Agent 时代的产品开发 · Anthropic PM 的工作怎么变了
协调被压缩,手艺被还回来
"One of the ironies of being a product manager in the age of AI is that my work feels more human than ever."
"The job of product management has always been a mix of craft and alignment. For most of my career, my week was occupied by the latter: meetings with cross-functional stakeholders and teammates, status reports, and ticket backlogs."
PM 工作过去被会议、汇报、ticket backlog 占满,Claude 把"对齐"那部分自动化掉,反而把时间还给真正重要的那一部分——手艺。这是这篇文章的反讽,也是核心论点。
文档里 spec 漂亮没用,直接 build 才暴露真问题
"A spec that reads elegantly in a doc can fall apart the first time you try to build against it."
"API design used to live in documents and comment threads; on the AI exponential, we build with what we ship. With Claude Code, I can sketch out an agent against pre-production versions of our API specs, and within an afternoon be running a real prototype end-to-end."
Jess 用 Claude Code 在 pre-prod API spec 上跑出真原型,几次重塑 API 抽象——这些问题文档评审照不出来,等用户反馈又太晚。"build with what we ship" 是 AI 指数曲线下的对齐方式。
PM 的工具栈分三层:发散 / 收敛 / 基建
"I use Claude and Claude Cowork for open-ended research and discovery. Once I have greater clarity on the job to be done, I use Claude Code to write and ship a custom agent for it, built atop of Managed Agents."
Claude / Cowork 处理还没收敛的早期探索;Claude Code 在 JTBD 清晰后写并 ship 定制 agent;Managed Agents 是底座。三层各管一段——可以直接迁移到任何产品团队的工作流。
PM 给自己造三个 agent:数据 · 情绪 · demo
"Now I spin up bespoke agents for any 'job to be done.'"
"Building one is simple: I load the Managed Agents skill in Claude Code and outline a quick sketch of what I'm looking for. Managed Agents sessions run in the cloud, so I can walk away and come back to find the work done."
三个真实例子——adoption analytics(查内部数据库找 outlier)、developer sentiment(扫开发者反馈、并行 fan out)、demo building(模板 → 定制成品 demo)。每一个,都对应一类原本 scale 不动的 ops 长尾。
没造过 agent 的 PM,本周就开始
"If you're a product manager and you haven't built an agent yet, that's where I'd start this week. The experiments and tools you've always wished existed are a single prompt and a few API calls away."
这是文章的呼吁,不只是 Anthropic 内部经验。一个 prompt + 几个 API call,过去想做做不动的工具,现在做得动了。门槛已经低到不该再观望。
One of the ironies of being a product manager in the age of AI is that my work feels more human than ever.
在 AI 时代当 PM,有一个反讽:我的工作反而比以前更"像人"了。
The job of product management has always been a mix of craft and alignment.
做 PM 这件事,一直是手艺(craft)和对齐(alignment)两件事的混合。
For most of my career, my week was occupied by the latter: meetings with cross-functional stakeholders and teammates, status reports, and ticket backlogs with my engineering team.
我大半职业生涯,每周时间几乎都被"对齐"吃光了:跨职能 stakeholder 和团队的会议、status report、和工程团队的 ticket backlog。
I got used to making instinctive, quick decisions followed by uphill battles advocating, convincing, and resourcing; shipping impactful products often felt more transactional than generative.
我习惯了凭直觉快速做决定,然后再一路爬坡:rally 人、说服人、抢资源;ship 真正有影响力的产品,常常感觉更像在做交易,而不是在创造。
With Claude, I can pressure test ideas, automate workflows, and get unstuck.
有了 Claude,我能压力测试想法、自动化工作流、把卡住的环节推开。
I'm finally spending real time with our users and my team on the part of the job that always mattered most: the craft.
我终于能把真正的时间还给用户、还给团队,花在这份工作里一直最重要的那一部分:手艺。
While these new workflows changed my day-to-day, the most meaningful shifts happened when we started developing Claude Managed Agents (currently in beta), a suite of composable APIs for building and deploying cloud-hosted agents at scale.
新的工作流改变了我的日常,但真正有意义的转变,发生在我们开始做 Claude Managed Agents(目前 beta)的时候——一套可组合的 API,用来在云端规模化构建和部署 agent。
In this post, I'll share how Managed Agents has changed the way I work as a product manager, and a few patterns you can borrow for your own workflows.
这篇文章,我会分享 Managed Agents 怎么改变了我作为 PM 的工作方式,以及几个你可以直接借鉴的模式。
API design used to live in documents and comment threads; on the AI exponential, we build with what we ship.
API 设计过去躺在文档和评论线里;但身处 AI 指数曲线上,我们是用"我们 ship 的东西本身"来 build。
A spec that reads elegantly in a doc can fall apart the first time you try to build against it.
一份在 doc 里读起来漂亮的 spec,你第一次拿它去 build,可能就崩了。
With Claude Code, I can sketch out an agent against pre-production versions of our API specs, and within an afternoon be running a real prototype end-to-end.
用 Claude Code,我能在 pre-prod 版本的 API spec 上把 agent 画出草稿,一个下午就跑通端到端的真原型。
We reshaped API abstractions and Claude Console UX several times based on what we learned building with our own primitives—changes that even a multi-week doc review would never have surfaced, and otherwise would've come up too late via user feedback.
我们好几次重塑了 API 抽象和 Claude Console UX,靠的是用自家 primitive 来 build 时学到的东西——这些发现,哪怕走几周的文档评审也照不出来,等用户反馈到位就晚了。
We still litigate shapes and run raw curl requests to make sure we're happy with the bare-metal experience, but Claude Code gets me from the basic "hello world" test to a functional agent in the same sitting.
我们仍然会扣 shape、跑 raw curl,确认裸金属体验过关;但 Claude Code 让我从 "hello world" 一气走到一个能用的 agent。
As I build these agents, I'm able to more concretely anticipate ways our harness and API can flex for the next wave of model and task evolution.
在 build 这些 agent 的过程里,我能更具体地预判出我们的 harness 和 API,要怎样去适应下一波模型和任务演进。
Initially, these prototypes were just for shaping the product, but they now are evolving my day-to-day work as well.
起初,这些原型只是为了塑形产品;但现在,它们也在演化我自己每天的工作方式。
My workflow as a PM now splits cleanly across our products.
我作为 PM 的工作流,现在很清晰地分到了几个我们自己的产品上。
I use Claude and Claude Cowork for open-ended research and discovery—the murky, early-stage exploration where I want an ongoing conversation.
我用 Claude 和 Claude Cowork 做开放式调研和发现——那种还很模糊、早期、需要持续对话的探索。
Once I have greater clarity on the job to be done, I use Claude Code to write and ship a custom agent for it, built atop of Managed Agents.
一旦对"要解决的活"(JTBD)有了更明确的判断,我就用 Claude Code 写并 ship 一个定制 agent,底座是 Managed Agents。
The two-pronged payoff has been the biggest unlock.
这种双向回报,是最大的解锁。
On one side, being able to build against my own product easily raises the ceiling on what I can imagine shipping next.
一边是:能轻松地用自家产品来 build,直接抬高了我对下一版能 ship 什么的想象上限。
On the other, once the product is live, the same development muscle lets me automate the long tail of operational work that used to stall in my backlog.
另一边是:产品上线后,同样的开发肌肉让我能自动化掉那些长尾的 ops 工作——它们以前永远卡在 backlog 里推不动。
Now I spin up bespoke agents for any "job to be done."
现在,任何一个"要解决的活",我都直接 spin 起一个定制 agent 来干。
Building one is simple: I load the Managed Agents skill in Claude Code and outline a quick sketch of what I'm looking for.
做一个很简单:在 Claude Code 里加载 Managed Agents skill,大致说一下我要的是什么样。
Developers can also use the latest version of Claude Code and built-in claude-api skill to build with Managed Agents—just prompt Claude with "start onboarding for managed agents in Claude API" to get started.
开发者也可以直接用最新版 Claude Code 和内置的 claude-api skill 来基于 Managed Agents 建 agent——直接告诉 Claude "start onboarding for managed agents in Claude API" 就行。
After invoking this skill, Claude builds the agent, explaining its integration steps along the way, so I can easily shift direction as needed.
skill 一旦被调起,Claude 会一边 build agent、一边解释每步是怎么集成的,我中途想换方向也能随时调。
An agent with persistent access to our internal databases and skills for understanding our data schemas runs queries to surface interesting outliers and patterns. With memory of prior runs, it can build on prior findings and continuously advance its perspective.
一个 agent 持续访问我们内部数据库,并配着读懂 schema 的 skill,跑查询、把有趣的 outlier 和 pattern 浮出来。它带 memory,能基于之前的发现继续深挖,持续推进自己的视角。
An agent with the pre-built web search tool and guidance on focus areas scans a specific list of domains for the latest developer feedback, reporting back on common themes. Since there is so much content to analyze, it fans out research to multiple agents in parallel, waits for results, and synthesizes findings.
一个带 web search tool 的 agent,按指定关注域扫一组特定网站的最新开发者反馈,回报共性主题。因为内容量太大,它把任务 fan out 给多个并行 agent,等结果回来再做综合。
An agent with access to demo GitHub repos, branding assets, and an event deck turns prebuilt templates into a polished demo tailored to the relevant audience, such as a conference or customer meeting.
一个 agent 能访问 demo 的 GitHub repo、品牌素材、活动 deck,把预制模板变成针对当前受众(如某个会议或某个客户)定制的成品 demo。
Managed Agents sessions run in the cloud, so I can walk away and come back to find the work done.
Managed Agents 的 session 跑在云上,所以我能直接走开,回来发现活已经干完了。
Processes that could never scale because every launch has its own quirks are now easy to automate using Managed Agents, and every agent run feels energizing instead of tedious.
那些过去永远 scale 不动——因为每次发布都有自己的怪癖——的流程,现在用 Managed Agents 很容易自动化,每跑一次都感觉有劲,而不是繁琐。
A year ago, all of this kind of work would've crawled along in cross-functional staffing requests, chaotic spreadsheets, or half-baked concepts I just never got to try out.
一年前,这类工作要么淹没在跨部门 staffing 需求里慢慢爬,要么散在混乱的电子表格里,要么是我永远抽不出空去试的半成品概念。
Now, with Claude and Managed Agents, I can scale myself, using my time to partner with my team on developing the most impactful products.
现在,有了 Claude 和 Managed Agents,我能 scale 我自己,把时间花在和团队一起做最有影响力的产品上。
My day now spans generating innovative ideas with customers, digging into murky and ambiguous problems with my engineering counterparts, and investing real creative energy in frontier product work.
我的一天现在横跨:和客户一起冒新点子、和工程对位的同事一起挖那些模糊不清的问题、把真正的创造力投入到前沿产品工作里。
If you're a product manager and you haven't built an agent yet, that's where I'd start this week.
如果你是 PM、还没自己造过 agent,本周就从这里开始。
The experiments and tools you've always wished existed are a single prompt and a few API calls away.
那些你一直希望存在的实验和工具,如今只差一个 prompt 和几个 API call 的距离。