跳到正文
原文
Tomasz Tunguz 博客·· 10 天前AI 评分50

OpenAI 首席人力官 KC 谈超越自动化的工作重构

Beyond Automation: KC on Reimagining Work at OpenAI

文章要点
  • 1OpenAI 首席人力官 KC 称 junior 与 senior 之分已失去意义,经验正快速贬值
  • 2他提出模型是大脑,Codex 或 Claude Code 是外骨骼,人应转向与人协作
  • 3他建议给每个职位加 agent manager 后缀,如销售转为销售智能体管理者
  • 4OpenAI 称其智能体约 88 小时给出 Navier-Stokes 千年奖问题解法,数学界仍有争议
决策要点

AI 能力曲线变陡后,企业真正的选择不是把旧任务做快,而是重设工作方式和人才评价标准。

  • 影响:按资历定级和评估人才的框架会逐渐失准,需改看近期产出;把系统更新类工作交给智能体后,人的价值向协调与人际能力转移;更快完成旧任务只提升利润,重设业务才改变增长轨迹。
  • 参考:可考虑重审岗位描述与晋升标准中对资历的依赖,改按近期产出评估人;值得盘点哪些记录与更新类工作可交给智能体,并评估团队是否需要 agent manager 类角色。
  • 依据:观点来自 OpenAI 首席人力官在 Office Hours 的现场分享,属个人判断;Navier-Stokes 解法为 OpenAI 官方口径,数学界仍有争议。
正文

Last week we welcomed KC, Chief People Officer at OpenAI, on the tremendous change inside organizations.

KC traced four years of model progress in three steps. Two years ago, the parlor trick was asking a model how many R’s are in strawberry. A year ago, reasoning models solved competition-level math. In September, OpenAI’s agents attacked the Navier-Stokes Millennium Prize problem & said they had a solution in 88 hours1.

That implies a choice for companies : accomplish the same tasks faster or reimagine what is possible. The first lifts earnings & profit. The second bends the arc of a business. At first the slopes look the same. Then the geometric compounding creates explosive separation.

In this future, the people team needs to define how we work.

First, in answer to the most common questions from attendees (how does one evaluate junior talent?) KC replied the words junior & senior lack meaning today.

“I think it’s incorrect to use the word junior and senior. I think we should completely get rid of that word, because for the world that we’re heading to, junior usually means that you don’t have experience that’s relevant for a particular context. But for the world we’re heading to, maybe all of us don’t have experience that’s relevant for that particular context.”

What one knows about their work as of two years ago matters less than what you knew three months ago, so experience fades fast.

Next, he drew his vision of modern work. The model is a brain. Around it sits the harness, Codex or Claude Code, which he called an exoskeleton : skills, tools, & a runtime that holds context across agents. Above that sit incarnations : an assistant, then specialists.

Most firms use people as middleware for their own systems. Do the work, then log the work. He calls it work on work.

“The moment you relegate the system updating to the agents, humans start working with humans.”

That reframes each of our titles : add a suffix : agent manager. I am not a salesperson but a sales agent manager.

The session closed on EQ (emotional intelligence), through a story about an oncologist’s funeral. A patient saw several doctors with the same diagnosis & plan. She picked this one for his bedside manner.

“The differentiator is how you show up as a human. It’s how you show up, how you act with integrity, how you are really humble and want to learn.”

When everyone can reach for intelligence beyond their own, that is what still separates us. KC thinks that is good news. It sends us back to work worth doing.

I’m grateful to KC for sharing his insights at Office Hours!


  1. OpenAI, “On the Navier-Stokes Millennium Prize Problem,” September 8, 2026, reporting roughly 10,000 concurrent agents & a solution about 88 hours after launch. The result remains contested among mathematicians. ↩︎

来源:Tomasz Tunguz 博客 · tomtunguz.com