Transcriptly

How Top 1% AI-Native Organizations Actually Make Money | Harvard Business School, Rem Koning

EO18:16Added Aug 31, 2026

Rembrand M. Koning, Associate Professor at Harvard Business School, explains how AI is reshaping the playing field for entrepreneurs around the world.00:00 I...

Watch on YouTube →
Contributed by 刘嘉琪

Transcript

Transcript format

  1. Intro

  2. 00:00I wrote this paper 3 years ago looking
  3. 00:02at whether chat GPT delivered over
  4. 00:05WhatsApp could help small business
  5. 00:06entrepreneurs in Kenya for entrepreneurs
  6. 00:08who before our study were struggling.
  7. 00:10They had lower baseline profits and
  8. 00:13revenues. They saw a 10% decline in
  9. 00:15their profits and revenues from talking
  10. 00:17to the AI. Conversely, when we look at
  11. 00:18the people who were performing really
  12. 00:20well, the revenues and profits were
  13. 00:21above the median before our experiment.
  14. 00:23We find that they actually did better.
  15. 00:25How it might look different now? I think
  16. 00:27if we took GPT52 claudis and we put it
  17. 00:30behind WhatsApp, we get exactly the same
  18. 00:32results. Why would we get the same
  19. 00:33results if we got these better models? I
  20. 00:35think there's a really interesting thing
  21. 00:36when we look at the history of business.
  22. 00:38One way that you could get an edge was
  23. 00:39better at allocating stuff. The classic
  24. 00:42one is like allocating capital. Warren
  25. 00:44Buffett is better at allocating tapel
  26. 00:46and Birkshshire Haway than anyone else.
  27. 00:47Other companies are really good at
  28. 00:48allocating talent. I think we're in a
  29. 00:50world where increasingly what matters is
  30. 00:52your ability to allocate intelligence.
  31. 00:54If you can work that out better than
  32. 00:55other people, I think you've got an edge
  33. 00:56in the market.
  34. 00:59Hi, my name is Rem Coning. I'm a
  35. 01:01professor at Harvard Business School and
  36. 01:02I study entrepreneurship and AI. And I
  37. 01:04just like helping entrepreneurs do
  38. 01:06better. It could be a mogul in Silicon
  39. 01:08Valley. It could be someone selling
  40. 01:09coconuts in Indonesia. How do we help
  41. 01:11entrepreneurs bring new products to
  42. 01:12markets, compete better, grow their
  43. 01:14firms? And most recently, thinking about
  44. 01:15the role of AI as something that's just
  45. 01:17going to unlock a crazy amount of
  46. 01:19entrepreneurial potential.

  47. Lesson 1: Your Organization isn't AI native yet

  48. 01:35how AI is changing the way we build
  49. 01:38firms, right? We hear a lot about AI
  50. 01:40native firms and I think there are big
  51. 01:41questions about who can build them, uh
  52. 01:43how we scale them, how it changes
  53. 01:45strategy. So the AI founder sprint is an
  54. 01:47initiative that came out of INSEAD. We
  55. 01:49got over 500 entrepreneurs from all over
  56. 01:51the world. A quarter from Africa, a
  57. 01:53quarter from Asia, a quarter from the
  58. 01:55Americas, and a quarter from Europe. All
  59. 01:57building around AI. You're seeing people
  60. 01:59building stuff for AI and maternal
  61. 02:01health in Africa. You're seeing people
  62. 02:03build new edte startups in uh India.
  63. 02:06You're seeing stuff come out of Kenya.
  64. 02:07You're seeing awesome startups uh coming
  65. 02:09out of Europe, in the United States. And
  66. 02:11really what we did was we tracked how
  67. 02:13all these founders were using AI. I'll
  68. 02:16give you guys a little bit of a preview
  69. 02:17of the results which is that when you
  70. 02:19teach founders to be AI native when you
  71. 02:22tell them to really think about where
  72. 02:24they can apply generative AI not just
  73. 02:26chat GPT and like claude but the vibe
  74. 02:28coding tools multimodal tools all the
  75. 02:31agents that people are really excited
  76. 02:33about now when you tell them to really
  77. 02:35think about where to use that in their
  78. 02:36firm to move it forward it helps
  79. 02:37entrepreneurs everywhere do better so if
  80. 02:39you're building in Nigeria you are able
  81. 02:41to get more done every week about 20%
  82. 02:44more you're more likely to get
  83. 02:45customers, you're more likely to launch
  84. 02:47a product, you're more likely to have
  85. 02:49more revenue. And what's really crazy is
  86. 02:51even though you're more successful,
  87. 02:52right, you're growing faster. What we
  88. 02:54see is that these same founders say that
  89. 02:57they want to raise less capital. So,
  90. 02:58their demand for raising funds drops by
  91. 03:00$250,000.
  92. 03:02We're seeing folks really reimagine the
  93. 03:04workflows in their firm and build custom
  94. 03:07AI solutions. So, maybe the bottleneck
  95. 03:09for you is getting customers. How can
  96. 03:11you build new AI systems that
  97. 03:12automatically build out a marketing
  98. 03:14strategy and a go to market plan and not
  99. 03:16just build the plan but then execute it?
  100. 03:18And so that was one of the things that
  101. 03:19was most exciting I think from the
  102. 03:20sprint was seeing how people were taking
  103. 03:22whatever the bottleneck for their firm
  104. 03:24was it could be marketing maybe it was
  105. 03:26product development and they were using
  106. 03:27the off-the-shelf tools but then
  107. 03:29building basically their own agents if
  108. 03:31you will right so instead of headcount
  109. 03:32suddenly we're scaling just with on
  110. 03:34demand compute and that completely
  111. 03:36changes the economics of a business and
  112. 03:38what's possible particularly for
  113. 03:39founders outside of Silicon Valley I
  114. 03:42think a great example of it gamma
  115. 03:48I think the first thing that's really
  116. 03:49important when you're thinking about AI
  117. 03:50native is to understand that the value
  118. 03:52comes from two places. One is the
  119. 03:55process. You use AI in your coding. You
  120. 03:57use AI to do customer support tickets.
  121. 03:58And I think that's what we're all really
  122. 04:00familiar with, which is that we're sort
  123. 04:01of using AI to make our work go faster
  124. 04:03or make our work better. And that gives
  125. 04:05Gamma an edge. It helps. But really the
  126. 04:07key to Gamma is the way they've embedded
  127. 04:09AI into their product. And I think the
  128. 04:11key for AI native is that you're not
  129. 04:13just using it to do the work. you're
  130. 04:15embedding it in the product so that the
  131. 04:17AI can directly do the work with the
  132. 04:19customer. You want to take you as the
  133. 04:22human out of the loop. I love humans.
  134. 04:24We're amazing. It's great being on this
  135. 04:26uh call. It's great talking to people. I
  136. 04:28love sharing stories. But the problem
  137. 04:29with humans is we don't scale
  138. 04:31particularly well. And so if you were
  139. 04:33trying to make Gamma pregenerative AI,
  140. 04:35they'd probably have to employ, I don't
  141. 04:36know, tens of thousands, hundreds of
  142. 04:38thousands of graphic designers. The
  143. 04:39economics of the company would collapse.
  144. 04:41but instead by putting AI into the
  145. 04:44product, what Gamma is able to do is
  146. 04:47scale with compute rather than
  147. 04:49headcount. And I think it's a really
  148. 04:50exciting thing that if you're trying to
  149. 04:52be an AI native founder, that's what you
  150. 04:54need to find. Where are those places you
  151. 04:55can create loops where the AI is working
  152. 04:57with a user or another AI or something
  153. 04:59on the website where your team doesn't
  154. 05:00even need to be involved? That's the key
  155. 05:02to building AI native organizations.

  156. Allocate Intelligence

  157. 05:05Allocating intelligence. I think there's
  158. 05:07a really interesting thing when we look
  159. 05:08at the history of business. one way that
  160. 05:10you could get an edge was better at
  161. 05:12allocating stuff. The classic one is
  162. 05:14like allocating capital. So you have uh
  163. 05:16Warren Buffett is better at allocating
  164. 05:18capital and Birkshshire Hathaway than
  165. 05:20anyone else. He knew where to put his
  166. 05:22money and that made amazing amazing
  167. 05:24returns. Other companies are really good
  168. 05:25at allocating talent. So if you look at
  169. 05:27a company like McKenzie, big consulting
  170. 05:29company, they're really good at working
  171. 05:30out who should become partners, who
  172. 05:32should they hire at the base of the
  173. 05:33pyramid, matching that talent with the
  174. 05:35right clients, what they're really good
  175. 05:36at. I think we're in a world where
  176. 05:38increasingly what matters is your
  177. 05:40ability to allocate intelligence. And
  178. 05:41what that means is you need to allocate
  179. 05:43what is done by different models. What
  180. 05:45are you going to have Claude do? What
  181. 05:46are you going to have Lovable do? What
  182. 05:48are you going to have Grock or Deepseek
  183. 05:50do? Right? Thinking about how you blend,
  184. 05:52how you orchestrate, how you allocate
  185. 05:54your product to these different sorts of
  186. 05:55intelligences is just incredibly
  187. 05:58important. But I think this is the the
  188. 06:00key, which is that you also need to work
  189. 06:01out how to allocate what's being done by
  190. 06:03the AI and what's being done by humans
  191. 06:06because at the end of the day, we still
  192. 06:08have some edge over some of these
  193. 06:09models. And even if they're better or
  194. 06:11faster at thinking, often we think
  195. 06:13differently. And when you're thinking
  196. 06:15about strategy and you're thinking about
  197. 06:16how to gain an edge in the market, it's
  198. 06:18not about necessarily doing something
  199. 06:19better. It's about doing something
  200. 06:21different. Doing something in a way
  201. 06:22nobody else can. And so if you can work
  202. 06:24out how you bring your human
  203. 06:26intelligence and you allocate jobs in
  204. 06:28the company to humans and the places
  205. 06:30where they can add value over and above
  206. 06:32the models or do things differently than
  207. 06:33the models can and then you work out how
  208. 06:35to blend those together. I think that's
  209. 06:37a place where we're going to see a
  210. 06:38source of advantage moving forward and
  211. 06:39it's a really exciting time to play cuz
  212. 06:41I think all of us are struggling how to
  213. 06:43allocate our own intelligence. like what
  214. 06:44should I have TGPT do and what should I
  215. 06:46do is a thing I know I struggle with
  216. 06:48every day but increasingly this is going
  217. 06:50to be a question at the strategic level
  218. 06:51for firms like what should you as an
  219. 06:53entrepreneur do right and what should
  220. 06:55you give to an AI and which AI should
  221. 06:57you give it to if you can work that out
  222. 06:58better than other people I think you've
  223. 07:00got an edge in the market
  224. 07:03is AI an equalizer an amplifier I'm

  225. Lesson 2: Equalizer or Amplifier? - Why some thrive while others fall behind

  226. 07:06going to say the standard professor
  227. 07:08answer which is it depends or maybe it's
  228. 07:10both um but let's get a little bit
  229. 07:12deeper we all now can code with lovable
  230. 07:14We can all build amazing decks with
  231. 07:15Gamma. All of us can use chat GPT and
  232. 07:17claude to get rid of typos and have a
  233. 07:19copy editor in our pocket. Wow, it is an
  234. 07:21equalizer, right? It is amazing. It
  235. 07:23moves everybody up, right? We can all do
  236. 07:26so much more. But here's the problem,
  237. 07:28right? I think that's for the existing
  238. 07:29work that we do. Whether you're applying
  239. 07:31AI over an existing task that you have
  240. 07:33or you're building a new sort of
  241. 07:35business, when you're building a new
  242. 07:36sort of business, a new sort of product,
  243. 07:38when you're thinking about how AI is
  244. 07:39going to change your firm, be it a small
  245. 07:41business or a tech startup, the returns
  246. 07:43to thinking about how AI can do this are
  247. 07:45going to be greatest for those who have
  248. 07:46the ability to do that. And those are
  249. 07:47going to be people who are already
  250. 07:48pretty good. Those are going to be
  251. 07:49people who've developed the judgment.
  252. 07:51Maybe they started a company before.
  253. 07:52They probably have a stronger technical
  254. 07:54background. Those are the people going
  255. 07:55to be able to imagine the really big
  256. 07:57wins and get those huge returns. They're
  257. 07:59going to see their judgment, their
  258. 08:00agency amplified.
  259. 08:02I wrote this paper 3 years ago whether
  260. 08:05chat GPT delivered over WhatsApp could
  261. 08:08help small business entrepreneurs in
  262. 08:09Kenya improve their business's
  263. 08:11performance. And what we found was
  264. 08:12really surprising, which was that four
  265. 08:15entrepreneurs who before our study were
  266. 08:17struggling, they had lower baseline
  267. 08:19profits and revenues, they saw a 10%
  268. 08:22decline in their profits and revenues
  269. 08:24from talking to the AI. It would have
  270. 08:26been better had we never given them the
  271. 08:27AI. And what we find is that the reason
  272. 08:29for them is that they ask the AI a lot
  273. 08:31of questions. They use it, they interact
  274. 08:33with it, but they get a lot of advice
  275. 08:35from the AI and they don't know how to
  276. 08:36pick the good advice from the bad
  277. 08:38advice. They don't have the judgment to
  278. 08:40separate what's good from bad, which
  279. 08:41might explain why they were low
  280. 08:42performing in the first place.
  281. 08:44Conversely, when we look at the people
  282. 08:45who are performing really well, their uh
  283. 08:47revenues and profits were above the
  284. 08:48median before our experiment, the better
  285. 08:50entrepreneurs, we find that they
  286. 08:52actually did better. And when we look at
  287. 08:53the chat logs, the reason we see is that
  288. 08:55they're asking kind of the same sorts of
  289. 08:57questions as the low performers, but
  290. 08:58they're then following up and following
  291. 09:00the advice that isn't the bad advice,
  292. 09:02it's the good advice. Unless you've
  293. 09:04developed the judgment, the mental
  294. 09:05models to actually know where to apply
  295. 09:07it, it can lead you down a road of slop.
  296. 09:10And that slop can actually lead you to
  297. 09:12make less money. And it's really
  298. 09:13interesting as we've been working on
  299. 09:15this paper, it's now almost 3 years old,
  300. 09:17how it might look different now than
  301. 09:19back then. I think if we took Claude
  302. 09:21Opus and we put it behind WhatsApp, we
  303. 09:23get exactly the same results. The issue
  304. 09:25is you still have the same problem,
  305. 09:26which is that the entrepreneur asks a
  306. 09:28question and the AI gives them four or
  307. 09:30five plausible things to do and you need
  308. 09:32to know which one's actually right for
  309. 09:34you. So that's the first thing. The
  310. 09:35second thing is I think if we were doing
  311. 09:37the study now, we probably wouldn't do
  312. 09:39it through a chatbot. I think a big
  313. 09:40mistake you're seeing entrepreneurs make
  314. 09:42is that we're still stuck in the chatbot
  315. 09:44world. Chad GPT launched, it was huge.
  316. 09:46It changed the way we thought about
  317. 09:47artificial intelligence, about how we
  318. 09:49use our computers. And it changed us and
  319. 09:51sort of locked us into this idea that
  320. 09:53there was a a chatbot for everything.
  321. 09:55We'll have a chatbot for Shopify. We'll
  322. 09:57have a chatbot for these entrepreneurs.
  323. 09:58Harvard Business School has internal
  324. 10:00chat bots. We'll just build chat bots
  325. 10:01everywhere. And it turns out we don't
  326. 10:04need more chatbots. So imagine you go to
  327. 10:06a chatbot and it says, "Oh, you need to
  328. 10:07update your website so you can get more
  329. 10:09sales." If I'm a Kenyan entrepreneur and
  330. 10:10I don't know how to code, how am I
  331. 10:12updating my website? So I think if I was
  332. 10:14doing this study today, the thing that
  333. 10:15would be really exciting is giving more
  334. 10:17agentic AI, right, to the entrepreneurs
  335. 10:20today. Could you give them the tools to
  336. 10:22build better websites or launch better
  337. 10:23marketing campaigns? A lot of these
  338. 10:25entrepreneurs, they're struggling. They
  339. 10:27don't have extra money. They don't have
  340. 10:28extra staff. They don't have extra
  341. 10:30people to do something for them. Could
  342. 10:31you build them virtual employees that
  343. 10:34help them run their business and expand
  344. 10:36the things that they're good at and get
  345. 10:38more done during the day? these are
  346. 10:39people who are working hard and often
  347. 10:41the constraint is just time. They don't
  348. 10:43have the time to do it. Could you use AI
  349. 10:45to help them do more in the time that
  350. 10:46they have? Um, so that's something we're
  351. 10:48exploring in some current work right
  352. 10:49now, which I'm really excited about. And
  353. 10:51I think the more we get in the mindset
  354. 10:53of it's not going to be we're going to
  355. 10:54having these conversations with the AI,
  356. 10:56but that we're going to be telling them
  357. 10:58and they're going to go take actions in
  358. 10:59the world on behalf of us, I think
  359. 11:01that's really exciting. It's going to
  360. 11:02open up a lot more modalities in terms
  361. 11:04of interfaces, right? and maybe that
  362. 11:06we're talking to them like how we're
  363. 11:07having a conversation right now. I think
  364. 11:09it's also a key unlock for startups all
  365. 11:11around the world. You're never going to
  366. 11:12build AI systems better than open AI or
  367. 11:14anthropics. Sorry, they're at the
  368. 11:15frontier. They've got billions of
  369. 11:17dollars in funding, but what you do have
  370. 11:19is better contextual knowledge of a
  371. 11:20workflow, of a situation, of how things
  372. 11:23work in different parts of the world.
  373. 11:24And I think that can give you a real
  374. 11:26edge because if you can get that right
  375. 11:27context into the AI, oh my god, what it
  376. 11:29can do is absolutely amazing. And I
  377. 11:31think you're seeing no better example of
  378. 11:32this right now than with skills in cloud
  379. 11:34code, right? So people are making these
  380. 11:36skills. What are these skills? They're
  381. 11:38kind of context. They're like little
  382. 11:39snippets of how to do a particular task
  383. 11:41where you've told the AI how to do it.
  384. 11:43You've given it the context for how to
  385. 11:44think about it. And then we can share
  386. 11:46these skills with everybody else in the
  387. 11:47world. And it's just wild to see how
  388. 11:50effective this is at making the models
  389. 11:52better.
  390. 11:56I think more people will be

  391. Lesson 3: The Next Trillion Dollar Businesses - The playing field is shifting

  392. 11:57entrepreneurs, right? more people are
  393. 11:59going to go out and build their own
  394. 12:00businesses and I think more people in
  395. 12:02companies are going to behave like
  396. 12:04entrepreneurs partly because we can all
  397. 12:06build now right and I think that's one
  398. 12:07of the things that's always attracted me
  399. 12:09to entrepreneurs whether you're building
  400. 12:10a car wash or building a a software
  401. 12:11company it's like you have to create
  402. 12:13something from nothing I think
  403. 12:14increasingly all of us are going to do
  404. 12:16it so the world is going to look a lot
  405. 12:17more entrepreneurial and then you look
  406. 12:18at developing markets and I think
  407. 12:20there's a opportunity for them to
  408. 12:22leapfrog like they've done with fintech
  409. 12:24right if you go to India India's payment
  410. 12:26infrastructure uh this thing called the
  411. 12:27UPI is light years ahead of the United
  412. 12:30States. You can pay with everything on
  413. 12:31your phone. It's unlocked billions of
  414. 12:33dollars in value, maybe even more. Um
  415. 12:36there's been a huge number of startups
  416. 12:37around it. They have these amazing
  417. 12:39financial g gateways. I think there's an
  418. 12:41opportunity with AI around this and sort
  419. 12:42of knowledge work around the world. But
  420. 12:44I think to do that, we need to make sure
  421. 12:45the AI systems have context from
  422. 12:47emerging markets. Do they know enough
  423. 12:49about the Kenyan entrepreneurs to guide
  424. 12:50them? I think thinking about what are
  425. 12:52the big knowledge problems in these
  426. 12:54places, particularly education, I think
  427. 12:56AI could have a profound effect there.
  428. 12:58I'm really excited to see more of the
  429. 13:00application layer come to emerging
  430. 13:03markets and I think there's reason to be
  431. 13:05optimistic, right? We've all familiar
  432. 13:06with the inference cost curves, right?
  433. 13:08The cost of, you know, calling a GPT4
  434. 13:10quality model just keeps going down
  435. 13:12exponentially, you know, to a year from
  436. 13:14now, two years from now, it's going to
  437. 13:16be basically free. when things become
  438. 13:17basically free, I think it opens up a
  439. 13:20lot of opportunity to work in markets
  440. 13:22where people just have less money to
  441. 13:24spend. I think to get like really
  442. 13:25concrete on the business side, I think
  443. 13:28it can give people labor that they
  444. 13:29couldn't hire otherwise. There's often a
  445. 13:31lack of experts. Like you just can't
  446. 13:33find someone to help you with marketing.
  447. 13:34They're not there. They didn't get that
  448. 13:35college training. If I can hire a
  449. 13:37virtual agent who's as good as a
  450. 13:38marketer in New York or Silicon Valley
  451. 13:40and I'm in Nairobi, holy moly, is that
  452. 13:43amazing. I think that'd be a really
  453. 13:44concrete one. help them do their
  454. 13:46marketing, help them export more, help
  455. 13:48them skill their workforce potentially.
  456. 13:50When we sort of step back, then there's
  457. 13:52even a bigger question. How is AI going
  458. 13:53to transform the economy? And the flip
  459. 13:55side of more entrepreneurship is that I
  460. 13:57think we're going to see a proliferation
  461. 13:59of software tackling problems that we
  462. 14:01never even imagined it could tackle.
  463. 14:02Some of these are going to be really
  464. 14:03deep, like AlphaFold. That's going to be
  465. 14:05awesome. Some of these though are going
  466. 14:07to be much more mundane. So, I'm in
  467. 14:10Thailand and there isn't a CRM for my
  468. 14:12restaurant business. Well, now
  469. 14:14somebody's going to take Lovable or
  470. 14:16Codeex or whatever it is and they're
  471. 14:18going to make the world's best CRM for
  472. 14:20Thai restaurant owners and that's going
  473. 14:22to solve this person's problem and help
  474. 14:24that business grow. And I think there's
  475. 14:25probably millions if not billions of
  476. 14:27other problems that software could start
  477. 14:30to solve. And I'm really excited about
  478. 14:32that because software is awesome.
  479. 14:34Marginal costs are low. It's incredibly
  480. 14:36scalable. It can turn knowledge into
  481. 14:38something that can help millions if not
  482. 14:40billions of people. And so I think what
  483. 14:42we're going to see is a world that looks
  484. 14:43more like the software economy. I think
  485. 14:45there's a downside to that. If we look
  486. 14:47at the past 20, 30, 40 years, software
  487. 14:50has also led to potentially a
  488. 14:51concentration in wealth. We have these
  489. 14:53sort of, you know, mega billionaires at
  490. 14:55the top who have these marketplace
  491. 14:57platform and network effects businesses
  492. 14:59like Facebook that really h have
  493. 15:01controlled a lot of the digital
  494. 15:02ecosystem. And so I think there's a big
  495. 15:04question on the policy side of how we
  496. 15:06prevent that from happening again. I
  497. 15:07think one thing that's really exciting
  498. 15:08about vibe coding and how it might
  499. 15:10change the economy is that I don't think
  500. 15:12we're in a world of network effects. I
  501. 15:14think we're in a world of small like
  502. 15:16kind of SAS applications. I think we're
  503. 15:18in a world of uh bootstrapping
  504. 15:20businesses that don't need that VC
  505. 15:22investment. And so I am a little bit
  506. 15:23hopeful that this might be something
  507. 15:25that's going to spread prosperity more
  508. 15:26broadly. Though I do think we need to
  509. 15:28think about uh what it might do to the
  510. 15:30concentration of wealth if more of the
  511. 15:32world starts looking like software and
  512. 15:33tech.
  513. 15:35The most dangerous assumption they make
  514. 15:38is that by building with AI, they have
  515. 15:40made something people want. And that is
  516. 15:42just not true. Just like traditional
  517. 15:43software, you can build with AI and
  518. 15:45nobody can want it. I'm seeing this
  519. 15:47basically loop where people get stuck
  520. 15:49cuz the AI tools are so fun. So you're
  521. 15:51like, Claude, make me something. Codeex,
  522. 15:52make me something. And then you're like,
  523. 15:54let's add this other feature. And then
  524. 15:55like, let's do another one. And a month
  525. 15:56goes by and you've built the most
  526. 15:58beautiful piece of software. It's crazy
  527. 16:00overengineered. And then you launch and
  528. 16:01nobody wants it. And so I think some of
  529. 16:03the traditional stuff around building
  530. 16:05software still holds. Get it in front of
  531. 16:07users, have users play with it, see what
  532. 16:09they want. I think a correlary of that
  533. 16:11is a little bit of AI goes a long way.
  534. 16:14Going back to the example of Gamma,
  535. 16:16really the core of their AI at the
  536. 16:18beginning was let's just have people
  537. 16:19write a couple of sentences and then
  538. 16:21we'll generate a deck. Everything else
  539. 16:22was traditional software. That one
  540. 16:24unlock, oh my gosh, led to an amazing
  541. 16:27business. Can you find that one place in
  542. 16:29someone's workflow where you can apply a
  543. 16:31little bit of AI, just a drop or two,
  544. 16:33and that unlocks how they can do
  545. 16:35something, changes a problem that was
  546. 16:37really hard for them, that's what you
  547. 16:39really want to find. So, I think finding
  548. 16:40the smallest point where you can use
  549. 16:42these AI systems is really valuable.
  550. 16:44Somehow, it's going to give them an
  551. 16:46insight without them having the earned
  552. 16:48insight, right? I think it's more
  553. 16:50important than ever for founders to have
  554. 16:51a real earned insight, to have the
  555. 16:53judgment, to have the taste, to know how
  556. 16:56and where to apply these tools. I think
  557. 16:58that's the thing you need to spend your
  558. 16:59time on, not thinking about how do you
  559. 17:01get access to the next foundation model,
  560. 17:02whatever it is. Use last generation's
  561. 17:05model. It's probably fine for the
  562. 17:06purposes of what you're building. What
  563. 17:08matters is can you figure out where and
  564. 17:10how to apply it? And that's a different
  565. 17:11skill set than necessarily just cranking
  566. 17:13through the AI engineering.
  567. 17:15There are two things that are changing.

  568. Next episode

  569. 17:17how we're learning is changing and those
  570. 17:19who are really going to accelerate are
  571. 17:21going to be those who just realize you
  572. 17:23can be learning way more than you were
  573. 17:25before and you just need to unlock that.
  574. 17:27My name is Drew Bent. I lead education
  575. 17:29at Anthropic. One of the things that I
  576. 17:31think holds us all back is we give AI
  577. 17:34tools pretty simple problems when we
  578. 17:36could be giving them much more complex
  579. 17:38problems. We are not elevating our
  580. 17:41ambition with what we can do with these
  581. 17:43AI tools. As the AI tools get smarter,
  582. 17:46where could I push you further? Where
  583. 17:47could you have pushed me further? And
  584. 17:49eventually, if we're looking ahead,
  585. 17:51there will be in some cases, I think,
  586. 17:53this inversion of control where actually
  587. 17:55the AI model is doing some of the
  588. 17:57highest level strategic thinking and
  589. 17:59then delegating to you, the human for
  590. 18:01the areas that require human taste,
  591. 18:04human agency.