Y Combinator CEO on Founder Psychology in the Age of AI

以下是内容的中文翻译： Y Combinator 总裁兼首席执行官盖瑞·谭（Gary Tan）分享了他的个人历程、对硅谷文化的洞察、AI的变革力量以及社区参与的重要性。他的职业生涯始于2003年，即互联网泡沫破裂之后，当时科技工作机会稀缺。尽管渴望在初创公司工作，他却加入了微软，他后来将此决定视为一种遗憾，因为它追逐的是“热门”而非“自己所知”。这次经历，加上他拒绝了Palantir的早期机会，在他心中根植了“别装腔作势”（don't larp）的哲学：要真诚，遵循自己的信念，并相信自己的直接经验，而不是被感知到的趋势或社会地位所左右。他强调，“我生活中所有美好的事物都有些‘小众狂热’的特质”，它们都始于一种与正统观念格格不入的信念。 在保罗·格雷厄姆（Paul Graham）和杰西卡·利文斯顿（Jessica Livingston）的领导下，Y Combinator通过为来自不同背景的聪明人进入科技生态系统提供途径，彻底改变了硅谷，建立了一个真正的社区，让创始人们能够真诚地分享脆弱并相互支持。 谈话随后转向了AI对初创公司和工作性质的深远影响。谭强调，“氛围编程”（vibe coding）和“代理编程”（agentic coding）让个人生产力提高了“400倍”，这要求创始人有更高的抱负。他指出，传统的SaaS模式正在演变，代码不再是珍贵之物；相反，真诚地追求解决看似不重要的问题，往往能产生最重要的结果。谭以他的个人项目，如dairieslist.org和G Brain为例，他在这些项目中“只是随便捣鼓”以学习。他提出了“一个Markdown文件就是一个员工”的概念，这意味着任何定义明确的业务流程都可以通过AI“技能文件”完美自动化，从而使小型团队实现快速增长（例如，2-3人团队和数百个智能代理实现0到1500万美元的年度经常性收入）。这种递归式自我改进适用于业务的各个方面，从工程到销售和客户支持。 谭认为，AI可以通过处理“建设性冲突”和官僚任务，极大地改善组织结构。他引用了Brex首席执行官佩德罗·弗兰切斯基（Pedro Francheschi）的例子，他使用智能代理监控内部沟通，为领导层提供了对组织问题和冲突的“洞察力”。这种范式转变解决了人类只能管理“七加减二”个事项的局限性，使企业能够实现“好1000倍”的产品和服务。他微软时期臭名昭著的“棒球棒”故事——他和一位同事不得不直接面对Windows团队以修复漏洞，因为电子邮件被忽视——说明了受限于人类的官僚主义的低效，而AI现在可以消除这种低效。谭认为初创公司“必须”采纳这种代理组织形式，而大公司则“难以”做到。他设想，未来中层官僚机构将由智能代理管理，将白领员工从“泥潭”中解脱出来，让他们能够从事更具目的性的工作。 展望未来，谭预测“下一代计算机”将是基于语音的，拥有庞大的上下文和记忆能力，扮演一个仁慈的“礼宾员”角色，理解个人的希望和恐惧。他提供了一种关于AI的“乐观（白药丸）”视角：虽然其潜力巨大，但由于人类机构、政府和大型企业固有的迟缓性，AI的社会融合速度将比许多人担心的要慢。这种“迟缓”是积极的，它允许社会有时间适应。 最后，谭深入探讨了地方政治领域，认为那里的问题“受限于人类协调”。他对政治的“乐观（白药丸）”看法是：相信法治和公务员的真诚。他分享了自己参与旧金山政治的经历，其动机源于针对亚裔美国人的犯罪上升以及限制性学校政策（例如，阻止初中代数教学）等问题。这段经历让他看到了地方组织抗衡“意识形态”力量并追究机构责任的力量。他倡导“地方行动”方针，相信解决社区层面的问题最终将带来更好的州和国家治理，从而培育一个科技抱负能够本地化并真正产生影响的社会。

Gary Tan, President and CEO of Y Combinator, shares his journey and insights into Silicon Valley culture, the transformative power of AI, and the importance of community engagement. His career began in 2003, post-dot-com bust, when tech jobs were scarce. Despite wanting to work at startups, he joined Microsoft, a decision he later framed as a regret of chasing "what was hot" rather than "what he knew." This experience, alongside turning down an early opportunity at Palantir, instilled in him the "don't larp" philosophy: be earnest, follow your conviction, and trust your direct experience over perceived trends or societal status. He emphasizes that "everything that's awesome in my life is kind of a cult," starting with a belief that flies in the face of orthodoxy. Y Combinator, under Paul Graham and Jessica Livingston, revolutionized Silicon Valley by providing a pathway for smart individuals from diverse backgrounds to access the tech ecosystem, creating a genuine community where founders can be vulnerable and supportive of each other. The conversation then shifts to the profound impact of AI on startups and the nature of work. Tan highlights "vibe coding" and "agentic coding" as enabling individuals to be "400 times" more productive, demanding higher ambition from founders. He notes that traditional SaaS models are evolving, and code is no longer precious; rather, it's the earnest pursuit of solutions to seemingly unimportant problems that often yields the most significant results. Tan exemplifies this with his personal projects like dairieslist.org and G Brain, where he "just messed around" to learn. He introduces the concept of a "markdown file is an employee," suggesting that any well-defined business process can be perfectly automated by AI "skill files," enabling small teams to achieve rapid growth (e.g., 0 to $15M ARR with 2-3 people and hundreds of agents). This recursive self-improvement applies to all aspects of a business, from engineering to sales and customer support. AI, Tan argues, can dramatically improve organizational structures by handling "productive conflict" and bureaucratic tasks. He cites Brex CEO Pedro Francheschi, who uses agents to monitor internal communications, providing leadership with "clairvoyance" into organizational issues and conflicts. This paradigm shift addresses the human limitation of managing only "seven plus or minus two" items, allowing businesses to achieve "1000 times better" products and services. The infamous "baseball bat" story from his Microsoft days—where he and a colleague had to confront a Windows team directly to fix bugs because emails were ignored—illustrates the inefficiency of human-bound bureaucracy that AI can now eliminate. Tan believes startups *must* adopt this agentic organization, while large corporations *struggle* to. He envisions a future where mid-level bureaucracy is managed by agents, unburdening white-collar workers from "morass" and freeing them for more intentional work. Looking ahead, Tan predicts the "next computer" will be voice-based, with vast context and memory, acting as a benevolent "concierge" that understands individual hopes and fears. He offers a "white pill" perspective on AI: while its potential is immense, its societal integration will be slower than many fear, due to the inherent slowness of human institutions, governments, and large companies. This "slowness" is a positive, allowing society to adapt. Finally, Tan delves into the realm of local politics, viewing problems there as "human coordination bound." His white pill for politics is a belief in the rule of law and the earnestness of public servants. He shares his engagement in San Francisco politics, motivated by issues like rising crime against Asian Americans and restrictive school policies (e.g., preventing middle school algebra). This experience showed him the power of local organization to counter "ideological" forces and hold institutions accountable. He advocates for an "act local" approach, believing that fixing community-level problems will ultimately lead to better state and national governance, fostering a society where technology's ambition can be localized and truly impactful.