort of like big open

114
00:04:25,720 --> 00:04:29,479
question of like okay well this new

115
00:04:27,680 --> 00:04:32,000
technology exist what should you do with

116
00:04:29,478 --> 00:04:33,279
it where is the value going to acrw

117
00:04:32,000 --> 00:04:35,120
where are the good opportunities for

118
00:04:33,279 --> 00:04:36,359
startups I was going through the list of

119
00:04:35,120 --> 00:04:38,360
like all the billion dollar companies

120
00:04:36,360 --> 00:04:41,720
that were created and I kind of had this

121
00:04:38,360 --> 00:04:43,840
realization that um you could kind of

122
00:04:41,720 --> 00:04:46,440
bucket the the different paths that

123
00:04:43,839 --> 00:04:48,439
people took into like three buckets um

124
00:04:46,439 --> 00:04:50,439
there's there's a first bucket that

125
00:04:48,439 --> 00:04:54,399
people started with which was like I

126
00:04:50,439 --> 00:04:57,360
would call them Obviously good ideas

127
00:04:54,399 --> 00:05:00,599
that could be Mass consumer products um

128
00:04:57,360 --> 00:05:03,199
so that's like docs photos email

129
00:05:00,600 --> 00:05:05,080
calendar chat all these things that like

130
00:05:03,199 --> 00:05:06,840
we used to do on our desktop with that

131
00:05:05,079 --> 00:05:10,279
obviously could be moved to the browser

132
00:05:06,839 --> 00:05:12,560
and mobile and the interesting thing is

133
00:05:10,279 --> 00:05:15,439
zero startups won in those categories

134
00:05:12,560 --> 00:05:17,120
100% of the value flow to incumbents

135
00:05:15,439 --> 00:05:19,478
right like Google Facebook Amazon they

136
00:05:17,120 --> 00:05:21,478
own all all those businesses folks

137
00:05:19,478 --> 00:05:23,279
forget that like Google Docs wasn't the

138
00:05:21,478 --> 00:05:24,680
only company that tried to bring

139
00:05:23,279 --> 00:05:26,439
Microsoft Office online there were like

140
00:05:24,680 --> 00:05:27,759
30 companies that tried to bring

141
00:05:26,439 --> 00:05:30,439
Microsoft Office online but they all

142
00:05:27,759 --> 00:05:33,240
lost Google one then there was a second

143
00:05:30,439 --> 00:05:36,000
category which was like Mass consumer

144
00:05:33,240 --> 00:05:39,439
ideas that were not obvious that nobody

145
00:05:36,000 --> 00:05:41,038
predicted um that's like uber instacart

146
00:05:39,439 --> 00:05:44,600
door Dash

147
00:05:41,038 --> 00:05:47,719
coinbase th Airbnb those ones those ones

148
00:05:44,600 --> 00:05:49,919
came out of left field like the the dot

149
00:05:47,720 --> 00:05:52,960
dot dot between XML HTTP request and

150
00:05:49,918 --> 00:05:54,318
Airbnb is like very not obvious yeah and

151
00:05:52,959 --> 00:05:55,799
so the incumbents didn't even try

152
00:05:54,319 --> 00:05:57,759
competing in those spaces until it was

153
00:05:55,800 --> 00:06:00,560
like too late and so startups are able

154
00:05:57,759 --> 00:06:02,560
to win there and then there's a third

155
00:06:00,560 --> 00:06:04,600
category which is all the B2B SAS

156
00:06:02,560 --> 00:06:08,120
companies and that's like 300 of them

157
00:06:04,600 --> 00:06:10,280
and so like Mo like by by by number of

158
00:06:08,120 --> 00:06:12,319
logos way more billion dollar companies

159
00:06:10,279 --> 00:06:14,478
were created in that third category than

160
00:06:12,319 --> 00:06:16,720
the first two I think one reason why

161
00:06:14,478 --> 00:06:19,758
that happened is like there is

162
00:06:16,720 --> 00:06:22,000
no like Microsoft of SAS like there is

163
00:06:19,759 --> 00:06:23,598
no company that somehow does like SAS

164
00:06:22,000 --> 00:06:25,800
for like every vertical and every

165
00:06:23,598 --> 00:06:27,159
product like for structural reasons it

166
00:06:25,800 --> 00:06:29,598
seems to be the case that like they're

167
00:06:27,160 --> 00:06:31,439
all different companies and that's why

168
00:06:29,598 --> 00:06:33,478
there so many of them I think Salesforce

169
00:06:31,439 --> 00:06:37,000
is probably like the first true SAS

170
00:06:33,478 --> 00:06:40,159
company um and i' I remember Mark benov

171
00:06:37,000 --> 00:06:42,560
coming to speak at YC and he tells the

172
00:06:40,160 --> 00:06:43,800
story as just very early on people just

173
00:06:42,560 --> 00:06:46,038
didn't believe you could build

174
00:06:43,800 --> 00:06:48,520
sophisticated Enterprise applications

175
00:06:46,038 --> 00:06:50,598
like over the cloud or via SAS it was

176
00:06:48,519 --> 00:06:51,959
just so um there was just like a

177
00:06:50,598 --> 00:06:53,639
perception issue right it was like no

178
00:06:51,959 --> 00:06:54,839
like you don't you buy like your box

179
00:06:53,639 --> 00:06:56,439
software and that's like the real

180
00:06:54,839 --> 00:06:58,119
software that you run the way we always

181
00:06:56,439 --> 00:07:00,079
do it it was it was quite contrar cuz

182
00:06:58,120 --> 00:07:02,160
the early web app sucked they were like

183
00:07:00,079 --> 00:07:03,719
via web where you had to be a Visionary

184
00:07:02,160 --> 00:07:04,800
like PG and understand that the browser

185
00:07:03,720 --> 00:07:06,720
was going to keep getting better and

186
00:07:04,800 --> 00:07:08,879
that eventually it' be good which feels

187
00:07:06,720 --> 00:07:10,440
like quite reminisent of today right

188
00:07:08,879 --> 00:07:11,720
where it's like the yeah the same thing

189
00:07:10,439 --> 00:07:13,199
like oh no like you won't be able to

190
00:07:11,720 --> 00:07:16,120
build like sophisticated Enterprise

191
00:07:13,199 --> 00:07:18,520
applications that use these llm or AI

192
00:07:16,120 --> 00:07:20,800
tools because they hallucinate or

193
00:07:18,519 --> 00:07:22,399
they're not perfect or they um they kind

194
00:07:20,800 --> 00:07:25,240
of like just toys but yeah that's like

195
00:07:22,399 --> 00:07:27,598
the early SAS story exactly the same and

196
00:07:25,240 --> 00:07:30,400
so when I think about the parallels with

197
00:07:27,598 --> 00:07:31,878
LMS I could easily imagine the same

198
00:07:30,399 --> 00:07:33,799
thing happening which is that there's a

199
00:07:31,879 --> 00:07:35,720
bunch of categories that are like Mass

200
00:07:33,800 --> 00:07:38,120
consumer applications that are obviously

201
00:07:35,720 --> 00:07:39,919
huge opportunities but probably the

202
00:07:38,120 --> 00:07:41,399
incumbents will win all of those so

203
00:07:39,918 --> 00:07:43,839
that's something like a per like a

204
00:07:41,399 --> 00:07:45,120
general purpose AI Voice Assistant that

205
00:07:43,839 --> 00:07:46,560
you you know you can ask it to do

206
00:07:45,120 --> 00:07:47,720
anything and it'll like go do that thing

207
00:07:46,560 --> 00:07:49,478
that's an obvious thing that should

208
00:07:47,720 --> 00:07:51,159
exist but like all the big players are

209
00:07:49,478 --> 00:07:53,120
going to be competing to be that thing

210
00:07:51,158 --> 00:07:55,199
right W Apple's a little slow on that

211
00:07:53,120 --> 00:07:58,199
one why is Siri so stupid still what

212
00:07:55,199 --> 00:07:59,759
year is it it makes no sense I mean it's

213
00:07:58,199 --> 00:08:02,800
like a count to that is like the very

214
00:07:59,759 --> 00:08:05,759
obvious thing is search and maybe Google

215
00:08:02,800 --> 00:08:08,158
will still win um on search but

216
00:08:05,759 --> 00:08:09,960
perplexity is definitely give them run

217
00:08:08,158 --> 00:08:11,639
for the money right yeah this is the

218
00:08:09,959 --> 00:08:12,758
classic innovators dilemma at the end of

219
00:08:11,639 --> 00:08:14,639
the day I mean you could argue going

220
00:08:12,759 --> 00:08:17,879
back to what you said about Uber or

221
00:08:14,639 --> 00:08:20,478
Airbnb these were actually really risky

222
00:08:17,879 --> 00:08:22,560
things from a regulatory standpoint so

223
00:08:20,478 --> 00:08:25,519
if you're Google and you have basically

224
00:08:22,560 --> 00:08:27,360
a guaranteed you know giant a pot of

225
00:08:25,519 --> 00:08:29,519
gold that you know sort of comes to you

226
00:08:27,360 --> 00:08:31,520
every single month like why would you

227
00:08:29,519 --> 00:08:33,759
endanger that pot of gold to sort of

228
00:08:31,519 --> 00:08:36,079
pursue these things that uh might be

229
00:08:33,759 --> 00:08:37,278
scary or might ruin the pot of gold I

230
00:08:36,080 --> 00:08:38,879
think that's I think that's like

231
00:08:37,278 --> 00:08:41,038
probably the primary reason why the

232
00:08:38,879 --> 00:08:42,479
incumbents didn't end up building those

233
00:08:41,038 --> 00:08:44,838
products and didn't even clone them even

234
00:08:42,479 --> 00:08:45,959
after they got big and it was obvious

235
00:08:44,839 --> 00:08:47,839
that they were going to work who will

236
00:08:45,958 --> 00:08:50,119
never launch an an Uber clone they never

237
00:08:47,839 --> 00:08:53,640
launch an Airbnb clone um I was

238
00:08:50,120 --> 00:08:54,799
listening to this uh talk by Travis and

239
00:08:53,639 --> 00:08:56,958
one of the things that he said that

240
00:08:54,799 --> 00:08:58,799
really stuck with me is that in the in

241
00:08:56,958 --> 00:08:59,799
the first years of uber he was very

242
00:08:58,799 --> 00:09:01,838
scared that he would was going to

243
00:08:59,799 --> 00:09:03,879
personally go to prison for like a long

244
00:09:01,839 --> 00:09:05,200
time like he was actually personally

245
00:09:03,879 --> 00:09:07,679
risking going to prison in order to

246
00:09:05,200 --> 00:09:09,600
build that company and so yeah no highly

247
00:09:07,679 --> 00:09:12,958
paid Google exactly was going to do that

248
00:09:09,600 --> 00:09:15,200
what do you think about um why the

249
00:09:12,958 --> 00:09:17,159
incumbents didn't go into B2B SAS is it

250
00:09:15,200 --> 00:09:20,000
part of the reason is that a lot of the

251
00:09:17,159 --> 00:09:22,519
use cases are very there's a very wide

252
00:09:20,000 --> 00:09:24,720
distribution I it's a great question I

253
00:09:22,519 --> 00:09:28,720
love to hear what you guys think my take

254
00:09:24,720 --> 00:09:31,000
is that it's just too hard to do that

255
00:09:28,720 --> 00:09:33,160
many things as a company like each B2B

256
00:09:31,000 --> 00:09:34,958
SAS company really requires like the

257
00:09:33,159 --> 00:09:37,439
people who are running the product in

258
00:09:34,958 --> 00:09:39,639
the business to be extremely deep in one

259
00:09:37,440 --> 00:09:42,120
domain and care very deeply about a lot

260
00:09:39,639 --> 00:09:44,000
of really obscure issues you know like

261
00:09:42,120 --> 00:09:45,679
take like Gusto for example like why

262
00:09:44,000 --> 00:09:47,120
didn't Google build a Gusto competitor

263
00:09:45,679 --> 00:09:48,838
well there's no one to Google who really

264
00:09:47,120 --> 00:09:50,959
understands payroll and has the patience

265
00:09:48,839 --> 00:09:52,959
to like deal with all the nuances of all

266
00:09:50,958 --> 00:09:55,559
these like stupid payroll regulations

267
00:09:52,958 --> 00:09:57,159
and like it's just like like it's just

268
00:09:55,559 --> 00:09:59,399
not worth it for them it's easier for

269
00:09:57,159 --> 00:10:02,199
them to just focus on like a few really

270
00:09:59,399 --> 00:10:04,159
huge categories in the B2B SAS world

271
00:10:02,200 --> 00:10:06,160
it's it's sort of about the unbundling

272
00:10:04,159 --> 00:10:09,039
bundling of software argument that comes

273
00:10:06,159 --> 00:10:11,399
up a lot as well I think and why didn't

274
00:10:09,039 --> 00:10:14,719
why did all these vertical B2B SAS

275
00:10:11,399 --> 00:10:18,759
products evolve versus just like Oracle

276
00:10:14,720 --> 00:10:21,920
or sap or um netet yeah netu just owning

277
00:10:18,759 --> 00:10:23,240
like everything um and I think it might

278
00:10:21,919 --> 00:10:25,519
be Al is another thing that's

279
00:10:23,240 --> 00:10:28,278
attributable to the shift to like SASS

280
00:10:25,519 --> 00:10:29,799
and the internet is in the old ways of

281
00:10:28,278 --> 00:10:31,759
selling software again like you had this

282
00:10:29,799 --> 00:10:34,199
box software that was really like

283
00:10:31,759 --> 00:10:35,919
expensive to install and you had like a

284
00:10:34,200 --> 00:10:38,079
whole ecosystem around it and anytime

285
00:10:35,919 --> 00:10:40,000
you wanted something custom like the

286
00:10:38,078 --> 00:10:41,679
integrators would just say oh no like we

287
00:10:40,000 --> 00:10:44,278
can like just build a UA custom like

288
00:10:41,679 --> 00:10:46,278
payroll feature or something like that

289
00:10:44,278 --> 00:10:48,679
and then Salesforce comes along with

290
00:10:46,278 --> 00:10:50,480
like a SAS solution and it just seems

291
00:10:48,679 --> 00:10:52,479
like it could never be as powerful or

292
00:10:50,480 --> 00:10:54,200
sophisticated as like the expensive

293
00:10:52,480 --> 00:10:56,759
Enterprise installation you just paid

294
00:10:54,200 --> 00:10:58,200
for but they Prov that it totally was

295
00:10:56,759 --> 00:11:00,278
the case and I think that just like

296
00:10:58,200 --> 00:11:02,600
opened the gates for all all of these

297
00:11:00,278 --> 00:11:04,399
like vertical sash solutions to emerge

298
00:11:02,600 --> 00:11:05,959
doing exactly what you're saying the

299
00:11:04,399 --> 00:11:08,278
other problem is that with a lot of this

300
00:11:05,958 --> 00:11:10,958
enterprise software if you're a user of

301
00:11:08,278 --> 00:11:13,600
Oracle and a netw suite because they're

302
00:11:10,958 --> 00:11:15,799
they have to cover so much ground the

303
00:11:13,600 --> 00:11:17,120
user experience is actually pretty bad

304
00:11:15,799 --> 00:11:19,359
they're trying to be jack of all trades

305
00:11:17,120 --> 00:11:21,078
but master of none yeah so ends up being

306
00:11:19,360 --> 00:11:24,320
a bit of a kitchen sync type of

307
00:11:21,078 --> 00:11:27,399
experience and this is where if you go

308
00:11:24,320 --> 00:11:29,560
and build a B2B SAS vertical company you

309
00:11:27,399 --> 00:11:31,519
could do literally a 10x better

310
00:11:29,559 --> 00:11:33,278
experience and more delightful because

311
00:11:31,519 --> 00:11:35,720
there's this Stark difference between

312
00:11:33,278 --> 00:11:37,600
consumer products and Enterprise user

313
00:11:35,720 --> 00:11:40,360
experience yeah well there's only uh

314
00:11:37,600 --> 00:11:45,079
what three price points in software it's

315
00:11:40,360 --> 00:11:48,278
uh $5 per seat $500 per seat or $55,000

316
00:11:45,078 --> 00:11:51,919
per seat and uh that Maps directly to

317
00:11:48,278 --> 00:11:54,278
Consumer SMB or Enterprise sales and

318
00:11:51,919 --> 00:11:56,719
then I think time in Memorial has taught

319
00:11:54,278 --> 00:11:58,919
us that in the past and this is less and

320
00:11:56,720 --> 00:12:00,000
less true uh with new software

321
00:11:58,919 --> 00:12:02,838
thankfully

322
00:12:00,000 --> 00:12:05,320
but Enterprise is terrible software

323
00:12:02,839 --> 00:12:08,040
because it's not the user buying it you

324
00:12:05,320 --> 00:12:10,639
know some high up Mucky muuk inside

325
00:12:08,039 --> 00:12:13,399
Fortune 1000 is the person who's getting

326
00:12:10,639 --> 00:12:16,399
whin and D for this you know Mega seven

327
00:12:13,399 --> 00:12:18,278
figure contract and you know they're

328
00:12:16,399 --> 00:12:20,839
going to choose something that maybe

329
00:12:18,278 --> 00:12:22,559
isn't that good actually for the end

330
00:12:20,839 --> 00:12:25,760
user the person who has to actually use

331
00:12:22,559 --> 00:12:28,439
the software day-to-day and um I'm sort

332
00:12:25,759 --> 00:12:31,958
of curious to see how this changes with

333
00:12:28,440 --> 00:12:33,760
llms actually I mean to date one of the

334
00:12:31,958 --> 00:12:35,638
more Salient things that we've seen for

335
00:12:33,759 --> 00:12:37,919
both SMB and enterprise software

336
00:12:35,639 --> 00:12:40,000
companies is that or all software

337
00:12:37,919 --> 00:12:42,278
companies all startups period is like

338
00:12:40,000 --> 00:12:44,559
you know there's a sense that as Revenue

339
00:12:42,278 --> 00:12:47,000
scales the number of people you have to

340
00:12:44,559 --> 00:12:51,159
hire scales with it and so when you look

341
00:12:47,000 --> 00:12:54,278
at unicorns uh even in today's YC

342
00:12:51,159 --> 00:12:56,198
portfolio uh it's quite routine to see a

343
00:12:54,278 --> 00:12:58,399
company that reached a hundred or $200

344
00:12:56,198 --> 00:13:02,639
million a year in Revenue but they have

345
00:12:58,399 --> 00:13:04,679
like 500 a th000 2,000 employees already

346
00:13:02,639 --> 00:13:06,278
and I'm just going to be very curious

347
00:13:04,679 --> 00:13:08,599
like uh even the advice that I'm

348
00:13:06,278 --> 00:13:11,159
starting to give companies that are you

349
00:13:08,600 --> 00:13:13,000
know a month or two out of the batch uh

350
00:13:11,159 --> 00:13:14,480
it's a it's feeling a little bit

351
00:13:13,000 --> 00:13:17,240
different than the kind of advice I

352
00:13:14,480 --> 00:13:20,079
would give last year or two years ago in

353
00:13:17,240 --> 00:13:23,839
the past you might say you know let me

354
00:13:20,078 --> 00:13:26,039
find the absolute smartest person uh in

355
00:13:23,839 --> 00:13:28,240
all of these other parts of the org like

356
00:13:26,039 --> 00:13:29,639
customer success or sales or different

357
00:13:28,240 --> 00:13:31,320
things like that

358
00:13:29,639 --> 00:13:33,919
and uh I want to find someone who I've

359
00:13:31,320 --> 00:13:35,320
worked with who is I know is great and

360
00:13:33,919 --> 00:13:37,360
then I'm going to go sit on their you

361
00:13:35,320 --> 00:13:39,399
know uh on their doorstep until they

362
00:13:37,360 --> 00:13:41,440
quit their jobs and come work for me and

363
00:13:39,399 --> 00:13:43,679
I want them to be someone who can you

364
00:13:41,440 --> 00:13:45,880
know build a team for me hire a lot of

365
00:13:43,679 --> 00:13:48,559
people that might still be true but I'm

366
00:13:45,879 --> 00:13:50,278
starting to sense that uh the Met is

367
00:13:48,559 --> 00:13:53,198
shifting a little bit like you actually

368
00:13:50,278 --> 00:13:56,078
might want to hire more really good

369
00:13:53,198 --> 00:13:59,120
software Engineers who understand large

370
00:13:56,078 --> 00:14:01,039
language models uh who can actually

371
00:13:59,120 --> 00:14:02,919
automate the specific things that you

372
00:14:01,039 --> 00:14:06,078
need that are the bottlenecks to your

373
00:14:02,919 --> 00:14:08,000
growth and so it might result in you

374
00:14:06,078 --> 00:14:10,278
know a very subtle but you know

375
00:14:08,000 --> 00:14:11,879
significant change in the way startups

376
00:14:10,278 --> 00:14:14,039
grow their businesses sort of post-

377
00:14:11,879 --> 00:14:16,799
product Market fit it means that I'm

378
00:14:14,039 --> 00:14:19,278
going to build llm systems that bring

379
00:14:16,799 --> 00:14:21,719
down my costs that cause me not to have

380
00:14:19,278 --> 00:14:22,679
to hire a thousand people I think we're

381
00:14:21,720 --> 00:14:24,800
right at the beginning of that

382
00:14:22,679 --> 00:14:26,519
Revolution right now I mean we talked

383
00:14:24,799 --> 00:14:29,679
about this in a previous episode we

384
00:14:26,519 --> 00:14:31,759
talked about there will be a future

385
00:14:29,679 --> 00:14:33,359
unicorn company that's only run if we

386
00:14:31,759 --> 00:14:36,120
take it to the limit with only 10

387
00:14:33,360 --> 00:14:37,480
employees that's completely plausible

388
00:14:36,120 --> 00:14:39,720
and they're writing the evals and the

389
00:14:37,480 --> 00:14:40,920
prompts does it I think what you're

390
00:14:39,720 --> 00:14:43,800
saying is like a trend that was already

391
00:14:40,919 --> 00:14:46,319
underway pre llms like I remember when I

392
00:14:43,799 --> 00:14:49,198
was running triple bite for example we

393
00:14:46,320 --> 00:14:51,920
needed to like build marketing or cust

394
00:14:49,198 --> 00:14:54,078
like user acquisition basically um and

395
00:14:51,919 --> 00:14:55,360
especially after we raised a series B

396
00:14:54,078 --> 00:14:57,399
the like traditional way you were

397
00:14:55,360 --> 00:14:59,159
supposed to do that is to like hire a

398
00:14:57,399 --> 00:15:01,799
marketing executive and build out like

399
00:14:59,159 --> 00:15:03,240
marketing team and um and just like

400
00:15:01,799 --> 00:15:06,159
basically spin up this machine to do

401
00:15:03,240 --> 00:15:10,039
like sales and marketing but I'd

402
00:15:06,159 --> 00:15:11,360
actually met like a y founder um Mike

403
00:15:10,039 --> 00:15:13,278
who was his company was basically

404
00:15:11,360 --> 00:15:15,278
building like a smart frying pan sounds

405
00:15:13,278 --> 00:15:17,480
like bizarre but like he was a MIT

406
00:15:15,278 --> 00:15:19,399
engineer yeah you remember this um he's

407
00:15:17,480 --> 00:15:21,199
an MIT engineer and to sell the smart

408
00:15:19,399 --> 00:15:22,399
frying pan he had to get really really

409
00:15:21,198 --> 00:15:25,359
good at understanding like paid

410
00:15:22,399 --> 00:15:26,759
advertising and um uh Google ads and

411
00:15:25,360 --> 00:15:28,560
just a whole bunch of stuff and so he he

412
00:15:26,759 --> 00:15:29,959
taken this Engineers mindset approach to

413
00:15:28,559 --> 00:15:32,958
it and remember just talking to him

414
00:15:29,958 --> 00:15:34,719
about it and realizing this would be so

415
00:15:32,958 --> 00:15:37,559
much better to have an MIT engineer

416
00:15:34,720 --> 00:15:39,079
working on like our marketing efforts

417
00:15:37,559 --> 00:15:41,559
than any of the marketing candidates

418
00:15:39,078 --> 00:15:43,599
I've spoken to and he was able to like

419
00:15:41,559 --> 00:15:45,159
scale us up to like we were spending

420
00:15:43,600 --> 00:15:47,680
like 1. like a million dollars a month

421
00:15:45,159 --> 00:15:49,240
on just marketing and various like

422
00:15:47,679 --> 00:15:51,479
campaigns and triple bite had great

423
00:15:49,240 --> 00:15:53,759
marketing like I remember like the Cal

424
00:15:51,480 --> 00:15:55,240
trained station takeover that you did

425
00:15:53,759 --> 00:15:57,159
all the like out of home stuff that you

426
00:15:55,240 --> 00:15:59,240
did it was like really high quality

427
00:15:57,159 --> 00:16:00,759
stuff it stuck with it you could tell

428
00:15:59,240 --> 00:16:03,680
was not being done by some like VP

429
00:16:00,759 --> 00:16:05,800
marketing person um and that was all mik

430
00:16:03,679 --> 00:16:07,198
and like the comment I would often get

431
00:16:05,799 --> 00:16:08,799
when people would ask me around that

432
00:16:07,198 --> 00:16:11,838
time like how big is triple bite and we

433
00:16:08,799 --> 00:16:12,958
were like 50 people and so much yeah

434
00:16:11,839 --> 00:16:14,319
yeah people be like I thought this's

435
00:16:12,958 --> 00:16:15,879
like hundreds of people I was like no

436
00:16:14,318 --> 00:16:17,639
it's all because if you put a really

437
00:16:15,879 --> 00:16:20,240
smart engineer on some of these like

438
00:16:17,639 --> 00:16:22,519
tasks they just find ways to make they

439
00:16:20,240 --> 00:16:23,959
find leverage and now like llms can go

440
00:16:22,519 --> 00:16:25,959
even Way Beyond like The Leverage you

441
00:16:23,958 --> 00:16:28,559
had which is pure software okay so

442
00:16:25,958 --> 00:16:31,359
here's my pitch for 300 vertical AI

443
00:16:28,559 --> 00:16:33,838
agent unicorns literally every company

444
00:16:31,360 --> 00:16:36,480
that is a SAS unicorn you could imagine

445
00:16:33,839 --> 00:16:39,480
there's a vertical AI unicorn equivalent

446
00:16:36,480 --> 00:16:42,240
in like some new universe cuz like most

447
00:16:39,480 --> 00:16:44,079
of these SAS unicorns beforehand there

448
00:16:42,240 --> 00:16:45,639
were some like box software company that

449
00:16:44,078 --> 00:16:47,198
was making the same thing that got

450
00:16:45,639 --> 00:16:49,000
disrupted by a SAS company and you could

451
00:16:47,198 --> 00:16:51,000
easily imagine the same thing happening

452
00:16:49,000 --> 00:16:53,360
again where now basically every every

453
00:16:51,000 --> 00:16:55,919
SAS company builds some software that

454
00:16:53,360 --> 00:16:57,199
some group of people use the vertical AI

455
00:16:55,919 --> 00:17:00,198
equivalent is just going to be the

456
00:16:57,198 --> 00:17:01,799
software plus the people in one product

457
00:17:00,198 --> 00:17:03,318
one thing might be just Enterprises in

458
00:17:01,799 --> 00:17:05,678
general right now are a little unsure

459
00:17:03,318 --> 00:17:07,519
about what exactly they like what agents

460
00:17:05,679 --> 00:17:08,880
they need and one approach I've seen

461
00:17:07,519 --> 00:17:11,439
from especially more experienced

462
00:17:08,880 --> 00:17:13,679
Founders like um Brett Taylor the CTO of

463
00:17:11,439 --> 00:17:15,400
Facebook started his company Sierra I

464
00:17:13,679 --> 00:17:17,919
don't know all the details but as far as

465
00:17:15,400 --> 00:17:20,798
I can tell it's essentially more like

466
00:17:17,919 --> 00:17:23,759
broadly about letting Enterprises like

467
00:17:20,798 --> 00:17:26,000
deploy these AI agents and spinning them

468
00:17:23,759 --> 00:17:27,480
up like custom for the Enterprise versus

469
00:17:26,000 --> 00:17:29,440
like oh hey we have like this specific

470
00:17:27,480 --> 00:17:31,759
agent to do this it's something I've

471
00:17:29,440 --> 00:17:33,679
seen from one of my companies called um

472
00:17:31,759 --> 00:17:36,720
uh Vector shift that funded about a year

473
00:17:33,679 --> 00:17:39,798
ago they're two really smart like

474
00:17:36,720 --> 00:17:41,120
Harvard computer scientists and it's a

475
00:17:39,798 --> 00:17:42,519
that what they found is that they're

476
00:17:41,119 --> 00:17:44,599
trying to build a platform to make it

477
00:17:42,519 --> 00:17:47,240
easy for Enterprises to build their own

478
00:17:44,599 --> 00:17:50,678
like use like no code or sdks to build

479
00:17:47,240 --> 00:17:53,038
their own like um internal llm powered

480
00:17:50,679 --> 00:17:54,600
agents but like Enterprises often don't

481
00:17:53,038 --> 00:17:57,400
know exactly what they want to use these

482
00:17:54,599 --> 00:17:59,759
things for and so bring it back I wonder

483
00:17:57,400 --> 00:18:01,080
if like in like the box software world

484
00:17:59,759 --> 00:18:02,759
you started off with just like a few

485
00:18:01,079 --> 00:18:04,399
vendors who just basically were trying

486
00:18:02,759 --> 00:18:06,079
to convince people to use software at

487
00:18:04,400 --> 00:18:08,320
all and it was just like it does

488
00:18:06,079 --> 00:18:09,839
everything um and then it gets more

489
00:18:08,319 --> 00:18:12,558
sophisticated and higher resolution and

490
00:18:09,839 --> 00:18:14,558
you get lots of like vertical SS players

491
00:18:12,558 --> 00:18:16,399
we go through that same period with llms

492
00:18:14,558 --> 00:18:18,240
where the early winners might just be

493
00:18:16,400 --> 00:18:20,120
these like general purpose hey like we

494
00:18:18,240 --> 00:18:22,679
like make it easy for you to do llm

495
00:18:20,119 --> 00:18:25,119
stuff and then it the vertical agents

496
00:18:22,679 --> 00:18:26,480
will come in over time or do you think

497
00:18:25,119 --> 00:18:28,639
there's reasons it's different now and

498
00:18:26,480 --> 00:18:29,400
the vertical agents will take off on day

499
00:18:28,640 --> 00:18:30,840
one

500
00:18:29,400 --> 00:18:32,840
yeah that's interesting because if you

501
00:18:30,839 --> 00:18:35,359
think about the history of SAS the

502
00:18:32,839 --> 00:18:38,480
consumer things worked first like 2005

503
00:18:35,359 --> 00:18:41,038
to 2010 was mostly consumer applications

504
00:18:38,480 --> 00:18:43,599
like email and chat and maps and people

505
00:18:41,038 --> 00:18:45,319
got people as individuals got used to

506
00:18:43,599 --> 00:18:48,079
using these tools themselves and I think

507
00:18:45,319 --> 00:18:49,798
that made it easier to sell SAS tools to

508
00:18:48,079 --> 00:18:51,798
companies because you know the same

509
00:18:49,798 --> 00:18:53,480
people are both employees and consumers

510
00:18:51,798 --> 00:18:54,879
yeah I I think the answer might just be

511
00:18:53,480 --> 00:18:58,279
like this is this is all just a

512
00:18:54,880 --> 00:19:00,360
continuation of software and just

513
00:18:58,279 --> 00:19:02,200
there's no reason it has to reset back

514
00:19:00,359 --> 00:19:05,000
like llms don't have to reset back to a

515
00:19:02,200 --> 00:19:07,038
few general purpose like Enterprise llm

516
00:19:05,000 --> 00:19:08,798
platforms doing everything because

517
00:19:07,038 --> 00:19:11,519
Enterprises have already been trained on

518
00:19:08,798 --> 00:19:14,400
like the value of Point Solutions and

519
00:19:11,519 --> 00:19:15,359
vertical Solutions um and like the user

520
00:19:14,400 --> 00:19:16,480
experience not going to be that

521
00:19:15,359 --> 00:19:19,879
different these things will just be a

522
00:19:16,480 --> 00:19:21,079
lot more powerful and so if Enterprises

523
00:19:19,880 --> 00:19:23,039
have already built the muscle of

524
00:19:21,079 --> 00:19:25,439
believing that like startups or vertical

525
00:19:23,038 --> 00:19:28,200
Solutions can be better than like Legacy

526
00:19:25,440 --> 00:19:31,240
broad platforms they are probably going

527
00:19:28,200 --> 00:19:33,360
to be willing to take a bet on a startup

528
00:19:31,240 --> 00:19:34,880
promising a very good vertical AI agent

529
00:19:33,359 --> 00:19:36,599
solution today and I feel like we're all

530
00:19:34,880 --> 00:19:39,000
seeing that in the batch now with some

531
00:19:36,599 --> 00:19:41,480
of our companies are getting faster

532
00:19:39,000 --> 00:19:44,279
Traction in Enterprises for these

533
00:19:41,480 --> 00:19:45,720
vertical AI agents than like we've ever

534
00:19:44,279 --> 00:19:47,960
seen before I think we're just early in

535
00:19:45,720 --> 00:19:50,798
the game right like all software sort of

536
00:19:47,960 --> 00:19:52,679
starts quite vertical and then as the

537
00:19:50,798 --> 00:19:56,279
industries actually get much more

538
00:19:52,679 --> 00:19:58,240
developed um then I mean I just answered

539
00:19:56,279 --> 00:19:59,798
my earlier question it's like you know

540
00:19:58,240 --> 00:20:02,640
why does company end up having a

541
00:19:59,798 --> 00:20:04,519
thousand employees it's actually that uh

542
00:20:02,640 --> 00:20:06,320
you know early early in the game

543
00:20:04,519 --> 00:20:08,558
everyone's making these specific point

544
00:20:06,319 --> 00:20:11,279
Solutions and then at some point you've

545
00:20:08,558 --> 00:20:13,319
got to go horizontal like you're already

546
00:20:11,279 --> 00:20:15,759
doing this crazy spend on sales and

547
00:20:13,319 --> 00:20:18,079
marketing and then the only way you can

548
00:20:15,759 --> 00:20:20,720
actually continue to grow once you sort

549
00:20:18,079 --> 00:20:22,399
of get 100% or you know some large

550
00:20:20,720 --> 00:20:25,480
majority of the market is you actually

551
00:20:22,400 --> 00:20:27,759
have to do like not just a point

552
00:20:25,480 --> 00:20:32,000
solution but things that sort of work

553
00:20:27,759 --> 00:20:33,759
together the other point of why the bull

554
00:20:32,000 --> 00:20:36,798
case for vertical AI agents could be

555
00:20:33,759 --> 00:20:39,440
even bigger than SAS is that SAS you

556
00:20:36,798 --> 00:20:41,558
still needed a operations team or set of

557
00:20:39,440 --> 00:20:43,038
people to operate the software in order

558
00:20:41,558 --> 00:20:44,798
to get all the workflows to be done I

559
00:20:43,038 --> 00:20:47,798
don't know approval workflows or you

560
00:20:44,798 --> 00:20:50,440
have to input the data the argument here

561
00:20:47,798 --> 00:20:52,000
is that you will get not only replacing

562
00:20:50,440 --> 00:20:54,519
all that set of SAS software so that

563
00:20:52,000 --> 00:20:56,400
would be like one to one mapping but is

564
00:20:54,519 --> 00:20:58,679
also going to eat all of the a lot of

565
00:20:56,400 --> 00:21:00,880
the payroll because we look a lot of the

566
00:20:58,679 --> 00:21:02,798
spend for companies big chunk is still a

567
00:21:00,880 --> 00:21:04,200
payroll and software's Tiny exactly they

568
00:21:02,798 --> 00:21:06,000
spend way more on employees than they do

569
00:21:04,200 --> 00:21:07,480
on software so it'll be these smaller

570
00:21:06,000 --> 00:21:11,640
companies that way more efficient that

571
00:21:07,480 --> 00:21:14,440
need way less humans to do random data

572
00:21:11,640 --> 00:21:16,200
entry or approvals or click the software

573
00:21:14,440 --> 00:21:18,080
I agree I think it's very possible the

574
00:21:16,200 --> 00:21:20,480
vertical equivalence will be 10 times as

575
00:21:18,079 --> 00:21:22,399
large as the SAS company that they are

576
00:21:20,480 --> 00:21:24,120
disrupting I mean there there's two case

577
00:21:22,400 --> 00:21:26,440
it could be that the vertical Point

578
00:21:24,119 --> 00:21:28,639
solution could be just big enough and

579
00:21:26,440 --> 00:21:30,440
you don't need to do that bro breath

580
00:21:28,640 --> 00:21:32,440
thing right it that could be a nice

581
00:21:30,440 --> 00:21:34,400
scenario should we give some examples I

582
00:21:32,440 --> 00:21:37,038
feel like we've all been working with so

583
00:21:34,400 --> 00:21:38,640
many vertical AI agent companies we've

584
00:21:37,038 --> 00:21:41,480
got like news from the

585
00:21:38,640 --> 00:21:44,000
front how it's actually going well your

586
00:21:41,480 --> 00:21:46,519
former uh head of product Aaron Cannon

587
00:21:44,000 --> 00:21:48,880
is working on a YC company called outset

588
00:21:46,519 --> 00:21:51,960
that I worked with and uh basically

589
00:21:48,880 --> 00:21:54,120
they're taking llms uh to the surveys

590
00:21:51,960 --> 00:21:56,558
and qual Trix space so qual Trix is

591
00:21:54,119 --> 00:21:58,918
almost certainly not really going to

592
00:21:56,558 --> 00:22:00,798
build the best of breed uh large

593
00:21:58,919 --> 00:22:02,400
language model with reasoning and then

594
00:22:00,798 --> 00:22:04,720
the funny thing about surveys is you

595
00:22:02,400 --> 00:22:06,600
know who's it actually for it's for

596
00:22:04,720 --> 00:22:08,600
people who run products for marketing

597
00:22:06,599 --> 00:22:10,319
teams it's for people who are trying to

598
00:22:08,599 --> 00:22:12,719
make sense of like what do our customers

599
00:22:10,319 --> 00:22:15,480
actually want and what are surveys like

600
00:22:12,720 --> 00:22:18,120
guess what that's language so um and

601
00:22:15,480 --> 00:22:20,240
then I feel like these types of

602
00:22:18,119 --> 00:22:23,639
businesses um actually have to thread

603
00:22:20,240 --> 00:22:27,278
this needle um because Enterprise and

604
00:22:23,640 --> 00:22:29,440
SMB software often is sold based on a

605
00:22:27,278 --> 00:22:32,400
particular person who who is the key

606
00:22:29,440 --> 00:22:34,798
decision maker and um you have to go

607
00:22:32,400 --> 00:22:36,320
high enough in the organization so that

608
00:22:34,798 --> 00:22:38,558
the people you're selling to are not

609
00:22:36,319 --> 00:22:40,200
afraid that their whole their job Andor

610
00:22:38,558 --> 00:22:42,759
their whole team's job is going to go

611
00:22:40,200 --> 00:22:45,038
away totally that's kind of the move

612
00:22:42,759 --> 00:22:46,278
that I seen that a lot of companies that

613
00:22:45,038 --> 00:22:48,400
sell need to do because if you're going

614
00:22:46,278 --> 00:22:50,720
to go and sell to the team that's going

615
00:22:48,400 --> 00:22:53,759
to get replaced by AI they're going to

616
00:22:50,720 --> 00:22:56,600
sabotage it man it just does not work so

617
00:22:53,759 --> 00:22:58,079
I think this is an interesting way that

618
00:22:56,599 --> 00:22:59,839
a lot of these are top down and you have

619
00:22:58,079 --> 00:23:02,480
to go through at some point even get the

620
00:22:59,839 --> 00:23:04,759
CEO to sign off on it a company I'm

621
00:23:02,480 --> 00:23:06,798
working with u MCH that's sort of

622
00:23:04,759 --> 00:23:08,000
essentially an AI agent but for at least

623
00:23:06,798 --> 00:23:11,158
where they're starting is like QA

624
00:23:08,000 --> 00:23:13,079
testing um they're getting really great

625
00:23:11,159 --> 00:23:15,760
traction right now and it's interesting

626
00:23:13,079 --> 00:23:17,960
because you remember a decade ago um why

627
00:23:15,759 --> 00:23:21,079
can't we worked with rainforest QA like

628
00:23:17,960 --> 00:23:24,519
rainforest was a QA as a service company

629
00:23:21,079 --> 00:23:26,399
and that they had this exact tension of

630
00:23:24,519 --> 00:23:29,558
where they couldn't actually replace

631
00:23:26,400 --> 00:23:31,600
your QA team and so they needed to build

632
00:23:29,558 --> 00:23:33,440
software that made the QA te more like

633
00:23:31,599 --> 00:23:34,639
efficient but really that obviously

634
00:23:33,440 --> 00:23:35,840
meant trying to replace as many of them

635
00:23:34,640 --> 00:23:39,000
as possible they couldn't replace the

636
00:23:35,839 --> 00:23:40,798
whole team and so they were always on

637
00:23:39,000 --> 00:23:42,319
the sort of like tight rope between

638
00:23:40,798 --> 00:23:43,759
trying to sell the software to like the

639
00:23:42,319 --> 00:23:47,079
head of engineering as like this will

640
00:23:43,759 --> 00:23:48,400
mean you'll need less QA people and

641
00:23:47,079 --> 00:23:49,960
great but then you also have to go sell

642
00:23:48,400 --> 00:23:51,400
that to the QA team who don't want to be

643
00:23:49,960 --> 00:23:53,278
replaced and so I think that was always

644
00:23:51,400 --> 00:23:56,200
like a friction for that business for

645
00:23:53,278 --> 00:23:59,038
how it could like scale and grow but now

646
00:23:56,200 --> 00:24:01,080
like mtic with AI can actually just

647
00:23:59,038 --> 00:24:02,960
replace the QA people so their pitch is

648
00:24:01,079 --> 00:24:04,319
not oh this like makes your QA people

649
00:24:02,960 --> 00:24:06,400
faster it's like this just means you

650
00:24:04,319 --> 00:24:07,918
don't need a QA team at all so they can

651
00:24:06,400 --> 00:24:09,960
just focus the sell onto like

652
00:24:07,919 --> 00:24:12,120
engineering and Engineering doesn't need

653
00:24:09,960 --> 00:24:14,440
buying from QA at this point and you can

654
00:24:12,119 --> 00:24:16,000
also go in I mean to start with you can

655
00:24:14,440 --> 00:24:18,080
go and sell to companies that don't even

656
00:24:16,000 --> 00:24:19,640
have big QA teams at the moment they

657
00:24:18,079 --> 00:24:21,000
just use something like mtic and then it

658
00:24:19,640 --> 00:24:22,120
will just like keep scaling with them

659
00:24:21,000 --> 00:24:25,480
scaling and they'll just never build a

660
00:24:22,119 --> 00:24:26,959
QA team ever yes that is a real life

661
00:24:25,480 --> 00:24:28,759
case study of what Diana was saying

662
00:24:26,960 --> 00:24:30,000
about why these vertical AI agent

663
00:24:28,759 --> 00:24:31,960
companies going to be 10 times as big as

664
00:24:30,000 --> 00:24:33,880
the SAS companies yeah I'm seeing this

665
00:24:31,960 --> 00:24:35,440
interesting now um like in recruiting

666
00:24:33,880 --> 00:24:38,720
too I had this exact same issue with

667
00:24:35,440 --> 00:24:40,278
triple vet where to build the software

668
00:24:38,720 --> 00:24:41,720
um to build software that makes it easy

669
00:24:40,278 --> 00:24:43,278
to like screen and hire software

670
00:24:41,720 --> 00:24:45,200
Engineers you need buying from both the

671
00:24:43,278 --> 00:24:47,000
engineering team that they're joining

672
00:24:45,200 --> 00:24:48,159
but also the recruiting team and

673
00:24:47,000 --> 00:24:49,519
effectively the software we were

674
00:24:48,159 --> 00:24:51,480
building was trying to replace the

675
00:24:49,519 --> 00:24:54,720
recruiters but we couldn't completely

676
00:24:51,480 --> 00:24:57,960
replace the recruiters but now with NYC

677
00:24:54,720 --> 00:24:59,278
and so the recruiters were always like

678
00:24:57,960 --> 00:25:01,000
oppos

679
00:24:59,278 --> 00:25:03,038
opposing it CU it was a threat to them

680
00:25:01,000 --> 00:25:06,440
yeah so it just always like friction on

681
00:25:03,038 --> 00:25:08,319
like how um on like how far you can get

682
00:25:06,440 --> 00:25:11,360
when the customer you're trying to sell

683
00:25:08,319 --> 00:25:13,918
to is worried about being replaced um

684
00:25:11,359 --> 00:25:15,959
but yeah I think now it's still early

685
00:25:13,919 --> 00:25:18,720
days but now with AI you can build

686
00:25:15,960 --> 00:25:20,319
things that do the whole stack like of

687
00:25:18,720 --> 00:25:22,120
recruiting we have a company we worked

688
00:25:20,319 --> 00:25:23,639
with last batch like Nico work with them

689
00:25:22,119 --> 00:25:25,239
a priora which is actually just doing

690
00:25:23,640 --> 00:25:27,240
like the full like technical screen the

691
00:25:25,240 --> 00:25:29,200
full initial recruiter screen and

692
00:25:27,240 --> 00:25:31,480
getting great traction so I think as

693
00:25:29,200 --> 00:25:32,679
those things keep going like they won't

694
00:25:31,480 --> 00:25:33,798
have they have the same thing you won't

695
00:25:32,679 --> 00:25:35,480
have the friction of oh I need to

696
00:25:33,798 --> 00:25:37,639
convince recruiters to use this you're

697
00:25:35,480 --> 00:25:39,480
probably just like not build a

698
00:25:37,640 --> 00:25:43,679
recruiting team in the same way that you

699
00:25:39,480 --> 00:25:46,440
used to I mean other example is even for

700
00:25:43,679 --> 00:25:49,240
de tool companies they have to do a lot

701
00:25:46,440 --> 00:25:51,558
of a developer support and I work with

702
00:25:49,240 --> 00:25:54,240
this company called cap. AI that

703
00:25:51,558 --> 00:25:58,599
basically buildt one of the best chat

704
00:25:54,240 --> 00:26:01,679
Bots that responds to a lot of the a lot

705
00:25:58,599 --> 00:26:03,879
of the technical details that are hard

706
00:26:01,679 --> 00:26:05,480
to answer and I think a lot of the

707
00:26:03,880 --> 00:26:08,440
companies that started using them they

708
00:26:05,480 --> 00:26:10,319
actually ended up having Dev rail teams

709
00:26:08,440 --> 00:26:12,720
that are a lot smaller because it

710
00:26:10,319 --> 00:26:14,678
ingests a lot of uh the developer

711
00:26:12,720 --> 00:26:17,919
documentations even the YouTube videos

712
00:26:14,679 --> 00:26:19,360
that Dev tools put up and even a lot of

713
00:26:17,919 --> 00:26:22,200
the chat history so it just keeps

714
00:26:19,359 --> 00:26:23,759
getting better and better and it's like

715
00:26:22,200 --> 00:26:26,000
gives really good answers actually it's

716
00:26:23,759 --> 00:26:28,640
one one of the best I've seen yeah I

717
00:26:26,000 --> 00:26:30,240
also worked with a customer support like

718
00:26:28,640 --> 00:26:32,600
an AI customer support agent company

719
00:26:30,240 --> 00:26:36,798
called Power help well actually we both

720
00:26:32,599 --> 00:26:39,558
did um last batch and I learned a couple

721
00:26:36,798 --> 00:26:42,720
interesting things from parel um the

722
00:26:39,558 --> 00:26:45,200
first is customer like AI agents for

723
00:26:42,720 --> 00:26:46,720
customer support was like the category

724
00:26:45,200 --> 00:26:48,399
that's like famously crowded where

725
00:26:46,720 --> 00:26:49,798
there's like supposedly like you know a

726
00:26:48,398 --> 00:26:52,278
100 of them and if you go and you Google

727
00:26:49,798 --> 00:26:54,679
like AI customer support agent you'll

728
00:26:52,278 --> 00:26:55,960
get like a 100 results on Google um but

729
00:26:54,679 --> 00:26:57,640
what I learned through working with

730
00:26:55,960 --> 00:27:00,759
parel is like it's actually kind of

731
00:26:57,640 --> 00:27:03,240
like like almost all of those

732
00:27:00,759 --> 00:27:05,599
companies are doing very simple like

733
00:27:03,240 --> 00:27:07,798
zero shot llm prompting that can't

734
00:27:05,599 --> 00:27:08,918
actually replace a real customer support

735
00:27:07,798 --> 00:27:10,839
team that does a lot of really

736
00:27:08,919 --> 00:27:12,759
complicated workflows it just kind of

737
00:27:10,839 --> 00:27:14,720
makes for like a nice demo like to

738
00:27:12,759 --> 00:27:16,480
actually replace a customer support team

739
00:27:14,720 --> 00:27:18,000
for like an at Scale company that has

740
00:27:16,480 --> 00:27:19,519
like 100 customer support reps that do

741
00:27:18,000 --> 00:27:21,038
lots of complicated things every day you

742
00:27:19,519 --> 00:27:22,558
need like really complicated software

743
00:27:21,038 --> 00:27:24,720
that does all the stuff that like Jake

744
00:27:22,558 --> 00:27:26,158
heler was talking about and there's

745
00:27:24,720 --> 00:27:27,720
there were only like three or four

746
00:27:26,159 --> 00:27:28,720
companies that were even attempting to

747
00:27:27,720 --> 00:27:30,720
do that

748
00:27:28,720 --> 00:27:32,120
and cumulative they had cumulatively

749
00:27:30,720 --> 00:27:33,640
they had like less than 1% Market

750
00:27:32,119 --> 00:27:35,719
penetration and so the market was just

751
00:27:33,640 --> 00:27:37,640
completely open I could also see that

752
00:27:35,720 --> 00:27:40,319
being another case of um hyper

753
00:27:37,640 --> 00:27:42,720
specialization or hyper verticalization

754
00:27:40,319 --> 00:27:45,240
like there's not going to be I mean

755
00:27:42,720 --> 00:27:47,640
maybe eventually there could be a single

756
00:27:45,240 --> 00:27:50,839
general purpose customer support agent

757
00:27:47,640 --> 00:27:52,480
software company but we're like in in

758
00:27:50,839 --> 00:27:54,278
you know that that'll be like a eighth

759
00:27:52,480 --> 00:27:56,558
or ninth inning kind of thing and we're

760
00:27:54,278 --> 00:27:58,278
literally in the first inning so you

761
00:27:56,558 --> 00:28:00,879
know instead you know you're going to

762
00:27:58,278 --> 00:28:03,440
have companies like gig ml that you know

763
00:28:00,880 --> 00:28:06,399
it's doing it for zepto doing 30,000

764
00:28:03,440 --> 00:28:08,840
tickets uh every single day and

765
00:28:06,398 --> 00:28:11,719
replacing a team of a thousand people

766
00:28:08,839 --> 00:28:14,000
and but it's very specific and it has

767
00:28:11,720 --> 00:28:16,240
you know it's not a general purpose demo

768
00:28:14,000 --> 00:28:19,159
Weare kind of thing like it's 10,000

769
00:28:16,240 --> 00:28:22,759
test cases in a very detailed uh eval

770
00:28:19,159 --> 00:28:25,760
set that you know is basically just for

771
00:28:22,759 --> 00:28:28,119
zepto and things like zepto yeah uh but

772
00:28:25,759 --> 00:28:30,158
if you are you know any of the other

773
00:28:28,119 --> 00:28:32,359
Marketplace companies you're probably

774
00:28:30,159 --> 00:28:34,799
going to use it cuz like that's a very

775
00:28:32,359 --> 00:28:36,759
well-defined kind of marketplace that's

776
00:28:34,798 --> 00:28:38,319
you know instant delivery Marketplace I

777
00:28:36,759 --> 00:28:40,679
think this is the kind of dynamic that

778
00:28:38,319 --> 00:28:42,720
led there to be like $300 billion do SAS

779
00:28:40,679 --> 00:28:44,559
companies rather than like one like1

780
00:28:42,720 --> 00:28:45,919
trillion do like meta SAS thing that

781
00:28:44,558 --> 00:28:47,440
provides all the software for the world

782
00:28:45,919 --> 00:28:50,240
it's just like the customers just

783
00:28:47,440 --> 00:28:51,679
require really heavily like tailored

784
00:28:50,240 --> 00:28:53,159
Solutions and it's hard to build one

785
00:28:51,679 --> 00:28:54,399
that like works for every everyone

786
00:28:53,159 --> 00:28:55,720
exactly I mean we already gave three

787
00:28:54,398 --> 00:28:57,359
examples of customer support but there

788
00:28:55,720 --> 00:28:59,798
are very different verticals it's like

789
00:28:57,359 --> 00:29:01,158
de tool comp need very different kind of

790
00:28:59,798 --> 00:29:04,158
support that you need and the training

791
00:29:01,159 --> 00:29:06,200
set to marketplaces very different right

792
00:29:04,159 --> 00:29:08,080
yeah I guess whether you have agents or

793
00:29:06,200 --> 00:29:10,360
real human beings working for you you

794
00:29:08,079 --> 00:29:13,240
end up with the same problem which is

795
00:29:10,359 --> 00:29:15,918
every company bumps up against CO's

796
00:29:13,240 --> 00:29:18,720
theory of the firm which says that any

797
00:29:15,919 --> 00:29:21,559
given firm will grow only so much to the

798
00:29:18,720 --> 00:29:23,798
point where it uh becomes inefficient to

799
00:29:21,558 --> 00:29:27,440
be larger than that and then that's why

800
00:29:23,798 --> 00:29:29,558
they sort of networks and ecosystems and

801
00:29:27,440 --> 00:29:31,519
you know a full blown economy you know

802
00:29:29,558 --> 00:29:34,079
like every firm will sort of specialize

803
00:29:31,519 --> 00:29:36,880
to do what it is particularly good at

804
00:29:34,079 --> 00:29:38,599
and then the limits the outer limits of

805
00:29:36,880 --> 00:29:42,880
what those firms can be it's actually

806
00:29:38,599 --> 00:29:44,480
based on uh your ability as a manager so

807
00:29:42,880 --> 00:29:46,799
yeah that that part a little bit breaks

808
00:29:44,480 --> 00:29:49,159
my brain because you know when we spend

809
00:29:46,798 --> 00:29:50,879
time with Parker Conrad at ripling uh

810
00:29:49,159 --> 00:29:52,760
one of his favorite points is actually

811
00:29:50,880 --> 00:29:55,278
well you know everyone's very obsessed

812
00:29:52,759 --> 00:29:57,798
with with the fact that the rocks can

813
00:29:55,278 --> 00:29:59,720
talk and you know maybe they can draw

814
00:29:57,798 --> 00:30:02,839
but the more interesting thing for him

815
00:29:59,720 --> 00:30:04,000
you know running HR It software that uh

816
00:30:02,839 --> 00:30:05,720
you know he spends a lot of time

817
00:30:04,000 --> 00:30:07,919
thinking about HR like actually the

818
00:30:05,720 --> 00:30:11,038
coolest thing about the LMS is that the

819
00:30:07,919 --> 00:30:13,720
rocks can read and from his perspective

820
00:30:11,038 --> 00:30:16,679
like you he's I think he has 3,000

821
00:30:13,720 --> 00:30:19,000
employees he still runs payroll for all

822
00:30:16,679 --> 00:30:20,320
3,000 employees through Rippling so I

823
00:30:19,000 --> 00:30:22,679
think he spends a lot of time thinking

824
00:30:20,319 --> 00:30:26,278
about like how can one person extend

825
00:30:22,679 --> 00:30:27,720
their ability as a manager and uh I

826
00:30:26,278 --> 00:30:28,519
think we're going to see a lot more

827
00:30:27,720 --> 00:30:31,278
there

828
00:30:28,519 --> 00:30:34,278
that would be an a reverse argument that

829
00:30:31,278 --> 00:30:37,278
if we're at this moment where uh tools

830
00:30:34,278 --> 00:30:40,720
for managers and CEOs are going to get

831
00:30:37,278 --> 00:30:42,119
much more powerful um oh it could it

832
00:30:40,720 --> 00:30:44,000
could it could increase the scale of the

833
00:30:42,119 --> 00:30:45,439
firm that you can run right and that's

834
00:30:44,000 --> 00:30:47,480
certainly what ripling is trying to do

835
00:30:45,440 --> 00:30:49,600
like he's attempting to build this like

836
00:30:47,480 --> 00:30:51,319
Suite of HR tools where if he wins he's

837
00:30:49,599 --> 00:30:53,119
going to eat a whole bunch of billion

838
00:30:51,319 --> 00:30:54,678
dollar SAS companies and like one one

839
00:30:53,119 --> 00:30:57,239
giant company it's very interesting

840
00:30:54,679 --> 00:30:59,159
point Gary I think what made me think

841
00:30:57,240 --> 00:31:03,120
about this is that with having all these

842
00:30:59,159 --> 00:31:06,159
AI SAS tools it's going to give the

843
00:31:03,119 --> 00:31:08,359
ability to all these leaders and all

844
00:31:06,159 --> 00:31:10,120
these Orcs to basically open the

845
00:31:08,359 --> 00:31:12,359
aperture of the context window of how

846
00:31:10,119 --> 00:31:14,599
much information they can parse because

847
00:31:12,359 --> 00:31:16,479
there a limit of how much uh humans we

848
00:31:14,599 --> 00:31:17,759
can have meaningful relationship there's

849
00:31:16,480 --> 00:31:20,360
like the whole thing with the D Mar

850
00:31:17,759 --> 00:31:22,879
number it's about 300 people that you

851
00:31:20,359 --> 00:31:25,719
150 that you can have a meaningful

852
00:31:22,880 --> 00:31:27,960
relationship with but with AI because

853
00:31:25,720 --> 00:31:30,159
all of these rocks now can read I think

854
00:31:27,960 --> 00:31:32,319
think we will be able to extend that

855
00:31:30,159 --> 00:31:35,159
dumbar limit that we have yeah I think

856
00:31:32,319 --> 00:31:38,558
uh Flo crell had this interesting post

857
00:31:35,159 --> 00:31:41,480
on Twitter that went viral around um I

858
00:31:38,558 --> 00:31:44,240
think someone had made a voice chat like

859
00:31:41,480 --> 00:31:46,360
just weekend project as a CEO but it

860
00:31:44,240 --> 00:31:50,278
would call uh all

861
00:31:46,359 --> 00:31:52,439
1,500 of their employees yeah and uh you

862
00:31:50,278 --> 00:31:53,798
know it was you know very short call

863
00:31:52,440 --> 00:31:56,840
like kind of sounded like it was from

864
00:31:53,798 --> 00:31:59,480
the CEO just asking kind of personally I

865
00:31:56,839 --> 00:32:02,558
mean it sort of reminds of um that scene

866
00:31:59,480 --> 00:32:04,079
in her where it zooms out and uh

867
00:32:02,558 --> 00:32:06,558
actually you know you're following the

868
00:32:04,079 --> 00:32:09,278
experience of one person using the her

869
00:32:06,558 --> 00:32:11,240
OS but actually that her OSS is actually

870
00:32:09,278 --> 00:32:13,038
speaking to 15 you know thousands or

871
00:32:11,240 --> 00:32:16,880
tens of thousands of people all at one

872
00:32:13,038 --> 00:32:19,398
time how many others

873
00:32:16,880 --> 00:32:21,760
8,316 yeah I mean large language models

874
00:32:19,398 --> 00:32:25,398
can talk and can have conversations and

875
00:32:21,759 --> 00:32:28,200
then to what extent can uh you know this

876
00:32:25,398 --> 00:32:31,000
power actually extend the capability of

877
00:32:28,200 --> 00:32:33,798
one or a few people to uh understand

878
00:32:31,000 --> 00:32:34,919
what's going on I I heard about that yuk

879
00:32:33,798 --> 00:32:36,278
it got it definitely got me thinking

880
00:32:34,919 --> 00:32:37,759
because as understood the product is

881
00:32:36,278 --> 00:32:39,599
something like it just it will call up

882
00:32:37,759 --> 00:32:41,558
all your employees and then your

883
00:32:39,599 --> 00:32:43,398
employees can just like ramble about

884
00:32:41,558 --> 00:32:45,119
what they've been doing and it will just

885
00:32:43,398 --> 00:32:46,839
extract the meaning out of it and give

886
00:32:45,119 --> 00:32:48,879
the CEO of like a like bullet point

887
00:32:46,839 --> 00:32:50,359
summary of here the most important stuff

888
00:32:48,880 --> 00:32:51,639
and there were a bunch of like SAS

889
00:32:50,359 --> 00:32:54,798
companies that attempted to do these

890
00:32:51,638 --> 00:32:56,558
sort of like weekly pulse Pulses from

891
00:32:54,798 --> 00:32:58,720
employees using like traditional SAS

892
00:32:56,558 --> 00:33:00,519
software but like that version is is

893
00:32:58,720 --> 00:33:02,600
literally a 100 times better than the

894
00:33:00,519 --> 00:33:07,159
pre- elm version of this idea but I

895
00:33:02,599 --> 00:33:08,959
wonder with like that particular tool um

896
00:33:07,159 --> 00:33:11,399
just like it's not it's going Beyond

897
00:33:08,960 --> 00:33:12,679
just like reading and summarizing like

898
00:33:11,398 --> 00:33:14,439
this this is the argument of like if

899
00:33:12,679 --> 00:33:16,080
writing is thinking then like there's

900
00:33:14,440 --> 00:33:19,080
actually just a huge amount

901
00:33:16,079 --> 00:33:20,439
of work that's involved in the effort of

902
00:33:19,079 --> 00:33:22,319
figuring out like who's an effective

903
00:33:20,440 --> 00:33:24,278
communicator and like what are the most

904
00:33:22,319 --> 00:33:25,960
important things to be like what what

905
00:33:24,278 --> 00:33:28,159
are the key things to be focused on as a

906
00:33:25,960 --> 00:33:30,200
company and I just wonder if that at

907
00:33:28,159 --> 00:33:31,720
some point do the llms do like they go

908
00:33:30,200 --> 00:33:33,720
beyond just like summarizing and reading

909
00:33:31,720 --> 00:33:37,159
and doing actual thinking at which point

910
00:33:33,720 --> 00:33:39,919
like who's actually running the the

911
00:33:37,159 --> 00:33:41,720
organization an interesting

912
00:33:39,919 --> 00:33:43,200
thought I guess the other thing that's

913
00:33:41,720 --> 00:33:45,919
kind of interesting about how Parker

914
00:33:43,200 --> 00:33:47,639
Conrad's thinking about it is um I found

915
00:33:45,919 --> 00:33:49,880
out about this recently off a an

916
00:33:47,638 --> 00:33:51,759
interview with Matt mcginness his coo

917
00:33:49,880 --> 00:33:54,720
that uh there are more than a hundred

918
00:33:51,759 --> 00:33:57,398
Founders who work at ripling now as sort

919
00:33:54,720 --> 00:34:00,120
of specific people who run like an

920
00:33:57,398 --> 00:34:02,158
entire SAS vertical inside Rippling it's

921
00:34:00,119 --> 00:34:03,918
super cool the way he's built the team

922
00:34:02,159 --> 00:34:04,880
har probably knows a lot about it

923
00:34:03,919 --> 00:34:06,919
because you've done a bunch of

924
00:34:04,880 --> 00:34:10,440
interviews with him um yeah I mean it's

925
00:34:06,919 --> 00:34:14,039
definitely very focused on uh recruiting

926
00:34:10,440 --> 00:34:17,720
Founders and I mean Parker like Rippling

927
00:34:14,039 --> 00:34:20,960
is essentially the the case against

928
00:34:17,719 --> 00:34:23,158
vertical like verticalization trying to

929
00:34:20,960 --> 00:34:23,159
uh

930
00:34:26,719 --> 00:34:31,078
horizontalization like lots of value and

931
00:34:29,239 --> 00:34:33,078
he wants to recruit Founders and teams

932
00:34:31,079 --> 00:34:34,599
that build on top of the platform like

933
00:34:33,079 --> 00:34:36,280
it's almost a little bit more sort of

934
00:34:34,599 --> 00:34:39,240
like amazones whereas like shared

935
00:34:36,280 --> 00:34:40,760
infrastructure um yeah I think every

936
00:34:39,239 --> 00:34:42,598
product that they've released I mean

937
00:34:40,760 --> 00:34:45,000
things like time tracking and whatnot I

938
00:34:42,599 --> 00:34:46,960
mean basically they launch a thing and

939
00:34:45,000 --> 00:34:49,480
it hits like multi-millions of dollars

940
00:34:46,960 --> 00:34:50,960
in ARR on day one of launching and

941
00:34:49,480 --> 00:34:53,119
that's exactly what we were talking

942
00:34:50,960 --> 00:34:55,760
about earlier like once you once you

943
00:34:53,119 --> 00:34:56,960
have a vertical once you have a tow hold

944
00:34:55,760 --> 00:34:58,920
what you're saying is Well I have to

945
00:34:56,960 --> 00:35:01,800
spend this money on sales and marketing

946
00:34:58,920 --> 00:35:05,358
anyway can I uh you know basically get

947
00:35:01,800 --> 00:35:06,920
higher LTV and hold my CAC constant and

948
00:35:05,358 --> 00:35:09,078
uh that's sort of what you if you look

949
00:35:06,920 --> 00:35:10,760
at all the top uh software companies

950
00:35:09,079 --> 00:35:12,079
today it's like that's what Oracle is

951
00:35:10,760 --> 00:35:14,440
that's what Microsoft is that's what

952
00:35:12,079 --> 00:35:17,039
Salesforce is ripling knock on wood

953
00:35:14,440 --> 00:35:20,280
going to be the next but um it's it's an

954
00:35:17,039 --> 00:35:22,719
interesting alternative to uh going from

955
00:35:20,280 --> 00:35:23,720
zero to one totally on your own do you

956
00:35:22,719 --> 00:35:25,279
guys want to talk about some of the

957
00:35:23,719 --> 00:35:27,799
voice companies that we have I think

958
00:35:25,280 --> 00:35:30,040
that's like an interesting like sub

959
00:35:27,800 --> 00:35:32,519
category of this of this stuff is like

960
00:35:30,039 --> 00:35:35,759
really blowing up now I have a company

961
00:35:32,519 --> 00:35:37,079
that I work with called Salient that

962
00:35:35,760 --> 00:35:40,200
basically

963
00:35:37,079 --> 00:35:42,760
does AI voice calling to automate a lot

964
00:35:40,199 --> 00:35:44,759
of that collection in the auto Ling

965
00:35:42,760 --> 00:35:46,400
space which tradition so they like call

966
00:35:44,760 --> 00:35:50,079
up people and they're like hey you owe

967
00:35:46,400 --> 00:35:52,599
$1,000 on your car yeah which actually

968
00:35:50,079 --> 00:35:54,519
up with that actually this kind of job

969
00:35:52,599 --> 00:35:57,760
is one of those butter passing job it

970
00:35:54,519 --> 00:35:59,559
kind of sucks because a lot of uh these

971
00:35:57,760 --> 00:36:01,839
low wage workers work in all these call

972
00:35:59,559 --> 00:36:04,480
centers and it's like a terrible boring

973
00:36:01,838 --> 00:36:05,838
job so very high churn and giant

974
00:36:04,480 --> 00:36:07,440
headcount to run these because there's

975
00:36:05,838 --> 00:36:10,039
just so many accounts with these banks

976
00:36:07,440 --> 00:36:14,480
that have to do that and this is a

977
00:36:10,039 --> 00:36:17,199
perfect task that AI could automate and

978
00:36:14,480 --> 00:36:19,519
what Salient has done is has been able

979
00:36:17,199 --> 00:36:21,159
to actually get very very accurate and

980
00:36:19,519 --> 00:36:23,199
it has been going live with a lot of big

981
00:36:21,159 --> 00:36:25,279
Banks which is super exciting and this

982
00:36:23,199 --> 00:36:28,000
was a company from last year and

983
00:36:25,280 --> 00:36:29,960
demonstrating that that part of it that

984
00:36:28,000 --> 00:36:32,079
they were able to get in because they

985
00:36:29,960 --> 00:36:33,880
sold through top down I guess the space

986
00:36:32,079 --> 00:36:35,760
feels like it's moving very quickly and

987
00:36:33,880 --> 00:36:38,318
that we have incredible companies that

988
00:36:35,760 --> 00:36:40,839
are voice infra companies like vapy and

989
00:36:38,318 --> 00:36:43,679
then people can sort of get started

990
00:36:40,838 --> 00:36:45,880
right away and Retail also I mean these

991
00:36:43,679 --> 00:36:48,358
companies that have reached pretty fast

992
00:36:45,880 --> 00:36:50,838
scale just because it's one of the more

993
00:36:48,358 --> 00:36:53,119
exciting like mindblowing things that

994
00:36:50,838 --> 00:36:56,400
you can get up and running within I mean

995
00:36:53,119 --> 00:36:58,880
literally the course of hours um and

996
00:36:56,400 --> 00:37:00,119
then some of the question that you know

997
00:36:58,880 --> 00:37:02,358
remains unanswered and we hope they

998
00:37:00,119 --> 00:37:04,880
figure it out is how do you hold on to

999
00:37:02,358 --> 00:37:08,318
them especially as you uh run into

1000
00:37:04,880 --> 00:37:11,559
things like the new open AI voice apis

1001
00:37:08,318 --> 00:37:13,239
um you know do you go direct like you

1002
00:37:11,559 --> 00:37:16,199
Pro it's probably way more work to try

1003
00:37:13,239 --> 00:37:19,039
to use the underlying apis off the bat

1004
00:37:16,199 --> 00:37:21,279
but these uh platforms are clearly low

1005
00:37:19,039 --> 00:37:23,159
bar and then the question is can you

1006
00:37:21,280 --> 00:37:25,440
keep raising the ceiling so that you can

1007
00:37:23,159 --> 00:37:27,118
hold on to customers forever har you

1008
00:37:25,440 --> 00:37:29,400
were making an interesting point earlier

1009
00:37:27,119 --> 00:37:31,880
about like how the apps that people have

1010
00:37:29,400 --> 00:37:34,000
built on top of LMS has changed from

1011
00:37:31,880 --> 00:37:35,400
like early 2023 when it started until

1012
00:37:34,000 --> 00:37:36,719
now voice which we were just talking

1013
00:37:35,400 --> 00:37:38,880
about as a great example of this I think

1014
00:37:36,719 --> 00:37:41,279
even if you went 6 months back it felt

1015
00:37:38,880 --> 00:37:43,599
like the voices were not realistic

1016
00:37:41,280 --> 00:37:45,560
enough yet the latency was too high like

1017
00:37:43,599 --> 00:37:48,359
there was it felt like we were probably

1018
00:37:45,559 --> 00:37:51,159
a ways off having AI voice apps that

1019
00:37:48,358 --> 00:37:53,358
could meaningfully like replace like

1020
00:37:51,159 --> 00:37:57,078
humans calling people up and like here

1021
00:37:53,358 --> 00:38:00,519
we are and yeah I was just zooming out

1022
00:37:57,079 --> 00:38:03,920
thinking back to the first YC batch

1023
00:38:00,519 --> 00:38:05,119
where llm powered apps first came in was

1024
00:38:03,920 --> 00:38:08,760
probably winter

1025
00:38:05,119 --> 00:38:11,000
2023 you know almost 2 years ago now and

1026
00:38:08,760 --> 00:38:13,560
the apps were essentially just things

1027
00:38:11,000 --> 00:38:15,480
that spat out some text and not even

1028
00:38:13,559 --> 00:38:17,759
like perfect T Rockit talk that's about

1029
00:38:15,480 --> 00:38:20,679
it yeah sort of more like copy editing

1030
00:38:17,760 --> 00:38:22,839
marketing edit email edits it was just a

1031
00:38:20,679 --> 00:38:24,598
kind of more like just like incremental

1032
00:38:22,838 --> 00:38:26,159
yeah like I I had a company I mean the

1033
00:38:24,599 --> 00:38:28,519
one that sticks in my head is a company

1034
00:38:26,159 --> 00:38:30,358
Speedy brand and all what they did is

1035
00:38:28,519 --> 00:38:32,920
make it very easy for like a small

1036
00:38:30,358 --> 00:38:35,199
business to just generate a Blog and

1037
00:38:32,920 --> 00:38:37,519
spit out content marketing um it's like

1038
00:38:35,199 --> 00:38:40,759
very obvious idea and it wasn't perfect

1039
00:38:37,519 --> 00:38:41,838
but it was pretty cool at the time and

1040
00:38:40,760 --> 00:38:43,319
that's what we've talked about a bunch

1041
00:38:41,838 --> 00:38:45,159
of the show but that's like the chat gbt

1042
00:38:43,318 --> 00:38:47,519
raer turned out around that time hey

1043
00:38:45,159 --> 00:38:49,480
like this is what an llm app looks like

1044
00:38:47,519 --> 00:38:52,199
it's just a chat GPT rapper it does very

1045
00:38:49,480 --> 00:38:53,639
basic spits out some text like it's

1046
00:38:52,199 --> 00:38:56,439
going to get crushed by openi in the

1047
00:38:53,639 --> 00:38:58,960
next release like and it did yeah well I

1048
00:38:56,440 --> 00:39:01,000
I I don't know if that one did but but

1049
00:38:58,960 --> 00:39:02,838
the that that first that first wave of

1050
00:39:01,000 --> 00:39:06,039
llm apps mostly did get crushed by the

1051
00:39:02,838 --> 00:39:07,759
next wave of GPT I feel like we've had

1052
00:39:06,039 --> 00:39:09,400
this sort of boiling of the Frog effect

1053
00:39:07,760 --> 00:39:11,240
where from our perspect it's sort of

1054
00:39:09,400 --> 00:39:13,480
like every three months things have just

1055
00:39:11,239 --> 00:39:14,598
kept getting progressively better and

1056
00:39:13,480 --> 00:39:17,119
now we're at this point where we're

1057
00:39:14,599 --> 00:39:19,160
talking about like full-on vertical AI

1058
00:39:17,119 --> 00:39:22,119
agents that are going to replace entire

1059
00:39:19,159 --> 00:39:23,838
teams and functions and Enterprises um

1060
00:39:22,119 --> 00:39:26,400
and just that progression is still

1061
00:39:23,838 --> 00:39:28,358
mindblowing to me like with two years in

1062
00:39:26,400 --> 00:39:30,400
which is still relatively early and the

1063
00:39:28,358 --> 00:39:33,358
rate of progress is just like unlike

1064
00:39:30,400 --> 00:39:34,760
anything we've seen before and I think

1065
00:39:33,358 --> 00:39:37,480
what's interesting to see is we

1066
00:39:34,760 --> 00:39:39,760
discussed this in the last episode is a

1067
00:39:37,480 --> 00:39:41,400
lot of the foundation models are kind of

1068
00:39:39,760 --> 00:39:43,680
coming head-to-head there used to be

1069
00:39:41,400 --> 00:39:46,960
only one player in town with open AI but

1070
00:39:43,679 --> 00:39:48,598
we've been seeing in the last batch this

1071
00:39:46,960 --> 00:39:51,400
has been changing Claude is a huge

1072
00:39:48,599 --> 00:39:53,960
Contender thank God it's like

1073
00:39:51,400 --> 00:39:57,400
competition is you know the the soil for

1074
00:39:53,960 --> 00:40:00,079
a very fertile Marketplace ecosystem uh

1075
00:39:57,400 --> 00:40:02,079
for which consumers will have choice and

1076
00:40:00,079 --> 00:40:04,200
uh Founders have a shot and that's the

1077
00:40:02,079 --> 00:40:06,200
world I want to live in so people are

1078
00:40:04,199 --> 00:40:08,559
watching and thinking about starting a

1079
00:40:06,199 --> 00:40:11,519
startup or maybe have already started

1080
00:40:08,559 --> 00:40:13,639
and uh they're hearing all of this how

1081
00:40:11,519 --> 00:40:16,519
do you know what the right vertical is

1082
00:40:13,639 --> 00:40:19,920
for you you got to find

1083
00:40:16,519 --> 00:40:21,559
some boring repeative admin work

1084
00:40:19,920 --> 00:40:23,800
somewhere and that seems to be like the

1085
00:40:21,559 --> 00:40:26,159
common threat across all of the stuff is

1086
00:40:23,800 --> 00:40:29,720
if you can find a boring repetitive

1087
00:40:26,159 --> 00:40:33,078
admin task um there is likely going to

1088
00:40:29,719 --> 00:40:35,439
be a billion dollar AI agent startup if

1089
00:40:33,079 --> 00:40:37,039
you keep digging deep enough into it but

1090
00:40:35,440 --> 00:40:39,159
it sounds like you should go after

1091
00:40:37,039 --> 00:40:42,199
something that you directly have some

1092
00:40:39,159 --> 00:40:44,078
sort of experience or relationship to

1093
00:40:42,199 --> 00:40:45,480
that is a common like there there's

1094
00:40:44,079 --> 00:40:47,720
definitely a Common Thread I've seen in

1095
00:40:45,480 --> 00:40:49,358
the companies that are that I'm seeing

1096
00:40:47,719 --> 00:40:50,639
promis with and another one just pops

1097
00:40:49,358 --> 00:40:52,480
into my head sweet spot I think I

1098
00:40:50,639 --> 00:40:54,159
mentioned on this before like they're

1099
00:40:52,480 --> 00:40:56,358
basically building an AI agent to bid on

1100
00:40:54,159 --> 00:40:58,519
government contracts and the way they

1101
00:40:56,358 --> 00:40:59,920
found that idea and a year ago was they

1102
00:40:58,519 --> 00:41:01,239
just had a friend whose full-time job

1103
00:40:59,920 --> 00:41:03,039
was to sit there on like a government

1104
00:41:01,239 --> 00:41:05,358
website like refreshing the page like

1105
00:41:03,039 --> 00:41:07,119
looking for new proposals to bid on and

1106
00:41:05,358 --> 00:41:08,519
they they were pivoting they're like ah

1107
00:41:07,119 --> 00:41:10,838
like that seems like something an llm

1108
00:41:08,519 --> 00:41:12,480
could do um a company from a recent

1109
00:41:10,838 --> 00:41:13,880
batch which pivoted into a new idea

1110
00:41:12,480 --> 00:41:15,240
that's getting great traction like

1111
00:41:13,880 --> 00:41:17,000
they're basically building an AI agent

1112
00:41:15,239 --> 00:41:18,759
to do um process like medical billing

1113
00:41:17,000 --> 00:41:21,119
for dental clinics and the way they

1114
00:41:18,760 --> 00:41:22,800
found the idea was um one of the

1115
00:41:21,119 --> 00:41:24,039
founders mother is a dentist and so he

1116
00:41:22,800 --> 00:41:25,760
just decided to go to work with her for

1117
00:41:24,039 --> 00:41:27,759
a day and just sit there seeing what she

1118
00:41:25,760 --> 00:41:29,280
did and she's like oh like all of that

1119
00:41:27,760 --> 00:41:31,040
like processing claim seems like really

1120
00:41:29,280 --> 00:41:32,240
boring like an llm should totally be

1121
00:41:31,039 --> 00:41:34,960
able to do that and he just started

1122
00:41:32,239 --> 00:41:36,879
writing software for like his mother's

1123
00:41:34,960 --> 00:41:40,079
dental clinic so I guess I mean in

1124
00:41:36,880 --> 00:41:41,559
robotics the classic Maxim is uh you the

1125
00:41:40,079 --> 00:41:42,720
robots that are going to be profitable

1126
00:41:41,559 --> 00:41:47,799
and that are going to work are going to

1127
00:41:42,719 --> 00:41:50,719
be um dirty and dangerous jobs and in

1128
00:41:47,800 --> 00:41:53,960
this case for vertical SAS look for

1129
00:41:50,719 --> 00:41:56,399
boring butter passing

1130
00:41:53,960 --> 00:41:58,599
jobs well with that we're out of time

1131
00:41:56,400 --> 00:42:01,130
for today we'll catch you on the light

1132
00:41:58,599 --> 00:42:12,449
cone next time

1133
00:42:01,130 --> 00:42:12,449
[Music]