返回列表
🧠 阿头学 · 💬 讨论题

Anthropic员工提出的"AI采用四阶段论"是一篇嵌套了组织演进叙事的产品推广

Anthropic相关人士提出的"AI采用四阶段论"本质上是将Claude特定功能嵌入组织演进叙事的产品推广,其关于"护栏优先于token"和"替代性ROI"的观察具有管理价值,但阶段模型本身缺乏实证支撑且存在明显的循环论证。
打开原文 ↗

2026-07-17 原文链接 ↗
阅读简报
双语对照
完整翻译
原文
讨论归档

核心观点

  • 阶段论概念悬空 作者在标题和推文中宣称存在“AI 采用的四个阶段”,却自始至终未给出各阶段的具体定义、划分标准或跃迁条件,使整篇论述缺乏可检验的框架基础。
  • 产品功能被包装为组织规律 文中将 Claude 特有的 /agent 视图、/loop、/batch、worktree 隔离等功能设定为进阶的必需要素,实质是将特定 SaaS 产品的操作手册升格为普适性的组织进化论。
  • “护栏 > Tokens”的组织判断成立 尽管存在明显营销动机,作者指出模型能力只是基础、真正的瓶颈在于信任机制(端到端验证、自动代码审查、权限隔离),这一点切中了企业级 AI 落地的真实痛点。
  • ROI 衡量需区分“活动”与“真实回报” 用“原本是否会投入人工工程时长”替代仪表盘使用量来评估 AI 价值,是纠偏当前 AI 采购中虚荣指标迷信的有效方法,但该计算方式忽略了 AI 引入的隐性成本(技术债务、幻觉修复、安全审查)。
  • “四阶段”经验局限于软件工程 作者的经验样本全部来自工程师群体的代码生成场景,却试图推广至所有团队与公司,对非技术部门(如销售、HR、传统制造)的组织阻力与信任结构差异缺乏解释力。

跟我们的关联

  • 对 ATou:在评估 Claude 或其他 AI Agent 工具时,可直接套用“替代工时 ROI”模型——不问“AI 多做了多少”,而问“如果不买这个工具,原计划排期里会投入多少人工小时”,以此筛除伪需求;下一步应要求供应商证明其“护栏”能力而非仅展示生成能力。
  • 对 Neta:若将 AI 视为投资标的或效率杠杆,需警惕被“10 倍工程师”轶事和“阶段论”叙事 inflated 的估值预期;下一步应重点尽调被投企业的 AI 采用是否建立了真实的验证与审计护栏,还是仅停留在购买 tokens 的“活动”层面。
  • 对 Uota:作为内容消费者与批判者,可将此文作为典型案例训练识别“思想领导力软广”的能力——即观察作者是否将特定产品功能(如 /loop、worktree 隔离)偷换为行业必然规律,以及是否利用未定义的抽象概念(四阶段)构建虚假权威。
  • 对 ATou/Neta/Uota 的下一步:在各自场景中建立“瓶颈-护栏”检查清单——在推进任何 AI 应用前,先明确当前阶段的信任瓶颈是“验证”“权限”还是“隔离”,再匹配工具功能,而非反向被产品路线图牵着走。

讨论引子

  • 如果“AI 采用的四个阶段”从未被定义,这类模糊框架为何能在社交媒体上获得广泛传播?这是否反映了 B 端 AI 决策者的认知焦虑?
  • 当 AI Agent 的“护栏”建设成本(自动化审查、沙箱隔离、幻觉修复)可能接近甚至超过其节省的人工时,“替代工时 ROI”模型应如何修正才能反映真实收益?
  • 个体“10 倍产出”在组织内引发的更可能是效仿还是恐惧与阻力?管理者应如何设计制度来消化这种效率不平等带来的团队张力?

我每天都会和其他公司的工程师交流,听到的总是同一件事:有一个人借助 Claude 把产出提升到了 10 倍,但组织里的其他人还没跟上。

观察团队采用 AI 的过程时,我一直在反复看到同样的 4 个阶段。

我把它们整理在这里了:AI 采用的阶段 https://t.co/kQnRAUMKpP


https://x.com/bcherny/status/2077929386146169269

穿过这些阶段并不存在唯一正确的路径。每个团队、每家公司都不一样。但在每一个阶段,光有 tokens 都不足以推动你继续前进:要走到下一个阶段,你需要找到并拆解下一组瓶颈,同时建立起下一组护栏。


https://x.com/bcherny/status/2077929390806073807

在实践中,这意味着要给 Claude 提供能够端到端验证自身工作的方式。这意味着为权限启用自动模式,默认开启自动化代码审查和安全审查,并使用能够让你同时管理多个 agent 的界面,比如 CLI 中的 Agent 视图、桌面应用、iOS 和 Android 应用,以及 Tag。

要进入更高阶段,则意味着 /loop、/batch、动态工作流,以及对子 agent 进行 worktree 隔离。重点不在某一个单独的功能,而在于配合合适的护栏使用合适的功能,让 Claude 能够以团队信任其输出的方式,把整类工作自动化。


https://x.com/bcherny/status/2077929397495959693

当你的团队已经买账之后,接下来该怎么追踪?使用量值得关注,比如看一个仪表盘,但它衡量的是活动,不是回报。更好的问题是:这件事你原本是否也会投入工程资源去做?如果答案是会,那会投入多少,又会消耗多少人工工程时长?这就是你的回报。


https://x.com/bcherny/status/2077929404219474148

更大的收益出现在修复和维护都在后台发生、而你的团队可以专注于构建新东西的时候。到了那时,你才会开始做那些之前甚至不在能力范围内的事情。

Anthropic 现在处在第 3 阶段,并且正朝着第 4 阶段推进。就我个人而言,我刚刚到达第 4 级。

I talk to engineers at other companies every day and hear the same thing: one person is 10x'ing their output with Claude but the rest of the org hasn't caught up.

Watching teams adopt AI, I keep seeing the same 4 steps.

I mapped them out here: Steps of AI Adoption https://t.co/kQnRAUMKpP


https://x.com/bcherny/status/2077929386146169269

There’s no one right path through the steps. Every team and company is different. But at each step, tokens aren’t enough to move you forward: to get to the next step, you need to find and break down the next set of bottlenecks, and build up the next set of guardrails.


https://x.com/bcherny/status/2077929390806073807

In practice that means giving Claude ways to verify its own work end to end. It means enabling auto mode for permissions, defaulting on automated code review and security review, and using interfaces that let you manage multiple agents at once (Agent view in CLI, Desktop app, iOS and Android apps, Tag).

To get to higher levels it means /loop, /batch, dynamic workflows, and worktree isolation for subagents. It's not about a single feature, but rather using the right features with the right guardrails that enable Claude to automate entire classes of work in a way that your team can trust the output.


https://x.com/bcherny/status/2077929397495959693

Once your teams are bought in, how do you track it? Usage is worth watching (e.g. a dashboard), but it measures activity, not return. A better question: would you have spent engineering effort on this anyway? If yes, how much and what would it have cost in manual eng-hours? That's your return.


https://x.com/bcherny/status/2077929404219474148

The bigger payoff comes when fixing and maintaining happens in the background and your teams can focus on building. That's when you start doing things that weren't even in range before.

Anthropic is on step 3 and pushing toward 4. Personally, I just hit level 4.

Curious where you are -- what step is your team on?

我每天都会和其他公司的工程师交流,听到的总是同一件事:有一个人借助 Claude 把产出提升到了 10 倍,但组织里的其他人还没跟上。

观察团队采用 AI 的过程时,我一直在反复看到同样的 4 个阶段。

我把它们整理在这里了:AI 采用的阶段 https://t.co/kQnRAUMKpP


https://x.com/bcherny/status/2077929386146169269

穿过这些阶段并不存在唯一正确的路径。每个团队、每家公司都不一样。但在每一个阶段,光有 tokens 都不足以推动你继续前进:要走到下一个阶段,你需要找到并拆解下一组瓶颈,同时建立起下一组护栏。


https://x.com/bcherny/status/2077929390806073807

在实践中,这意味着要给 Claude 提供能够端到端验证自身工作的方式。这意味着为权限启用自动模式,默认开启自动化代码审查和安全审查,并使用能够让你同时管理多个 agent 的界面,比如 CLI 中的 Agent 视图、桌面应用、iOS 和 Android 应用,以及 Tag。

要进入更高阶段,则意味着 /loop、/batch、动态工作流,以及对子 agent 进行 worktree 隔离。重点不在某一个单独的功能,而在于配合合适的护栏使用合适的功能,让 Claude 能够以团队信任其输出的方式,把整类工作自动化。


https://x.com/bcherny/status/2077929397495959693

当你的团队已经买账之后,接下来该怎么追踪?使用量值得关注,比如看一个仪表盘,但它衡量的是活动,不是回报。更好的问题是:这件事你原本是否也会投入工程资源去做?如果答案是会,那会投入多少,又会消耗多少人工工程时长?这就是你的回报。


https://x.com/bcherny/status/2077929404219474148

更大的收益出现在修复和维护都在后台发生、而你的团队可以专注于构建新东西的时候。到了那时,你才会开始做那些之前甚至不在能力范围内的事情。

Anthropic 现在处在第 3 阶段,并且正朝着第 4 阶段推进。就我个人而言,我刚刚到达第 4 级。

I talk to engineers at other companies every day and hear the same thing: one person is 10x'ing their output with Claude but the rest of the org hasn't caught up.

Watching teams adopt AI, I keep seeing the same 4 steps.

I mapped them out here: Steps of AI Adoption https://t.co/kQnRAUMKpP


https://x.com/bcherny/status/2077929386146169269

There’s no one right path through the steps. Every team and company is different. But at each step, tokens aren’t enough to move you forward: to get to the next step, you need to find and break down the next set of bottlenecks, and build up the next set of guardrails.


https://x.com/bcherny/status/2077929390806073807

In practice that means giving Claude ways to verify its own work end to end. It means enabling auto mode for permissions, defaulting on automated code review and security review, and using interfaces that let you manage multiple agents at once (Agent view in CLI, Desktop app, iOS and Android apps, Tag).

To get to higher levels it means /loop, /batch, dynamic workflows, and worktree isolation for subagents. It's not about a single feature, but rather using the right features with the right guardrails that enable Claude to automate entire classes of work in a way that your team can trust the output.


https://x.com/bcherny/status/2077929397495959693

Once your teams are bought in, how do you track it? Usage is worth watching (e.g. a dashboard), but it measures activity, not return. A better question: would you have spent engineering effort on this anyway? If yes, how much and what would it have cost in manual eng-hours? That's your return.


https://x.com/bcherny/status/2077929404219474148

The bigger payoff comes when fixing and maintaining happens in the background and your teams can focus on building. That's when you start doing things that weren't even in range before.

Anthropic is on step 3 and pushing toward 4. Personally, I just hit level 4.

Curious where you are -- what step is your team on?

📋 讨论归档

讨论进行中…