1
00:00:00,100 --> 00:00:01,966
welcome to money is freedom

2
00:00:01,966 --> 00:00:02,866
a podcast

3
00:00:02,866 --> 00:00:06,066
exploring how finance and freedom connect in our lives

4
00:00:07,033 --> 00:00:08,833
created from thoughtful research

5
00:00:08,833 --> 00:00:11,300
and narrated by Notebook lmaai

6
00:00:11,400 --> 00:00:13,266
this series brings you clear

7
00:00:13,266 --> 00:00:15,966
meaningful insights into finance and beyond

8
00:00:16,066 --> 00:00:18,500
in the state of AI 2025

9
00:00:18,566 --> 00:00:20,666
Bessemer Venture Partners highlights

10
00:00:21,066 --> 00:00:25,800
2025 as first light following the 2023 AI Big Bang

11
00:00:25,800 --> 00:00:29,566
mass release of Chat GPT revealing emerging AI

12
00:00:29,566 --> 00:00:33,000
galaxies of foundational companies and best practices

13
00:00:33,600 --> 00:00:37,100
this episode explores new benchmarks for AI startups

14
00:00:37,233 --> 00:00:39,766
categorizing them into two archetypes

15
00:00:40,233 --> 00:00:41,333
supernovas

16
00:00:41,400 --> 00:00:45,000
which are explosively scaling with unprecedented growth

17
00:00:45,000 --> 00:00:47,366
often trading distribution for profit

18
00:00:47,433 --> 00:00:50,466
with approximately 25% gross margins

19
00:00:50,600 --> 00:00:53,600
and shooting stars which are fast growing

20
00:00:53,600 --> 00:00:55,200
capital efficient companies

21
00:00:55,300 --> 00:00:56,766
achieving strong product

22
00:00:56,766 --> 00:01:00,700
market fit and approximately 60% gross margins

23
00:01:00,900 --> 00:01:02,266
welcome back to the deep dive

24
00:01:02,266 --> 00:01:04,900
uh your shortcut to being well informed

25
00:01:05,000 --> 00:01:07,600
today we're navigating the well

26
00:01:07,600 --> 00:01:09,033
the rapidly expanding

27
00:01:09,033 --> 00:01:10,833
universe of artificial intelligence

28
00:01:10,833 --> 00:01:12,300
that's right specifically

29
00:01:12,300 --> 00:01:14,300
we're looking through the lens of Bessemer

30
00:01:14,300 --> 00:01:17,833
Venture Partners The State of AI 2025 report

31
00:01:17,833 --> 00:01:19,100
really important one you know

32
00:01:19,100 --> 00:01:21,766
if 2023 was the AI Big Bang

33
00:01:21,766 --> 00:01:26,200
as they put it then 2025 feels more like first light

34
00:01:26,200 --> 00:01:27,300
hmm first light

35
00:01:27,300 --> 00:01:28,300
I like that yeah

36
00:01:28,300 --> 00:01:30,566
the initial chaos seems to be subsiding a bit

37
00:01:30,566 --> 00:01:31,866
and we're starting to glimpse

38
00:01:31,866 --> 00:01:33,666
you know foundational companies

39
00:01:33,666 --> 00:01:36,566
emerging best practices clearer paths forward

40
00:01:36,566 --> 00:01:37,833
yeah things solidifying

41
00:01:37,833 --> 00:01:39,900
and the report's message is well

42
00:01:39,900 --> 00:01:41,433
pretty unequivocal AI

43
00:01:41,433 --> 00:01:44,166
is driving the biggest wave of technology change

44
00:01:44,166 --> 00:01:45,466
we have ever witnessed

45
00:01:45,466 --> 00:01:48,000
and that's precisely why this deep dive into Best

46
00:01:48,000 --> 00:01:51,200
Summer's report is so crucial for any founder

47
00:01:51,200 --> 00:01:52,433
investor or really

48
00:01:52,433 --> 00:01:52,833
anyone

49
00:01:52,833 --> 00:01:55,666
just trying to separate the genuine innovation from

50
00:01:55,666 --> 00:01:57,600
let's face it the sheer volume of hype

51
00:01:57,600 --> 00:01:59,500
there is a lot of hype oh yeah

52
00:01:59,566 --> 00:02:01,400
there's still plenty of wild

53
00:02:01,400 --> 00:02:03,466
unpredictability in dark matter out there

54
00:02:03,466 --> 00:02:05,766
as they call it but Besemer

55
00:02:06,200 --> 00:02:09,200
having deployed over $1 billion into AI

56
00:02:09,200 --> 00:02:11,366
native startups since 2023

57
00:02:11,833 --> 00:02:12,600
well

58
00:02:12,600 --> 00:02:16,000
they're kind of uniquely positioned to map this out

59
00:02:16,000 --> 00:02:18,833
$1 billion definitely gives them perspective exactly

60
00:02:18,833 --> 00:02:20,733
so their mission with this report

61
00:02:21,400 --> 00:02:23,166
share concrete benchmarks

62
00:02:23,500 --> 00:02:26,566
tour roadmaps across five key AI areas

63
00:02:26,866 --> 00:02:29,933
surface those unanswered questions that dark matter

64
00:02:30,100 --> 00:02:33,266
and importantly offer five bold predictions

65
00:02:33,433 --> 00:02:35,233
it's like they're handing us a constellation chart

66
00:02:35,233 --> 00:02:36,100
for the future then

67
00:02:36,100 --> 00:02:37,066
that's a great way to put it

68
00:02:37,066 --> 00:02:38,066
a constellation chart okay

69
00:02:38,066 --> 00:02:39,100
let's unpack this

70
00:02:39,166 --> 00:02:41,266
the report kicks off with a really fascinating

71
00:02:41,266 --> 00:02:43,066
look at what great actually means

72
00:02:43,066 --> 00:02:44,166
for AI startups

73
00:02:44,166 --> 00:02:45,766
now in 2025

74
00:02:45,766 --> 00:02:48,266
it's moving beyond just traditional sauce metrics right

75
00:02:48,266 --> 00:02:51,166
they highlight this tale of two AI startups

76
00:02:51,166 --> 00:02:53,033
introducing what they call supernovas

77
00:02:53,033 --> 00:02:55,166
and shooting stars it's a brilliant distinction

78
00:02:55,166 --> 00:02:57,266
really captures the two main types of uh

79
00:02:57,266 --> 00:02:58,566
explosive growth recently

80
00:02:58,566 --> 00:02:59,600
yeah first up

81
00:02:59,800 --> 00:03:04,300
the AI supernovas these are companies showing truly um

82
00:03:04,366 --> 00:03:06,166
unprecedented velocity

83
00:03:06,300 --> 00:03:08,666
sprinting to around 40 million dollar a R R

84
00:03:08,666 --> 00:03:10,033
in their first year of commercialization

85
00:03:10,033 --> 00:03:11,866
it's one 40 million year one

86
00:03:11,866 --> 00:03:15,466
and then hitting a staggering 125 million dollar a R R

87
00:03:15,466 --> 00:03:16,933
by year two wow

88
00:03:16,966 --> 00:03:17,266
now

89
00:03:17,266 --> 00:03:19,400
what's crucial to understand is that these supernovas

90
00:03:19,400 --> 00:03:21,533
often operate with low gross margins

91
00:03:21,700 --> 00:03:24,300
sometimes even negative really negative

92
00:03:24,300 --> 00:03:25,900
yeah they're actively trading profit

93
00:03:25,900 --> 00:03:27,033
for that rapid distribution

94
00:03:27,033 --> 00:03:28,600
for market share but get this

95
00:03:28,600 --> 00:03:31,233
their revenue per full time employee is incredible

96
00:03:31,233 --> 00:03:34,066
about $1.13 million a R

97
00:03:34,066 --> 00:03:36,100
R per F t e whoa

98
00:03:36,100 --> 00:03:36,900
which is like

99
00:03:36,900 --> 00:03:39,200
four to five times higher than a typical song

100
00:03:39,200 --> 00:03:42,000
you gotta be super efficient in that sense extremely

101
00:03:42,166 --> 00:03:43,833
but the report does caution

102
00:03:43,833 --> 00:03:44,066
you know

103
00:03:44,066 --> 00:03:46,766
this hyper growth can sometimes signal vulnerability

104
00:03:46,866 --> 00:03:48,066
maybe low switching costs

105
00:03:48,066 --> 00:03:50,100
or what they call thin wrapper concerns

106
00:03:50,100 --> 00:03:51,833
dazzling but potentially fragile

107
00:03:51,833 --> 00:03:53,400
that's a sharp warning so

108
00:03:53,400 --> 00:03:56,766
are these supernovas just like a necessary early stage

109
00:03:56,766 --> 00:03:57,833
or is there a real danger

110
00:03:57,833 --> 00:03:59,700
many won't last past that fragile phase

111
00:03:59,700 --> 00:04:01,566
with those low margins well

112
00:04:01,600 --> 00:04:03,033
it's probably a bit of both

113
00:04:03,033 --> 00:04:05,400
isn't it the supernovas definitely push the boundaries

114
00:04:05,400 --> 00:04:07,733
they prove out entirely new categories

115
00:04:07,833 --> 00:04:09,900
but yes the risk is real

116
00:04:10,000 --> 00:04:10,600
many might

117
00:04:10,600 --> 00:04:12,433
struggle to convert that rapid adoption

118
00:04:12,433 --> 00:04:13,966
into sustainable profit

119
00:04:13,966 --> 00:04:16,000
if they don't build strong defensibility

120
00:04:16,300 --> 00:04:17,366
they set the pace

121
00:04:17,500 --> 00:04:20,266
but maybe the long term winners evolve differently

122
00:04:20,266 --> 00:04:21,666
which brings us to the second type

123
00:04:21,666 --> 00:04:23,066
I guess exactly

124
00:04:23,266 --> 00:04:24,966
then you have the AI shooting stars

125
00:04:24,966 --> 00:04:26,800
these are more like stellar sauce companies

126
00:04:26,800 --> 00:04:28,633
just accelerated OK

127
00:04:28,633 --> 00:04:30,366
they achieve strong product market fit

128
00:04:30,366 --> 00:04:32,933
great retention impressive expansion

129
00:04:33,166 --> 00:04:35,166
they grow faster than traditional saws

130
00:04:35,166 --> 00:04:36,633
but at rates that are you know

131
00:04:36,633 --> 00:04:38,566
more capital efficient more sustainable

132
00:04:38,566 --> 00:04:40,466
makes sense what do those benchmarks look like

133
00:04:40,466 --> 00:04:43,433
so approximately $3 million AR in year one

134
00:04:43,433 --> 00:04:46,066
growing to maybe $12 million in year two

135
00:04:46,100 --> 00:04:48,166
then $40 million in year three

136
00:04:48,166 --> 00:04:50,766
and hitting around $103 million in year four

137
00:04:50,766 --> 00:04:52,700
still very fast growth oh absolutely

138
00:04:52,700 --> 00:04:55,300
but crucially they maintain healthy gross margins

139
00:04:55,433 --> 00:04:57,000
typically around 60%

140
00:04:57,233 --> 00:05:00,433
and they achieve about $164,000 a R

141
00:05:00,433 --> 00:05:02,566
R per FTE in their first year

142
00:05:02,566 --> 00:05:05,300
okay so a different profile entirely right

143
00:05:05,300 --> 00:05:07,866
and this growth trajectory is so significant

144
00:05:08,000 --> 00:05:10,266
best summer actually coined a new term for it

145
00:05:10,300 --> 00:05:11,666
Q2 T3

146
00:05:11,666 --> 00:05:12,766
Q2 T three yeah

147
00:05:12,766 --> 00:05:14,866
stands for quadruple quadruple triple

148
00:05:14,866 --> 00:05:17,100
triple triple growth over five years

149
00:05:17,100 --> 00:05:18,566
quadruple quadruple triple

150
00:05:18,566 --> 00:05:19,266
triple triple

151
00:05:19,266 --> 00:05:21,366
that's serious acceleration

152
00:05:21,600 --> 00:05:22,800
does this mean the rules

153
00:05:22,800 --> 00:05:24,633
for scaling have fundamentally changed

154
00:05:24,633 --> 00:05:26,566
for all startups now not just AI ones

155
00:05:26,566 --> 00:05:27,400
that's a great point

156
00:05:27,400 --> 00:05:29,466
because the report really emphasizes that

157
00:05:29,466 --> 00:05:31,300
while supernovas grab headlines

158
00:05:31,300 --> 00:05:33,833
shooting stars are the most important benchmark to aim

159
00:05:33,833 --> 00:05:35,200
for scalable

160
00:05:35,200 --> 00:05:36,600
capital efficient growth

161
00:05:36,600 --> 00:05:38,366
that's now uniquely achievable

162
00:05:38,366 --> 00:05:41,366
absolutely I think speed just matters more than ever

163
00:05:41,433 --> 00:05:43,566
AI has undeniably unlocked

164
00:05:43,566 --> 00:05:45,533
faster product development cycles

165
00:05:45,666 --> 00:05:48,700
quicker go to market more rapid distribution

166
00:05:48,766 --> 00:05:52,166
that q2t3 isn't just some pie in the sky dream anymore

167
00:05:52,166 --> 00:05:54,800
it's becoming an achievable benchmark for many

168
00:05:54,800 --> 00:05:57,266
so the bar's been raised for everyone pretty much

169
00:05:57,266 --> 00:05:59,700
but the tools to reach it are also more powerful now

170
00:05:59,700 --> 00:06:01,566
okay so with those benchmarks set

171
00:06:01,566 --> 00:06:03,166
the report then shifts gears

172
00:06:03,166 --> 00:06:04,533
it takes us on this tour

173
00:06:04,633 --> 00:06:07,433
through what they call the roadmaps of the AI cosmos

174
00:06:07,433 --> 00:06:08,566
yeah I like that framing

175
00:06:08,566 --> 00:06:11,033
it's where we see these early galaxies of AI

176
00:06:11,033 --> 00:06:14,500
forming but also bump into that dark matter

177
00:06:14,500 --> 00:06:18,000
those big unresolved questions still shaping everything

178
00:06:18,000 --> 00:06:19,700
precisely and they start with the foundation

179
00:06:19,700 --> 00:06:21,466
yeah AI infrastructure right

180
00:06:21,466 --> 00:06:24,033
and here you can clearly see galaxies forming

181
00:06:24,033 --> 00:06:26,500
in the model layer you've got the big players

182
00:06:26,500 --> 00:06:28,500
open AI anthropic Gemini

183
00:06:28,500 --> 00:06:29,500
Llama Xai

184
00:06:29,500 --> 00:06:31,766
dominating headlines usual suspects

185
00:06:31,766 --> 00:06:33,100
but it's not just them right

186
00:06:33,100 --> 00:06:36,000
the open source models like Kimi Deepseek

187
00:06:36,100 --> 00:06:39,166
Quinn Mixtrell are also showing real strength

188
00:06:39,166 --> 00:06:40,800
sometimes in specialized areas

189
00:06:40,900 --> 00:06:44,266
what does that dual track mean for the ecosystem

190
00:06:44,266 --> 00:06:44,566
well

191
00:06:44,566 --> 00:06:47,500
it creates this really vibrant competitive landscape

192
00:06:47,500 --> 00:06:48,766
which is generally good

193
00:06:48,800 --> 00:06:51,400
the report calls this phase AI infrastructure

194
00:06:51,400 --> 00:06:52,633
second act okay

195
00:06:52,633 --> 00:06:53,366
second act yeah

196
00:06:53,366 --> 00:06:54,666
the first act was you know

197
00:06:54,666 --> 00:06:56,100
algorithmic breakthroughs

198
00:06:56,100 --> 00:06:58,166
proving AI could solve problems

199
00:06:58,300 --> 00:07:00,666
now the shift is from just solving problems

200
00:07:00,666 --> 00:07:02,266
to building systems that can define

201
00:07:02,266 --> 00:07:04,833
measure and solve problems with experience

202
00:07:04,833 --> 00:07:07,300
clarity and purpose more operational

203
00:07:07,300 --> 00:07:09,166
more integrated exactly

204
00:07:09,166 --> 00:07:11,466
so we're seeing new infrastructure tools emerge

205
00:07:11,466 --> 00:07:12,766
not just for scale

206
00:07:12,900 --> 00:07:15,666
but for grounding AI in real world context

207
00:07:15,800 --> 00:07:18,633
for continuous learning think reinforcement learning

208
00:07:18,633 --> 00:07:20,466
environments like fleet and matrices

209
00:07:20,666 --> 00:07:23,300
or new evaluation frameworks from companies like Big

210
00:07:23,300 --> 00:07:24,600
Spin Dot AI

211
00:07:24,833 --> 00:07:28,366
kill an AI and also the rise of compound AI systems

212
00:07:28,366 --> 00:07:30,200
integrating knowledge retrieval

213
00:07:30,200 --> 00:07:32,633
memory planning that brings up the dark matter

214
00:07:32,633 --> 00:07:34,900
they mention here the bitter lesson of AI

215
00:07:34,900 --> 00:07:36,200
which is this idea well

216
00:07:36,200 --> 00:07:37,200
you explain it right

217
00:07:37,200 --> 00:07:39,533
the bitter lesson essentially poses the question

218
00:07:39,700 --> 00:07:41,800
will simple scalable architectures

219
00:07:41,800 --> 00:07:43,833
fed with vast amounts of data always

220
00:07:43,833 --> 00:07:44,900
went out in the long run

221
00:07:44,900 --> 00:07:47,100
or will more complex technique

222
00:07:47,100 --> 00:07:48,633
specific approaches from betting

223
00:07:48,633 --> 00:07:52,233
context and domain expertise ultimately prove superior

224
00:07:52,233 --> 00:07:53,666
it's still an open debate

225
00:07:53,733 --> 00:07:55,233
so for a founder building here

226
00:07:55,233 --> 00:07:55,800
the challenge is

227
00:07:55,800 --> 00:07:58,000
how to best embed that domain knowledge

228
00:07:58,000 --> 00:08:00,200
is it about the data or the techniques

229
00:08:00,200 --> 00:08:01,300
it's really both the

230
00:08:01,300 --> 00:08:04,000
how of getting specific intelligence into your model

231
00:08:04,000 --> 00:08:06,866
effectively is still a massive area of innovation

232
00:08:06,866 --> 00:08:08,900
you probably can't just throw more generic

233
00:08:08,900 --> 00:08:10,200
data at every single problem

234
00:08:10,200 --> 00:08:11,666
and expect the best outcome

235
00:08:11,666 --> 00:08:13,666
technique matters okay

236
00:08:14,033 --> 00:08:17,033
from infrastructure it's a natural jump to the builders

237
00:08:17,033 --> 00:08:18,300
the developers and

238
00:08:18,300 --> 00:08:20,566
the report sees a huge shift in

239
00:08:20,566 --> 00:08:22,400
how they'll even interact with these systems

240
00:08:22,400 --> 00:08:23,733
doesn't it oh absolutely

241
00:08:23,900 --> 00:08:25,633
in developer platforms and tooling

242
00:08:25,633 --> 00:08:28,066
the big shift they highlight is natural language

243
00:08:28,066 --> 00:08:29,866
becoming the new programming interface

244
00:08:29,866 --> 00:08:31,500
prompts as programs exactly

245
00:08:31,500 --> 00:08:32,933
prompts are programs

246
00:08:33,000 --> 00:08:35,866
and LLMs are acting like a new kind of computer

247
00:08:35,866 --> 00:08:37,200
it's a paradigm shift

248
00:08:37,400 --> 00:08:40,500
AI isn't just an incremental change for developer tools

249
00:08:40,633 --> 00:08:41,466
it's a completely

250
00:08:41,466 --> 00:08:43,433
new way to think about software development

251
00:08:43,433 --> 00:08:45,833
and within that they talk about galaxies forming

252
00:08:45,833 --> 00:08:47,266
around something called the Model

253
00:08:47,266 --> 00:08:49,366
Context Protocol MCP

254
00:08:49,500 --> 00:08:51,866
described as the USB C of AI

255
00:08:51,866 --> 00:08:53,333
what's that about huh

256
00:08:53,500 --> 00:08:57,166
the USB C of AI it's becoming a universal way for AI

257
00:08:57,166 --> 00:08:59,600
agents to access external APIs tools

258
00:08:59,600 --> 00:09:02,066
real time data think of it as a standard plug

259
00:09:02,066 --> 00:09:03,766
OK it allows for persistent memory

260
00:09:03,766 --> 00:09:06,866
cross sessions complex multi tool workflows

261
00:09:06,966 --> 00:09:08,866
fine grain permissions this

262
00:09:08,866 --> 00:09:11,366
means agents can actually chain tasks together

263
00:09:11,400 --> 00:09:14,533
reason over live systems and truly act for users

264
00:09:14,566 --> 00:09:15,966
not just assist passively

265
00:09:15,966 --> 00:09:18,333
so they can actually do things right

266
00:09:18,400 --> 00:09:20,500
and key enablers here things like Prefex

267
00:09:20,500 --> 00:09:23,966
Fast MCP and tools like arcade and keycard

268
00:09:24,000 --> 00:09:26,866
it's all about empowering agents to handle complex

269
00:09:26,866 --> 00:09:28,933
tasks securely and effectively

270
00:09:29,033 --> 00:09:31,100
letting them plug and play with different tools

271
00:09:31,100 --> 00:09:33,866
this seems to connect to the next bit of dark matter

272
00:09:33,866 --> 00:09:36,166
memory context and beyond

273
00:09:36,600 --> 00:09:38,100
you mentioned memory earlier

274
00:09:38,100 --> 00:09:40,433
can you break down the difference between context

275
00:09:40,433 --> 00:09:42,500
and memory as the report sees it

276
00:09:42,500 --> 00:09:43,900
yeah it's a critical distinction

277
00:09:43,900 --> 00:09:47,000
context is the data the model sees during inference

278
00:09:47,000 --> 00:09:49,066
basically it's immediate working memory

279
00:09:49,066 --> 00:09:50,933
for that specific interaction

280
00:09:51,100 --> 00:09:52,866
okay like what's in the prompt window

281
00:09:52,866 --> 00:09:53,900
sort of yeah

282
00:09:53,900 --> 00:09:54,766
memory though

283
00:09:54,766 --> 00:09:57,733
is the information retained across interactions

284
00:09:57,833 --> 00:10:00,033
this is what allows for multistep reasoning

285
00:10:00,033 --> 00:10:02,666
personalization continuity over time for an agent

286
00:10:02,666 --> 00:10:03,300
ah okay

287
00:10:03,300 --> 00:10:04,833
so memory builds up exactly

288
00:10:04,833 --> 00:10:07,366
and the leading stacks are combining short term memory

289
00:10:07,800 --> 00:10:09,533
via large context Windows

290
00:10:09,833 --> 00:10:12,066
long term memory using vector databases

291
00:10:12,066 --> 00:10:13,300
maybe memory OBIS

292
00:10:13,566 --> 00:10:16,666
and semantic memory through techniques like hybrid REEs

293
00:10:16,666 --> 00:10:17,700
and the report says this

294
00:10:17,700 --> 00:10:19,966
context and memory might be the new moats

295
00:10:19,966 --> 00:10:21,766
for AI application founders

296
00:10:21,900 --> 00:10:23,333
that's the key insight

297
00:10:23,366 --> 00:10:26,066
when your product accumulates intelligence about a user

298
00:10:26,066 --> 00:10:27,666
or their specific world

299
00:10:27,666 --> 00:10:30,766
it creates what they call emotional switching costs

300
00:10:30,900 --> 00:10:32,500
emotional switching costs yeah

301
00:10:32,500 --> 00:10:34,033
those accumulated insights

302
00:10:34,033 --> 00:10:36,166
become incredibly sticky assets

303
00:10:36,400 --> 00:10:38,366
it's not just what the AI knows generally

304
00:10:38,366 --> 00:10:38,700
but what

305
00:10:38,700 --> 00:10:41,866
it remembers and understands about you specifically

306
00:10:41,866 --> 00:10:45,033
that's hard to replicate that makes a lot of sense okay

307
00:10:45,033 --> 00:10:47,600
moving on to horizontal and enterprise AI

308
00:10:47,600 --> 00:10:48,866
this is where it gets really interesting

309
00:10:48,866 --> 00:10:50,700
I think the report suggests

310
00:10:50,700 --> 00:10:53,033
AI is starting to disrupt massive

311
00:10:53,033 --> 00:10:55,466
horizontal systems of record

312
00:10:55,766 --> 00:10:58,533
Saures like Salesforce SAP

313
00:10:58,566 --> 00:11:00,166
stuff that seemed untouchable

314
00:11:00,200 --> 00:11:02,566
it's a potential seismic shift for sure

315
00:11:02,566 --> 00:11:04,500
we're seeing galaxies forming

316
00:11:04,500 --> 00:11:06,366
as these traditional systems of record

317
00:11:06,366 --> 00:11:08,333
might get replaced by what the report calls

318
00:11:08,400 --> 00:11:10,200
systems of action systems of action

319
00:11:10,200 --> 00:11:11,966
how does AI enable that well

320
00:11:11,966 --> 00:11:14,700
AI's ability to structure unstructured data

321
00:11:14,700 --> 00:11:16,866
like emails or call transcripts

322
00:11:16,900 --> 00:11:18,800
and its ability to generate code on demand

323
00:11:18,800 --> 00:11:20,700
that dramatically speeds up migration

324
00:11:20,700 --> 00:11:22,266
away from legacy systems

325
00:11:22,266 --> 00:11:24,333
those painful multi year implementations

326
00:11:24,433 --> 00:11:25,600
maybe they become obsolete

327
00:11:25,600 --> 00:11:27,566
replaced by by agentic workflows

328
00:11:27,566 --> 00:11:29,133
doing the rote data entry

329
00:11:29,233 --> 00:11:32,500
think CRM tools like day Dot AI or Adio

330
00:11:32,500 --> 00:11:33,200
automatically

331
00:11:33,200 --> 00:11:35,766
logging customer interactions from everywhere

332
00:11:36,000 --> 00:11:38,633
or AI native ERPs like Everest

333
00:11:38,633 --> 00:11:41,500
DOS roulette automating finance tasks

334
00:11:41,500 --> 00:11:42,833
so what are the key unlocks

335
00:11:42,833 --> 00:11:44,500
for founders trying to build these

336
00:11:44,500 --> 00:11:45,866
the report highlights a few

337
00:11:45,866 --> 00:11:46,666
using AI

338
00:11:46,666 --> 00:11:49,600
as a Trojan horse feature to get access to data flows

339
00:11:49,833 --> 00:11:52,300
achieving like 90% faster implementation

340
00:11:52,300 --> 00:11:53,700
thanks to code generation

341
00:11:53,700 --> 00:11:56,600
doing data migrations in maybe a day using AI

342
00:11:56,600 --> 00:11:58,633
and delivering a clear 10 x ROI

343
00:11:58,633 --> 00:12:00,100
compared to the old systems

344
00:12:00,200 --> 00:12:01,700
it's a completely different value proposition

345
00:12:01,700 --> 00:12:02,700
a total reinvention

346
00:12:02,700 --> 00:12:05,166
we're seeing similar wedges in HR recruiting

347
00:12:05,400 --> 00:12:08,466
enterprise search F P N a too right yes

348
00:12:08,466 --> 00:12:09,633
similar patterns are emerging

349
00:12:09,633 --> 00:12:11,100
across those horizontal categories

350
00:12:11,100 --> 00:12:13,600
but OK the dark matter here is enterprise

351
00:12:13,600 --> 00:12:15,400
ERP and the long tail of soar

352
00:12:16,066 --> 00:12:18,766
so while disrupting Salesforce might be happening

353
00:12:18,766 --> 00:12:22,233
replacing a giant complex manufacturing ERP system

354
00:12:22,233 --> 00:12:25,100
that's further off much further off according to report

355
00:12:25,666 --> 00:12:27,166
while there's exciting stuff in AI

356
00:12:27,166 --> 00:12:30,433
native accounting and ERP for SMBs

357
00:12:30,433 --> 00:12:32,366
and maybe mid market

358
00:12:32,366 --> 00:12:34,866
large scale enterprise ERP replacement

359
00:12:34,866 --> 00:12:36,733
especially for complex industries

360
00:12:36,766 --> 00:12:38,500
with intricate supply chains

361
00:12:38,766 --> 00:12:41,366
that's still years away why is that so much harder

362
00:12:41,466 --> 00:12:44,533
just the sheer inertia the existing deep integrations

363
00:12:44,600 --> 00:12:46,766
regulatory hurdles the complexity

364
00:12:47,000 --> 00:12:47,933
it's immense

365
00:12:48,166 --> 00:12:50,833
and then there's the long tail of other source

366
00:12:50,833 --> 00:12:54,100
identity platforms computer aided dispatch systems

367
00:12:54,200 --> 00:12:55,500
content management systems

368
00:12:55,500 --> 00:12:57,900
disrupting those is likely a decade long journey

369
00:12:57,900 --> 00:12:59,633
patience and capital needed there

370
00:12:59,633 --> 00:13:01,100
a lot of both got it

371
00:13:01,100 --> 00:13:02,633
now let's shift to vertical AI

372
00:13:02,633 --> 00:13:04,466
the report seems really bullish here

373
00:13:04,466 --> 00:13:08,266
noting how AI is tackling those technophobic verticals

374
00:13:08,266 --> 00:13:10,300
that traditional saws kind of struggled with

375
00:13:10,300 --> 00:13:13,133
what's AI's edge in these specific industries

376
00:13:13,200 --> 00:13:16,266
AI's edge is its ability to handle high value tasks

377
00:13:16,266 --> 00:13:17,500
that are often multimodal

378
00:13:17,500 --> 00:13:19,100
dealing with images voice

379
00:13:19,100 --> 00:13:21,566
sensor data or heavily language based

380
00:13:21,633 --> 00:13:23,033
these were often manual service

381
00:13:23,033 --> 00:13:25,033
intensive processes that old school

382
00:13:25,033 --> 00:13:26,466
sauce couldn't automate effectively

383
00:13:26,466 --> 00:13:27,700
okay so it cracks problem

384
00:13:27,700 --> 00:13:29,266
sauce couldn't exactly

385
00:13:29,266 --> 00:13:30,966
and we're seeing galaxies forming

386
00:13:30,966 --> 00:13:32,700
with really concrete examples

387
00:13:32,700 --> 00:13:34,966
of vertical specific workflow automation

388
00:13:34,966 --> 00:13:36,366
like what okay

389
00:13:36,366 --> 00:13:39,500
in healthcare a bridge automates clinical notes

390
00:13:39,500 --> 00:13:42,200
smartdicks health hospitals find missed revenue

391
00:13:42,366 --> 00:13:44,000
legal even up generates

392
00:13:44,000 --> 00:13:46,233
demand packages for personal injury cases

393
00:13:46,233 --> 00:13:48,766
Ivo reviews contracts education

394
00:13:48,900 --> 00:13:50,400
brisk teaching in magic school

395
00:13:50,400 --> 00:13:52,400
are building AI tools for teachers

396
00:13:52,466 --> 00:13:55,800
real estate Elise AI automates property management

397
00:13:55,800 --> 00:13:57,900
communications home services

398
00:13:57,966 --> 00:14:00,500
Hatch uses AI for customer service teams

399
00:14:00,600 --> 00:14:02,233
Rilla analyzes sales calls

400
00:14:02,233 --> 00:14:03,666
lots of specific examples

401
00:14:03,666 --> 00:14:07,000
are there patterns to these successful vertical AI

402
00:14:07,000 --> 00:14:07,766
companies yeah

403
00:14:07,766 --> 00:14:09,433
the report identifies three key patterns

404
00:14:09,433 --> 00:14:11,500
1 they start with a compelling wedge

405
00:14:11,500 --> 00:14:13,600
often a language heavy or multimodal task

406
00:14:13,600 --> 00:14:15,400
maybe involving voice or audio

407
00:14:15,400 --> 00:14:16,866
deeply embedded in a workflow

408
00:14:16,866 --> 00:14:17,866
hmm 2

409
00:14:17,866 --> 00:14:20,200
context is key they need deep domain expertise

410
00:14:20,200 --> 00:14:22,100
and build data models specific to that vertical

411
00:14:22,100 --> 00:14:23,666
3 they're built for value

412
00:14:23,666 --> 00:14:24,500
delivering clear

413
00:14:24,500 --> 00:14:27,500
often 10 x productivity gains or massive ROI right away

414
00:14:27,500 --> 00:14:30,000
makes sense but then there's the dark matter

415
00:14:30,433 --> 00:14:32,966
integrating with or competing against legacy systems

416
00:14:32,966 --> 00:14:36,300
in these verticals dealing with incumbents waking up

417
00:14:36,300 --> 00:14:38,433
and whether data moats are truly defensible

418
00:14:38,433 --> 00:14:39,233
and fragmented

419
00:14:39,233 --> 00:14:42,100
often privacy sensitive industries like healthcare

420
00:14:42,300 --> 00:14:43,800
how do founders navigate that

421
00:14:43,800 --> 00:14:45,966
it's definitely tricky that's the challenge

422
00:14:46,033 --> 00:14:47,600
deeply embedding into workflows

423
00:14:47,600 --> 00:14:49,166
helps create switching costs

424
00:14:49,400 --> 00:14:51,800
building truly proprietary data insights

425
00:14:51,866 --> 00:14:54,166
especially from unique data types like voice

426
00:14:54,166 --> 00:14:56,400
in that vertical builds defensibility

427
00:14:56,866 --> 00:14:58,300
but yeah it's a constant battle

428
00:14:58,300 --> 00:15:00,100
against the inertia of the old systems

429
00:15:00,100 --> 00:15:02,000
and incumbents who are starting to react

430
00:15:02,166 --> 00:15:04,166
you have to deliver undeniable value

431
00:15:04,166 --> 00:15:05,700
undeniable value okay

432
00:15:05,866 --> 00:15:08,200
we've covered enterprise verticals infrastructure

433
00:15:08,200 --> 00:15:10,366
what about us the everyday users

434
00:15:10,600 --> 00:15:12,700
the report also looks at consumer AI

435
00:15:12,700 --> 00:15:15,366
noting a shift beyond just basic productivity

436
00:15:15,366 --> 00:15:17,866
absolutely the consumer space is evolving

437
00:15:17,900 --> 00:15:20,633
we're seeing usage move into deeper areas like therapy

438
00:15:20,633 --> 00:15:22,366
companionship self growth

439
00:15:22,433 --> 00:15:25,400
but the core galaxies forming are still around AI

440
00:15:25,400 --> 00:15:26,966
assistance for everyday tasks

441
00:15:26,966 --> 00:15:29,066
and interestingly creation

442
00:15:29,066 --> 00:15:30,966
assistance and creation yeah

443
00:15:31,033 --> 00:15:32,800
general purpose LLMs like Chat

444
00:15:32,800 --> 00:15:33,966
GPT and Gemini

445
00:15:33,966 --> 00:15:35,300
are becoming daily habits

446
00:15:35,300 --> 00:15:37,266
for literally hundreds of millions

447
00:15:37,800 --> 00:15:40,133
voice interaction is also getting much bigger

448
00:15:40,266 --> 00:15:41,466
and then you have perplexity

449
00:15:41,466 --> 00:15:43,833
emerging as a breakout darling for AI

450
00:15:43,833 --> 00:15:47,300
native search I've heard a lot about perplexity right

451
00:15:47,300 --> 00:15:49,500
and their agentic browser concept

452
00:15:49,500 --> 00:15:52,700
comet could potentially redefine how we think about

453
00:15:52,700 --> 00:15:53,800
ambient agents

454
00:15:53,800 --> 00:15:56,000
AI just working for you in the background

455
00:15:56,433 --> 00:15:56,866
plus

456
00:15:56,866 --> 00:16:00,266
AI is dramatically lowering the barrier to creation

457
00:16:00,866 --> 00:16:02,266
think create dot X y

458
00:16:02,266 --> 00:16:05,466
z for building apps Suno and audio for Music

459
00:16:05,466 --> 00:16:07,966
Moon Valley Runway for multimedia

460
00:16:07,966 --> 00:16:11,100
exactly making creation accessible to way more people

461
00:16:11,166 --> 00:16:13,500
and we are seeing those purpose built AI assistants too

462
00:16:13,500 --> 00:16:14,566
especially in mental health

463
00:16:14,566 --> 00:16:16,566
Rosebud Finch although areas like email

464
00:16:16,566 --> 00:16:18,366
calendar seem tougher due to trust issues

465
00:16:18,366 --> 00:16:20,966
OK so what's the dark matter in consumer AI

466
00:16:20,966 --> 00:16:23,100
clear unsolved consumer pain points

467
00:16:23,100 --> 00:16:24,066
like what well

468
00:16:24,066 --> 00:16:25,433
it seems many obvious

469
00:16:25,433 --> 00:16:27,966
high value consumer use cases are still underserved

470
00:16:27,966 --> 00:16:30,000
because the underlying agentic infrastructure

471
00:16:30,000 --> 00:16:31,166
the tech that lets AI

472
00:16:31,166 --> 00:16:33,300
take complex actions safely and reliably

473
00:16:33,300 --> 00:16:34,166
is still maturing

474
00:16:34,166 --> 00:16:36,500
so the tech isn't quite there for some big ideas

475
00:16:36,633 --> 00:16:38,200
not quite robust enough yet

476
00:16:38,200 --> 00:16:39,966
for things hiding in plain sight

477
00:16:40,100 --> 00:16:41,500
like a truly personalized

478
00:16:41,500 --> 00:16:45,566
end to end travel concierge that books everything or

479
00:16:45,566 --> 00:16:47,800
smart shopping agents that don't just find deals

480
00:16:47,800 --> 00:16:50,466
but actually browse compare complex options

481
00:16:50,466 --> 00:16:53,066
and securely checkout for you across different sites

482
00:16:53,066 --> 00:16:54,700
that would be amazing right

483
00:16:54,900 --> 00:16:56,566
but it raises the big question

484
00:16:56,600 --> 00:16:58,700
who will ultimately own these powerful

485
00:16:58,700 --> 00:17:01,533
agentic use cases will it be the browser

486
00:17:01,600 --> 00:17:03,300
a general LLM assistant

487
00:17:03,400 --> 00:17:07,300
or a new wave of specialized purpose built agent apps

488
00:17:07,300 --> 00:17:10,400
a battleground for the next generation of consumer tech

489
00:17:10,400 --> 00:17:12,600
exactly the answer will likely determine

490
00:17:12,600 --> 00:17:14,633
the next big consumer platform winners

491
00:17:14,633 --> 00:17:16,866
okay that's a fascinating space to watch

492
00:17:16,866 --> 00:17:17,866
and speaking of the future

493
00:17:17,866 --> 00:17:21,600
let's pivot to Bessemer's five big predictions for AI

494
00:17:21,600 --> 00:17:22,766
in the coming years

495
00:17:22,766 --> 00:17:24,966
these are based on their partner's consensus

496
00:17:25,000 --> 00:17:26,666
yeah these are their boldest calls

497
00:17:26,666 --> 00:17:28,500
prediction number one the browser

498
00:17:28,500 --> 00:17:31,466
will emerge as a dominant interface for agentic AI

499
00:17:31,466 --> 00:17:32,866
we touch on this with comet

500
00:17:32,866 --> 00:17:34,000
why the browser

501
00:17:34,000 --> 00:17:36,966
they argue these new agentic browsers like comet or DIA

502
00:17:36,966 --> 00:17:38,066
aren't just add ons

503
00:17:38,200 --> 00:17:41,266
they're embedding AI right at the operating layer

504
00:17:41,400 --> 00:17:43,233
this allows for multi step automation

505
00:17:43,233 --> 00:17:46,066
intelligent actions across different websites and tabs

506
00:17:46,066 --> 00:17:48,100
seamlessly so it's deeply integrated

507
00:17:48,100 --> 00:17:51,400
exactly it's ubiquity in that deep integration make it

508
00:17:51,400 --> 00:17:52,000
in their view

509
00:17:52,000 --> 00:17:54,533
the most capable and inevitable interface layer

510
00:17:54,600 --> 00:17:55,500
for agents

511
00:17:55,500 --> 00:17:58,000
they even say the new browser wars have begun

512
00:17:58,000 --> 00:18:01,000
the browser as the iOS interesting

513
00:18:01,033 --> 00:18:02,233
okay prediction 2

514
00:18:02,233 --> 00:18:04,766
2026 will be the year of generative video

515
00:18:04,766 --> 00:18:07,300
we had images hitting maturity in 2024

516
00:18:07,300 --> 00:18:09,700
voice really taking off in 2025

517
00:18:09,866 --> 00:18:11,700
video is next it feels like video

518
00:18:11,700 --> 00:18:13,800
generation is moving incredibly fast

519
00:18:13,800 --> 00:18:16,266
it is models like Google's VO3

520
00:18:16,266 --> 00:18:18,466
cling from China open AI's Sora

521
00:18:18,466 --> 00:18:19,500
Moon valley's Marie

522
00:18:19,500 --> 00:18:21,266
they're rapidly improving in quality

523
00:18:21,266 --> 00:18:23,166
controllability realism

524
00:18:23,266 --> 00:18:24,700
they see it nearing a tipping point

525
00:18:24,700 --> 00:18:26,233
for real commercial viability

526
00:18:26,233 --> 00:18:27,433
at scale soon

527
00:18:27,433 --> 00:18:30,200
beyond just cool demos but what are the hurdles

528
00:18:30,200 --> 00:18:31,633
besides just making it look good

529
00:18:31,633 --> 00:18:33,966
well there are big questions about market structure

530
00:18:34,200 --> 00:18:35,800
will the big labs dominate

531
00:18:35,800 --> 00:18:37,100
can open source compete

532
00:18:37,233 --> 00:18:39,500
what about real time video generation

533
00:18:39,833 --> 00:18:42,000
and maybe most importantly

534
00:18:42,166 --> 00:18:43,566
the IP and copyright

535
00:18:43,566 --> 00:18:45,833
issues are getting exponentially more complex

536
00:18:45,833 --> 00:18:46,733
with video

537
00:18:46,866 --> 00:18:49,500
that's a major hurdle for widespread commercial use

538
00:18:49,500 --> 00:18:51,433
illegal and ethical minefield

539
00:18:51,433 --> 00:18:52,300
pretty much okay

540
00:18:52,300 --> 00:18:54,033
prediction three evils

541
00:18:54,033 --> 00:18:54,800
and data lineage

542
00:18:54,800 --> 00:18:56,700
will become a critical catalyst for AI

543
00:18:56,700 --> 00:18:59,200
product development this one sounds a bit technical

544
00:18:59,200 --> 00:19:01,466
but it's crucial evaluations okay

545
00:19:01,466 --> 00:19:03,566
why so important because enterprises

546
00:19:03,566 --> 00:19:05,900
aren't just looking for raw performance anymore

547
00:19:05,900 --> 00:19:09,000
they need confidence they need to trust the AI

548
00:19:09,300 --> 00:19:11,866
public benchmarks aren't enough for specific

549
00:19:11,866 --> 00:19:13,666
real world business applications

550
00:19:13,666 --> 00:19:15,266
so they need custom testing

551
00:19:15,300 --> 00:19:18,633
exactly the prediction is a big shift in 2025

552
00:19:18,633 --> 00:19:20,866
2026 towards private

553
00:19:20,866 --> 00:19:23,933
grounded and trusted internal evaluation suites

554
00:19:24,200 --> 00:19:26,133
Taylor to a company's own data

555
00:19:26,166 --> 00:19:29,000
their specific risks their industry regulations

556
00:19:29,033 --> 00:19:30,266
this sounds like the boring

557
00:19:30,266 --> 00:19:33,000
but essential plumbing that makes AI trustworthy

558
00:19:33,000 --> 00:19:34,400
especially in regulated fields

559
00:19:34,400 --> 00:19:36,000
like finance or healthcare

560
00:19:36,000 --> 00:19:37,100
that's a perfect way to put it

561
00:19:37,100 --> 00:19:38,100
for founders

562
00:19:38,200 --> 00:19:40,900
the real differentiator won't just be accuracy scores

563
00:19:40,900 --> 00:19:43,100
it'll be proving exactly how

564
00:19:43,100 --> 00:19:44,966
when and why your model works

565
00:19:44,966 --> 00:19:47,100
reliably in their specific environment

566
00:19:47,500 --> 00:19:48,800
tooling for this is emerging

567
00:19:48,800 --> 00:19:51,566
multi metric evils synthetic data environments

568
00:19:51,566 --> 00:19:53,100
from companies like Brain Trust

569
00:19:53,200 --> 00:19:54,600
Lane Chain Arclex

570
00:19:54,600 --> 00:19:56,600
you need built in evil infrastructure

571
00:19:56,600 --> 00:19:58,700
got it prediction four

572
00:19:58,966 --> 00:20:00,433
this one feels a bit SCI fi

573
00:20:00,433 --> 00:20:01,200
ah yeah

574
00:20:01,200 --> 00:20:04,233
a new AI native social media giant could emerge

575
00:20:04,233 --> 00:20:05,500
whoa like what

576
00:20:05,500 --> 00:20:08,133
well they draw parallels to past tech shifts

577
00:20:08,266 --> 00:20:10,000
PHP enabled Facebook mobile birth

578
00:20:10,000 --> 00:20:12,566
Instagram mobile video fueled TikTok

579
00:20:13,100 --> 00:20:14,266
what fundamental AI

580
00:20:14,266 --> 00:20:16,700
capability enables the next social giant

581
00:20:16,700 --> 00:20:17,900
what are the possibilities

582
00:20:17,900 --> 00:20:19,800
they speculate it can take different forms

583
00:20:19,900 --> 00:20:21,266
maybe a network where AI

584
00:20:21,266 --> 00:20:23,933
agents manage a lot of our social interactions for us

585
00:20:24,133 --> 00:20:26,900
or maybe platforms populated by emotionally intelligent

586
00:20:26,900 --> 00:20:28,133
AI influencers

587
00:20:28,166 --> 00:20:29,833
or even clones of real people

588
00:20:29,833 --> 00:20:31,666
building on things like character dot AI

589
00:20:31,666 --> 00:20:34,566
or replica AI agents managing our social lives

590
00:20:34,566 --> 00:20:37,766
or platforms full of Ais that's a wild thought

591
00:20:37,766 --> 00:20:39,800
could that make things like echo chambers worse

592
00:20:39,800 --> 00:20:41,866
or maybe offer different kinds of connection

593
00:20:41,866 --> 00:20:43,600
that's the huge unanswered question

594
00:20:43,600 --> 00:20:45,000
isn't it it's a real tension

595
00:20:45,066 --> 00:20:47,333
AI could personalize connections deeply

596
00:20:47,400 --> 00:20:49,566
or it could create incredibly sophisticated

597
00:20:49,566 --> 00:20:51,033
manipulation or echo chambers

598
00:20:51,033 --> 00:20:54,133
yeah or just feel uncanny

599
00:20:54,300 --> 00:20:56,633
the breakthroughs in voice memory

600
00:20:56,633 --> 00:20:58,333
generative media are the fuel

601
00:20:58,600 --> 00:21:01,866
but the societal implications are massive and unknown

602
00:21:01,866 --> 00:21:03,666
definitely one to watch okay

603
00:21:03,666 --> 00:21:05,300
the final prediction No. 5

604
00:21:05,400 --> 00:21:09,100
the incumbents strike back as AI MNA heats up

605
00:21:09,266 --> 00:21:09,766
basically

606
00:21:09,766 --> 00:21:12,200
after a couple years of watching startups disrupt

607
00:21:12,366 --> 00:21:13,300
the big enterprise

608
00:21:13,300 --> 00:21:16,766
giants are now aggressively buying AI capabilities

609
00:21:16,966 --> 00:21:19,233
so the big guys are waking up and opening their wallets

610
00:21:19,233 --> 00:21:21,433
yep especially in vertical software

611
00:21:21,433 --> 00:21:22,666
healthcare logistics

612
00:21:22,666 --> 00:21:25,233
legal tech they're requiring AI native companies

613
00:21:25,233 --> 00:21:26,666
not just to add features

614
00:21:26,666 --> 00:21:28,166
but often as a way to fundamentally

615
00:21:28,166 --> 00:21:29,700
reinvent their value proposition

616
00:21:29,700 --> 00:21:31,000
so for founders

617
00:21:31,666 --> 00:21:34,300
does this change the build versus sell calculation

618
00:21:34,300 --> 00:21:37,566
is the window closing or are new exit doors opening

619
00:21:37,566 --> 00:21:39,333
it definitely shifts the landscape

620
00:21:39,633 --> 00:21:42,866
it means founders need really strong technical modes

621
00:21:43,033 --> 00:21:44,333
strong data modes

622
00:21:44,466 --> 00:21:46,700
they need to be thinking about MNA strategy earlier

623
00:21:46,700 --> 00:21:47,500
perhaps

624
00:21:47,700 --> 00:21:50,266
and understanding where the incumbents are weak

625
00:21:50,600 --> 00:21:52,100
so they can position themselves

626
00:21:52,100 --> 00:21:53,766
as the essential missing piece

627
00:21:54,000 --> 00:21:55,166
it's both a threat

628
00:21:55,200 --> 00:21:58,200
and potentially a huge opportunity for strategic exits

629
00:21:58,400 --> 00:21:59,166
wow okay

630
00:21:59,166 --> 00:22:01,233
those are some really thought provoking predictions

631
00:22:01,233 --> 00:22:03,433
yeah so let's try to bring this all together

632
00:22:03,433 --> 00:22:04,900
what does this mean for you

633
00:22:04,900 --> 00:22:06,400
the listener maybe the builder

634
00:22:06,400 --> 00:22:07,600
the innovator out there

635
00:22:07,600 --> 00:22:10,000
what are the absolute key takeaways from this

636
00:22:10,000 --> 00:22:13,500
deep dive into the state of AI 2025

637
00:22:13,500 --> 00:22:14,266
yeah Best summer

638
00:22:14,266 --> 00:22:16,366
helpfully boils it down to 10 crucial points

639
00:22:16,366 --> 00:22:18,333
for founders first

640
00:22:18,366 --> 00:22:21,266
really understand those two AI startup archetypes

641
00:22:21,500 --> 00:22:23,366
are you aiming for supernova velocity

642
00:22:23,366 --> 00:22:25,300
or sustainable shooting star growth

643
00:22:25,433 --> 00:22:27,433
know your path okay second

644
00:22:27,433 --> 00:22:29,866
remember that memory and context are the new modes

645
00:22:30,033 --> 00:22:31,500
build products that learn

646
00:22:31,500 --> 00:22:32,866
adapt personalize

647
00:22:32,866 --> 00:22:34,066
that creates real stickiness

648
00:22:34,066 --> 00:22:36,466
got it memory in context third

649
00:22:36,466 --> 00:22:39,100
systems of action are replacing systems of record

650
00:22:39,100 --> 00:22:41,533
don't just sprinkle AI on old workflows

651
00:22:41,566 --> 00:22:43,233
use it to completely reimagine them

652
00:22:43,233 --> 00:22:46,166
reimagine not just enhance exactly fourth

653
00:22:46,400 --> 00:22:48,900
start with an AI wedge solve a narrow

654
00:22:48,900 --> 00:22:52,066
painful high friction problem with obvious 10 x value

655
00:22:52,066 --> 00:22:54,066
then expand from there find that wedge

656
00:22:54,066 --> 00:22:55,400
what's fifth fifth

657
00:22:55,400 --> 00:22:58,300
the browser is your canvas for agenic AI

658
00:22:58,700 --> 00:23:00,066
think about how your application

659
00:23:00,066 --> 00:23:01,166
can leverage the browser

660
00:23:01,233 --> 00:23:03,400
as this emerging AI operating layer

661
00:23:03,400 --> 00:23:06,100
okay browser as the platform 6th

662
00:23:06,100 --> 00:23:09,166
private continuous evaluation is mission critical

663
00:23:09,166 --> 00:23:10,666
especially for enterprise

664
00:23:10,700 --> 00:23:13,300
build in ways to prove your AI is trusted

665
00:23:13,300 --> 00:23:16,500
explainable and performs reliably in their world

666
00:23:16,500 --> 00:23:18,466
build in trust make sense

667
00:23:18,466 --> 00:23:19,166
seventh

668
00:23:19,166 --> 00:23:21,666
speed of implementation is a strategic advantage

669
00:23:21,766 --> 00:23:23,800
use AI to crush on boarding times

670
00:23:23,800 --> 00:23:25,800
and make switching to your solution painless

671
00:23:25,800 --> 00:23:27,300
that undermines vendor lock in

672
00:23:27,300 --> 00:23:30,100
faster time to value precisely eighth

673
00:23:30,100 --> 00:23:31,966
vertical AI is the new sauce

674
00:23:31,966 --> 00:23:34,400
those supposedly technophobic industries

675
00:23:34,566 --> 00:23:36,700
they're adopting AI fast

676
00:23:36,700 --> 00:23:39,000
when you solve their core high value problems

677
00:23:39,000 --> 00:23:41,800
with clear ROI go deep in verticals ninth

678
00:23:42,000 --> 00:23:43,433
comments are awake and acquisitive

679
00:23:43,433 --> 00:23:46,500
build strong notes know your value and be MNA ready

680
00:23:46,500 --> 00:23:48,566
if that's your goal be prepared for MNA

681
00:23:48,566 --> 00:23:50,433
and the last one and finally

682
00:23:50,433 --> 00:23:52,600
10th and maybe the most important

683
00:23:52,600 --> 00:23:55,400
taste and judgment are your differentiators

684
00:23:55,766 --> 00:23:58,800
in a world overflowing with AI capabilities

685
00:23:58,900 --> 00:24:01,366
human insight into what should be built

686
00:24:01,500 --> 00:24:03,200
what users truly need

687
00:24:03,200 --> 00:24:05,733
what constitutes a great product experience

688
00:24:05,800 --> 00:24:07,833
it's becoming the ultimate differentiator

689
00:24:07,833 --> 00:24:09,166
taste and judgment I like that

690
00:24:09,166 --> 00:24:10,833
it's not just about the tech itself

691
00:24:10,833 --> 00:24:12,833
not at all it's about applying it wisely

692
00:24:12,833 --> 00:24:15,166
this deep dive really hammers home

693
00:24:15,166 --> 00:24:18,233
just how dynamic this AI universe is

694
00:24:18,233 --> 00:24:20,200
the rules are changing so fast

695
00:24:20,200 --> 00:24:22,633
but the opportunities for real innovation

696
00:24:22,633 --> 00:24:23,966
for those with the right insight

697
00:24:23,966 --> 00:24:26,166
the right judgment they seem immense

698
00:24:26,400 --> 00:24:27,233
thank you for joining us

699
00:24:27,233 --> 00:24:30,166
on this exploration of the state of AI 2025

700
00:24:30,166 --> 00:24:32,300
absolutely and maybe a final thought

701
00:24:32,300 --> 00:24:34,000
building on the report's conclusion

702
00:24:34,000 --> 00:24:36,033
they say you don't just need a better model

703
00:24:36,033 --> 00:24:37,966
you need a better model of the world hmm

704
00:24:38,100 --> 00:24:40,633
the companies that win next won't do more AI

705
00:24:40,633 --> 00:24:42,900
they'll do the right AI at the right altitude

706
00:24:42,900 --> 00:24:44,100
with the right outcome

707
00:24:44,500 --> 00:24:46,400
so as you navigate this AI cosmos

708
00:24:46,400 --> 00:24:47,433
maybe reflect on that

709
00:24:47,433 --> 00:24:49,766
what is the right AI as the right level

710
00:24:49,800 --> 00:24:51,766
for the specific problem you're trying to solve

711
00:24:51,766 --> 00:24:53,733
and the value you truly wanna create