Every 公司 CEO Dan Shipper 分享反直觉预测：AI 不会消灭工作，而是将工作分化为代理辅助和 AI 原生环境，同时 SaaS 股票仍值得买入。
Dan Shipper, CEO of Every, shares contrarian predictions: AI won't kill jobs but will bifurcate work into agent-assisted and AI-native environments, while SaaS stocks remain a buy.
要点 · TL;DR
AI 不会取代工作；人类将管理智能体并创造新价值。 AI won't replace jobs; humans will manage agents and create new value.
工作将分化为个人智能体和公司超级智能体两种模式。 Work bifurcates into personal agents and a super agent for the company.
每个人都将拥有至少一个可以委派工作的智能体，大部分工作将在 Codex 或 Claude Co-work 中完成。 Everyone will have at least one agent to delegate work to, and most work will happen in Codex or Claude Co-work.
智能体需要一个关心它们的人才能良好运作；自动化需要人类监督。 Agents need a human who cares about them to function well; automation requires human oversight.
SaaS 不会消亡；智能体会增加 SaaS 用户，公司应该买入 SaaS 股票。 SaaS is not dying; agents increase SaaS users, and companies should buy SaaS stocks.
产品经理和全栈设计师将蓬勃发展，因为他们可以直接使用 AI 工具交付产品。 PMs and full-stack designers will thrive because they can ship products directly using AI tools.
AI 失业末日不会发生；人类仍然需要管理智能体并创造新价值。 The AI job apocalypse is not happening; humans are still needed to manage agents and create new value.
反共识 · Contrarian takes
命令行界面已过时；像 Codex 和 Claude Co-work 这样的图形界面是未来的工作界面。 CLIs are over; GUIs like Codex and Claude Co-work are the future of work surfaces.
公司应该为整个公司配备一个超级智能体，而不是为每个员工配备个人智能体。 Companies should have one super agent for the whole company, not personal agents for each employee.
SaaS 公司不应将 AI 构建到产品中；用户将通过 Codex 自带 AI。 SaaS companies should not build AI into their products; users will bring their own AI via Codex.
基准测试高估了 AI 的自主性；现实任务仍需要人类设定框架和监督。 Benchmarks overestimate AI autonomy; real-world tasks still require human framing and oversight.
对于内部文档和电子邮件，AI 生成的内容通常比人类写得更好。 AI-generated writing is often better than human writing for internal documents and emails.
AI 的前沿不在旧金山，而在 AI 与真实人类做某事的交汇处。 The edge of AI is not in San Francisco but wherever AI meets a real human doing something.
本期章节 · Chapters（共 33）
开场与丹的独特视角Opening and Dan's unique perspective
写作作为未来显化方式Writing as a way to manifest the future
赞助商：WorkOSSponsor: WorkOS
对话结构：三类预测Structure of the conversation: three buckets of predictions
预测：工作将分化为两种代理使用模式Prediction: work will bifurcate into two modes of using agents
个人代理 vs 超级代理Personal agents vs super agent
编码代理的演进Evolution of Coding Agents
Codex 作为统一工作空间Codex as a Unified Workspace
带浏览器的AI代理 vs 浏览器中的AIAI Agent with Browser vs. AI in Browser