Mike Krieger：AI 时代产品构建，构建简单了但直觉仍关键Mike Krieger: In the AI Era, Building Is Easy but Intuition Remains Key
软件构建从未如此简单——AI 能让你在几小时内从零打造出完整产品——但知道该构建什么、该砍掉什么的直觉，仍然需要真实世界的经验和刻意的迭代来培养。
Mike Krieger 是 Instagram 的联合创始人，目前在 Anthropic Labs 工作。他从推出 Instagram（此前从失败产品 Bourbon 转向）的经历出发，解释了 2026 年产品开发的新现实。过去，数月构建和用户反馈塑造了最终的简洁体验；如今，Claude 能在短短两小时内重建完整的 Bourbon，甚至预见性地添加了 Instagram 后来的滤镜等功能。然而，模型擅长堆叠功能，却不擅长知道何时简化，这容易导致“室内树”——过度构建的产品，缺少逐步真实用户反馈带来的强化力量。
Krieger 强调，一旦你知道要构建什么，实际构建就变得微不足道，但磨练你对世界的看法和干预方式仍需时间。在 Anthropic，他们采用快速重写，推出最小可用版本，如他们的 co-work 工具仅用 10 天推出，并培养 agent-native 产品，让 AI 能原生交互并修改自身环境。“我感觉这就是 2026 年软件设计的艺术与科学，”他指出，强调在 vibe coding 和快速迭代时代，平衡稳健性、打磨度和灵活性的重要性。
团队现在将能写代码的设计师与工程师配对成小型、高信念驱动的组，保持协调轻量以实现快速转向。招聘仍重视深厚的系统专长来确保架构强度，而产品感和品味比以往更重要。对构建者的启示：AI 加速了执行，但也压缩了删除、重写和贴近用户的必要性——不要让速度创造出泛化或脆弱的产品。
Mike Krieger 是 Instagram 的联合创始人，目前在 Anthropic Labs 工作。他从推出 Instagram（此前从失败产品 Bourbon 转向）的经历出发，解释了 2026 年产品开发的新现实。过去，数月构建和用户反馈塑造了最终的简洁体验；如今，Claude 能在短短两小时内重建完整的 Bourbon，甚至预见性地添加了 Instagram 后来的滤镜等功能。然而，模型擅长堆叠功能，却不擅长知道何时简化，这容易导致“室内树”——过度构建的产品，缺少逐步真实用户反馈带来的强化力量。
Krieger 强调，一旦你知道要构建什么，实际构建就变得微不足道，但磨练你对世界的看法和干预方式仍需时间。在 Anthropic，他们采用快速重写，推出最小可用版本，如他们的 co-work 工具仅用 10 天推出，并培养 agent-native 产品，让 AI 能原生交互并修改自身环境。“我感觉这就是 2026 年软件设计的艺术与科学，”他指出，强调在 vibe coding 和快速迭代时代，平衡稳健性、打磨度和灵活性的重要性。
团队现在将能写代码的设计师与工程师配对成小型、高信念驱动的组，保持协调轻量以实现快速转向。招聘仍重视深厚的系统专长来确保架构强度，而产品感和品味比以往更重要。对构建者的启示：AI 加速了执行，但也压缩了删除、重写和贴近用户的必要性——不要让速度创造出泛化或脆弱的产品。
The Takeaway: Building software has never been easier—AI lets you go from zero to a complete product in hours—but developing the intuition to know what to build and what to cut still takes real-world experience and deliberate iteration.
Mike Krieger, co-founder of Instagram and now working at Anthropic Labs, draws from launching Instagram after pivoting from a failed product called Bourbon to explain the new realities of product development in 2026. Previously, months of building and user feedback shaped the final simple experience; today, Claude can rebuild Bourbon feature-complete in just two hours, even anticipating future additions like filters. Yet models are great at piling on features but poor at knowing when to simplify, risking "indoor trees"—overbuilt products without the strengthening forces of incremental real-user exposure.
Krieger stresses that once you know what to build, construction is trivial, but honing your worldview and interventions on the world still demands time. At Anthropic, they've embraced quick rewrites, launching minimal viable versions like their co-work tool in 10 days, and fostering agent-native products where AI can natively interact with and modify its own environment. "I feel like that is the art and science of software design in 2026," he notes, pointing to the balance of robustness, polish, and flexibility in an era of vibe coding and rapid iteration.
Teams now pair designers who code with engineers in small, conviction-driven groups, keeping coordination light to enable fast pivots. Hiring still values deep systems expertise for architectural strength, while product sense and taste matter more than ever. The lesson for builders: AI accelerates execution but compresses the need to delete, rewrite, and stay close to users—don't let speed create generic or fragile products.
查看原文 →
Mike Krieger, co-founder of Instagram and now working at Anthropic Labs, draws from launching Instagram after pivoting from a failed product called Bourbon to explain the new realities of product development in 2026. Previously, months of building and user feedback shaped the final simple experience; today, Claude can rebuild Bourbon feature-complete in just two hours, even anticipating future additions like filters. Yet models are great at piling on features but poor at knowing when to simplify, risking "indoor trees"—overbuilt products without the strengthening forces of incremental real-user exposure.
Krieger stresses that once you know what to build, construction is trivial, but honing your worldview and interventions on the world still demands time. At Anthropic, they've embraced quick rewrites, launching minimal viable versions like their co-work tool in 10 days, and fostering agent-native products where AI can natively interact with and modify its own environment. "I feel like that is the art and science of software design in 2026," he notes, pointing to the balance of robustness, polish, and flexibility in an era of vibe coding and rapid iteration.
Teams now pair designers who code with engineers in small, conviction-driven groups, keeping coordination light to enable fast pivots. Hiring still values deep systems expertise for architectural strength, while product sense and taste matter more than ever. The lesson for builders: AI accelerates execution but compresses the need to delete, rewrite, and stay close to users—don't let speed create generic or fragile products.