演讲者首先分享了一个个人轶事，关于她在夏天休息期间没有关注人工智能新闻。回来后，她注意到虽然新的AI模型和技术已经涌现，但构建优秀产品和发展业务的基本挑战依然存在。这让她开始质疑对AI工具的强烈关注：我们是真的在创造更好的产品、更有效地赋能团队，还是仅仅对工具本身越来越着迷？ 她批评了行业长期以来对“软件工厂”的痴迷——即便是AI出现之前，也痴迷于扩大和优化组织产出。这种方法通常涉及招聘更多专业角色、创建更多流程，并依赖实验来指导决策，所有这些都旨在增加产出。然而，演讲者认为“更多产出并不意味着更好”；真正的目标应该是为客户构建更好的体验。她警告说，组织不应成为仅仅外包思维、只关注效率的“软件工厂”，提醒听众“产出不是产品”。客户购买的不是代码行或实验；他们购买的是他们想要和需要的东西。 演讲者提出，制作产品会产生两个关键结果：产品本身以及从其创作中获得的学习。从历史上看，构建产品的过程本身就产生了学习——在设计选择中挣扎、提出问题并与客户互动。她指出，当前的危险是过度自动化，尤其是借助AI，可能会切断这种与学习的直接联系。伟大的公司之所以能持续打造出优秀产品，是因为它们培养了强大的团队文化和背景，这种文化和背景建立在对自身领域、客户、技术的理解，以及拥有卓越的判断力和品味之上。这创造了一种“复利式学习”效应。 虽然AI可以自动化那些学习价值不高的重复性任务，例如Linear利用AI调查和修复bug，但演讲者强调，节省下来的时间应该被重新投入。团队不应该外包一切，而应该投入更多时间去理解客户问题、探索新想法并磨练他们的判断力。产品组织的目标是保持一个持续的学习循环，将客户背景和洞察带给整个团队。当深思熟虑地使用时，AI可以成为*构建*这种背景的强大工具，而不是取代它。例如，Linear聚合了来自不同来源的客户反馈，让演讲者能够收到关于特定客户洞察（如AI工作流程）的个性化“每日简报”。 为了培养集体判断力和品味，Linear采用了两项关键实践：“质量星期三”（Quality Wednesday）和“功能烘焙”（Feature Roasts）。“质量星期三”鼓励每位团队成员每周花时间识别并修复产品中的一个小质量缺陷，从而培养他们对细节的洞察力。这些发现会被分享，让每个人都能学习。“功能烘焙”是可选会议，新功能开发团队会邀请其他人对他们的工作进行批评，提供原始、未经筛选的反馈。这个过程帮助开发团队更真实地了解用户感知，并培训每个人产品质量和用户体验的含义。关键在于这些活动是*一起*进行的，促进了讨论和共享学习。 演讲者总结道，AI的效率不应该仅仅带来更多的执行。相反，它提供了一个机会，将时间重新导向更深入的客户理解、探索、批判和反思。未来，产品组织可能会从繁重的执行工作转向更抽象的活动，专注于团队改进和组织学习。核心思想是，公司要对其工作形成一种“复利式理解”。这需要一个共享的背景空间——关于客户、产品思维和正在进行的工作——供人类和AI代理访问。 最终，演讲者认为“背景即产品”。虽然工具将持续改进，自动化也将不断扩展，但重点应该转移到自动化“已知事物”上，同时投入更多时间构建集体理解。她提出，直觉并非魔法，而是通过与客户和产品工作的互动而获得的训练有素的学习。人们拥有的背景越好，理解越深入，他们创造的产品就越优秀。

The speaker began by sharing a personal anecdote about taking a summer break from following AI news. Upon returning, they observed that while new AI models and techniques had emerged, the fundamental challenge of building great products and growing a business remained. This led them to question the intense focus on AI tools: are we truly creating better products and empowering our teams more effectively, or are we simply becoming more obsessed with the tools themselves? They critiqued the industry's long-standing obsession with "software factories" – scaling and optimizing organizational output, even before the advent of AI. This approach often involves hiring more specialized roles, creating more processes, and relying on experiments to guide decision-making, all aimed at increasing output. However, the speaker argued that "more output isn't better"; the true goal should be to build better experiences for customers. They warned against organizations becoming mere "software factories" that outsource thinking and focus solely on efficiency, reminding listeners that "the output is not the product." Customers don't buy lines of code or experiments; they buy something they want and need. The speaker posits that making products yields two key outcomes: the product itself and the learning derived from its creation. Historically, the process of building products inherently generated learning—struggling with design choices, asking questions, and engaging with customers. The current danger, they suggest, is that over-automation, particularly with AI, risks severing this direct connection to learning. Great companies consistently build great products because they cultivate a strong team culture and context built around understanding their space, customers, technology, and possessing strong judgment and taste. This creates a "compounding learning" effect. While AI can automate repeatable tasks that offer little learning, like Linear's use of AI to investigate and fix bugs, the speaker emphasized that the time saved should be reinvested. Instead of outsourcing everything, teams should dedicate more time to understanding customer problems, exploring new ideas, and refining their judgment. The goal for product organizations is to maintain a continuous learning loop, bringing customer context and insights to the entire team. AI, when used thoughtfully, can be a powerful tool for *building* this context, not replacing it. Linear, for example, aggregates customer feedback from various sources, allowing the speaker to receive personalized "daily briefings" on specific customer insights, such as AI workflows. To foster collective judgment and taste, Linear employs two key practices: "Quality Wednesday" and "Feature Roasts." Quality Wednesday encourages every team member to spend time each week identifying and fixing a small quality defect in the product, training their eye for detail. These findings are shared, allowing everyone to learn. Feature Roasts are optional meetings where teams building new features invite others to critique their work, providing raw, unfiltered feedback. This process helps the development team gain a more realistic understanding of user perception and trains everyone on what quality and user experience mean for the product. The key is that these activities are done *together*, fostering discussion and shared learning. The speaker concluded by arguing that AI's efficiency shouldn't just lead to more execution. Instead, it offers an opportunity to redirect time towards deeper customer understanding, exploration, critique, and reflection. In the future, product organizations might shift away from execution-heavy work towards more abstract activities focused on team improvement and organizational learning. The core idea is that companies develop a "compounding understanding" of their work. This requires a shared space for context—about customers, product thinking, and ongoing work—accessible to both people and AI agents. Ultimately, the speaker believes that "context becomes the product." While tools will continue to improve and automation will expand, the focus should shift to automating "known things" while investing more time in building collective understanding. Intuition, they posit, is not magic but trained learning derived from engagement with customers and product work. The better the context people have, and the better their understanding, the better the products they will create.