刘宁Ning Liu
← 文章← Articles

AI落地缺的是什么?Anthropic投1亿美元培训FDE,只答了一半What Is AI Adoption Missing? Anthropic's $100 Million FDE Programme Answers Only Half

2026-10-16

10 月 2 日,Anthropic 宣布投 1 亿美元,计划到 2027 年底培训 1 万名 FDE。

FDE 一般认为最早来自 Palantir,原文是 Forward Deployed Engineer,国内多译作「前沿部署工程师」;Anthropic 这次用的是 Frontier Deployed Engineer,缩写一样。这两年,OpenAI、Anthropic 这些模型公司,几家大的云厂商,国内的腾讯、阿里云,都在招这个岗位。

这次公告有两点值得注意。一是 Anthropic 给的理由:企业用 AI 的速度很快,但能让它在真实业务里跑起来的人,成了最难找的人才。二是学员不是 Anthropic 自己的员工,是各家公司提名的工程师,集中培训几天后,回到本公司,用 12 周牵头一个真实的项目。首批里有麦肯锡、德勤这样的咨询公司,也有摩根士丹利、诺和诺德这样的企业。

FDE 做的是什么

简单说,就是派到客户那里,把 AI 用进客户的业务。先摸清客户实际怎么干活、数据怎么流,找出 AI 能接手的环节;再动手写代码,把模型接进客户现有的系统;上线前做评测,上线后接着盯,出了问题负责修;最后把客户的共性需求带回自己公司的产品团队。

为什么现在这么缺

模型越来越强,也越来越容易买到。但把它放进一家公司的业务里,难处大多不在模型本身:业务方说不清要什么,技术团队不了解业务,数据散在各个系统里,做出来以后也没人负责持续改进。

我一直说,传统企业做 AI,七分在业务和组织,三分在技术。FDE 这么火,说明模型公司自己也看到了这一点。

厂商派来的 FDE,和企业自己的人

现在大多数 FDE 是厂商派来的。他们懂模型,也写得了代码,很有用。但他们是厂商的人,主要任务是让自家产品在客户那里用起来、用得更多。也有评论文章直接指出,一些云厂商的 FDE 项目本质上是销售项目,帮客户的同时,也把客户绑在自己的平台上。

所以,Anthropic 说缺的是能让 AI 在真实业务里跑起来的人,这只答了一半。另一半在企业自己这边:还得有一个拿主意的人。我在之前的文章里把这份职责叫 AI BP:像 HRBP、财务 BP 一样挂在业务旁边,把业务的需求翻译成 AI 能接的任务,再把 AI 能做什么讲回给业务。Palantir 自己也是两种人配合,写代码的 FDE,和偏产品、偏业务的部署策略师。AI BP 更接近后者,不一定写代码。

它跟 FDE 干的是同一类活,区别在站在哪一边:厂商派来的 FDE,管的是把它的产品用起来;企业自己的 AI BP,管的是这件事该不该做、用谁家的、做到什么算好,以及厂商走了以后谁接着管。

对传统企业来说

Anthropic 这次培训的是各家公司自己的工程师,要求有扎实的软件工程功底。多数传统企业未必有这样的人,也不用急,工程能力可以从外面找,厂商的 FDE 也可以用。

不能缺的是企业自己这一头。这个人可以由内部的业务骨干兼着,也可以先请一位外部的人帮着带一段,关键是他站在公司这边。对多数传统企业来说,先定下这一侧由谁负责,比先找厂商要一个 FDE 更要紧。

On 2 October, Anthropic announced that it would invest $100 million to train 10,000 FDEs by the end of 2027.

FDE is generally thought to have started at Palantir, where it stands for Forward Deployed Engineer. Anthropic uses Frontier Deployed Engineer, with the same abbreviation. Over the past two years, model companies such as OpenAI and Anthropic, several major cloud providers and, in China, Tencent and Alibaba Cloud have all been hiring for the role.

Two things about the announcement stand out. The first is the reason Anthropic gave: companies are adopting AI quickly, but the people who can make it work in real business have become the hardest talent to find. The second is that the trainees are not Anthropic employees. They are engineers nominated by their own companies, who train together for a few days and then go back to lead a real project at their own company for 12 weeks. The first cohort includes consulting firms such as McKinsey and Deloitte, and companies such as Morgan Stanley and Novo Nordisk.

What an FDE does

Put simply, an FDE is sent to a customer to bring AI into the customer's business. They first learn how the customer actually works and how its data flows, and find the steps AI can take over. Then they write code to connect the model to the customer's existing systems. They run evaluations before launch, keep watching after launch and fix problems when they come up. Finally, they take the needs that customers have in common back to their own company's product team.

Why they are so scarce now

Models keep getting stronger and easier to buy. But when you put one into a company's business, most of the difficulty is not in the model itself: the business side cannot say clearly what it wants, the technical team does not understand the business, data is scattered across different systems, and once something is built, no one is responsible for improving it.

I have long said that when traditional companies do AI, it is seventy percent business and organisation and thirty percent technology. The FDE boom shows that the model companies have seen this too.

The vendor's FDE and the company's own person

Most FDEs today are sent by vendors. They understand the models, they can write code, and they are very useful. But they work for the vendor, and their main job is to get the vendor's product used, and used more, at the customer. One commentary has said outright that some cloud providers' FDE programmes are essentially sales programmes: while they help customers, they also tie those customers to their own platforms.

So Anthropic says what is missing are people who can make AI work in real business, and that is only half the answer. The other half lies inside the company: someone also has to make the call. In an earlier article I called this role the AI BP. Like an HR or finance business partner, it sits alongside the business, turns business needs into tasks AI can take on, and explains back to the business what AI can do. Palantir itself pairs two kinds of people: FDEs who write code, and deployment strategists who lean towards product and business. The AI BP is closer to the latter and does not necessarily write code.

It does the same kind of work as an FDE. The difference is which side it stands on. A vendor's FDE is responsible for getting its product used. A company's own AI BP is responsible for whether something should be done at all, whose product to use, what counts as good enough, and who carries on after the vendor leaves.

For traditional companies

What Anthropic is training are companies' own engineers, and it requires solid software engineering skills. Most traditional companies may not have such people, but there is no need to rush: engineering capability can be found outside, and vendors' FDEs can be used too.

What a company cannot do without is its own side of this. The role can be taken on by an experienced person in the business alongside their existing job, or an outside person can be brought in to lead it for a while. What matters is that they stand on the company's side. For most traditional companies, deciding who is responsible on their own side matters more than asking a vendor for an FDE first.

刘宁 Ning Liu|帮传统企业把 AI 跑通的人。工程师出身:浙大计算机本科、中科大硕士,在微软、MicroStrategy 做过工程师,2018 年起带队做企业 AI 落地,主导过集团级 AI 中台。技术之外,还有两段经历:长江商学院 MBA(含密歇根 Ross 海外交换),能把技术和公司战略对齐;创过业,又在教育、地产等传统行业做了 8 年 CTO,从技术配合业务,做到技术带着业务走。现为企业提供 AI 落地诊断、战略规划、项目落地 Sprint(带企业团队一起交付)与 Fractional AI Lead(长期在位的 AI 负责人)服务。聊聊你的 AI 落地 →
Ning Liu — Your Fractional CTO for AI. An engineer by background (Zhejiang University, USTC; Microsoft, MicroStrategy), group AI-platform lead, hands-on with enterprise AI since 2018. Beyond the engineering, two things: an MBA from CKGSB (with an exchange semester at Michigan Ross), so I can align technology with company strategy; and I founded a company, then spent 8 years as a CTO in traditional industries such as education and real estate — moving from tech that supports the business to tech that leads it. I offer AI Audit, Strategy, Delivery Sprint and Fractional AI Lead. Let's talk about your AI →