Google Search 产品负责人 Robby Stein：AI 时代如何成为优秀的产品经理

简要说明Brief Description

Robby Stein 回顾了自己在 Instagram 和 Google 搜索领域近二十年的产品工作经历，阐释了当今优秀产品经理所应具备的特质。随着 AI 让小型团队能够构建功能更丰富的产品，他认为，产品经理的核心能力在于判断力：理解人们的真实需求，诊断产品为何未能达到预期，并塑造出运行可靠且经过周到思考的体验。他通过 Instagram Stories 和 Reels、Google 的 AI 体验，以及用于评估产品质量的工具等案例，阐述了这一方法。Robby Stein reflects on nearly two decades of product work at Instagram and Google Search to explain what makes a strong product manager today. As AI makes it possible for small teams to build far more, he argues that a PM’s central skill is judgment: understanding people’s real needs, diagnosing why a product falls short, and shaping an experience that works reliably and feels thoughtful. He illustrates this approach with examples from Instagram Stories and Reels, Google’s AI experiences, and tools for evaluating product quality.

目录Table of Contents

• 产品经理的核心能力是判断力
  The product manager’s craft is judgment
• 了解人们以及他们雇用产品来完成的工作
  Understand people and the jobs they are hiring products to do
• 找出根本原因，然后不断迭代，逐步实现产品契合度
  Find root causes, then iterate toward product fit
• 匠心意味着消除痛点，创造愉悦。
  Craft means removing pain and creating delight
• AI 时代产品经理的实用行动指南
  A practical playbook for product managers in the AI era

产品经理的核心能力是判断力The product manager’s craft is judgment

在构建面向消费者的产品近 20 年后，Robby Stein 见证过被数十亿人使用的产品，也见证过无人问津的产品。在这些成功与失败中，他注意到团队打造优秀产品的方式存在一些可重复的模式。他的经历包括构建 Instagram Stories 及 Reels、信息流排序和直接消息等相关体验，以及致力于探索 AI 时代 Google 搜索应有的形态：让用户可以随心提问，并从整个 Web 中获取有帮助的信息，包括文本、图片、视频和音频。After nearly 20 years building consumer products, Robby Stein has seen both products used by billions of people and products used by nobody. Across the successes and failures, he has noticed repeatable patterns in how teams create great products. His experience includes building Instagram Stories and related experiences such as Reels, feed ranking, and direct messaging, and working on what Google Search should become in the AI era: a way to ask anything and get helpful information from across the web, including text, images, video, and audio.

这场演讲讨论的是产品的技艺：如何打造人们喜爱的产品，以及 AI 如何改变能够从事这项工作的人群。过去需要整个团队耗费数年才能完成的任务，如今越来越可以由一个人借助少量智能体或 AI 助手来尝试完成。这一转变改变了产品经理所能带来的价值。传统上，产品经理负责组织工作、推动进展，并帮助团队完成任务。这些技能依然重要，但当几乎任何东西都可以被构建出来时，更重要的贡献在于决定什么值得构建，以及它是否足够好。The talk is about the craft of product: how to build something people love, and how AI changes who can do that work. Tasks that once took whole teams years can increasingly be attempted by an individual with a handful of agents or AI assistants. That shift changes the value a product manager brings. PMs have traditionally organized work, created momentum, and helped teams get things done. Those skills still matter, but when almost anything can be built, the more important contribution is deciding what is worth building and whether it is good.

因此，这门技艺在于判断力与品味。产品经理可能需要把许多事情都做得相当不错，但应当在决策方面做到卓越。这意味着要研究如何做出好的决策，同时接受许多决策会出错这一事实。Stein 将他的方法概括为三个部分：深入理解人，诊断根本原因，以及为最终体验注入技艺。The craft is therefore judgment and taste. A PM may need to do many things reasonably well, but should become exceptional at making decisions. That means studying how to make good decisions while accepting that many decisions will be wrong. Stein frames his approach as three parts: understanding people deeply, diagnosing root causes, and bringing craft to the finished experience.

了解人们以及他们雇用产品来完成的工作Understand people and the jobs they are hiring products to do

产品探索始于人及其根本需求。克莱顿·克里斯坦森的《与运气竞争》阐述了“待完成的任务”这一理念：人们并不是简单地使用产品，而是“雇用”产品来完成某件事。要发现这项任务，理解这个人的故事会很有帮助，而不只是询问他们想要哪些功能。Product discovery starts with people and their fundamental needs. Clayton Christensen’s Competing Against Luck describes the “jobs to be done” idea: people do not simply use products; they hire them to accomplish something. To discover that job, it helps to understand the person’s story rather than ask only what features they want.

在一次与该框架开发者共同参加的 Instagram 外出活动中，Stein 的团队通过回忆最近一次重大购买来进行练习。他回忆的是购买一张床。采访者不应只抽象地询问哪些功能重要，而应重现当时的情境：这个人在哪里、在做什么、和谁在一起，以及是什么促使他们做出这一决定。At an Instagram offsite with people who developed the framework, Stein’s team practiced by recalling a major recent purchase. His was a bed. Instead of asking only which features mattered in the abstract, the interviewer should reconstruct the moment: where the person was, what they were doing, who was with them, and what led them to make the decision.

斯坦因和他的妻子试睡了几张床。他的妻子让他躺到其中一张上，然后选中了它，因为她需要一张不会因丈夫翻身而把自己吵醒的床。传统的产品路线图可能会侧重于降温、可持续性或价格。这些特性固然重要，但在这个具体故事中，决定性的需求是一张能够隔绝动作影响的床。如果不详细询问购买过程，产品团队可能永远无法发现这一需求。“待完成的工作”是指客户购买某种产品或服务时试图实现的进展。Stein and his wife tried several beds. His wife asked him to flop onto one, then chose it because she needed a bed that would not wake her when he moved. A conventional product roadmap might focus on cooling, sustainability, or price. Those features can matter, but the decisive need in this particular story was a bed that isolated movement. Without asking about the purchase in detail, the product team might never uncover that need. The job to be done is the progress a customer is trying to make when they hire a product or service.

同样的洞察也适用于数字产品。在 Google 早期的 AI 体验中，团队发现人们来这里是为了获取灵感，其中包括寻求购买决策方面的帮助。纯文本聊天机器人并不适合回答“该买什么样的沙发”这类问题：用文字描述沙发很困难。团队很早就投入了多模态理解、检索和知识系统，因此人们可以询问某款沙发，查看视觉结果，通过“祖母绿”或“毛茸茸”等细节来完善需求，并在持续的对话中探索抱枕等相关物品。用户真正需要的是一段获取灵感的体验，而不只是一个文本答案。The same kind of insight applies to digital products. Early in Google’s AI experiences, the team saw that people came for inspiration, including help deciding what to buy. Text-only chatbots were poorly suited to a question such as what kind of couch to get: describing a couch in words is difficult. The team invested early in multimodal understanding, retrieval, and knowledge systems so people could ask about a couch, see visual results, refine the request with details such as “emerald green” or “fuzzy,” and explore related items such as throw pillows through a continuing conversation. The underlying need was an inspirational journey, not simply an answer in text.

另一个例子来自早期对 Google 搜索中 AI Mode 的探索。人们可以询问某个街区或美食，系统也能提供相关信息，但它缺少人们对 Google 所期待的东西：有关真实地点的准确、权威信息。用户希望能在地图上看到地点、评分、营业时间、菜系和价格。将这些细节与对话式体验结合起来，提升了产品的实用性，团队也发现用户对此给出了很高的评价。同样，重要的洞察在于，人们为什么使用这个产品，以及他们期待产品帮助自己完成什么。Another example came from early exploration of AI Mode in Google Search. People could ask about a neighborhood or food, and the system could provide information, but it lacked what people expected from Google: accurate, authoritative information about real places. Users wanted to see locations on a map, ratings, opening hours, cuisine, and price. Bringing those details together with a conversational experience made the product more useful, and the team saw strong user ratings for it. Again, the important insight was why people came to the product and what they expected it to help them do.

AI 还可能让大规模发现这些洞见变得更加容易。Stein 对利用访谈记录来识别待完成的任务感兴趣，并希望构建一个采用相同方法论来访谈人们的代理：询问他们为何使用某款产品，以及他们试图解决的根本问题。技术可以帮助扩展研究，但问题仍需揭示行为背后的人类情境。AI may also make it easier to find these insights at scale. Stein is interested in using interview transcripts to identify jobs to be done and in building an agent that interviews people with the same methodology: asking why they use a product and what underlying problem they are trying to solve. The technology can help extend research, but the questions still need to uncover the human context behind the behavior.

找出根本原因，然后不断迭代，逐步实现产品契合度Find root causes, then iterate toward product fit

即使最初的愿景很明确，产品的第一个版本通常也不太完善。团队通过找出体验中的具体问题、理解这些问题、确定优先级、加以修复，然后重复这一过程来取得进展。每次更新都会让产品更接近团队所追求的契合状态。Stein 将这一循环比作模型训练：每一轮评估和修改都提供了进一步改进的机会。Even with a clear initial vision, the first version of a product is usually not very good. Teams make progress by identifying the specific problems in the experience, understanding them, ranking them, fixing them, and then repeating the process. Each update moves the product closer to the fit the team is seeking. Stein compares the cycle to model training: each round of evaluation and changes provides another opportunity to improve.

Instagram Stories 说明了，询问人们为什么没有做预期的事情，可以揭示真正的障碍。Stories 曾经历过多个内部版本，这些版本被整合进信息流，但效果并不好。团队希望人们分享日常生活中的片段，但许多人没有采用这一体验，也不愿意自在地分享。团队没有假设他们需要更好的创作工具，而是询问人们为什么不发帖，并对答案进行了归类。一个主要主题是受众：人们不希望前任、老师、阿姨或姐妹看到自己发布的滑稽或私密内容。Instagram Stories illustrates how asking why people are not doing the intended thing can reveal the real obstacle. Stories had gone through internal versions that were integrated into the feed and did not work well. The team wanted people to share moments from their day, but many people did not adopt the experience or feel comfortable sharing. Rather than assume they needed better creative tools, the team asked why they were not posting and grouped the answers. A major theme was the audience: people did not want an ex, teacher, aunt, or sister to see a goofy or personal post.

随后，团队对数千人开展了一项大规模调查，以量化这些主题。调查得出的结论是，除非产品解决受众负担的问题，否则增加创意功能也无法解决这一难题。团队花了两年时间探索不同的构想。有些版本将私密分享融入个人主页和动态中，但这些版本令人困惑：人们不明白为什么要把一张未经修饰的个人照片放在精心制作的动态帖子之间。这个概念只有存在于 Stories 中时才奏效。团队砍掉了其他方案，将其重新作为一种 Stories 体验推出，这样一来就合情合理了。The team then ran a large survey with thousands of people to quantify the themes. It concluded that unless the product addressed the audience burden, adding creative features would not solve the problem. The team spent two years exploring different ideas. Some versions mixed private sharing into the profile and feed, but they were confusing: people did not understand why they would place a raw personal photo among polished feed posts. The concept worked only when it lived inside Stories. The team cut the other permutations and relaunched it as a Stories experience, where it made sense.

Reels 提供了另一个例子。在巴西的一次上线过程中，团队最初认为，那些制作搞笑视频和舞蹈视频的人不会希望这些帖子永远留在自己的个人主页上。这次上线举步维艰。实际上，那些投入时间编排舞蹈的人希望视频能够长期保留、触达他们所在的学校，并有可能走红。一些人希望成为娱乐内容创作者，或围绕自己的内容打造业务。团队大幅改变了产品形式，使视频得以长期存在，而不是短暂呈现，从而让产品与创作者的需求保持一致。Reels provides another example. During a launch in Brazil, the team initially assumed people making goofy videos and dances would not want those posts to remain on their profiles forever. The launch struggled. In practice, people who invested time in making a dance wanted the video to live on, reach their school, and potentially go viral. Some hoped to become entertainers or build businesses around their content. The team substantially changed the format so videos were persistent rather than ephemeral, aligning the product with what creators wanted.

AI 可以帮助团队更高效地运行这一反馈闭环。在 Google，选择加入的用户可以分享他们的使用体验。模型可以帮助整理由此产生的定性评论，并识别反复出现的问题。例如，一条关于背包的回复可能提供了正确的尺寸和比较信息，却没有询问将要背包的孩子的情况。背包可能太重，或者尺寸不合适。更具协作性的对话会询问缺失的背景信息，帮助用户做出正确的决定。理解人们并不总是知道自己需要提供哪些信息，是帮助他们取得进展的关键。团队投入了这一方法，并在参与度、使用率、价值和帮助性方面都看到了提升。AI can help teams run this feedback loop more efficiently. At Google, users who have opted in can share feedback about their experience. Models can help organize the resulting qualitative comments and identify recurring issues. For example, a response about backpacks might provide correct dimensions and comparisons but fail to ask about the child who will carry one. The backpack could be too heavy or the wrong size. A more collaborative conversation would ask for the missing context and help the user reach the right decision. Understanding that people do not always know what information they need to provide is central to helping them make progress. The team invested in this approach and saw gains in engagement, usage, value, and helpfulness.

匠心意味着消除痛点，创造愉悦体验Craft means removing pain and creating delight

当产品准备好触达更多客户时，剩下的适配与完善工作就显得至关重要。Stein 将工艺描述为一种感受：制作产品的人是否真正用心于此。两件事会塑造这种印象：产品是否运行良好且不会给用户带来困扰，以及它带给人们怎样的感受。产品除了实现基本功能外，还可以带来愉悦与兴奋。Once a product is ready to reach more customers, the remaining fit and finish matter. Stein describes craft as the feeling that the people who made a product cared about it. Two things shape that impression: whether the product works well without causing user pain, and how it makes people feel. A product can create joy and excitement as well as perform its basic function.

AI 可以帮助团队大规模检查产品质量。Stein 使用 Google 的 Antigravity 系统构建了一个内部工具，该工具利用智能体与产品进行交互：输入一个问题、提出后续问题、记录使用体验、截取屏幕截图，并根据详细的评分标准评估结果。评估可以测试数学内容是否正确呈现、是否出现异常字符，以及当用户询问生物发光等视觉相关内容时，系统是否提供了适当的视觉呈现，而不只是文字。AI can help teams inspect product quality at scale. Stein built an internal tool with Google’s Antigravity system that uses an agent to interact with a product: it enters a question, asks a follow-up, records the experience, captures screenshots, and evaluates the results against a detailed rubric. The evaluation can test whether mathematical content renders correctly, whether strange characters appear, and whether a question about something visual, such as bioluminescence, receives an appropriate visual treatment rather than text alone.

这让寻找最佳响应和发现故障都变得更加容易。智能体可以识别出出现故障或不符合规范的体验，并越来越多地帮助修复这些问题。团队不必仅依赖 QA 团队或个人去尝试产品的每个部分，而是可以使用智能体来扩展重复性评估，并将发现的问题纳入持续改进循环。This makes it easier to find both the best responses and breakages. Agents can identify experiences that are broken or off spec, and increasingly help fix them. Instead of relying only on QA teams or individuals to try every part of a product, teams can use agents to scale repetitive evaluation and feed the findings into a continuous improvement loop.

匠心也体现在那些让体验充满生命力的细节中。Google 为 AI 时代重新设计了搜索框，使其能够随着用户的使用不断拓展，并接受来自文件、视频和照片的上下文信息。这一设计既要让人一眼认出是 Google，又要传达 AI 已经融入其中并正在提供帮助。色彩渐变代表了这种能力。光标会循环呈现 Google 的各种颜色，这一细节已被用户注意到并分享。当用户点击搜索框时，它会通过动效、加载环和触觉反馈做出响应。这些细微的设计选择运用动效、色彩和触感传递出用心，也赋予产品一种人性化的特质。它们很容易被忽略，但共同塑造了体验的个性。Craft also comes through in details that make an experience feel alive. Google redesigned its search box for an AI era in which it can grow with the user and accept context from files, videos, and photos. The design needed to feel recognizably like Google while signaling that AI was present and helping. A color gradient represents that capability. The cursor cycles through Google’s colors, a detail users have noticed and shared. When someone taps the box, it responds with motion, a loading ring, and haptic feedback. These small choices use motion, color, and touch to communicate care and give the product a human quality. They are easy to overlook, but they contribute to the experience’s character.

AI 时代产品经理的实用指南A practical playbook for product managers in the AI era

产品工作的基本原则并没有改变，但 AI 让我们能够以新的方式运用这些原则。首先，深入了解用户，识别他们真正感受到的需求和问题。然后，严谨地判断哪些问题必须解决，才能让产品朝着目标推进。最后，打造一种简单易用、运行可靠，并能让用户在使用时感到愉悦的体验。The fundamentals of product work have not changed, but AI makes it possible to apply them in new ways. First, understand people deeply and identify the needs and problems they actually feel. Then be rigorous about determining which issues must be fixed for the product to work toward its goal. Finally, build an experience that is simple to use, works reliably, and makes people feel good about using it.

这就是为什么产品管理和做出正确决策的能力如今如此重要。构建产品有许多行之有效的方法，每个团队都必须决定哪种方法适合自身的情况。在选择要构建什么时，少花些时间去追逐有趣的功能、新技术或另一种产品分发方式，多花些时间去了解屏幕另一端的人。That is why product management and the ability to make good decisions are so important now. There are many valid ways to build products, and each team has to decide what fits its situation. When choosing what to build, spend less time reaching for an interesting feature, a new technology, or another way to distribute the product. Spend more time understanding the people on the other side of the screen.