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 00:00:16,766 --> 00:00:19,766 welcome to episode 43 Today 9 00:00:19,766 --> 00:00:20,966 we're taking a deep dive 10 00:00:20,966 --> 00:00:23,200 into the future of artificial intelligence 11 00:00:23,400 --> 00:00:25,366 pulling insights from a recent AI 12 00:00:25,366 --> 00:00:28,933 Ascent 2,025 keynote from Sequoia Capital 13 00:00:28,966 --> 00:00:32,133 the venture capital powerhouse based in Silicon Valley 14 00:00:33,200 --> 00:00:34,700 Sequoia discussed the current 15 00:00:34,700 --> 00:00:36,800 state and future potential of AI 16 00:00:37,100 --> 00:00:39,933 highlighted the rapid pace of AI development 17 00:00:39,966 --> 00:00:43,000 and its potential to significantly disrupt and expand 18 00:00:43,000 --> 00:00:43,800 markets 19 00:00:43,800 --> 00:00:46,900 drawing parallels to the impact of the cloud transition 20 00:00:47,400 --> 00:00:49,766 tune in as we unpack what's happening now 21 00:00:49,766 --> 00:00:52,466 and what's coming next in the world of AI 22 00:00:52,466 --> 00:00:55,266 today we're jumping straight into Sequoia Capital's AI 23 00:00:55,266 --> 00:00:57,566 Ascent 2025 keynote 24 00:00:57,866 --> 00:01:01,233 we know you want the core insights on AI's future fast 25 00:01:01,233 --> 00:01:02,766 without getting lost in the weeds 26 00:01:02,766 --> 00:01:05,000 yeah and that's exactly what we aim to deliver 27 00:01:05,000 --> 00:01:06,666 this event it um 28 00:01:06,800 --> 00:01:09,300 it brought together some really sharp people in AI 29 00:01:09,300 --> 00:01:11,566 right lots of different perspectives definitely 30 00:01:11,600 --> 00:01:13,300 but there was this common thread 31 00:01:13,300 --> 00:01:15,466 you know this idea of a massive 32 00:01:15,466 --> 00:01:17,733 trillion dollar opportunity in AI 33 00:01:17,900 --> 00:01:20,066 but also how do you actually navigate it 34 00:01:20,066 --> 00:01:22,066 it's moving so quickly exactly 35 00:01:22,066 --> 00:01:23,666 so our goal today is simple 36 00:01:24,066 --> 00:01:26,033 pull out those key insights for you 37 00:01:26,033 --> 00:01:27,633 we'll look at where AI is now 38 00:01:27,633 --> 00:01:29,366 where it seems to be heading mid to long term 39 00:01:29,366 --> 00:01:31,900 and crucially what are some well 40 00:01:31,900 --> 00:01:33,266 actionable takeaways let's 41 00:01:33,266 --> 00:01:35,000 start right there with the so what 42 00:01:35,000 --> 00:01:36,666 why should you listening now 43 00:01:36,666 --> 00:01:38,833 really care deeply about this AI shift 44 00:01:38,833 --> 00:01:40,433 they used a pretty interesting analogy 45 00:01:40,433 --> 00:01:42,633 comparing it to the cloud transition right 46 00:01:42,633 --> 00:01:43,466 but they immediately 47 00:01:43,466 --> 00:01:45,600 pointed out a massive difference in scale 48 00:01:45,633 --> 00:01:46,400 you might think OK 49 00:01:46,400 --> 00:01:49,000 AI another service layer like the cloud was 50 00:01:49,033 --> 00:01:50,966 but the numbers tell a different story 51 00:01:51,166 --> 00:01:53,100 the cloud market big as it got 52 00:01:53,100 --> 00:01:54,800 maybe $400 billion 53 00:01:54,833 --> 00:01:56,766 was roughly the size of the software market 54 00:01:56,766 --> 00:01:58,066 when it started exactly 55 00:01:58,066 --> 00:02:00,500 now contrast that with AI 56 00:02:00,700 --> 00:02:04,100 the projection is that the market for AI 57 00:02:04,100 --> 00:02:05,500 and this includes services 58 00:02:05,500 --> 00:02:07,400 and how it changes software itself 59 00:02:07,466 --> 00:02:09,500 could be an order of magnitude bigger 60 00:02:09,600 --> 00:02:12,266 right from the start an order of magnitude bigger 61 00:02:12,266 --> 00:02:13,766 that's that's huge 62 00:02:13,766 --> 00:02:14,800 it really is 63 00:02:14,800 --> 00:02:17,100 and it's not just about adding a new service 64 00:02:17,233 --> 00:02:19,133 the key inside I think 65 00:02:19,166 --> 00:02:19,866 was that AI 66 00:02:19,866 --> 00:02:22,100 could fundamentally change what software is 67 00:02:22,100 --> 00:02:23,566 how so well 68 00:02:23,566 --> 00:02:25,933 think about moving from selling tools 69 00:02:25,966 --> 00:02:29,400 which fits neatly in a software budget to selling uh 70 00:02:29,400 --> 00:02:30,700 actual outcomes 71 00:02:30,766 --> 00:02:32,633 that might pull from completely different budgets 72 00:02:32,633 --> 00:02:34,800 like labor budgets ah okay 73 00:02:34,800 --> 00:02:36,500 so you're not buying the shovel anymore 74 00:02:36,500 --> 00:02:38,266 you're buying the finished hole 75 00:02:38,266 --> 00:02:40,200 so to speak something like that yeah 76 00:02:40,266 --> 00:02:42,033 it potentially expands the whole pie 77 00:02:42,033 --> 00:02:43,433 not just slices it differently 78 00:02:43,433 --> 00:02:45,966 it really reframes where value gets created 79 00:02:45,966 --> 00:02:47,433 now beyond just the size 80 00:02:47,433 --> 00:02:49,066 they hammered home the speed 81 00:02:49,166 --> 00:02:50,433 the acceleration they did 82 00:02:50,433 --> 00:02:51,833 they use this layer cake analogy 83 00:02:51,833 --> 00:02:53,033 which I thought was quite effective 84 00:02:53,033 --> 00:02:54,033 explain that a bit 85 00:02:54,033 --> 00:02:56,366 so the idea is that each big tech wave 86 00:02:56,366 --> 00:02:57,966 like the internet mobile 87 00:02:57,966 --> 00:03:00,700 cloud builds on the previous layers 88 00:03:01,266 --> 00:03:03,100 AI is the latest layer OK 89 00:03:03,100 --> 00:03:04,700 makes sense but two key things here 90 00:03:04,700 --> 00:03:07,000 first the ingredients for this AI layer 91 00:03:07,000 --> 00:03:08,666 the computing power the networks 92 00:03:08,666 --> 00:03:09,966 the data the talent 93 00:03:10,100 --> 00:03:11,466 they're already pretty mature 94 00:03:11,466 --> 00:03:14,833 they're in place AI isn't some distant dream right 95 00:03:14,833 --> 00:03:16,333 the foundations are solid 96 00:03:16,366 --> 00:03:19,866 and second the pace of these layers is accelerating 97 00:03:19,966 --> 00:03:21,866 each new wave arrives faster 98 00:03:21,866 --> 00:03:23,633 and spreads quicker than the last 99 00:03:23,633 --> 00:03:25,700 so AI isn't just bigger it's happening much 100 00:03:25,700 --> 00:03:28,766 much faster that classic hockey stick curve 101 00:03:28,766 --> 00:03:30,266 but maybe even steeper this time 102 00:03:30,266 --> 00:03:33,333 precisely and they dug into why it's adopting so fast 103 00:03:33,333 --> 00:03:34,700 what's really fueling it okay 104 00:03:34,700 --> 00:03:36,666 what are the drivers three main things 105 00:03:36,666 --> 00:03:38,733 people know about it people want it 106 00:03:38,866 --> 00:03:41,300 and people can get it simple but powerful 107 00:03:41,300 --> 00:03:42,800 awareness first yeah 108 00:03:42,800 --> 00:03:44,000 you can thank things like chat 109 00:03:44,000 --> 00:03:46,666 GPT for making AI incredibly visible 110 00:03:46,666 --> 00:03:48,800 almost overnight to the general public 111 00:03:49,066 --> 00:03:51,100 it's not some obscure tech anymore true 112 00:03:51,100 --> 00:03:53,266 then the desire amplified by 113 00:03:53,266 --> 00:03:55,633 you know massive online communities right 114 00:03:55,633 --> 00:03:57,533 Reddit X places like that 115 00:03:57,666 --> 00:03:59,900 they act as global accelerators 116 00:03:59,900 --> 00:04:01,733 for discovering and wanting new tech 117 00:04:01,900 --> 00:04:03,966 the buzz built incredibly fast 118 00:04:03,966 --> 00:04:07,400 a built in demand engine and finally accessibility 119 00:04:07,900 --> 00:04:10,500 nearly everyone's online the pipes 120 00:04:10,500 --> 00:04:14,000 the infrastructure it's already there for AI to spread 121 00:04:14,000 --> 00:04:15,566 unlike the early cloud days 122 00:04:15,566 --> 00:04:17,166 where you had to build out the infrastructure 123 00:04:17,166 --> 00:04:18,666 and convince people exactly 124 00:04:18,766 --> 00:04:20,866 AI landed on a stage that was already 125 00:04:20,866 --> 00:04:24,033 built globally it really shows how tech distribution 126 00:04:24,033 --> 00:04:25,766 itself has changed okay 127 00:04:25,766 --> 00:04:28,400 so massive opportunity accelerating fast 128 00:04:28,400 --> 00:04:32,500 got it but where do you actually play to win that 129 00:04:32,500 --> 00:04:33,600 the million dollar question 130 00:04:33,600 --> 00:04:35,900 or I guess the trillion dollar question here right 131 00:04:35,900 --> 00:04:37,433 and they acknowledged look 132 00:04:37,433 --> 00:04:39,000 maybe the completely blank canvas 133 00:04:39,000 --> 00:04:40,633 isn't quite as blank as a year ago 134 00:04:40,633 --> 00:04:42,666 but that white space is still enormous 135 00:04:42,666 --> 00:04:44,500 so plenty room still absolutely 136 00:04:44,500 --> 00:04:46,166 if you look back at previous tech shifts 137 00:04:46,166 --> 00:04:47,166 internet mobile 138 00:04:47,166 --> 00:04:49,000 most of the value wasn't actually in the core 139 00:04:49,000 --> 00:04:51,866 infrastructure it was in the applications built on top 140 00:04:51,866 --> 00:04:53,500 e commerce social media 141 00:04:53,500 --> 00:04:54,966 all those apps on our phones 142 00:04:54,966 --> 00:04:57,266 exactly that's where the bulk of the value landed 143 00:04:57,266 --> 00:04:59,233 mm hmm and the initial thought was 144 00:04:59,233 --> 00:05:01,366 AI would follow that same pattern 145 00:05:01,633 --> 00:05:03,900 build applications on top of the core models 146 00:05:03,900 --> 00:05:05,066 but there's a wrinkle now 147 00:05:05,066 --> 00:05:07,633 right with the foundation models themselves 148 00:05:07,633 --> 00:05:08,633 getting so capable 149 00:05:08,633 --> 00:05:10,566 that's the key evolution they pointed out 150 00:05:10,566 --> 00:05:12,000 these big foundation models 151 00:05:12,000 --> 00:05:13,866 yeah they're not just static platforms anymore 152 00:05:13,866 --> 00:05:15,266 they're getting better and better 153 00:05:15,266 --> 00:05:17,566 and starting to do things that look a lot like 154 00:05:17,566 --> 00:05:18,833 applications themselves 155 00:05:18,833 --> 00:05:22,266 making it more competitive at that app layer definitely 156 00:05:22,566 --> 00:05:25,400 so for startups or anyone building here 157 00:05:25,500 --> 00:05:27,100 the advice was pretty clear 158 00:05:27,100 --> 00:05:29,966 which was focus customer back 159 00:05:30,633 --> 00:05:31,366 really 160 00:05:31,366 --> 00:05:34,900 really understand a specific customer's pain point 161 00:05:35,000 --> 00:05:36,666 go deep into a vertical industry 162 00:05:36,666 --> 00:05:38,366 or a specific business function 163 00:05:38,366 --> 00:05:40,166 solve a real specific problem 164 00:05:40,166 --> 00:05:42,233 yes tackle something complex 165 00:05:42,233 --> 00:05:45,833 maybe niche where AI can offer clear value and 166 00:05:45,833 --> 00:05:46,800 maybe for now 167 00:05:46,800 --> 00:05:49,200 that solution even involves having a human in the loop 168 00:05:49,200 --> 00:05:51,666 it doesn't have to be pure AI magic from day one 169 00:05:51,666 --> 00:05:52,766 practical advice 170 00:05:52,766 --> 00:05:54,266 they mentioned something called the Leoni 171 00:05:54,266 --> 00:05:55,700 Merchandising cycle yeah 172 00:05:55,700 --> 00:05:58,466 it was a quick mention but basically a framework 173 00:05:58,466 --> 00:06:02,033 reminding everyone that even with AI 174 00:06:02,033 --> 00:06:04,633 you still need to do the basics of building a business 175 00:06:04,633 --> 00:06:05,700 you know develop the product 176 00:06:05,700 --> 00:06:07,000 engineer it market it 177 00:06:07,000 --> 00:06:08,266 sell it support it 178 00:06:08,666 --> 00:06:10,500 the fundamentals don't disappear right 179 00:06:10,500 --> 00:06:11,700 the tech is exciting 180 00:06:11,700 --> 00:06:13,966 but the business building still matters 181 00:06:13,966 --> 00:06:14,666 and this ties 182 00:06:14,666 --> 00:06:17,666 into how you build a sustainable advantage 183 00:06:17,900 --> 00:06:19,900 a moat there's this tension 184 00:06:19,900 --> 00:06:23,266 right between focusing purely on the coolest tech 185 00:06:23,266 --> 00:06:24,766 the tech out view 186 00:06:25,000 --> 00:06:27,200 versus starting with the customer need 187 00:06:27,200 --> 00:06:28,266 the customer back view 188 00:06:28,266 --> 00:06:30,800 and they argued for the customer approach strongly 189 00:06:31,066 --> 00:06:32,433 they argued that long term 190 00:06:32,433 --> 00:06:34,066 defensibility in AI 191 00:06:34,066 --> 00:06:35,833 is increasingly going to come from that 192 00:06:35,833 --> 00:06:37,600 deep customer understanding 193 00:06:37,633 --> 00:06:39,566 and the unique value you build around it 194 00:06:39,566 --> 00:06:41,433 how do you build that moat practically 195 00:06:41,433 --> 00:06:42,666 any examples yeah 196 00:06:42,666 --> 00:06:43,800 they offered a few strategies 197 00:06:43,800 --> 00:06:46,200 things like building an end to end solution 198 00:06:46,266 --> 00:06:47,300 not just a point tool 199 00:06:47,300 --> 00:06:49,766 becoming indispensable to the workflow exactly 200 00:06:49,766 --> 00:06:51,566 or building data and the other big one was 201 00:06:51,566 --> 00:06:53,566 getting really industry specific 202 00:06:53,766 --> 00:06:56,033 like open evidence in healthcare 203 00:06:56,033 --> 00:06:57,466 Harvey in legal 204 00:06:57,666 --> 00:07:00,933 deep domain expertise creates a strong moat 205 00:07:01,133 --> 00:07:03,366 it's all about that connection with the user 206 00:07:03,366 --> 00:07:04,666 okay that makes sense 207 00:07:04,666 --> 00:07:07,033 now they also talked about specific things 208 00:07:07,033 --> 00:07:09,433 AI companies need to watch out for this idea 209 00:07:09,433 --> 00:07:13,233 of revenue vibe versus real revenue caught my attention 210 00:07:13,233 --> 00:07:14,433 yeah let's unpack that 211 00:07:14,433 --> 00:07:16,766 revenue vibe is like the initial excitement 212 00:07:16,766 --> 00:07:19,200 the buzz maybe some early sign UPS 213 00:07:19,200 --> 00:07:20,766 it looks good on the surface 214 00:07:20,766 --> 00:07:22,466 it might be superficial exactly 215 00:07:22,466 --> 00:07:25,000 real revenue or maybe better put 216 00:07:25,000 --> 00:07:28,200 real engagement is about sustainable user behavior 217 00:07:28,200 --> 00:07:30,000 are people actually adopting it 218 00:07:30,000 --> 00:07:32,000 using it regularly sticking with it 219 00:07:32,000 --> 00:07:34,900 long term retention that's what matters 220 00:07:34,900 --> 00:07:38,000 so look past the hype look at the core metrics 221 00:07:38,000 --> 00:07:40,200 absolutely and tied very closely to that 222 00:07:40,200 --> 00:07:42,233 is what they call good vibes with customers 223 00:07:42,233 --> 00:07:44,366 yeah which is basically trust 224 00:07:44,366 --> 00:07:45,833 trust seems huge right now 225 00:07:45,833 --> 00:07:46,733 it's Paramount 226 00:07:46,800 --> 00:07:49,300 especially now while the tech is still finding its feet 227 00:07:49,500 --> 00:07:51,666 if your users trust you trust your vision 228 00:07:51,666 --> 00:07:53,366 trust that you'll iterate and improve 229 00:07:53,366 --> 00:07:55,233 that's maybe even more valuable than your initial 230 00:07:55,233 --> 00:07:56,766 feature set interesting 231 00:07:56,766 --> 00:07:59,300 build trust first what about margins 232 00:07:59,500 --> 00:08:01,533 AI can be expensive to run 233 00:08:01,566 --> 00:08:03,033 they address that yes 234 00:08:03,033 --> 00:08:05,000 initial gross margins might be lower 235 00:08:05,000 --> 00:08:06,166 because of compute costs 236 00:08:06,166 --> 00:08:08,500 the a cost per token to run the models right 237 00:08:08,500 --> 00:08:10,866 but the expectation 238 00:08:10,966 --> 00:08:14,800 or maybe the hope is that margins improve over time 239 00:08:14,966 --> 00:08:16,966 two reasons the underlying cost 240 00:08:16,966 --> 00:08:18,966 per token is generally trending down 241 00:08:19,100 --> 00:08:21,666 and companies should be able to charge more 242 00:08:21,666 --> 00:08:22,966 as they deliver more value 243 00:08:22,966 --> 00:08:24,833 moving from tools to those outcomes 244 00:08:24,833 --> 00:08:25,500 we talked about 245 00:08:25,500 --> 00:08:28,666 value based pricing as the AI gets better potentially 246 00:08:28,666 --> 00:08:29,566 yes yeah 247 00:08:29,566 --> 00:08:32,200 and finally they circled back to the data flywheel idea 248 00:08:32,200 --> 00:08:33,200 with a key condition right 249 00:08:33,200 --> 00:08:34,000 yes a 250 00:08:34,000 --> 00:08:36,766 data flywheel isn't magic just because you have data 251 00:08:36,800 --> 00:08:39,466 it only counts if that flywheel demonstrably improves 252 00:08:39,466 --> 00:08:42,133 a key business metric better product 253 00:08:42,166 --> 00:08:44,500 better user experience higher retention 254 00:08:44,700 --> 00:08:46,033 something tangible otherwise 255 00:08:46,033 --> 00:08:47,366 it's just data sitting there 256 00:08:47,366 --> 00:08:48,900 got it data needs to do work 257 00:08:48,900 --> 00:08:50,466 okay shifting gear slightly 258 00:08:50,466 --> 00:08:52,966 they talked about the sheer momentum behind AI 259 00:08:52,966 --> 00:08:54,866 this feeling of inevitability 260 00:08:54,866 --> 00:08:57,133 yeah they use the phrase nature hates a vacuum 261 00:08:57,500 --> 00:08:57,833 it said 262 00:08:57,833 --> 00:09:00,033 there's this tremendous sucking sound in the market 263 00:09:00,033 --> 00:09:02,633 pulling AI in everywhere that's a vivid image 264 00:09:02,633 --> 00:09:04,933 what was the point the main point was 265 00:09:04,966 --> 00:09:08,000 don't get overly distracted by macroeconomic headwinds 266 00:09:08,000 --> 00:09:09,333 yes those exist 267 00:09:09,766 --> 00:09:13,366 but the fundamental pull of this technological shift 268 00:09:13,366 --> 00:09:16,433 this AI adoption wave is potentially so strong 269 00:09:16,433 --> 00:09:18,366 it overrides typical economic cycles 270 00:09:18,366 --> 00:09:21,166 so the tech trend is the bigger force right now 271 00:09:21,166 --> 00:09:22,333 that was the argument 272 00:09:22,366 --> 00:09:25,466 the implication being you need urgency 273 00:09:25,466 --> 00:09:28,200 you need velocity if you're sitting on the sidelines 274 00:09:28,200 --> 00:09:30,366 someone else is probably eating your lunch 275 00:09:30,366 --> 00:09:32,900 or inventing a new kind of lunch entirely 276 00:09:33,233 --> 00:09:35,533 a clear call for speed wait 277 00:09:35,566 --> 00:09:38,000 okay then they did a kind of year in review 278 00:09:38,033 --> 00:09:39,566 looking back at progress 279 00:09:39,566 --> 00:09:41,600 from both the customer side and the tech side right 280 00:09:41,600 --> 00:09:44,200 the customer back in technology out perspectives 281 00:09:44,266 --> 00:09:46,166 and the customer data was pretty compelling 282 00:09:46,166 --> 00:09:47,033 what stood out 283 00:09:47,033 --> 00:09:50,166 engagement specifically for AI native apps 284 00:09:50,166 --> 00:09:52,200 things built around AI from the start 285 00:09:52,400 --> 00:09:54,366 they showed how the ratio of daily users 286 00:09:54,366 --> 00:09:56,233 to monthly users for platforms like Chat 287 00:09:56,233 --> 00:09:58,600 GPT has shot up it's approaching 288 00:09:58,600 --> 00:10:01,100 the stickiness of established platforms like Reddit 289 00:10:01,100 --> 00:10:02,833 so people aren't just trying it once 290 00:10:02,833 --> 00:10:04,766 they're integrating it into their routines 291 00:10:04,800 --> 00:10:05,300 it seems 292 00:10:05,300 --> 00:10:10,533 so it suggests AI is delivering real ongoing value 293 00:10:10,600 --> 00:10:11,966 not just novelty 294 00:10:12,033 --> 00:10:14,700 and they mentioned real world impact expanding too 295 00:10:14,700 --> 00:10:16,100 yeah beyond just chat 296 00:10:16,266 --> 00:10:17,966 areas like advertising education 297 00:10:17,966 --> 00:10:20,566 healthcare they name dropped open evidence again 298 00:10:20,566 --> 00:10:23,666 suggesting AI is making real diagnostic differences 299 00:10:23,900 --> 00:10:26,566 and then there was the her moment for voice AI 300 00:10:26,566 --> 00:10:27,233 ah yes 301 00:10:27,233 --> 00:10:28,433 like the movie meaning 302 00:10:28,433 --> 00:10:30,700 meaning voice generation tech has gotten so good 303 00:10:30,700 --> 00:10:32,600 it's basically crossed the uncanny valley 304 00:10:32,600 --> 00:10:35,900 it sounds human which opens up well 305 00:10:35,900 --> 00:10:38,600 a ton of possibility huge implications there 306 00:10:38,600 --> 00:10:40,733 any other standout applications 307 00:10:40,766 --> 00:10:43,500 coding that came up as a major breakout area 308 00:10:43,500 --> 00:10:46,466 tools like anthropics Claude 3.5 sonnet 309 00:10:46,466 --> 00:10:48,100 finding real product market fit 310 00:10:48,100 --> 00:10:50,366 making developers significantly more productive 311 00:10:50,366 --> 00:10:52,666 that's a big one changing how soccer gets built 312 00:10:52,666 --> 00:10:53,600 absolutely now 313 00:10:53,600 --> 00:10:56,133 on the technology outside the pure tech advances 314 00:10:56,200 --> 00:10:58,033 they did note that maybe the pace 315 00:10:58,033 --> 00:10:59,400 of just making the biggest models 316 00:10:59,400 --> 00:10:59,966 even bigger 317 00:10:59,966 --> 00:11:02,766 that pre training race might be slowing down a bit 318 00:11:02,766 --> 00:11:04,400 some low hanging fruit might be picked 319 00:11:04,400 --> 00:11:07,633 okay a potential plateau in the sheer scale potentially 320 00:11:07,633 --> 00:11:09,900 but they quickly highlighted key areas 321 00:11:09,900 --> 00:11:11,366 where breakthroughs are happening 322 00:11:11,433 --> 00:11:13,666 such as better reasoning AI 323 00:11:13,666 --> 00:11:16,233 getting better at actually thinking through problems 324 00:11:16,233 --> 00:11:17,966 not just pattern matching open AI 325 00:11:17,966 --> 00:11:19,366 I gotta mention there okay 326 00:11:19,366 --> 00:11:22,033 also big strides in synthetic data 327 00:11:22,033 --> 00:11:24,900 creating artificial data to train AI 328 00:11:24,900 --> 00:11:28,566 which is crucial better tool use AI using 329 00:11:28,566 --> 00:11:30,633 external software and APIs more effectively 330 00:11:30,633 --> 00:11:31,100 and better 331 00:11:31,100 --> 00:11:33,666 AI scaffolding techniques to build more complex 332 00:11:33,666 --> 00:11:35,466 reliable AI systems 333 00:11:35,466 --> 00:11:37,466 so refinement and capability building 334 00:11:37,466 --> 00:11:39,466 not just raw size exactly 335 00:11:39,466 --> 00:11:41,933 pushing towards more sophisticated tasks 336 00:11:42,033 --> 00:11:43,700 and they also praised innovation 337 00:11:43,700 --> 00:11:46,200 happening right at the edge of research and product 338 00:11:46,466 --> 00:11:49,266 mentioning deep research in Notebook LM 339 00:11:49,266 --> 00:11:50,866 which is now called Hugh I believe 340 00:11:50,866 --> 00:11:53,033 taking cutting edge ideas and making them usable 341 00:11:53,033 --> 00:11:55,433 right which brings us back to that key question 342 00:11:55,433 --> 00:11:58,600 yeah in this whole AI stack from the chips to the apps 343 00:11:58,666 --> 00:12:00,266 where does the most value end up 344 00:12:00,266 --> 00:12:01,800 and their answer was still 345 00:12:01,800 --> 00:12:03,566 still firmly at the application layer 346 00:12:03,800 --> 00:12:05,700 despite the foundation models getting better 347 00:12:05,700 --> 00:12:06,233 they believe 348 00:12:06,233 --> 00:12:08,500 companies solving specific customer problems 349 00:12:08,766 --> 00:12:11,166 r v open evidence for the examples again 350 00:12:11,166 --> 00:12:13,100 are where the biggest value capture will be 351 00:12:13,100 --> 00:12:14,600 even with the increased competition 352 00:12:14,600 --> 00:12:16,300 from the models themselves seems 353 00:12:16,300 --> 00:12:18,900 so the opportunity to build targeted 354 00:12:18,900 --> 00:12:22,166 user loved applications is still seen as massive 355 00:12:22,366 --> 00:12:24,900 though they did give a nod to Jensen Huang at Nvidia 356 00:12:24,900 --> 00:12:27,233 of course can't ignore the hardware layer right now 357 00:12:27,233 --> 00:12:29,066 he's kind of the king of the stack for the moment 358 00:12:29,066 --> 00:12:30,400 you could say that yeah 359 00:12:30,400 --> 00:12:32,366 but then focusing back on apps 360 00:12:32,400 --> 00:12:35,833 they pointed to the first real cohort of AI killer apps 361 00:12:35,833 --> 00:12:37,200 emerging who made the list 362 00:12:37,200 --> 00:12:38,166 Chat GPT 363 00:12:38,166 --> 00:12:41,066 actually Harvey Glean Sierra Cursor 364 00:12:41,233 --> 00:12:43,033 a Bridge Listen Labs 365 00:12:43,033 --> 00:12:45,300 Open evidence a diverse group 366 00:12:45,300 --> 00:12:47,233 but all delivering significant value 367 00:12:47,233 --> 00:12:50,200 and looking forward they see more agent first companies 368 00:12:50,200 --> 00:12:51,433 yes companies 369 00:12:51,433 --> 00:12:53,566 where the core product is an AI agent 370 00:12:54,266 --> 00:12:56,800 moving beyond today's often brittle prototypes 371 00:12:56,800 --> 00:12:57,900 towards really robust 372 00:12:57,900 --> 00:13:00,766 reliable agents that can perform complex tasks 373 00:13:00,766 --> 00:13:02,833 how do we get to those robust agents 374 00:13:02,833 --> 00:13:04,566 two main paths are discussed 375 00:13:04,566 --> 00:13:06,300 one involves clever orchestration 376 00:13:06,300 --> 00:13:08,900 coordinating multiple AI models and tools 377 00:13:09,066 --> 00:13:11,166 with really strong testing and evaluation 378 00:13:11,166 --> 00:13:12,666 to ensure reliability okay 379 00:13:12,666 --> 00:13:13,900 managing complexity 380 00:13:13,900 --> 00:13:16,400 the other path is creating agents that are fine tuned 381 00:13:16,400 --> 00:13:18,700 specifically for an end to end task 382 00:13:18,700 --> 00:13:19,966 really specialized 383 00:13:19,966 --> 00:13:22,633 Langchain and Openai were mentioned as enablers here 384 00:13:22,633 --> 00:13:25,966 and this leads to the idea of vertical agents 385 00:13:25,966 --> 00:13:29,000 exactly this seemed like a really big prediction 386 00:13:29,000 --> 00:13:31,000 agents trained with deep expertise 387 00:13:31,000 --> 00:13:33,000 in just one specific domain 388 00:13:33,000 --> 00:13:34,200 like the examples they gave 389 00:13:34,200 --> 00:13:35,433 yeah really striking ones 390 00:13:35,433 --> 00:13:36,600 an AI agent 391 00:13:36,600 --> 00:13:39,700 beating human penetration testers at cyber security 392 00:13:39,866 --> 00:13:43,400 AI outperforming humans in complex Devops 393 00:13:43,400 --> 00:13:44,866 or network troubleshooting 394 00:13:45,266 --> 00:13:48,766 it suggests a future of highly specialized AI experts 395 00:13:48,766 --> 00:13:50,700 wow AI specialist 396 00:13:50,700 --> 00:13:53,600 which then connects to this idea of an abundance era 397 00:13:53,833 --> 00:13:55,533 code was the first example 398 00:13:55,633 --> 00:13:58,766 if AI makes certain types of labor like basic coding 399 00:13:58,800 --> 00:14:01,500 super cheap and abundant then scarcity shifts elsewhere 400 00:14:01,500 --> 00:14:03,366 right to things like taste judgment 401 00:14:03,366 --> 00:14:05,166 quality control creativity 402 00:14:05,300 --> 00:14:07,400 fundamentally changes the economic equation 403 00:14:07,400 --> 00:14:08,800 that's a profound shift 404 00:14:08,800 --> 00:14:10,466 so taking that agent idea further 405 00:14:10,466 --> 00:14:12,633 what were the mid to long term predictions 406 00:14:12,633 --> 00:14:15,766 it moved from individual agents to agent swarms 407 00:14:15,766 --> 00:14:18,166 mm hmm the idea that agents won't just work alone 408 00:14:18,166 --> 00:14:19,766 but will collaborate coordinate 409 00:14:19,766 --> 00:14:21,466 form interconnected systems 410 00:14:21,466 --> 00:14:22,800 okay teams of agents 411 00:14:22,800 --> 00:14:26,733 and that evolves into the concept of an agent economy 412 00:14:27,166 --> 00:14:27,633 a future 413 00:14:27,633 --> 00:14:29,866 where these agents aren't just passing information 414 00:14:29,866 --> 00:14:32,333 but are actually transferring resources 415 00:14:32,366 --> 00:14:33,766 making transactions 416 00:14:33,866 --> 00:14:36,100 operating with some understanding of trust 417 00:14:36,233 --> 00:14:38,300 like autonomous economic actors 418 00:14:38,300 --> 00:14:40,800 agents doing business with each other and with us 419 00:14:40,800 --> 00:14:43,766 that's the vision but they cautioned 420 00:14:43,866 --> 00:14:44,500 getting there 421 00:14:44,500 --> 00:14:47,033 requires solving some major technical challenges 422 00:14:47,033 --> 00:14:48,633 like what first 423 00:14:48,633 --> 00:14:50,066 persistent identity 424 00:14:50,566 --> 00:14:52,566 agents need a stable sense of self 425 00:14:52,566 --> 00:14:53,100 memory 426 00:14:53,100 --> 00:14:56,300 history to build reliable interactions over time 427 00:14:56,400 --> 00:14:58,200 they can't just reset every conversation 428 00:14:58,200 --> 00:14:59,433 they need to remember things 429 00:14:59,433 --> 00:15:01,566 learn exactly second 430 00:15:01,566 --> 00:15:03,300 seamless communication protocols 431 00:15:03,700 --> 00:15:06,500 we need the equivalent of tcpip for agents 432 00:15:06,900 --> 00:15:09,133 standard ways for them to exchange information 433 00:15:09,166 --> 00:15:10,900 value and signals of trust 434 00:15:11,033 --> 00:15:13,300 they mentioned MCP as a potential standard 435 00:15:13,300 --> 00:15:15,266 emerging a common language for agents 436 00:15:15,266 --> 00:15:16,700 makes sense and third 437 00:15:16,766 --> 00:15:18,966 maybe most critical security 438 00:15:19,466 --> 00:15:21,466 if agents are making autonomous transactions 439 00:15:21,466 --> 00:15:23,700 trust and security become absolutely vital 440 00:15:23,800 --> 00:15:24,300 you need to know 441 00:15:24,300 --> 00:15:26,100 the agent you're dealing with is legitimate 442 00:15:26,100 --> 00:15:29,066 and secure that's a huge challenge and opportunity 443 00:15:29,066 --> 00:15:29,800 absolutely 444 00:15:29,800 --> 00:15:32,666 trust in a world run partly by autonomous agents 445 00:15:33,500 --> 00:15:34,966 okay beyond the tech 446 00:15:35,100 --> 00:15:37,800 does this require us to change how we think 447 00:15:37,866 --> 00:15:41,900 definitely they highlighted three key mindset shifts 448 00:15:41,900 --> 00:15:45,033 the first is adopting a stochastic mindset 449 00:15:45,033 --> 00:15:46,966 sarcastic meaning probabilistic 450 00:15:46,966 --> 00:15:48,633 not deterministic moving 451 00:15:48,633 --> 00:15:49,866 away from expecting computers 452 00:15:49,866 --> 00:15:51,166 to give the exact same answer 453 00:15:51,166 --> 00:15:53,800 every time like 2 + 2 is 4 454 00:15:53,866 --> 00:15:56,233 AI responses might vary slightly 455 00:15:56,233 --> 00:15:59,500 be probabilistic but still correct within context 456 00:15:59,500 --> 00:16:01,666 we need to get comfortable with that uncertainty right 457 00:16:01,666 --> 00:16:03,066 it's not traditional programming 458 00:16:03,066 --> 00:16:04,633 second a management mindset 459 00:16:04,633 --> 00:16:06,000 as we work more with agents 460 00:16:06,000 --> 00:16:07,866 our role shifts from doing the task 461 00:16:07,866 --> 00:16:10,766 to managing the agent understanding its capabilities 462 00:16:10,766 --> 00:16:12,066 its limitations guiding it 463 00:16:12,066 --> 00:16:13,766 overseeing its work less operator 464 00:16:13,766 --> 00:16:15,566 more manager precisely and third 465 00:16:15,566 --> 00:16:17,400 embracing increased leverage 466 00:16:17,400 --> 00:16:19,266 but also accepting less certainty 467 00:16:19,366 --> 00:16:21,433 agents can amplify our output massively 468 00:16:21,433 --> 00:16:22,700 let us scale faster with 469 00:16:22,766 --> 00:16:23,933 your people but that comes with 470 00:16:23,933 --> 00:16:25,733 relying on these probabilistic systems 471 00:16:25,733 --> 00:16:28,133 managing that inherent uncertainty and risk 472 00:16:28,133 --> 00:16:30,200 more power more potential variants 473 00:16:30,266 --> 00:16:31,700 a trade off it seems 474 00:16:31,700 --> 00:16:33,766 so ultimately the picture they painted 475 00:16:33,766 --> 00:16:35,666 was one of massive transformation 476 00:16:35,666 --> 00:16:38,266 driven by these agents exponential leverage 477 00:16:38,266 --> 00:16:39,300 faster scaling 478 00:16:39,300 --> 00:16:42,066 reduced human workforce in some areas perhaps 479 00:16:42,133 --> 00:16:43,200 and eventually maybe 480 00:16:43,200 --> 00:16:44,333 these individual agents 481 00:16:44,333 --> 00:16:46,366 merge into something even more complex 482 00:16:46,366 --> 00:16:48,900 like neural networks of neural networks 483 00:16:48,900 --> 00:16:51,533 hmm fundamentally changing work companies 484 00:16:51,533 --> 00:16:53,666 the economy it's a pretty radical vision 485 00:16:53,666 --> 00:16:54,733 it really is yeah 486 00:16:54,733 --> 00:16:57,600 so to sum up this deep dive into the keynote 487 00:16:57,700 --> 00:17:01,400 it's a huge AI opportunity moving incredibly fast 488 00:17:01,700 --> 00:17:03,300 the tech is advancing rapidly 489 00:17:03,300 --> 00:17:04,566 especially around reasoning 490 00:17:04,566 --> 00:17:06,666 tool use and specialized agents 491 00:17:06,966 --> 00:17:09,566 the value seems to be accruing at the application layer 492 00:17:09,566 --> 00:17:11,233 focusing on customer problems 493 00:17:11,233 --> 00:17:12,066 and the longer view 494 00:17:12,066 --> 00:17:14,533 points towards this transformative agent economy 495 00:17:14,533 --> 00:17:16,633 exactly which brings us back to you 496 00:17:16,633 --> 00:17:17,433 our listener 497 00:17:17,466 --> 00:17:19,466 you've now got a handle on these key insights 498 00:17:19,466 --> 00:17:20,866 the scale the speed 499 00:17:20,933 --> 00:17:22,900 the focus on applications and agents 500 00:17:22,900 --> 00:17:25,066 you understand the shift from tools to outcomes 501 00:17:25,066 --> 00:17:27,166 the importance of trust the rise of 502 00:17:27,466 --> 00:17:29,666 AI in this potential abundance era 503 00:17:29,666 --> 00:17:32,166 so maybe the final thought to leave you with is this 504 00:17:32,933 --> 00:17:35,833 as these AI agents become more integrated 505 00:17:35,933 --> 00:17:38,966 more autonomous operating within this agent economy 506 00:17:39,466 --> 00:17:41,533 what happens next what are the challenges 507 00:17:41,533 --> 00:17:44,233 the opportunities maybe the weird side effects 508 00:17:44,233 --> 00:17:45,500 that weren't even discussed 509 00:17:45,500 --> 00:17:45,733 yeah 510 00:17:45,733 --> 00:17:48,766 what does it really mean for trust on a societal level 511 00:17:48,766 --> 00:17:51,533 for human creativity what does work even look like 512 00:17:51,533 --> 00:17:54,366 when your collaborators are increasingly AI agents 513 00:17:54,366 --> 00:17:55,566 lots to think about 514 00:17:55,566 --> 00:17:57,733 definitely a landscape that will keep evolving