Anthropic 研究实验室产品主管分享公司早期挣扎、Opus 3 与 Claude Code 的突破，以及 AI 时代产品管理角色的演变。
Anthropic's Head of Product for Research and Labs shares the story of the company's early struggles, the breakthrough with Opus 3 and Claude Code, and the changing role of product management in the age of AI.
要点 · TL;DR
Claude 的编码能力，尤其是 Claude Code 和 Opus 45，推动了其采用，并使 Anthropic 成为领导者。 Claude's coding capabilities, especially Claude Code and Opus 45, drove its adoption and positioned Anthropic as a leader.
Anthropic 的自力更生文化和小团队实验培养了快速创新和适应性。 Anthropic's bootstrapped culture and small-team experimentation foster rapid innovation and adaptability.
评估取代了传统的产品需求文档，成为核心产品管理工具，实现数据驱动决策。 Evals replace traditional PRDs as core product management tools, enabling data-driven decisions.
核心观点 · Key points
编码成为关键差异化因素；Claude Code 和 Opus 45 推动了采用率。 Coding became a key differentiator; Claude Code and Opus 45 drove adoption.
自举式文化和小团队实验有助于创新。 Bootstrapped culture and small team experimentation foster innovation.
评估取代了传统产品需求文档，成为核心产品管理工具。 Evals replace traditional PRDs as core product management tools.
模型的指数级增长要求适应性和第一性原理思考。 Exponential model growth demands adaptability and first-principles thinking.
用户反馈需转化为可执行的具体细节，供研究人员使用。 User feedback must be translated into actionable specifics for researchers.
反共识 · Contrarian takes
安全和对齐使Claude能够提出异议，从而更有用。 Safety and alignment make Claude more useful by enabling pushback.
AI写作目前生硬，但优先级低于智能体能力。 AI writing is currently stiff but deprioritized over agentic capabilities.
最大化令牌消耗就是活在未来并获取超额收益。 Maximizing token spend is living in the future and capturing alpha.
产品经理必须像写测试驱动开发那样写评估，而不仅仅是文档。 PMs must write evals like test-driven development, not just documents.
管理者必须保持动手并交付产品，才能有效领导。 Managers must stay hands-on and ship products to lead effectively.
本期章节 · Chapters（共 39）
早期与文化Early days and culture
赞助商消息Sponsor message
Opus 3：转折点Opus 3 as an inflection point
Opus 45与Claude Code协同Opus 45 and Claude Code synergy
编码预测与指数曲线Coding predictions and exponential curve
指数内部：适应性与第一原理Inside the Exponential: Adaptability and First Principles
涌现能力与不可预测性Emergent Capabilities and Unpredictability
产品悬置与发现Product Overhang and Discovery
Token消耗与实验Token Spending and Experimentation
实验与实验室团队Experimentation and Labs team
研究团队与研究员工作Research team and researcher work
用户反馈与可操作性User Feedback and Actionability
成功研究员特质What Makes a Successful Researcher
雄心与向前兼容Ambition and Forward Compatibility
前沿模型能力影响Impact of Frontier Model Capabilities
模型访问反馈循环与包容性Model access feedback loop and inclusiveness
赞助广告Sponsor advertisement
产品角色变化与第一原理思维Product role change and first principles thinking
评估驱动产品开发Eval-driven product development
PRD仍相关PRDs still relevant
AI世界产品经理技能Skills for product managers in AI world
动手实验与乐趣Hands-on experimentation and joy
深入与优先级Go deep and prioritize
产品管理中使用ClaudeUsing Claude in product management
用Claude作辅导工具Using Claude as a coaching tool
对AI萎缩与过度依赖的担忧Fear of AI atrophy and over-reliance
反驳与校准的价值The value of pushback and alignment
用思考伙伴保护思维Protecting your thinking with a thinking partner
AI写作与识别AIAI Writing and Knowing It's AI
AI写作不足与投资Why AI Writing Falls Short and Investment
人类大脑仍宝贵之处Where Human Brains Remain Valuable
AI时代养育子女Raising children in the AI era
避免倦怠与建设强大团队Avoiding burnout and building strong teams