--- author: EO date: '2026-03-06' guest: '' layout: post.njk source: https://www.youtube.com/watch?v=I08ZeY7QcrE speaker: EO tags: - entrepreneurship - artificial-intelligence - business-strategy - emerging-markets - ai-native-firms title: AI如何重塑创业格局：从硅谷到全球的智能增长新篇章 summary: 本文深入探讨了AI技术对创业的颠覆性影响，分析了AI作为赋能者和放大器的双重角色，强调了“智能分配”和“领域知识”在AI时代的重要性。文章指出，AI不仅降低了创业门槛，更催生了AI原生企业和新兴市场的跨越式发展，但也提醒创业者需警惕过度依赖AI工具而忽视用户需求，并关注AI可能带来的财富集中问题。 insight: '' draft: true series: '' category: ai-application area: tech-engineering project: [] people: [] companies_orgs: - Harvard Business School - INSEAD - Gamma - OpenAI - Anthropic products_models: - ChatGPT - GPT-4o - Claude - Lovable - Codeex - Grock - Deepseek - Vibe coding tools - Multimodal tools media_books: [] status: evergreen --- ### AI对创业者的初期影响与智能分配的兴起 这篇论文是我三年前撰写的，旨在研究通过 WhatsApp 交付的 ChatGPT 是否能帮助肯尼亚的小型企业创业者。对于研究开始前本就挣扎于低利润和低营收的企业家来说，他们在使用 AI 后，利润和营收反而下降了 10%。反之，那些在实验前表现优异、营收和利润高于中位数的创业者，在使用 AI 后表现得更好。 那么，这在今天会有何不同？我认为，即使我们现在将 GPT-4o 或 Claude 等更先进的模型部署在 WhatsApp 背后，也可能得到相同的结果。为什么会这样？回溯商业史，获得优势的一种方式是善于配置资源。经典案例是资本配置，如沃伦·巴菲特在伯克希尔·哈撒韦公司对资本的配置能力无人能及。其他公司则擅长人才的配置。我认为，我们正处在一个日益重要的时代，那就是**智能分配**（Allocating Intelligence: 指有效分配计算资源、模型能力和人类智慧以达成商业目标的能力）的能力。如果你能比他人更擅长这一点，你就能在市场中获得优势。 你好，我是 Rem Koning，哈佛商学院教授，研究创业精神与人工智能。我热衷于帮助各种规模的企业家，无论是硅谷的巨头还是印度尼西亚的椰子小贩，让他们做得更好。我致力于思考如何帮助企业家将新产品推向市场、提升竞争力并发展壮大他们的企业。而最近，我尤其关注人工智能（AI）如何能够释放出惊人的创业潜力。
Original English I wrote this paper 3 years ago looking at whether chat GPT delivered over WhatsApp could help small business entrepreneurs in Kenya for entrepreneurs who before our study were struggling. They had lower baseline profits and revenues. They saw a 10% decline in their profits and revenues from talking to the AI. Conversely, when we look at the people who were performing really well, the revenues and profits were above the median before our experiment. We find that they actually did better. How it might look different now? I think if we took GPT52 claudis and we put it behind WhatsApp, we get exactly the same results. Why would we get the same results if we got these better models? I think there's a really interesting thing when we look at the history of business. One way that you could get an edge was better at allocating stuff. The classic one is like allocating capital. Warren Buffett is better at allocating tapel and Birkshshire Haway than anyone else. Other companies are really good at allocating talent. I think we're in a world where increasingly what matters is your ability to allocate intelligence. If you can work that out better than other people, I think you've got an edge in the market. Hi, my name is Rem Coning. I'm a professor at Harvard Business School and I study entrepreneurship and AI. And I just like helping entrepreneurs do better. It could be a mogul in Silicon Valley. It could be someone selling coconuts in Indonesia. How do we help entrepreneurs bring new products to markets, compete better, grow their firms? And most recently, thinking about the role of AI as something that's just going to unlock a crazy amount of entrepreneurial potential.
### AI原生企业：重塑商业模式与全球创业版图 那么，人工智能（AI）是如何改变我们构建企业的方式的呢？我们经常听到“AI原生企业”（AI-native firms）的说法，但关于谁能构建它们、如何规模化以及它如何改变战略，仍然存在许多疑问。为此，我们启动了“AI创始人冲刺”（AI founder sprint）项目，这是一项源自欧洲工商管理学院（INSEAD）的倡议。我们吸引了来自全球的 500 多名创业者参与，其中四分之一来自非洲，四分之一来自亚洲，四分之一来自美洲，以及四分之一来自欧洲，他们都围绕 AI 进行创新。 项目显示，你看到了人们正在为 AI 构建各种应用，比如在非洲开发辅助母婴健康的 AI 工具，在印度创立新的教育科技（EdTech）初创公司，以及在肯尼亚涌现出的各类创新项目。欧洲和美国也诞生了许多优秀的初创企业。我们跟踪了这些创始人如何使用 AI。初步结果表明，当你教导创始人成为“AI原生”时，让他们深入思考如何应用生成式 AI——不仅仅是 ChatGPT 或 Claude 这样的聊天机器人，还包括 Vibe 编码工具、多模态工具以及当下令人兴奋的各类智能体（agents）——并将其融入他们的企业以推动发展，这会帮助全球各地的创业者表现得更好。例如，在尼日利亚创业的企业家，每周能完成更多工作，平均效率提升约 20%，他们获得客户、成功发布产品以及增加收入的可能性都更高。 更有趣的是，尽管这些创始人取得了更大的成功、增长更快，但他们却表示希望融资的需求减少了。他们的融资需求下降了约 250,000 美元。这表明，他们正在重新构想企业的工作流程，并构建定制化的 AI 解决方案。比如，如果企业面临获客瓶颈，他们可以构建新的 AI 系统来自动制定营销策略和上市计划，甚至执行这些计划。这正是“AI创始人冲刺”项目中最令人兴奋的部分之一：看到人们如何根据自己企业的瓶颈（无论是营销还是产品开发），利用现成的工具，并在此基础上构建自己的智能体（agents）。这意味着，原本需要依靠人力（headcount）才能实现规模化的业务，现在可以通过按需计算（on-demand compute）来实现，这彻底改变了企业的经济模型和发展可能性，尤其为硅谷以外的创始人带来了新的机遇。一个很好的例子是 **Gamma**，这家公司展示了AI如何改变商业模式。
Original English how AI is changing the way we build firms, right? We hear a lot about AI native firms and I think there are big questions about who can build them, uh how we scale them, how it changes strategy. So the AI founder sprint is an initiative that came out of INSEAD. We got over 500 entrepreneurs from all over the world. A quarter from Africa, a quarter from Asia, a quarter from the Americas, and a quarter from Europe. All building around AI. You're seeing people building stuff for AI and maternal health in Africa. You're seeing people build new edte startups in uh India. You're seeing stuff come out of Kenya. You're seeing awesome startups uh coming out of Europe, in the United States. And really what we did was we tracked how all these founders were using AI. I'll give you guys a little bit of a preview of the results which is that when you teach founders to be AI native when you tell them to really think about where they can apply generative AI not just chat GPT and like claude but the vibe coding tools multimodal tools all the agents that people are really excited about now when you tell them to really think about where to use that in their firm to move it forward it helps entrepreneurs everywhere do better so if you're building in Nigeria you are able to get more done every week about 20% more you're more likely to get customers, you're more likely to launch a product, you're more likely to have more revenue. And what's really crazy is even though you're more successful, right, you're growing faster. What we see is that these same founders say that they want to raise less capital. So, their demand for raising funds drops by $250,000. We're seeing folks really reimagine the workflows in their firm and build custom AI solutions. So, maybe the bottleneck for you is getting customers. How can you build new AI systems that automatically build out a marketing strategy and a go to market plan and not just build the plan but then execute it? And so that was one of the things that was most exciting I think from the sprint was seeing how people were taking whatever the bottleneck for their firm was it could be marketing maybe it was product development and they were using the off-the-shelf tools but then building basically their own agents if you will right so instead of headcount suddenly we're scaling just with on demand compute and that completely changes the economics of a business and what's possible particularly for founders outside of Silicon Valley I think a great example of it gamma
### AI原生产品：价值创造与规模化革命 我认为，在思考 AI 原生（AI-native）时，首先要理解价值来源于两个方面：一是**流程优化**，二是**产品创新**。流程优化是指你在日常工作中利用 AI，比如用于编写代码或处理客户支持工单，从而提高工作效率或质量。这正是我们目前所熟知的，利用 AI 使工作更快、更好。这确实能为 **Gamma** 这样的公司带来优势。但真正关键在于，AI 原生企业的核心在于将 AI **嵌入产品本身**（embedding AI into their product），使得 AI 能够直接与客户互动并完成工作。 这意味着，你需要设法将**人类（human）**从核心的运营循环中移除。我热爱人类，我们都很棒，沟通交流也很有意义，但我必须指出，人类的**扩展性（scale）**并不好。如果 Gamma 在生成式 AI 出现之前就想做现在的业务，他们可能需要雇佣数以万计甚至数十万的平面设计师，这会让公司的经济模型彻底崩溃。但通过将 AI 嵌入产品，Gamma 实现了以计算能力（compute）而非人力（headcount）为基础的规模化。因此，如果你想成为一家 AI 原生企业，就必须找到那些能够创造 AI 与用户、或其他 AI 或网站环节自主协作的闭环（loops）的场景，而无需你的团队介入。这才是构建 AI 原生组织的根本。
Original English I think the first thing that's really important when you're thinking about AI native is to understand that the value comes from two places. One is the process. You use AI in your coding. You use AI to do customer support tickets. And I think that's what we're all really familiar with, which is that we're sort of using AI to make our work go faster or make our work better. And that gives Gamma an edge. It helps. But really the key to Gamma is the way they've embedded AI into their product. And I think the key for AI native is that you're not just using it to do the work. you're embedding it in the product so that the AI can directly do the work with the customer. You want to take you as the human out of the loop. I love humans. We're amazing. It's great being on this uh call. It's great talking to people. I love sharing stories. But the problem with humans is we don't scale particularly well. And so if you were trying to make Gamma pregenerative AI, they'd probably have to employ, I don't know, tens of thousands, hundreds of thousands of graphic designers. The economics of the company would collapse. but instead by putting AI into the product, what Gamma is able to do is scale with compute rather than headcount. And I think it's a really exciting thing that if you're trying to be an AI native founder, that's what you need to find. Where are those places you can create loops where the AI is working with a user or another AI or something on the website where your team doesn't even need to be involved? That's the key to building AI native organizations.
### 智能分配的深度解析：AI与人类智慧的协同优势 我们再次深入探讨**智能分配**。回溯商业史，企业获得优势的方式之一是更善于**配置**各类资源。例如，在资本配置方面，沃伦·巴菲特通过其在伯克希尔·哈撒韦的卓越能力，精准地将资金投入到最具潜力的领域，带来了惊人的回报。而在人才配置方面，像麦肯锡这样的顶尖咨询公司，则精于识别和选拔人才，并将其匹配给最合适的客户项目。 如今，我们正进入一个**分配智能**日益重要的时代。这意味着你需要精细地分配不同 AI 模型的工作。例如，**Claude** 应该做什么？**Lovable** 适合哪类任务？**Grock** 或 **Deepseek** 又该承担何种角色？如何将这些不同的智能体融合、编排（orchestrate）并分配到你的产品中，是极其关键的。 但同样重要的是，你必须明确**AI与人类（AI and humans）**之间的任务分配。尽管 AI 在思考速度和效率上可能超越我们，但人类的思维方式依然独特。在制定战略、寻求市场优势时，关键不在于做得“更好”，而在于做得“**不同**”。因此，你需要找到人类智能的独特价值所在，将其与 AI 的能力相结合。将工作分配给那些能够超越模型、提供独特见解，或以模型无法企及的方式解决问题的人类员工。然后，你需要找到方法将这两者巧妙地融合在一起。
Original English Allocating intelligence. I think there's a really interesting thing when we look at the history of business. one way that you could get an edge was better at allocating stuff. The classic one is like allocating capital. So you have uh Warren Buffett is better at allocating capital and Birkshshire Hathaway than anyone else. He knew where to put his money and that made amazing amazing returns. Other companies are really good at allocating talent. So if you look at a company like McKenzie, big consulting company, they're really good at working out who should become partners, who should they hire at the base of the pyramid, matching that talent with the right clients, what they're really good at. I think we're in a world where increasingly what matters is your ability to allocate intelligence. And what that means is you need to allocate what is done by different models. What are you going to have Claude do? What are you going to have Lovable do? What are you going to have Grock or Deepseek do? Right? Thinking about how you blend, how you orchestrate, how you allocate your product to these different sorts of intelligences is just incredibly important. But I think this is the the key, which is that you also need to work out how to allocate what's being done by the AI and what's being done by humans because at the end of the day, we still have some edge over some of these models. And even if they're better or faster at thinking, often we think differently. And when you're thinking about strategy and you're thinking about how to gain an edge in the market, it's not about necessarily doing something better. It's about doing something different. Doing something in a way nobody else can. And so if you can work out how you bring your human intelligence and you allocate jobs in the company to humans and the places where they can add value over and above the models or do things differently than the models can.
### AI作为赋能者与放大器：机遇与陷阱并存 AI究竟是“**平均器**”（equalizer）还是“**放大器**”（amplifier）？教授的回答是：可能两者兼而有之。如今，我们都能使用 **Lovable** 编写代码，用 **Gamma** 制作精美的演示文稿，还能让 **ChatGPT** 和 **Claude** 充当我们的私人编辑，消除错别字。从这个角度看，AI 确实是一个惊人的“平均器”，它极大地提升了每个人的能力上限，让我们可以做更多事情。 然而，问题在于，这更多地体现在我们已有的工作上。当我们谈论构建一个全新的产品或企业时，AI 带来的回报将主要流向那些**有能力**利用 AI 的人。这些人通常是那些已经具备相当优势、拥有判断力、可能已有创业经验或更强技术背景的人。他们更有可能构想出巨大的成功，获得丰厚的回报，他们的判断力和能动性（agency）将被 AI 极大地放大。AI 并非简单地为所有人提供同等机会，而是可能加剧现有优势。
Original English is AI an equalizer an amplifier I'm going to say the standard professor answer which is it depends or maybe it's both um but let's get a little bit deeper we all now can code with lovable We can all build amazing decks with Gamma. All of us can use chat GPT and claude to get rid of typos and have a copy editor in our pocket. Wow, it is an equalizer, right? It is amazing. It moves everybody up, right? We can all do so much more. But here's the problem, right? I think that's for the existing work that we do. Whether you're applying AI over an existing task that you have or you're building a new sort of business, when you're building a new sort of business, a new sort of product, when you're thinking about how AI is going to change your firm, be it a small business or a tech startup, the returns to thinking about how AI can do this are going to be greatest for those who have the ability to do that. And those are going to be people who are already pretty good. Those are going to be people who've developed the judgment. Maybe they started a company before. They probably have a stronger technical background. Those are the people going to be able to imagine the really big wins and get those huge returns. They're going to see their judgment, their agency amplified.
### 从聊天机器人到智能体：AI赋能新兴市场与知识工作 研究曾发现，对于那些在 AI 出现前就已挣扎的肯尼亚创业者，使用 ChatGPT 后营收反而下降了 10%。根本原因在于，他们虽然与 AI 互动，但缺乏辨别 AI 建议好坏的**判断力（judgment）**。相反，表现优异的创业者则能更好地采纳 AI 的有效建议，并将其转化为实际行动。这提示我们，AI 并非万能药，缺乏“**挣得的洞察（earned insight）**”可能导致反效果。 教授认为，现在我们犯的一个大错误是，仍然被困在“聊天机器人”（chatbot）的思维模式中。ChatGPT 的出现改变了我们对 AI 的认知，但也让我们误以为万物皆可通过聊天机器人解决。事实上，我们**不需要更多的聊天机器人**。更有效的方式是提供“**智能体AI**”（agentic AI）。想象一下，你能否为创业者提供构建更优网站或启动更强营销活动的工具？能否为他们构建虚拟员工，承担日常业务运作，放大他们的强项，让他们在有限的时间内完成更多工作？这正是我们当前正在探索的方向——AI 从被动对话者转变为能主动采取行动的助手。 这对于全球初创企业来说是一大机遇。你无法在基础模型上与 OpenAI 或 Anthropic 竞争，但你拥有对特定工作流程、区域性情况或世界不同地区运作方式的**情境化知识（contextual knowledge）**。将这种宝贵的背景知识注入 AI，将能产生惊人的效果。这正是当前“云代码”（cloud code）中“技能”（skills）概念的体现：将特定任务的执行方式和上下文信息打包，让 AI 能更高效地完成任务。
Original English I wrote this paper 3 years ago whether chat GPT delivered over WhatsApp could help small business entrepreneurs in Kenya improve their business's performance. And what we found was really surprising, which was that four entrepreneurs who before our study were struggling, they had lower baseline profits and revenues, they saw a 10% decline in their profits and revenues from talking to the AI. It would have been better had we never given them the AI. And what we find is that the reason for them is that they ask the AI a lot of questions. They use it, they interact with it, but they get a lot of advice from the AI and they don't know how to pick the good advice from the bad advice. They don't have the judgment to separate what's good from bad, which might explain why they were low performing in the first place. Conversely, when we look at the people who are performing really well, their uh revenues and profits were above the median before our experiment, the better entrepreneurs, we find that they actually did better. And when we look at the chat logs, the reason we see is that they're asking kind of the same sorts of questions as the low performers, but they're then following up and following the advice that isn't the bad advice, it's the good advice. Unless you've developed the judgment, the mental models to actually know where to apply it, it can lead you down a road of slop. And that slop can actually lead you to make less money. And it's really interesting as we've been working on this paper, it's now almost 3 years old, how it might look different now than back then. I think if we took Claude Opus and we put it behind WhatsApp, we get exactly the same results. The issue is you still have the same problem, which is that the entrepreneur asks a question and the AI gives them four or five plausible things to do and you need to know which one's actually right for you. So that's the first thing. The second thing is I think if we were doing the study now, we probably wouldn't do it through a chatbot. I think a big mistake you're seeing entrepreneurs make is that we're still stuck in the chatbot world. Chad GPT launched, it was huge. It changed the way we thought about artificial intelligence, about how we use our computers. And it changed us and sort of locked us into this idea that there was a a chatbot for everything. We'll have a chatbot for Shopify. We'll have a chatbot for these entrepreneurs. Harvard Business School has internal chat bots. We'll just build chat bots everywhere. And it turns out we don't need more chatbots. So imagine you go to a chatbot and it says, "Oh, you need to update your website so you can get more sales." If I'm a Kenyan entrepreneur and I don't know how to code, how am I updating my website? So I think if I was doing this study today, the thing that would be really exciting is giving more agentic AI, right, to the entrepreneurs today. Could you give them the tools to build better websites or launch better marketing campaigns? A lot of these entrepreneurs, they're struggling. They don't have extra money. They don't have extra staff. They don't have extra people to do something for them. Could you build them virtual employees that help them run their business and expand the things that they're good at and get more done during the day? these are people who are working hard and often the constraint is just time. They don't have the time to do it. Could you use AI to help them do more in the time that they have? Um, so that's something we're exploring in some current work right now, which I'm really excited about. And I think the more we get in the mindset of it's not going to be we're going to having these conversations with the AI, but that we're going to be telling them and they're going to go take actions in the world on behalf of us, I think that's really exciting. It's going to open up a lot more modalities in terms of interfaces, right? and maybe that we're talking to them like how we're having a conversation right now. I think it's also a key unlock for startups all around the world. You're never going to build AI systems better than open AI or anthropics. Sorry, they're at the frontier. They've got billions of dollars in funding, but what you do have is better contextual knowledge of a workflow, of a situation, of how things work in different parts of the world. And I think that can give you a real edge because if you can get that right context into the AI, oh my god, what it can do is absolutely amazing. And I think you're seeing no better example of this right now than with skills in cloud code, right? So people are making these skills. What are these skills? They're kind of context. They're like little snippets of how to do a particular task where you've told the AI how to do it. You've given it the context for how to think about it. And then we can share these skills with everybody else in the world. And it's just wild to see how effective this is at making the models better.
### AI驱动经济转型：新兴市场的跨越式发展与软件经济的未来 AI 的发展预示着更多人将成为创业者，在公司内部也可能出现更多“创业者精神”。这得益于 AI 降低了从无到有创造事物的门槛。展望未来，世界将更加**创业化**。 特别是在**发展中市场（developing markets）**，AI 提供了**跨越式发展（leapfrog）**的机遇，正如他们在金融科技（Fintech）领域所做的那样。以印度的 UPI 支付系统为例，其先进程度远超美国，极大地解锁了数十亿美元的价值，催生了大量初创企业。AI 在知识工作领域同样具有巨大潜力，但前提是 AI 系统必须拥有来自新兴市场的**情境化知识**。例如，AI 是否足够了解肯尼亚的企业家来提供指导？AI 在教育等关键知识领域的影响将是深远的。 随着**推理成本（inference cost）**的急剧下降，AI 模型的使用将变得几乎免费，这将为那些资金有限的市场打开巨大的机遇。从商业角度看，AI 可以提供企业无法负担的人力支持，弥补专家（如营销专家）的缺乏。设想一下，如果身处内罗毕的创业者能够雇佣一位和纽约或硅谷顶级营销专家一样出色的**虚拟代理（virtual agent）**，其价值不可估量。AI 可以帮助他们开展市场营销、拓展出口业务、提升劳动力技能。 更宏观地看，AI 将深刻改变整个经济。它将带来软件应用的大规模涌现，解决我们过去从未想象过的问题。这些问题可能深奥如 **AlphaFold** 在生命科学领域的突破，也可能 mundane 如为泰国餐馆老板开发一款 CRM 系统。AI 驱动的软件拥有低边际成本和高可扩展性，能够将知识转化为惠及亿万人的力量。这将使世界经济更趋向于**软件经济（software economy）**。 然而，这其中也潜藏着挑战。过去几十年，软件发展在一定程度上导致了财富的**集中（concentration）**，形成了少数巨头企业控制数字生态系统。AI 是否会重蹈覆辙，加剧财富分配不均？值得庆幸的是，AI 驱动的**小型、自举（bootstrapping）的 SaaS 应用**模式，可能有助于**更广泛地 spread prosperity**。但我们仍需警惕 AI 经济对财富集中的潜在影响。
Original English I think more people will be entrepreneurs, right? more people are going to go out and build their own businesses and I think more people in companies are going to behave like entrepreneurs partly because we can all build now right and I think that's one of the things that's always attracted me to entrepreneurs whether you're building a car wash or building a a software company it's like you have to create something from nothing I think increasingly all of us are going to do it so the world is going to look a lot more entrepreneurial and then you look at developing markets and I think there's a opportunity for them to leapfrog like they've done with fintech right if you go to India India's payment infrastructure uh this thing called the UPI is light years ahead of the United States. You can pay with everything on your phone. It's unlocked billions of dollars in value, maybe even more. Um there's been a huge number of startups around it. They have these amazing financial g gateways. I think there's an opportunity with AI around this and sort of knowledge work around the world. But I think to do that, we need to make sure the AI systems have context from emerging markets. Do they know enough about the Kenyan entrepreneurs to guide them? I think thinking about what are the big knowledge problems in these places, particularly education, I think AI could have a profound effect there. I'm really excited to see more of the application layer come to emerging markets and I think there's reason to be optimistic, right? We've all familiar with the inference cost curves, right? The cost of, you know, calling a GPT4 quality model just keeps going down exponentially, you know, to a year from now, two years from now, it's going to be basically free. when things become basically free, I think it opens up a lot of opportunity to work in markets where people just have less money to spend. I think to get like really concrete on the business side, I think it can give people labor that they couldn't hire otherwise. There's often a lack of experts. Like you just can't find someone to help you with marketing. They're not there. They didn't get that college training. If I can hire a virtual agent who's as good as a marketer in New York or Silicon Valley and I'm in Nairobi, holy moly, is that amazing. I think that'd be a really concrete one. help them do their marketing, help them export more, help them skill their workforce potentially. When we sort of step back, then there's even a bigger question. How is AI going to transform the economy? And the flip side of more entrepreneurship is that I think we're going to see a proliferation of software tackling problems that we never even imagined it could tackle. Some of these are going to be really deep, like AlphaFold. That's going to be awesome. Some of these though are going to be much more mundane. So, I'm in Thailand and there isn't a CRM for my restaurant business. Well, now somebody's going to take Lovable or Codeex or whatever it is and they're going to make the world's best CRM for Thai restaurant owners and that's going to solve this person's problem and help that business grow. And I think there's probably millions if not billions of other problems that software could start to solve. And I'm really excited about that because software is awesome. Marginal costs are low. It's incredibly scalable. It can turn knowledge into something that can help millions if not billions of people. And so I think what we're going to see is a world that looks more like the software economy. I think there's a downside to that. If we look at the past 20, 30, 40 years, software has also led to potentially a concentration in wealth. We have these sort of, you know, mega billionaires at the top who have these marketplace platform and network effects businesses like Facebook that really h have controlled a lot of the digital ecosystem. And so I think there's a big question on the policy side of how we prevent that from happening again. I think one thing that's really exciting about vibe coding and how it might change the economy is that I don't think we're in a world of network effects. I think we're in a world of small like kind of SAS applications. I think we're in a world of uh bootstrapping businesses that don't need that VC investment. And so I am a little bit hopeful that this might be something that's going to spread prosperity more broadly. Though I do think we need to think about uh what it might do to the concentration of wealth if more of the world starts looking like software and tech.
### AI创业的误区与“挣得的洞察”：小步快跑，以终为始 最危险的假设之一是：仅仅因为用 AI 构建了产品，人们就会想要它。这与传统的软件开发一样，**构建 AI 产品不代表它就一定有市场需求**。很多人陷入了一个误区：AI 工具本身非常有趣，容易让人沉迷其中，不断添加功能，最终构建出“过度工程化”（overengineered）却无人问津的产品。 因此，经典的软件开发原则依然适用：**将其推向用户，让他们试用，了解他们的真实需求**。AI 的应用也应遵循“**少即是多**”的原则。以 Gamma 为例，其 AI 核心最初仅是将用户的几句话转化为演示文稿，其余部分则为传统软件。就是这个小小的 AI 突破，成就了一个伟大的商业。寻找那个能通过引入少量 AI（just a drop or two）就能极大地改变用户工作流程、解决他们长期难题的点，才是关键。 比仅仅依赖 AI 工具更重要的是，**创业者必须拥有“挣得的洞察（earned insight）”**，具备**判断力（judgment）**和**品味（taste）**，清楚 AI 工具的适用场景。与其纠结如何获取最新的基础模型，不如思考如何将现有模型用好。因为对于大多数应用场景而言，上一代模型也足够强大。**核心竞争力在于能否理解在哪里以及如何应用 AI**，这是一种与单纯“AI 工程”不同的技能。
Original English The most dangerous assumption they make is that by building with AI, they have made something people want. And that is just not true. Just like traditional software, you can build with AI and nobody can want it. I'm seeing this basically loop where people get stuck cuz the AI tools are so fun. So you're like, Claude, make me something. Codeex, make me something. And then you're like, let's add this other feature. And then like, let's do another one. And a month goes by and you've built the most beautiful piece of software. It's crazy overengineered. And then you launch and nobody wants it. And so I think some of the traditional stuff around building software still holds. Get it in front of users, have users play with it, see what they want. I think a correlary of that is a little bit of AI goes a long way. Going back to the example of Gamma, really the core of their AI at the beginning was let's just have people write a couple of sentences and then we'll generate a deck. Everything else was traditional software. That one unlock, oh my gosh, led to an amazing business. Can you find that one place in someone's workflow where you can apply a little bit of AI, just a drop or two, and that unlocks how they can do something, changes a problem that was really hard for them, that's what you really want to find. So, I think finding the smallest point where you can use these AI systems is really valuable. Somehow, it's going to give them an insight without them having the earned insight, right? I think it's more important than ever for founders to have a real earned insight, to have the judgment, to have the taste, to know how and where to apply these tools. I think that's the thing you need to spend your time on, not thinking about how do you get access to the next foundation model, whatever it is. Use last generation's model. It's probably fine for the purposes of what you're building. What matters is can you figure out where and how to apply it? And that's a different skill set than necessarily just cranking through the AI engineering.
### AI驱动的学习变革与控制权的颠覆 在学习方式上，AI 带来了深刻的变革。那些能够意识到并解锁**持续学习（learning way more）**能力的人，将获得加速发展的优势。**Drew Bent**，Anthropic 的教育负责人，指出我们常常给 AI 工具设定过于简单的问题，而错失了用它们解决更复杂挑战的机会。我们未能充分提升我们使用 AI 工具的能力。 随着 AI 工具的日益智能化，它们将可能在某些情况下实现**控制权的逆转（inversion of control）**。届时，AI 模型将承担部分高层次的战略思考，并将需要人类**品味（taste）**和**能动性（agency）**的任务委派给人类。这预示着未来人机协作将进入一个新阶段，AI 不仅是助手，更可能成为战略规划的共同制定者。
Original English >> There are two things that are changing. how we're learning is changing and those who are really going to accelerate are going to be those who just realize you can be learning way more than you were before and you just need to unlock that. My name is Drew Bent. I lead education at Anthropic. One of the things that I think holds us all back is we give AI tools pretty simple problems when we could be giving them much more complex problems. We are not elevating our ambition with what we can do with these AI tools. As the AI tools get smarter, where could I push you further? Where could you have pushed me further? And eventually, if we're looking ahead, there will be in some cases, I think, this inversion of control where actually the AI model is doing some of the highest level strategic thinking and then delegating to you, the human for the areas that require human taste, human agency.