OpenAI Codex 产品与工程负责人 Andrew Ambrosino 探讨 AI 如何颠覆产品开发流程，让实现变得廉价，品味成为新瓶颈。
Andrew Ambrosino, product and engineering lead for Codex at OpenAI, discusses how AI is inverting the product development process, making implementation cheap and taste the new bottleneck.
要点 · TL;DR
当实现变得廉价时，策展和品味成为新的瓶颈。 Curation and taste are the new bottlenecks as implementation becomes cheap.
产品团队应从自上而下的规划转向自下而上的探索。 Product teams should shift from top-down planning to bottoms-up exploration.
AI 在设计上表现不佳，因为难以评判品味且缺乏新颖性。 AI struggles with design due to difficulty in grading taste and lack of novelty.
核心观点 · Key points
实现不再昂贵；策展和品味成为新的瓶颈。 Implementation is no longer expensive; curation and taste are the new bottlenecks.
产品团队必须从自上而下的规划转向由品味驱动的自下而上探索。 Product teams must shift from top-down planning to bottoms-up exploration with taste-driven curation.
AI 在设计上仍然糟糕，因为品味难以评分且缺乏新颖性。 AI is still bad at design due to difficulty in grading taste and lack of novelty.
设计流程并未消亡而是转型：原型和生产代码现在充当设计产物。 The design process is not dead but transformed: prototypes and production code now serve as design artifacts.
角色融合正在发生，但专业性和最佳实践仍然重要。 Role collapse is happening, but specialties and best practices remain important.
反共识 · Contrarian takes
PRD 并未消亡；媒介必须匹配阶段：文档用于清晰，原型用于测试。 PRDs are not dead; the medium must match the stage: documents for clarity, prototypes for testing.
同一产品形态可能仅因模型时机而非功能而失败或成功。 The same product shape can fail or succeed based solely on model timing, not features.
Codex 虽为开发者设计，却作为通用知识工作工具成功。 Codex succeeded as a general knowledge work tool despite being designed for developers.
AI 模型增加代码复杂性；我们需要更擅长删除代码的模型。 AI models increase code complexity; we need models that are better at deleting code.
最好的桌面应用是连接其他工具的大本营，而非单一超级应用。 The best desktop app is a home base that connects to other tools, not a monolithic super app.
本期章节 · Chapters（共 35）
Codex 使用与设计挑战Codex Usage and Design Challenges
引言与背景Introduction and Context
AI 改变产品工作AI Changing Product Work
赞助商消息Sponsor Message
90 个原型 vs 文档90 prototypes vs. documents
AI 时代设计与软件工程Design vs. Software Engineering for AI
人类创造力 vs AI 模式Human Creativity vs. AI Patterns
设计流程已死？Design Process Dead?
AI 时代的设计流程Design process in the age of AI
团队组成与技能Team Composition and Skills
产品工作中的区域防守Zone Defense in Product Work
高能动性与品味High Agency and Taste
快速变化世界中的规划Planning in a Fast-Moving World
AI 时代的产品规划Product Planning in the Age of AI
赞助商插播：MercurySponsor Break: Mercury
为未来模型构建Building for Future Models
雄心与时机Ambition and Timing
自下而上的探索与颠覆Bottoms-up exploration and disruption
AI 在编程中的演进Progression of AI in coding
自主软件开发挑战Autonomous software development challenges
能否实现完全自主？Will we reach full autonomy?
产品领导者个人使用 AIPersonal use of AI as a product leader
设置每日简报工作流Setting up the daily brief workflow
训练应用与设置自动化Coaching the app and setting up automations
设计浏览器形态的挑战Challenges in designing the browser shape
Codex 与超级应用愿景Vision for Codex and the super app concept
Codex 编辑视频的故事Codex editing video story
失败角落：职业失败Fail corner: career failures
闪电轮：书籍推荐Lightning round: book recommendations
能动性与心态Agency and Mindset
近期最爱影视剧Favorite Recent Movie or TV Show
最喜欢的产品发现Favorite Product Discovery
人生格言与角色韧性Life Motto and Role Toughness
结束语Closing Remarks
关于 AI 与流程的更多见解Additional Insights on AI and Process