Brett Taylor discusses the shift to AI agents and outcome-based pricing, sharing his biggest mistake and the mindset behind his success.
要点 · TL;DR
AI 的下一个大机会是基于结果的定价的智能体，而不是基础模型。 The next big AI opportunity is agents with outcome-based pricing, not building foundation models.
成功来自灵活的自我认同、重新定位角色、知识诚实和从失败中学习。 Success comes from flexible identity, reframing roles, intellectual honesty, and learning from losses.
AI 时代要理解系统而非只写代码；为可验证性设计语言；让孩子把 AI 当作工具来用。 In the AI era, understand systems over coding; design languages for verifiability; treat AI as a utility for kids.
核心观点 · Key points
AI 市场将转向智能体和基于结果的定价，这显然是构建和销售软件的正确方式。 The AI market will shift towards agents and outcomes-based pricing, which is the obvious correct way to build and sell software.
前沿模型开发由超大规模企业和大型实验室主导，因为需要巨额资本支出，不适合初创公司。 Frontier model development is dominated by hyperscalers and large labs due to massive capex requirements, making it unsuitable for startups.
应用型 AI，尤其是智能体，将成为新的应用形态，为初创公司在特定业务领域提供长尾机会。 Applied AI, especially agents, will become the new app, with long-tail opportunities for startups in specific business domains.
计算机科学教育对于学习系统思维仍然有价值，即使编码转变为操作代码生成机器。 Computer science education remains valuable for learning systems thinking, even as coding transforms into operating code-generating machines.
AI 智能体通过自主完成任务可以显著提高生产力，使结果可衡量，并实现基于价值的定价。 AI agents can drive significant productivity gains by autonomously completing jobs, making outcomes measurable and enabling value-based pricing.
反共识 · Contrarian takes
编程的未来将需要为 AI 生成设计的新语言或系统，优先考虑可验证性而非人类易用性。 The future of programming will require new languages or systems designed for AI generation, prioritizing verifiability over human ergonomics.
编码智能体经常产生错误的代码，修复它可能比自己写代码更难，因此生产力提升需要根本原因分析。 Coding agents often produce incorrect code, and fixing it can be harder than writing code yourself, so productivity gains require root-cause analysis.
基于结果的定价优于基于使用的定价，因为 token 不是衡量价值的好方法，尤其是在 AI 领域。 Outcomes-based pricing is superior to usage-based pricing because tokens are not a good measure of value, especially in AI.
直接销售在 AI 领域正在复兴，与最近的产品驱动增长趋势相反，因为许多 AI 产品的购买者和使用者不同。 Direct sales is making a comeback in AI, contrary to the recent trend of product-led growth, because many AI products have different buyers and users.
AI 应该像谷歌搜索一样被视为一种工具，而不是像上瘾的手机，应该鼓励孩子用它来学习。 AI should be treated as a utility like Google Search, not like addictive mobile phones, and kids should be encouraged to use it for learning.
本期章节 · Chapters（共 41）
开场与介绍Introduction and Opening
Google Local发布与初期失败Google Local Launch and Initial Failure
转向谷歌地图Pivoting to Google Maps
谷歌地图的产品教训Product Lessons from Google Maps
主持人承认失败与成功Host Acknowledges Failure and Success
介绍与提问Introduction and Question
灵活身份与建造者心态Flexible Identity and Builder Mindset
多面创始人技能Multifaceted Founder Skills
谢丽尔·桑德伯格的指导Sheryl Sandberg's Mentorship
转折点The Turning Point
重新定义工作Reframing the Job
在新职责中寻找快乐Finding Joy in New Responsibilities
有影响力的工作与Cheryl建议Impactful Work and Cheryl's Advice
知识诚实与错误叙事Intellectual Honesty and Incorrect Storytelling
从失败中学习：FriendFeedLearning from Failures: FriendFeed
从失败竞争中学到的Learning from a lost competition
判断该听谁建议的困难The difficulty of knowing whose advice to take
建议与独立思考Advice and Independent Thinking
学编程与计算机科学Learning to Code and Computer Science
学编程与理解系统Learning to Code vs. Understanding Systems
编程抽象的演变Evolution of Programming Abstractions
AI对编程语言的影响Impact of AI on Programming Languages
为AI代码生成设计Designing for AI Code Generation
软件开发演变Software Development Evolution
赞助商VantaSponsor: Vanta
AI时代教孩子Teaching Kids in the Age of AI
教育与AIEducation and AI
代理公司是高利润SaaSAgent Companies as High-Margin SaaS
初创应专注代理而非基础模型Startups Should Focus on Agents, Not Foundation Models