返回列表
🧠 阿头学 · 💬 讨论题 · 💰投资

Ramp 的招聘观：用“尖峰能力”而不是标准履历组队

这篇文章最有价值的判断是“招聘要围绕具体问题寻找尖峰能力”，但它也明显把少数成功个案包装成方法论，带有强烈的创始人品牌 PR 色彩。
打开原文 ↗

2026-03-31 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

  • 品味先于招聘 作者判断招聘质量首先取决于管理者是否真的做过一线、见过好坏，因为没有品味的人无法定义“好候选人”；这个判断站得住，但也意味着招聘权不该轻易下放给只会流程的人。
  • 岗位不是抽象优秀，而是场景匹配 文中区分“零容错系统的完美主义者”和“快速迭代的交付者”，这个判断非常准确，因为多数招聘失败不是人差，而是把错误类型的人放进了错误问题。
  • 斜率比截距更值得下注 作者主张优先招成长曲线陡、痴迷度高、履历非传统的人，而不是市场已经高价定价的大厂资深者；这在早期组织里通常是对的，因为创业公司买不起成熟履历，也未必需要成熟履历。
  • 共识型招聘会筛出平庸 “六个同意等于否定”的判断很锋利，本质是在反对低风险、低争议的人才选择；这对打造高密度团队有效，但前提是面试官真有识别能力，否则只会把偏见伪装成魄力。
  • 文章把反常识包装得过头了 例如“能卖烂产品的人才是真销售”这个判断有洞察，但也有危险，因为它可能混淆真实销售能力与过度承诺、误导客户的能力，这里不能不加区分地吹捧。

跟我们的关联

  • 对 ATou 意味着什么、下一步怎么用 ATou 如果在搭早期团队，不该先写一份“大而全 JD”，而该先定义当前业务最缺的唯一能力；下一步可以把每个关键岗位压缩成“一个必须有、两个加分项”的版本。
  • 对 Neta 意味着什么、下一步怎么用 Neta 如果在做研究、产品或内容筛选，这篇文章提醒你别迷信履历和头衔，而要识别“长期痴迷 + 高斜率”的信号；下一步可以建立一套非传统能力证据清单，如作品、社区声誉、持续输出。
  • 对 Uota 意味着什么、下一步怎么用 Uota 若在观察组织与个体关系，这篇文章的核心不是招聘技巧，而是“组织如何给怪才定价”；下一步可以追问：什么样的组织能消化黑羊，什么样的组织只会被黑羊撕裂。
  • 对三者共同意味着什么、下一步怎么用 三者都该警惕“品味”这个词被滥用，因为它既可能是高级判断，也可能是未经校准的主观偏好；下一步应把“品味”拆成可验证信号：作品质量、问题定义能力、冲突处理方式、学习速度。

讨论引子

1. “斜率比截距重要”在哪些岗位成立，在哪些岗位其实是危险口号？ 2. “强烈同意 hire”如何避免沦为创始人个人偏好，而不是高质量判断？ 3. 能把烂产品卖出去，究竟是销售能力强，还是组织容忍了错误激励？

Ramp 会招高中辍学生、Minecraft 模组作者、奥运选手，以及那些把客户讨厌的产品卖出去的人

Ramp 联合创始人 Eric Glyman 做过一场分享，讲 Ramp 怎么看招聘。Ramp 的规模很快就要突破 1,500 人。他们一半的支出投向工程、产品和设计。他们的 SDR 背的指标是最接近的竞争对手的 4 倍。这家公司用一条简单的测试在运转：这件事能为客户省时间或省钱吗？能，就上线；不能，就不做。

他们就是用这种方式，搭出了背后的团队。

在你招任何人之前，你得先有品味

Ramp 几乎从不招所谓的纯管理者。他们想要的是做过一线的人，卖过单、写过文案、设计过界面，而且不是偶尔做一次，是长期稳定地做过。

Eric 到现在还在写文案，还在看营销素材，还会参与设计决策。理由很简单：亲自下场做事，才会长出品味。而品味，才是把人招对的关键。

他提到东京 99 岁的寿司大师小野二郎。有人问二郎的秘诀是什么。他的回答是：要做出美味的食物，你得先吃过美味的食物。你必须训练出能分辨好坏的味觉。没有好的品味，就做不出好的食物。

不要从做寿司开始，先从品尝寿司开始。

在你开任何一个岗位之前，先检查三件事：

  1. 这个人做过这份工作吗？懂不懂这门手艺？

  2. 需求清不清楚？人和问题匹不匹配？

  3. 他有没有判断力，能不能分辨好坏？

三条里有任何一条打不上钩，就还没到该招人的时候。

不同的问题，需要不同的人

早期的 Ramp，有两类工程师画像。

第一类是完美主义者。Ramp 发信用卡。你刷卡刷不出来，这一点都不好笑。技术栈里有些部分必须极低延迟、极快响应，对错误零容忍。要的人就是那种会为此较真到极致的人。

第二类是快速交付的人。学得快，天天发，甚至按小时发。拿到客户反馈，10 分钟就修掉。他们可能有点粗糙，在对的场景里这没问题。比如收据匹配有时 0.5 秒，有时 10 秒，不算危机，只要能检测到并及时响应。

光说好还不够。要问的是：对什么场景才算好？

你做的是人的生意，不是技术的生意

你搭的团队，就是你搭的公司。尤其在早期，你手里钱不多，却要在全世界最卷的市场里竞争，也就是人才市场。

Eric 的框架来自《点球成金》。Billy Bean 用很小的预算运营奥克兰运动家队。老办法是找全面型强打者。Bean 的做法是：找一组非常具体的指标，把它们叠在一起，拼出你要的结果。他组了支看起来很怪的队，有些人打得好但接不住球，结果赢得远超所有人的预期。

别招全面型。要拼的是一组尖峰能力，让它们一起产出你想要的结果。

去哪里找被低估的人才

招外星人。 Ramp 通过杰出人才签证（O-1/H-1B）招了 20–30 个人。大多数公司不愿意花时间做担保，所以优秀的人反而竞争更小。愿意把这件事做到底，就能接触到一大池别人直接忽略的人才。

更早发现人才。 Ramp 提供冬季实习、秋季实习、春季实习。他们认识 Calvin Lee 时他 18 岁，高中辍学，在 MIT，当时作为大一新生想找一个四周的冬季实习。几乎没人会为这种人竞争。他们一直跟他保持联系，建立起真实的关系，等时机到了，说服他加入而不是去 Citadel。如果等到常规的招聘窗口，他早就会被市场定价进去了。

找黑羊。 Ramp 有位非常出色的工程师有自闭症，上的是社区大学，履历一点都不显眼。但他在 Minecraft 社区很出名，搭过复杂的私服，细节控到极致。这种能力和搭复杂数据库直接相关。非传统履历往往藏着极强的能力，要找的是那种痴迷的证据。

按品味招人。 Eric 提到 Rick Rubin 接受 Anderson Cooper 采访的片段。Cooper 问：你会操作调音台吗？Rubin 说不会。你了解音乐吗？不。那你拿钱是为了什么？Rubin 的回答是：我对自己品味的信心，以及把我感受到的东西表达出来的能力，已经被证明能帮到很多艺术家。

早期的招人，应该找有强烈观点的人，甚至有点固执也没关系。泛泛的通才很常见。有明确立场、把一门手艺做得很具体的人很稀缺。

小心用血统来招人

很多人会说某家公司销售很强，就从那儿挖人。Eric 的反驳是：要再往下挖一层。

这就像从 1990 年代的芝加哥公牛招人，和从 2010 年代的芝加哥公牛招人，名字一样，水准却完全不同。这个人是把某个职能从零搭起来的人，还是只是跟着成熟体系做过规模化运转？是不是在对的时代？

还有一种硬伤：如果产品好到能自卖自夸，比如 Figma，那里的销售到底谁真厉害很难看出来。Eric 早期最强的销售 Max，之前在 Namely 的榜单上排第一。Namely 在几年里把营收做到 7,000–8,000 万美元，但 NPS 是 -40。怎么把大家讨厌的东西卖出去？那才是真销售。把这种人带到产品很强的公司里，看会发生什么。

把筛选标准简化到极致

大多数岗位描述都会列一长串要求。每加一条要求，就会缩小候选池、抬高价格。你可能会把原本几万人的池子，硬生生筛到只剩五个人。

早期的经验法则是：只选一个你要的特质。也许两个，但多半一个就够。

当你只在找一个尖峰，就能把搜索范围铺得更广，评估得更快，也能在真正对业务重要的那一个维度上，直接对比候选人。

把岗位描述写得短到离谱。让评估一个人的表现变得容易，只看一个指标。

斜率比截距更重要

一个有 10 年经验的人，在价值曲线上起点更高。这是截距，它是已知的，也早被市场定价了。18 岁的实习生起点更低。但如果实习生的斜率很陡，做过游戏、拿过比赛、对一门手艺痴迷，那两条线很快就会交叉。

每次都按斜率招人。零缺陷不是目标。只要尖峰是真的、轨迹够陡，有点怪也没关系。

把面试设计成测不同的东西

很多公司会做六轮面试，问的都是同一类问题的不同版本。比如讲讲你的背景，讲讲你最自豪的项目。结果一半时间都浪费在测同一个音符上。

Ramp 的做法是，每一轮面试都要提取不同的信号。有一轮可能就是辩论，他们会抓住你说的一句话，直接跟你唱反调。目的不是坑你，而是看你怎么处理分歧。因为当一群意志坚定、固执、尖峰明显的人放在一起，争论是必然的。

要有明确意图：这一小时到底想拿到什么信号？

要么强烈同意，要么算了

Ramp 把候选人分成四档：强烈否定、否定、同意、强烈同意。

六个同意等于否定。想要的是有人，尤其是招聘经理，会拍桌子：我要把这个人带成功。这种笃定很重要。来自对的人那一个强烈同意，可以压过其他犹豫，因为那意味着有人愿意为这次选择押上自己的信誉。

没人拍桌子，就继续找。

关于面试的不舒服真相

花八小时面试一个人，所有会议、晚餐、复盘。等他真的入职开始干活，两个工作日之内，你对和他共事的了解就会超过整个面试流程给你的信息量。

大多数人并不擅长面试。真正的工作会很快暴露现实。

最好的策略，尤其在早期，是从推荐里招人。你或你信得过的人，跟某个人一起干过、观察过、一起做过东西很多年。这是任何面试流程都复制不了的信息优势。

规模化之后会变什么

核心抽样。 规模变大后，要深入组织。别只和直接下属做 1:1，也要去找他们的下下级。一路往下钻，看看现实是不是符合你的假设。

创始人审核。 Ramp 从 Facebook 学到这一点。很长一段时间里，Mark Zuckerberg 审核了公司前 2,000 人的每一次招聘。在 Ramp，任何人入职之前，Eric、Kareem 和 Colin 都要签字。这能让门槛保持一致，也有助于成交：创始人看过你的材料，想要你，这是很强的信号。

按职能衡量。 销售岗位看回本周期。随着规模增长，G&A 在总费用中的占比应该下降。R&D 的产出要很多年才能衡量，所以就跟踪输入指标，比如生产力和吞吐量，看看下一位招聘到底是在加速还是拖慢。

和高信念的人发生冲突时

如果招的是黑羊和创始人型的人，就得预期会吵。Eric 的看法是：辩论的目标不是赢过对方，是一起更快接近真相。铁磨铁能更锋利，但前提是自我足够低，能把自己和立场分开。

他引用 @FoundersPodcast 的一句话：容忍天才。特别厉害的人往往固执，也有点不合群。这是独特产出的成本。

不管谁是老板，最好的想法都该赢。做对了，会吸引更深度的思考者。

感谢 @eglyman 和 @tryramp 团队。这里是我对他那场分享的笔记，我把它写出来是为了让更多人能学到。如果你是创始人，或者打算做创始人，去看完整视频吧，花一小时很值得。

Ramp hires high school dropouts, Minecraft modders, Olympians, and people who sold products their customers hated

Ramp 会招高中辍学生、Minecraft 模组作者、奥运选手，以及那些把客户讨厌的产品卖出去的人

Eric Glyman, co-founder of Ramp, gave a talk on how Ramp thinks about hiring. Ramp is about to cross 1,500 people. Half their spend goes to engineering, product, and design. Their SDRs carry 4x the quota of their closest competitor. The company runs on a simple test: does this save customers time or money? If yes, ship it. If not, don't.

Ramp 联合创始人 Eric Glyman 做过一场分享，讲 Ramp 怎么看招聘。Ramp 的规模很快就要突破 1,500 人。他们一半的支出投向工程、产品和设计。他们的 SDR 背的指标是最接近的竞争对手的 4 倍。这家公司用一条简单的测试在运转：这件事能为客户省时间或省钱吗？能，就上线；不能，就不做。

Here's how they built the team behind that.

他们就是用这种方式，搭出了背后的团队。

Before You Hire Anyone, You Need Taste

在你招任何人之前，你得先有品味

Ramp almost never hires "pure managers." They want people who've done the work sold the deals, written the copy, designed the interface. Not once, but consistently.

Ramp 几乎从不招所谓的纯管理者。他们想要的是做过一线的人，卖过单、写过文案、设计过界面，而且不是偶尔做一次，是长期稳定地做过。

Eric still writes copy. Still reviews marketing collateral. Still gets involved in design decisions. The reason is simple: doing the work gives you taste. And taste is what lets you hire well.

Eric 到现在还在写文案，还在看营销素材，还会参与设计决策。理由很简单：亲自下场做事，才会长出品味。而品味，才是把人招对的关键。

He references Jiro Ono, the 99-year-old sushi master in Tokyo. Someone asks Jiro his secret. His answer: "In order to make delicious food, you need to eat delicious food. You must develop a palate capable of discerning good. Without good taste, you can't make good food."

他提到东京 99 岁的寿司大师小野二郎。有人问二郎的秘诀是什么。他的回答是：要做出美味的食物，你得先吃过美味的食物。你必须训练出能分辨好坏的味觉。没有好的品味，就做不出好的食物。

You don't start by making sushi. You start by tasting it.

不要从做寿司开始，先从品尝寿司开始。

Before you open a single role, check three things:

在你开任何一个岗位之前，先检查三件事：

  1. Has this person done the work? Do they know the craft?
  1. 这个人做过这份工作吗？懂不懂这门手艺？
  1. Is there a clear need and a person-problem match?
  1. 需求清不清楚？人和问题匹不匹配？
  1. Do they have judgment and discernment?
  1. 他有没有判断力，能不能分辨好坏？

If you can't check all three, you're not ready to hire.

三条里有任何一条打不上钩，就还没到该招人的时候。

Different Problems Need Different People

不同的问题，需要不同的人

In the early days, Ramp had two archetypes of engineer.

早期的 Ramp，有两类工程师画像。

The first: perfectionists. Ramp issues credit cards. If you swipe and it doesn't work, that's not funny. Parts of the stack need very low latency, very fast response time, zero tolerance for error. You want people who obsess over that.

第一类是完美主义者。Ramp 发信用卡。你刷卡刷不出来，这一点都不好笑。技术栈里有些部分必须极低延迟、极快响应，对错误零容忍。要的人就是那种会为此较真到极致的人。

The second: fast shippers. People who learn quickly, ship daily, sometimes hourly. Get customer feedback, fix it in 10 minutes. They can be a little sloppy that's okay in the right context. If a receipt match takes half a second sometimes and 10 seconds other times, that's not a crisis. As long as you detect it and respond.

第二类是快速交付的人。学得快，天天发，甚至按小时发。拿到客户反馈，10 分钟就修掉。他们可能有点粗糙，在对的场景里这没问题。比如收据匹配有时 0.5 秒，有时 10 秒，不算危机，只要能检测到并及时响应。

"Good" isn't enough. Ask: good for what context?

光说好还不够。要问的是：对什么场景才算好？

You're in the People Business, Not the Tech Business

你做的是人的生意，不是技术的生意

The team you build is the company you build. Especially early on, when you have very few dollars and you're competing in the world's most competitive market the market for people.

你搭的团队，就是你搭的公司。尤其在早期，你手里钱不多，却要在全世界最卷的市场里竞争，也就是人才市场。

Eric's framework comes from Moneyball. Billy Bean ran the Oakland A's on a tiny budget. The old way was hiring well-rounded sluggers. Bean's approach: find very specific stats that stack together to produce the outcome you need. He assembled a strange-looking team guys who could hit but couldn't catch and they won far more than anyone expected.

Eric 的框架来自《点球成金》。Billy Bean 用很小的预算运营奥克兰运动家队。老办法是找全面型强打者。Bean 的做法是：找一组非常具体的指标，把它们叠在一起，拼出你要的结果。他组了支看起来很怪的队，有些人打得好但接不住球，结果赢得远超所有人的预期。

Don't hire well-rounded. Assemble a set of spikes that produce the outcome you need.

别招全面型。要拼的是一组尖峰能力，让它们一起产出你想要的结果。

Where to Find Mispriced Talent

去哪里找被低估的人才

Hire aliens. Ramp has hired 20–30 people through extraordinary ability visas (O-1/H-1B). Most companies won't put in the time to sponsor. That means less competition for excellent people. If you're willing to do the work, you get access to a talent pool most people ignore.

招外星人。 Ramp 通过杰出人才签证（O-1/H-1B）招了 20–30 个人。大多数公司不愿意花时间做担保，所以优秀的人反而竞争更小。愿意把这件事做到底，就能接触到一大池别人直接忽略的人才。

Spot talent early. Ramp offers winter internships, fall internships, spring internships. They met Calvin Lee when he was 18 a high school dropout at MIT looking for a four-week winter internship as a freshman. Nobody competes for that. They kept up with him, built a real relationship, and when it came time, convinced him to join instead of going to Citadel. If they'd waited until the normal recruiting window, he'd have been priced in.

更早发现人才。 Ramp 提供冬季实习、秋季实习、春季实习。他们认识 Calvin Lee 时他 18 岁，高中辍学，在 MIT，当时作为大一新生想找一个四周的冬季实习。几乎没人会为这种人竞争。他们一直跟他保持联系，建立起真实的关系，等时机到了，说服他加入而不是去 Citadel。如果等到常规的招聘窗口，他早就会被市场定价进去了。

Look for Black Sheep. One of Ramp's extraordinary engineers has autism, went to community college. Not an obvious resume. But he was very well known in the Minecraft community - he built complex private servers that showed extreme attention to detail. That skill was directly relevant to building complex databases. Non-traditional resumes can hide extreme capability. Look for evidence of obsession.

找黑羊。 Ramp 有位非常出色的工程师有自闭症，上的是社区大学，履历一点都不显眼。但他在 Minecraft 社区很出名，搭过复杂的私服，细节控到极致。这种能力和搭复杂数据库直接相关。非传统履历往往藏着极强的能力，要找的是那种痴迷的证据。

Hire for taste. Eric references Rick Rubin's interview with Anderson Cooper. Cooper asks: "Do you know how to work a soundboard?" Rubin says no. "Do you know anything about music?" No. "What are you being paid for?" Rubin's answer: "The confidence in my taste and my ability to express what I feel has proven helpful for many artists."

按品味招人。 Eric 提到 Rick Rubin 接受 Anderson Cooper 采访的片段。Cooper 问：你会操作调音台吗？Rubin 说不会。你了解音乐吗？不。那你拿钱是为了什么？Rubin 的回答是：我对自己品味的信心，以及把我感受到的东西表达出来的能力，已经被证明能帮到很多艺术家。

Early hires should have a strong, sometimes stubborn, point of view. Generalists are common. People with particular opinions who've built a specific craft are rare.

早期的招人，应该找有强烈观点的人，甚至有点固执也没关系。泛泛的通才很常见。有明确立场、把一门手艺做得很具体的人很稀缺。

Be Careful with Lineage Hiring

小心用血统来招人

People say things like "that company is known for great sales, hire from there." Eric's pushback: dig one layer deeper.

很多人会说某家公司销售很强，就从那儿挖人。Eric 的反驳是：要再往下挖一层。

It's the difference between recruiting from the 1990s Chicago Bulls and the 2010s Chicago Bulls. Same name, very different caliber. Was this person the one who built the function, or did they just operate it at scale? Was it the right era?

这就像从 1990 年代的芝加哥公牛招人，和从 2010 年代的芝加哥公牛招人，名字一样，水准却完全不同。这个人是把某个职能从零搭起来的人，还是只是跟着成熟体系做过规模化运转？是不是在对的时代？

There's also a hard negative. If a product is so good it sells itself like Figma you can't tell who's genuinely great at sales there. Eric's best early salesperson was Max, who was top of his leaderboard at Namely. Namely had scaled to $70–80M in revenue in a few years with an NPS of -40. How do you sell something people hate? That's a real salesperson. Bring them to a company with a great product and watch what happens.

还有一种硬伤：如果产品好到能自卖自夸，比如 Figma，那里的销售到底谁真厉害很难看出来。Eric 早期最强的销售 Max，之前在 Namely 的榜单上排第一。Namely 在几年里把营收做到 7,000–8,000 万美元，但 NPS 是 -40。怎么把大家讨厌的东西卖出去？那才是真销售。把这种人带到产品很强的公司里，看会发生什么。

Simplify What You're Selecting For

把筛选标准简化到极致

Most job descriptions have a long checklist of requirements. Every new requirement narrows your pool and raises your price. You might narrow from tens of thousands of potential candidates down to five.

大多数岗位描述都会列一长串要求。每加一条要求，就会缩小候选池、抬高价格。你可能会把原本几万人的池子，硬生生筛到只剩五个人。

Early-stage heuristic: pick one thing you're looking for. Maybe two. Probably one.

早期的经验法则是：只选一个你要的特质。也许两个，但多半一个就够。

If you're looking for just one spike, you can search much more broadly, assess much more quickly, and compare candidates on the one vector that actually matters for your business.

当你只在找一个尖峰，就能把搜索范围铺得更广，评估得更快，也能在真正对业务重要的那一个维度上，直接对比候选人。

Write unreasonably short job descriptions. Make it easy to assess someone's performance on one metric.

把岗位描述写得短到离谱。让评估一个人的表现变得容易，只看一个指标。

Slope Over Intercept

斜率比截距更重要

Someone with 10 years of experience starts higher on the value curve. That's intercept it's known, and it's priced in. The 18-year-old intern starts lower. But if the intern is on a steep slope built a video game, won a competition, obsessed over a craft the lines cross fast.

一个有 10 年经验的人，在价值曲线上起点更高。这是截距，它是已知的，也早被市场定价了。18 岁的实习生起点更低。但如果实习生的斜率很陡，做过游戏、拿过比赛、对一门手艺痴迷，那两条线很快就会交叉。

Hire for slope every time. "Zero defects" is not the goal. Weirdness is fine if the spike is real and the trajectory is steep.

每次都按斜率招人。零缺陷不是目标。只要尖峰是真的、轨迹够陡，有点怪也没关系。

Design Your Interviews to Measure Different Thing

把面试设计成测不同的东西

Most companies run six interviews that all ask variations of the same question. "Tell me about your background." "Tell me about a project you're proud of." Half the time is wasted measuring the same note.

很多公司会做六轮面试，问的都是同一类问题的不同版本。比如讲讲你的背景，讲讲你最自豪的项目。结果一半时间都浪费在测同一个音符上。

At Ramp, each interview is designed to extract a different signal. One might be a debate they'll find something you say and disagree with you. The goal isn't to trick you. It's to see how you handle disagreement, because if you put a bunch of determined, stubborn, spiky people together, debates are inevitable.

Ramp 的做法是，每一轮面试都要提取不同的信号。有一轮可能就是辩论，他们会抓住你说的一句话，直接跟你唱反调。目的不是坑你，而是看你怎么处理分歧。因为当一群意志坚定、固执、尖峰明显的人放在一起，争论是必然的。

Be intentional: what signal are we trying to extract from this hour?

要有明确意图：这一小时到底想拿到什么信号？

Strong Yes or Bust

要么强烈同意，要么算了

Ramp grades candidates on four levels: strong no, no, yes, strong yes.

Ramp 把候选人分成四档：强烈否定、否定、同意、强烈同意。

Six "yes" votes is a no. You want someone especially the hiring manager to pound the table. "I'm going to make this person successful." That conviction matters. One strong yes from the right person can overwhelm other hesitations, because it means someone is putting themselves on the line.

六个同意等于否定。想要的是有人，尤其是招聘经理，会拍桌子：我要把这个人带成功。这种笃定很重要。来自对的人那一个强烈同意，可以压过其他犹豫，因为那意味着有人愿意为这次选择押上自己的信誉。

If nobody's pounding the table, keep looking.

没人拍桌子，就继续找。

The Uncomfortable Truth About Interviews

关于面试的不舒服真相

Spend eight hours interviewing a candidate. All the meetings, the dinner, the retro. Within two business days of them actually starting, you'll have more information about what it's like to work with them than your entire interview process gave you.

花八小时面试一个人，所有会议、晚餐、复盘。等他真的入职开始干活，两个工作日之内，你对和他共事的了解就会超过整个面试流程给你的信息量。

Most people aren't that good at interviews. Real work reveals reality fast.

大多数人并不擅长面试。真正的工作会很快暴露现实。

The best tactic, especially early: hire from referrals. You or someone you trust has worked with, observed, and built things with a person for years. That's asymmetric information no interview process can replicate.

最好的策略，尤其在早期，是从推荐里招人。你或你信得过的人，跟某个人一起干过、观察过、一起做过东西很多年。这是任何面试流程都复制不了的信息优势。

What Changes at Scale

规模化之后会变什么

Core sampling. As you grow, go deep into the org. Don't just do one-on-ones with direct reports. Go to their skip levels. Drill down to see if reality matches your assumptions.

核心抽样。 规模变大后，要深入组织。别只和直接下属做 1:1，也要去找他们的下下级。一路往下钻，看看现实是不是符合你的假设。

Founder review. Ramp learned this from Facebook. For a long time, Mark Zuckerberg reviewed every hire up through 2,000 people. At Ramp, before anyone gets hired, Eric, Kareem, and Colin all sign off. It keeps the bar consistent, and it helps close candidates "the founders reviewed your packet and want you here" is a strong signal.

创始人审核。 Ramp 从 Facebook 学到这一点。很长一段时间里，Mark Zuckerberg 审核了公司前 2,000 人的每一次招聘。在 Ramp，任何人入职之前，Eric、Kareem 和 Colin 都要签字。这能让门槛保持一致，也有助于成交：创始人看过你的材料，想要你，这是很强的信号。

Measure by function. For sales roles, track payback period. G&A should shrink as a percentage of expenses as you grow. R&D output takes years to measure, so track input metrics productivity and throughput to see if the next hire speeds things up or slows things down.

按职能衡量。 销售岗位看回本周期。随着规模增长，G&A 在总费用中的占比应该下降。R&D 的产出要很多年才能衡量，所以就跟踪输入指标，比如生产力和吞吐量，看看下一位招聘到底是在加速还是拖慢。

On Conflict with High Conviction People

和高信念的人发生冲突时

If you hire black sheep and founder types, expect fights. Eric's take: the goal of a debate isn't to outsmart the other person. It's to get to the truth faster together. Iron sharpens iron, but only if egos are low enough to separate yourself from your position.

如果招的是黑羊和创始人型的人，就得预期会吵。Eric 的看法是：辩论的目标不是赢过对方，是一起更快接近真相。铁磨铁能更锋利，但前提是自我足够低，能把自己和立场分开。

He quotes a line from @FoundersPodcast : "Tolerate genius." People who are particularly good are stubborn and a little out there. That's the cost of unique output.

他引用 @FoundersPodcast 的一句话：容忍天才。特别厉害的人往往固执，也有点不合群。这是独特产出的成本。

Best ideas should win, regardless of who's the boss. Done right, it attracts deeper thinkers.

不管谁是老板，最好的想法都该赢。做对了，会吸引更深度的思考者。

All credit to @eglyman and the @tryramp team. These are my notes from his talk I just wrote them up so more people could learn from it. If you're a founder or planning to be one, watch the full video. It's worth the hour!

感谢 @eglyman 和 @tryramp 团队。这里是我对他那场分享的笔记，我把它写出来是为了让更多人能学到。如果你是创始人，或者打算做创始人，去看完整视频吧，花一小时很值得。

Ramp hires high school dropouts, Minecraft modders, Olympians, and people who sold products their customers hated

Eric Glyman, co-founder of Ramp, gave a talk on how Ramp thinks about hiring. Ramp is about to cross 1,500 people. Half their spend goes to engineering, product, and design. Their SDRs carry 4x the quota of their closest competitor. The company runs on a simple test: does this save customers time or money? If yes, ship it. If not, don't.

Here's how they built the team behind that.

Before You Hire Anyone, You Need Taste

Ramp almost never hires "pure managers." They want people who've done the work sold the deals, written the copy, designed the interface. Not once, but consistently.

Eric still writes copy. Still reviews marketing collateral. Still gets involved in design decisions. The reason is simple: doing the work gives you taste. And taste is what lets you hire well.

He references Jiro Ono, the 99-year-old sushi master in Tokyo. Someone asks Jiro his secret. His answer: "In order to make delicious food, you need to eat delicious food. You must develop a palate capable of discerning good. Without good taste, you can't make good food."

You don't start by making sushi. You start by tasting it.

Before you open a single role, check three things:

  1. Has this person done the work? Do they know the craft?

  2. Is there a clear need and a person-problem match?

  3. Do they have judgment and discernment?

If you can't check all three, you're not ready to hire.

Different Problems Need Different People

In the early days, Ramp had two archetypes of engineer.

The first: perfectionists. Ramp issues credit cards. If you swipe and it doesn't work, that's not funny. Parts of the stack need very low latency, very fast response time, zero tolerance for error. You want people who obsess over that.

The second: fast shippers. People who learn quickly, ship daily, sometimes hourly. Get customer feedback, fix it in 10 minutes. They can be a little sloppy that's okay in the right context. If a receipt match takes half a second sometimes and 10 seconds other times, that's not a crisis. As long as you detect it and respond.

"Good" isn't enough. Ask: good for what context?

You're in the People Business, Not the Tech Business

The team you build is the company you build. Especially early on, when you have very few dollars and you're competing in the world's most competitive market the market for people.

Eric's framework comes from Moneyball. Billy Bean ran the Oakland A's on a tiny budget. The old way was hiring well-rounded sluggers. Bean's approach: find very specific stats that stack together to produce the outcome you need. He assembled a strange-looking team guys who could hit but couldn't catch and they won far more than anyone expected.

Don't hire well-rounded. Assemble a set of spikes that produce the outcome you need.

Where to Find Mispriced Talent

Hire aliens. Ramp has hired 20–30 people through extraordinary ability visas (O-1/H-1B). Most companies won't put in the time to sponsor. That means less competition for excellent people. If you're willing to do the work, you get access to a talent pool most people ignore.

Spot talent early. Ramp offers winter internships, fall internships, spring internships. They met Calvin Lee when he was 18 a high school dropout at MIT looking for a four-week winter internship as a freshman. Nobody competes for that. They kept up with him, built a real relationship, and when it came time, convinced him to join instead of going to Citadel. If they'd waited until the normal recruiting window, he'd have been priced in.

Look for Black Sheep. One of Ramp's extraordinary engineers has autism, went to community college. Not an obvious resume. But he was very well known in the Minecraft community - he built complex private servers that showed extreme attention to detail. That skill was directly relevant to building complex databases. Non-traditional resumes can hide extreme capability. Look for evidence of obsession.

Hire for taste. Eric references Rick Rubin's interview with Anderson Cooper. Cooper asks: "Do you know how to work a soundboard?" Rubin says no. "Do you know anything about music?" No. "What are you being paid for?" Rubin's answer: "The confidence in my taste and my ability to express what I feel has proven helpful for many artists."

Early hires should have a strong, sometimes stubborn, point of view. Generalists are common. People with particular opinions who've built a specific craft are rare.

Be Careful with Lineage Hiring

People say things like "that company is known for great sales, hire from there." Eric's pushback: dig one layer deeper.

It's the difference between recruiting from the 1990s Chicago Bulls and the 2010s Chicago Bulls. Same name, very different caliber. Was this person the one who built the function, or did they just operate it at scale? Was it the right era?

There's also a hard negative. If a product is so good it sells itself like Figma you can't tell who's genuinely great at sales there. Eric's best early salesperson was Max, who was top of his leaderboard at Namely. Namely had scaled to $70–80M in revenue in a few years with an NPS of -40. How do you sell something people hate? That's a real salesperson. Bring them to a company with a great product and watch what happens.

Simplify What You're Selecting For

Most job descriptions have a long checklist of requirements. Every new requirement narrows your pool and raises your price. You might narrow from tens of thousands of potential candidates down to five.

Early-stage heuristic: pick one thing you're looking for. Maybe two. Probably one.

If you're looking for just one spike, you can search much more broadly, assess much more quickly, and compare candidates on the one vector that actually matters for your business.

Write unreasonably short job descriptions. Make it easy to assess someone's performance on one metric.

Slope Over Intercept

Someone with 10 years of experience starts higher on the value curve. That's intercept it's known, and it's priced in. The 18-year-old intern starts lower. But if the intern is on a steep slope built a video game, won a competition, obsessed over a craft the lines cross fast.

Hire for slope every time. "Zero defects" is not the goal. Weirdness is fine if the spike is real and the trajectory is steep.

Design Your Interviews to Measure Different Thing

Most companies run six interviews that all ask variations of the same question. "Tell me about your background." "Tell me about a project you're proud of." Half the time is wasted measuring the same note.

At Ramp, each interview is designed to extract a different signal. One might be a debate they'll find something you say and disagree with you. The goal isn't to trick you. It's to see how you handle disagreement, because if you put a bunch of determined, stubborn, spiky people together, debates are inevitable.

Be intentional: what signal are we trying to extract from this hour?

Strong Yes or Bust

Ramp grades candidates on four levels: strong no, no, yes, strong yes.

Six "yes" votes is a no. You want someone especially the hiring manager to pound the table. "I'm going to make this person successful." That conviction matters. One strong yes from the right person can overwhelm other hesitations, because it means someone is putting themselves on the line.

If nobody's pounding the table, keep looking.

The Uncomfortable Truth About Interviews

Spend eight hours interviewing a candidate. All the meetings, the dinner, the retro. Within two business days of them actually starting, you'll have more information about what it's like to work with them than your entire interview process gave you.

Most people aren't that good at interviews. Real work reveals reality fast.

The best tactic, especially early: hire from referrals. You or someone you trust has worked with, observed, and built things with a person for years. That's asymmetric information no interview process can replicate.

What Changes at Scale

Core sampling. As you grow, go deep into the org. Don't just do one-on-ones with direct reports. Go to their skip levels. Drill down to see if reality matches your assumptions.

Founder review. Ramp learned this from Facebook. For a long time, Mark Zuckerberg reviewed every hire up through 2,000 people. At Ramp, before anyone gets hired, Eric, Kareem, and Colin all sign off. It keeps the bar consistent, and it helps close candidates "the founders reviewed your packet and want you here" is a strong signal.

Measure by function. For sales roles, track payback period. G&A should shrink as a percentage of expenses as you grow. R&D output takes years to measure, so track input metrics productivity and throughput to see if the next hire speeds things up or slows things down.

On Conflict with High Conviction People

If you hire black sheep and founder types, expect fights. Eric's take: the goal of a debate isn't to outsmart the other person. It's to get to the truth faster together. Iron sharpens iron, but only if egos are low enough to separate yourself from your position.

He quotes a line from @FoundersPodcast : "Tolerate genius." People who are particularly good are stubborn and a little out there. That's the cost of unique output.

Best ideas should win, regardless of who's the boss. Done right, it attracts deeper thinkers.

All credit to @eglyman and the @tryramp team. These are my notes from his talk I just wrote them up so more people could learn from it. If you're a founder or planning to be one, watch the full video. It's worth the hour!

📋 讨论归档

讨论进行中…