AI时代,入职如何不再「从零交接」

marsbitPublished on 2026-05-17Last updated on 2026-05-17

Abstract

本文以作者加入Ramp公司的入职经历为切入点,探讨了在AI时代企业如何解决新员工“从零开始”的漫长适应期问题。文章指出,高速运转的公司不能依赖新人缓慢阅读文档和询问同事来获取上下文,而AI工具若各自为战,也无法发挥真正价值。 核心观点在于,企业需要构建一个持续更新、可信赖的“公司大脑”——一个系统化的知识底座。这个系统应自动吸收并沉淀会议记录、文档、Slack讨论、客户反馈和产品决策等所有内部信号,使其成为新员工和各类AI Agent统一、可随时查询的上下文来源。 作者分享了自己在Ramp前100天的实践:建立以Obsidian为核心、由Claude驱动的个人知识库,集成会议转录、笔记等内容;利用工具自动化归档信息;并在此基础上开发能自动生成会议议程、提炼产品动态等小型技能库。这些组件共同构成了一个可扩展的上下文层。 文章强调,当前企业AI应用多停留在为特定任务构建孤立“聊天机器人”的阶段,缺乏一个共享的、理解公司整体运作的“大脑”。真正的转型在于优先构建这个上下文基础设施,让入职、AI协作乃至客户接入都能从同一套丰富的背景知识开始,从而极大提升效率,让“快速上手”成为常态。 最终目标是,当“熟悉业务”这个阶段因其成本极低而不再被特别提及时,就意味着企业成功建立了高效的知识交接与复用系统。

编者按:AI 正在进入企业,但真正的问题并不是「要不要用 agent」,而是这些 agent 能不能理解公司本身。

本文以作者加入 Ramp 后的 100 天为线索,讨论了一个更底层的问题:高速运转的公司,不能只靠新人慢慢读文档、问同事、补上下文,也不能让每个 AI 工具各自为战。真正重要的是搭建一个持续更新的「公司大脑」,把会议、文档、Slack 讨论、客户反馈和产品决策沉淀下来,让新人和 agent 都能从同一套上下文出发。

当上下文被系统化,入职不再只是漫长的适应过程,AI 也不再只是一个个孤立工具。企业 AI 的价值,最终可能不在于部署多少 agent,而在于公司能否先建立起一个可信、可读、可复用的知识底座。

以下为原文:

在 4×100 米接力赛中,胜负往往不是由全程决定,而是被压缩在一段 20 米的交接区里。跑者必须在高速状态下完成接棒:接棒者起跑太早,接力棒会掉到地上;起跑太晚,交棒者不得不减速,整支队伍也会在瞬间失去优势。如果交接动作本身不够精准——手位、角度、时机任何一个环节出错——结果同样可能是掉棒。

一支队伍可以拥有全场最快的选手,却依然输在这 20 米里。速度重要,交接也重要。真正决定胜负的,是二者能否同时成立。

我见过的每一次岗位交接,本质上都像一场接力赛,只不过其中一名选手还停在起跑器上。新人周一入职,一切从零开始;组织却不会因此减速,仍然以原有节奏向前运转。于是,新人只能靠读文档、潜伏在 Slack 里、反复问同样几个问题,再花上三个月时间摸清组织的运行模式,直到自己终于变得「有用」。

我们通常把这段差距视为时间问题,仿佛只要足够久,新人自然会跟上。但事实并非如此。这个差距要么由系统解决,要么就会持续存在。

上下文,才是组织真正的交接系统

我加入 Ramp 大约 100 天了。在此之前,我在 Plaid 工作了五年,熟悉每一个产品、每一个客户故事,以及每一个决策背后的背景。我可以不假思索地讲出这些故事。但来到 Ramp 后,我对这一切几乎一无所知。

而产品营销的核心,恰恰是讲故事。如果你不知道故事里的角色、情节和前因后果,就不可能真正讲好这个故事。

从第一天起,我的目标就是搭建一个 AI-native 的产品营销组织。但要在缺乏上下文的情况下做到这一点,我首先必须扩展自己的知识底座——也就是支撑所有工作的「上下文层」。

Ramp 是一家以速度著称的公司。这里没有「下个季度再慢慢跟上」的空间。公司每周都在发布、迭代、推进。你要么跟上节奏,要么就会变成组织运行中的额外成本。

与此同时,我还在经历另一层 onboarding。Ramp 已经很快,但 AI 的变化更快,而我必须同时学习一家新公司和一个新的工作环境。我不是工程师,上一次打开终端还是大学计算机课。也就是说,我既要补上组织语境,又要适应新的 AI 工作方式,而这两件事彼此叠加,让难度进一步放大。

最终让我从这种压力中脱身的,不是完成某篇具体文章、某次产品发布,或某个工作流,而是把「上下文」本身当作交付物。只要上下文层搭建正确,后续所有工作都会变得更低成本。

于是,我开始构建真正可扩展的东西:一个能像优秀 wiki 帮助研究者一样,帮助我快速补课的系统。到第三周,它已经能基于我的笔记起草内容;到第八周,它已经能总结我没有参加的会议。学习和补课并没有消失,但随着系统不断填充,它们的成本开始一天比一天低。

这个想法的个人版本,其实已经出现一段时间了。曾任特斯拉 AI 负责人、OpenAI 创始成员之一的 Karpathy,在 4 月写过一篇文章,描述了他所说的「个人 LLM 知识库」:一个存放原始输入的文件夹,包括论文、文章、转录稿和个人笔记;一个在这些材料之上生成 wiki 的 LLM;再用 Obsidian 这样的编辑器作为前端。当资料积累到大约 100 篇文章时,LLM 就可以围绕个人语料库回答复杂问题,而不再需要复杂的检索技巧。

他的判断是:这里有机会诞生一个真正出色的新产品,而不是一堆临时拼凑的脚本。

个人版本今天已经存在了。但公司版本还没有。这正是问题所在。

大致来说,我在入职前 100 天搭建的是这样一套系统。它们都还不算精致,但共同构成了组织内部的「结缔组织」。

核心是一个 Obsidian vault,由 Claude 读取和写入。我接触过的会议转录、文档、公开观点和个人笔记,都会进入这个知识库。当我问「Geoff 和我三周前关于首页到底决定了什么」时,它会从这个 vault 中寻找答案,而不是依赖模型本身的泛化记忆。

为了持续给这个 vault 输入内容,Granola 会默认记录每一场会议,并在夜间归档转录稿。于是,我周一错过的会议,到周三就已经可以被查询。为了让公司其他人也能跟上,我选择公开工作——大多数我正在搭建的内容,会先出现在 #team-pmm 或相关发布项目频道里,然后才进入 Notion 文档。构建过程本身,就是一种同步机制。

在这个 vault 之上,还有一个小型的命名技能库,agent 可以按需调用。一个技能可以根据我和某个人最近四次会议生成议程;另一个技能可以扫描 Slack 中一周的产品动态,并转化为文章选题。每个技能大约是 200 行 markdown,用来替代过去需要手动完成的一类工作。

此外,我还基于 Ramp 的内部应用平台搭建了一个动态产品路线图。它读取的是同一套上下文层,因此它不会过期,因为它从一开始就不是静态文档。还有一份每天早上 8 点发到我 Slack 私信里的晨间摘要:昨天上线了什么、哪里卡住了、哪些事情需要我回应。这些内容在我睡觉时已经被整理好。

单独看,这些东西都不算惊艳。但放在一起,它们给出了一个可运行的答案:如果一家公司也拥有 Karpathy 所说的那种 wiki,它会是什么样子?

你可以称它为 wiki、图谱、上下文层,或者公司大脑。名字并不重要,功能才重要。它必须能够吸收公司已经产生的所有信号:会议、Slack 讨论、文档、代码、转录稿、客户电话和关键决策,并且在不依赖人工手动维护的情况下持续保持更新。它也必须成为每一位新员工、每一个新 agent 开始工作之前,首先读取的东西。

如果明天有一位新员工入职,他第一天应该读什么?如果真实答案是一份 2024 年的 Notion 文档,外加一个已经失效的 Confluence 链接,那本质上就是在让他从静止状态接棒。

从单点工具到公司大脑,AI 的真正缺口

今天,AI 进入企业的主要方式,仍然依赖 forward-deployed engineers。无论是 OpenAI、Anthropic,还是大型咨询公司,都会选择在模型之上搭建具体工作流。

这些工作是真实的,也有价值。但它们仍然停留在企业 AI 的「聊天机器人时代」:围绕特定任务封装出来的窄工具,单独看有用,却没有被接入一个能够持续复利的系统。

真正的「公司大脑」还没有出现。客服 agent 和 HR onboarding agent 可能是在不同月份、由不同团队分别搭建出来的。它们彼此并不知道上一次全员会决定了什么,不知道公司如何理解自己的市场,也不知道销售负责人在上一次管理层 offsite 上提出了什么判断。每个 agent 都只是一个有具体职责的聊天机器人,但它们并不共享同一个大脑。

这就是当前最大的缺口。而在实验室之外,几乎没有多少人在围绕这个问题构建产品。

如果你在 2026 年要组建一个团队或创办一家公司,操作顺序已经不同于 2022 年。先写上下文文件,再安装工具。记录每一场会议。先搭建 wiki,再搭建 dashboard。交付技能,而不是幻灯片。让新员工第一天阅读 wiki,第二天就开始为它贡献内容。招聘和晋升那些能让「公司大脑」持续运转的人,也要重用那些真正会读取公司大脑的 agent。

上下文不是副项目。它是让所有 AI 投资真正产生回报的基础设施。

我现在正在 Ramp 搭建其中的一部分:wiki、技能库、从同一个上下文层读取信息的应用,以及持续给它输入内容的组织机制。它还很小,也很早期。如果你也在其他地方尝试构建公司级版本,我很想交流经验。比一个值得信任的大脑更有用的,是两个大脑出现在同一个房间里。

回到接力赛。真正的胜利条件,不是最干净的交接,也不是最快的一棒,而是二者在同一段 20 米里同时发生。

新员工读取公司大脑,然后开始冲刺。新 agent 读取公司大脑,然后开始工作。新客户接入公司大脑,然后从第一天起就进入运行状态。

当「ramp-up」这个词不再有意义时,我们就知道自己做对了。

Related Questions

Q这篇文章的核心论点是什么?即,在AI时代,企业解决新员工入职和AI工具效能问题的关键是什么?

A文章的核心论点是,解决新员工入职困难和AI工具各自为战问题的关键,不在于部署更多孤立的智能体(agent),而在于企业首先需要构建一个统一的、持续更新的“公司大脑”。这个“公司大脑”是一个可信任、可读取、可复用的知识底座(或称为上下文层),它系统性地沉淀公司所有的会议、文档、讨论、决策和客户反馈。新员工和新AI工具都能从这个共同的上下文出发,从而摆脱“从零开始”的缓慢适应过程,实现高速无缝的“交接”。

Q作者在入职Ramp后,为解决自身“上下文缺失”的问题,具体搭建了哪几个核心组件?

A作者为解决“上下文缺失”问题,搭建了几个核心组件,共同构成了组织内部的“结缔组织”: 1. **核心知识库(Obsidian vault)**:由Claude读取和写入,存储所有会议转录、文档、笔记等原始材料,作为查询答案的来源。 2. **自动化内容输入系统(Granola)**:自动记录并归档会议转录稿,确保知识库持续更新。 3. **命名技能库**:包含多个小型、具体的AI技能(约200行Markdown代码),如根据近期会议生成议程、将Slack动态转为选题等,替代重复性手动工作。 4. **动态产品路线图**:基于内部应用平台搭建,从同一上下文层读取信息,因此能保持实时更新。 5. **个人晨间摘要**:每天早上8点自动生成并发送到Slack,总结前一日的工作进展、阻塞点和待办事项。

Q文章中提到的“公司大脑”需要具备哪两个关键功能?

A文章中提到的“公司大脑”需要具备两个关键功能: 1. **吸收与沉淀**:能够自动吸收公司产生的所有信号,包括会议、Slack讨论、文档、代码、客户反馈和关键决策等,并将其沉淀为结构化的知识。 2. **持续更新与可读取**:必须在不需要人工手动维护的情况下持续保持更新,并且要成为每一位新员工、每一个新AI工具开始工作前首先读取的“唯一真相源”。

Q作者认为当前企业应用AI的普遍模式存在什么主要缺陷?他用什么比喻来描述这种缺陷?

A作者认为当前企业应用AI的普遍模式是围绕特定任务(如客服、HR入职)封装出孤立的、窄功能的工具或智能体。这些工具虽然单独看有用,但彼此割裂,没有共享一个统一的“公司大脑”,因此无法理解公司的整体背景、最新决策和市场判断。 他用“聊天机器人时代”来比喻这种缺陷,意指这些AI应用就像一个个功能单一、互不关联的早期聊天机器人,缺乏一个能够持续产生复利的协同系统。

Q根据文章,未来(如2026年)组建团队或创办公司的“操作顺序”应该发生怎样的根本性改变?

A未来的“操作顺序”应该发生根本性改变,从“先安装工具,再积累知识”转变为“先构建上下文(知识底座),再安装工具”。具体包括: 1. 首先撰写和定义公司的上下文文件。 2. 记录每一场会议,先搭建动态的、活的Wiki(公司大脑),再基于此搭建数据看板(Dashboard)。 3. 注重交付可复用的“技能”,而不是静态的演示文稿。 4. 让新员工第一天就阅读并理解这个“公司大脑”,第二天就开始为其贡献内容。 5. 在招聘和晋升时,优先选择那些擅长维护和运用“公司大脑”的人才和AI工具。

Related Reads

The Paradox of Automation: The Stronger the AI, the Busier Humans Become

The Paradox of Automation: The more powerful AI becomes, the more work humans have to do. This article, based on observations from AI-heavy company Every, argues that while AI agents automate tasks like coding, writing, and customer service, they don't eliminate human jobs. Instead, they transform work and create *more* demand for human expertise. AI commoditizes "yesterday's human capabilities" by cheaply generating code, text, and images from past data. This leads to an abundance of similar, generic outputs. Consequently, what becomes scarce and valuable is human judgment in the present moment: knowing *what* is worth doing, *why*, and *how* to do it well. The article identifies two collaboration models: "Agent employees" for delegated tasks and "human-AI collaboration" within tools like Claude Code for complex work. In both cases, humans are essential to set direction, judge quality, and maintain systems. As AI makes execution cheap, human roles shift from executors to designers, reviewers, and meaning-makers. The author addresses "benchmark anxiety" by explaining that AI excels within specific, human-defined problem "frames." As AI masters one frame (e.g., code rewriting), new, more complex frames emerge (e.g., deciding *when* to rewrite). This creates an ongoing cycle where AI chases the frames, but humans remain the "framers." Even with advanced AGI, this dynamic may persist as long as AI lacks true human-like agency and self-directed purpose. The core paradox holds: automation amplifies the need for the very human judgment it seems to replace.

marsbit13m ago

The Paradox of Automation: The Stronger the AI, the Busier Humans Become

marsbit13m ago

a16z: 7 Charts to Understand How Tokenization is Changing the Nature of Assets

"a16z: 7 Charts on How Tokenization is Changing the Nature of Assets" Tokenized Assets (or Real-World Assets - RWA) are transforming asset forms, liquidity, and financial system construction. The market recently surpassed $30 billion, stabilizing around $34 billion (excluding stablecoins), representing a tenfold increase in less than two years, driven by clearer regulations, mature institutional infrastructure, and increased financial institution adoption. The primary driver of recent growth is tokenized U.S. Treasury bonds. These offer investors efficient, flexible digital access to yield-bearing assets and improve institutional operations like settlement and collateral management. Other asset classes show varied growth: asset-backed credit leads, followed by niche financial assets (e.g., reinsurance, mining notes), while venture capital took longer to scale. Market segmentation shows high concentration. In commodities, tokenized gold dominates (~$5 billion), as its standardized, storable nature fits tokenization well. Bonds are the largest category ($15.2B), but only ~5% are used in DeFi protocols. Conversely, smaller niches like reinsurance tokens see high (~84%) on-chain utilization, highlighting a core industry divide: most current tokenized assets are merely digitized records for easier holding/transfer, lacking the "composability" (free combination/interaction) that is key to blockchain-native finance. The ecosystem is distributed across multiple blockchains, with Ethereum hosting over half the value ($15.7B), followed by BNB Chain, Solana, and others. Future market size predictions vary widely (e.g., $2-$30 trillion by 2030+), but all indicate massive potential from the current small base. Tokenized assets currently represent minuscule fractions of their global counterparts (e.g., 0.01% of global bonds). The current phase focuses on digitizing straightforward assets. The next challenge is to bring more complex financial components on-chain and deeply integrate tokenized assets into composable, internet-native financial infrastructure.

链捕手54m ago

a16z: 7 Charts to Understand How Tokenization is Changing the Nature of Assets

链捕手54m ago

a16z: How Tokenization is Transforming the Nature of Assets in 7 Charts

"Tokenized Assets: How Tokenization Changes the Nature of Assets" by a16z Crypto The market for tokenized assets, excluding stablecoins, has grown from under $3 billion two years ago to over $340 billion today. US Treasury bonds are the primary growth driver, allowing investors to hold yield-bearing assets digitally and enabling more efficient settlement. Other key sectors include private credit (growing fastest), commodities (dominated by gold), and niche financial assets. However, the market remains concentrated in tokenized US Treasuries and gold. A critical insight is that most tokenized assets currently lack "composability." While the total market is large, only a small fraction is actively used within DeFi protocols. For instance, only about 5% of tokenized bonds and a low percentage of tokenized gold are utilized on-chain. In contrast, assets like reinsurance and private credit tokens show much higher on-chain usage rates (84% and 33%, respectively). This highlights a divide: many tokenized assets are merely digital records on a blockchain without enabling new, programmable financial applications. The Pantera Capital Token Native Index indicates over 70% of tokenized assets have minimal on-chain native functionality. Ethereum remains the dominant blockchain for tokenized assets (over $150B), but the ecosystem is diversifying across chains like BNB Chain, Solana, and Stellar, based on factors like cost and compliance. Major institutions forecast massive future growth, with predictions for the tokenized asset market ranging from $2 trillion to over $30 trillion by the early 2030s. However, compared to the global financial system (e.g., ~$140T bonds, multi-trillion dollar gold market), tokenized assets currently represent a tiny fraction (0.01% or less). The conclusion is that while tokenization has begun by digitizing and streamlining settlement for simpler assets, the next phase involves bringing more complex financial instruments on-chain and deeply integrating them into composable, internet-native financial infrastructure.

Odaily星球日报1h ago

a16z: How Tokenization is Transforming the Nature of Assets in 7 Charts

Odaily星球日报1h ago

The Revived Codex, Carrying OpenAI's Hopes for IPO

This article analyzes the intense recent development of OpenAI's Codex, positioning it as a crucial component for OpenAI's impending IPO. Over the past two months, Codex has seen a rapid series of major updates focused on integrating into real enterprise workflows. Key new features include enhanced context capture (Appshots, file previews, built-in browser), long-running task execution ("Goal Mode"), remote operation (phone control, lock-screen access), and enterprise management tools (plugin sharing, access tokens, automated risk review). These updates aim to make Codex a comprehensive AI workbench that can "see the scene, push tasks, and manage risks." The author argues that while ChatGPT proves OpenAI's massive user base and API provides foundational revenue, Codex represents OpenAI's clearest path to demonstrating tangible, high-value commercial viability. It targets developers and engineering teams—a segment already accustomed to paying for efficiency gains in costly software development cycles. This is critical because, despite higher overall revenue, OpenAI's adjusted operating margins remain deeply negative, highlighting the challenge of outrunning immense compute costs. The pressure is amplified by competitor Anthropic's success with Claude Code, which has shown that a focused approach on high-value enterprise and developer workflows can lead to a path toward profitability. Codex's aggressive evolution is thus seen as OpenAI's strategic move to capture a similar enterprise-ready, revenue-generating narrative essential for its market debut. In essence, "ChatGPT proved OpenAI has users. Codex needs to prove OpenAI is a business that can make money."

marsbit2h ago

The Revived Codex, Carrying OpenAI's Hopes for IPO

marsbit2h ago

a16z: 7 Charts to Understand How Tokenization Is Changing the Nature of Assets

a16z: 7 Charts on How Tokenization is Transforming the Nature of Assets Tokenized Assets, often referred to as "real-world assets" (RWA), are altering the form, flow, and structure of the financial system. The market recently surpassed $30 billion (excluding stablecoins), driven largely by tokenized U.S. Treasuries. These offer investors digital, yield-bearing assets with efficient settlement. Growth varies significantly by asset class. Asset-backed credit leads in speed, followed by niche financial assets, while venture capital and active strategies took longer to scale. U.S. Treasuries and commodities dominate, holding about two-thirds of the current market share. Within commodities, gold tokenization dominates entirely due to its standardization and historical appeal in crypto. The ecosystem is spread across multiple blockchains. Ethereum holds over half the market, with others like BNB Chain, Solana, and Stellar holding significant shares. However, a key insight is that most tokenized assets currently lack "composability." While the total market is large, only a small fraction (e.g., 5% of tokenized bonds) is used within DeFi protocols. Many tokens are simply digital records of off-chain assets, not natively programmable financial building blocks. In contrast, smaller categories like reinsurance tokens see very high on-chain usage. Looking ahead, forecasts for the tokenized asset market by 2030 range from $2 trillion to over $30 trillion, representing immense potential growth from today's ~$340 billion base. Yet, relative to global markets (e.g., $140T+ in bonds), tokenization's penetration remains minuscule (<0.02%). The current phase focuses on digitizing straightforward assets for efficiency. The next major challenge is bringing more complex financial instruments on-chain and integrating tokenized assets into truly composable, internet-native financial infrastructure.

marsbit2h ago

a16z: 7 Charts to Understand How Tokenization Is Changing the Nature of Assets

marsbit2h ago

Trading

Spot
Futures

Hot Articles

Ethena: Building a New Era of Web3‑Native Digital Dollars

Ethena is an Ethereum‑based synthetic dollar protocol that delivers crypto‑native monetary solutions, including USDe, a synthetic dollar, and sUSDe, a globally accessible U.S. dollar savings asset.

53.0k Total ViewsPublished 2026.03.16Updated 2026.03.16

Ethena: Building a New Era of Web3‑Native Digital Dollars

Discussions

Welcome to the HTX Community. Here, you can stay informed about the latest platform developments and gain access to professional market insights. Users' opinions on the price of ERA (ERA) are presented below.

活动图片