来自 OpenAI 的 Akshay 探讨了想法和品味如何成为新的瓶颈，以及 ChatGPT Work 如何体现将代码魔力带给每个人的使命。
Akshay from OpenAI discusses how ideas and taste are the new bottlenecks, and how ChatGPT Work embodies the mission of bringing code's magic to everyone.
要点 · TL;DR
有了 AI，构建变得容易；真正的分水岭是品味和想法。 With AI, building is easy; the differentiators are taste and ideas.
简化 AI 选择：默认设置应适合大多数人，而不是提供无数选项。 Simplify AI choices: default settings should fit most users, not force endless options.
衡量明确目标，而非代理指标，才能避免把动作当成进展。 Measure prescriptive goals, not proxy metrics, to avoid mistaking motion for progress.
核心观点 · Key points
用AI构建的瓶颈在于想法和品味，而非技术能力，因为现在谁都能上手构建。 The bottleneck for building with AI is ideas and taste, not technical ability, since anyone can build now.
默认模型设置应最适合大多数用户，选项应简化而非增加。 Default model settings should be the best for most users; choices should be simplified, not multiplied.
智能体为每个人解锁能力，而不仅仅是开发者，OpenAI内部Codex的采用就是例证。 Agents unlock capabilities for everyone, not just developers, as seen by Codex adoption inside OpenAI.
团队必须区分忙碌动作与真正进展；真正进展需要明确的目标，而非替代指标。 Teams must distinguish motion from progress; real progress needs prescriptive goals rather than proxy metrics.
记忆与积累的上下文让ChatGPT显得个性化，这是其跨产品价值的关键。 Memory and accumulated context make ChatGPT feel personal and are central to its value across products.
个人与职业生产力正在模糊，工具不应把用户束缚在僵化角色里。 Personal and professional productivity are blurring; tools should not box users into rigid roles.
反共识 · Contrarian takes
Sites已成为知识工作产物，在内部财务协作中取代幻灯片和电子表格。 Sites emerged as a knowledge work artifact, replacing slide decks and spreadsheets for internal finance collaboration.
用LLM生成绩效评估背景不是垃圾内容，而是智能体式搜索，能找出主管遗漏的亮点。 LLM-generated performance review context is not slop but agentic search that can surface wins supervisors missed.
更多人应使用Terra这类轻量模型，因为Soul等重型模型会耗尽算力且可能过度。 More people should use lighter models like Terra, because heavy models like Soul exhaust capacity and may be overkill.
人们应重新尝试几个月前放弃的能力；模型进步快于预期。 People should re-test capabilities they gave up on months ago; models improve faster than expectations.
我们希望用户不用选择自己的AGI版本；产品应自动将他们路由到正确的智能体模式。 We want users not to choose their AGI variant; the product should route them to the right agentic mode automatically.
合并Codex和ChatGPT看似反直觉，但这是为了避免把用户困在分离的产品身份中。 Merging Codex and ChatGPT was counterintuitive but intentional to avoid boxing users into separate product identities.
本期章节 · Chapters（共 12）
想法与品味Ideas and taste
赞助商消息Sponsor message
欢迎AkshayWelcome Akshay
无代码根基与ChatGPT WorkNo-code roots and ChatGPT Work
加入OpenAIJoining OpenAI
企业经验教训Enterprise lessons
FDE与产品对比FDE versus product
AI采用与机遇AI adoption and opportunity
ChatGPT Work发布ChatGPT Work launch
产品定位与个人效率Product positioning and personal productivity
Codex与ChatGPT Work工具Codex and ChatGPT Work harness