第一财经 2026-06-25 17:34:59 听新闻
作者：程程 责编：潘寅茹
AI extracts the key points for you—read the essentials in 10 seconds
AI quick read
节省
{{aisd}}
AI-generated Disclaimer
McCardel, who spent five years as an FDE at Palantir, said: "Extreme speed, high expectations for forward deployed actions, and intensity are not for everyone."
The Forward Deployed Engineer (FDE), a role that has existed for over a decade, is now becoming one of the hottest jobs in Silicon Valley.
Data from the online recruitment platform Indeed shows that in April 2026, the number of FDE job postings surged 729% year over year to 5,330, with Anthropic, OpenAI, Palantir, Stripe, and Google Cloud all hiring for this type of engineer. According to AlphaSense, a market intelligence and investment research search platform, the number of listed companies mentioning the role on earnings calls jumped from 8 in the previous year to 50 in 2025.
Last month, OpenAI established OpenAI Deployment with an initial investment of over $4 billion, dispatching FDEs into enterprises to provide customized AI building services, while also bringing in about 150 FDEs to the new company through the acquisition of the AI consulting firm Tomoro. Anthropic, meanwhile, partnered with Blackstone, Goldman Sachs, and others to establish an enterprise services joint venture valued at $1.5 billion, likewise adopting the FDE on-site model to drive deep deployment of the Claude model.
The role also comes with generous compensation. A 2026 survey of 1,500 FDEs by Perspective AI showed that the median total annual compensation for senior FDEs at frontier labs reached $485,000, rising to as high as $725,000 at the senior staff level, with total compensation ranges at top labs falling between $350,000 and $550,000.
What signals does Silicon Valley's massive demand for FDEs send? In response to this question, Tan Yinliang, a professor of decision sciences and management information systems at the China Europe International Business School (CEIBS) who focuses primarily on the economics and governance of artificial intelligence, told Yicai that this trend indicates the focus of competition among large model companies is shifting from "whose model is stronger" to "who can turn models into enterprise productivity systems." Technological leadership still matters, but it is becoming an entry ticket; what will truly determine the outcome of the second half is enterprise sales capability, industry solutions capability, engineering delivery capability, ecosystem partnership capability, and the ability to continuously demonstrate AI return on investment.
What do FDEs mainly do?
FDE is not an entirely new position; the American big data analytics and AI platform company Palantir pioneered this model. Wall Street analysts view FDEs as Palantir's "secret weapon." Dan Ives, managing director and senior analyst at Wedbush Securities, stated in a report sent to Yicai that the company's core value lies in the specialized capabilities provided by FDEs—working alongside customers in the field, building products in real time, while observing customer pain points and seeing changes immediately.
Most software companies follow a process of gathering customer requirements, designing solutions, building products, and deploying at scale, but this approach hit obstacles with the defense and intelligence agency clients Palantir sought to win. These institutional clients not only struggled to clearly articulate their requirements but also could not openly share their data, and even for similar counterterrorism projects, workflows differed greatly between agencies.
"We found that what each customer wanted was different. Rather than making a separate version for each one, or adding a bunch of features that only applied to a specific customer, it was better to make the product a highly customizable platform. That meant we had to send employees on-site to deeply understand users and make local adaptations." Bob McGrew, an early Palantir executive and now OpenAI's chief research officer, revealed in a media interview last September that the company's chief technology officer, Sham Sankar, invented the FDE strategy: having FDEs enter with the existing product to fill the gap between product capabilities and actual needs, first paving a "gravel road," while the backend product and engineering teams were responsible for rolling it out to 5 to 10 customers, turning that road into a "paved superhighway."
Internally, Palantir divides FDEs into two key roles: Echo teams and Delta teams. Echo teams are on-site analysts responsible for understanding customer needs, identifying which existing product features apply and which can be improved, while also managing customer relationships. The ideal candidates for these roles need domain expertise—for example, former Army officers—along with a rebellious spirit and the ability to identify shortcomings in existing practices. The other team is called Delta, composed of software engineers whose core capability is building "prototypes"—delivering in software form within extremely short timeframes rather than pursuing absolute code correctness. By 2016, the number of FDEs Palantir employed had already exceeded that of traditional software engineers.
When asked about the difference between the standard SaaS (software as a service) model and the FDE model, McGrew said the former favors simple, repeatable contracts and pricing, and because the marginal cost of deployment is very low, SaaS teams are happy with small contracts. But under the FDE model, product requirements gradually become more complex, and contract sizes grow accordingly.
However, some have raised new voices. Marc Andreessen, founder of the venture capital firm a16z, described the invention of this role in a blog post as "title arbitrage"—creating a new job title to name a new force emerging within institutions and a new capability under development. He argued that what Palantir is actually betting on is that "deep software integration with large enterprises and governments will be very important in the future. This work deserves a corresponding elevation in visibility."
Barry McCardel worked as an FDE at Palantir for about five years before founding the AI analytics platform Hex. In his view, it is not a glamorous job. "Doing this work means spending a lot of time on airplanes, sleeping in three-star hotels somewhere in the middle of America, and working in a dim, windowless conference room without enough power outlets," he told the media. The work itself is also extremely challenging: limited resources, small teams, yet the need to solve an unprecedented problem under intense time pressure. McCardel added that customers often don't end up using the projects they spent time developing. "Extreme speed, high expectations for forward deployed actions, and intensity are not for everyone," he said.
What signals does this send?
Notably, the FDE, which has existed for over a decade, is not a new role, but the timing of its explosion in Silicon Valley closely overlaps with the period from the birth of AI agents to their commercialization. What signals does this change send?
In terms of market demand: enterprises are seeking AI transformation, but purchasing AI does not equal completing AI transformation. The real difficulty lies in integrating AI into CRM, ERP, data warehouses, knowledge bases, permission systems, and approval workflows, and getting it actually used within the organization. When Thomas Kurian, CEO of Google Cloud, posted a job listing for FDEs last month, he said the role was nothing new for Google Cloud, but demand from customers and partners for Google's enterprise AI products and for Google engineers to help them embrace agent development is growing rapidly.
Meanwhile, large model companies are also targeting enterprise deployment. Almost simultaneously with the mass hiring of FDEs, OpenAI and Anthropic have been poaching talent from the software giant Salesforce. According to profiles on the professional social platform LinkedIn, nearly a hundred Salesforce employees have moved to these two AI giants so far this year, mainly in sales and marketing roles. Analysts say the two giants value these employees' deep connections with Fortune 500 companies and expect these seasoned professionals to help enterprises accelerate the deployment of their products.
Tan Yinliang believes the popularity of the FDE role shows that AI commercialization is undergoing a shift from "selling model capabilities" to "selling business outcomes." In terms of business models, as models gradually become infrastructure, simply selling underlying API calls easily devolves into a price war. If AI agents can be embedded into enterprise processes and directly improve sales, customer service, R&D, risk control, or operational efficiency, it becomes easier to build sustainable annual recurring revenue (ARR).
"Looking further ahead, AI companies are transforming from model companies into a new generation of enterprise software companies. The key to future competition is not just model strength, but who possesses industry knowledge, customer relationships, sales networks, and delivery capabilities to turn AI from a demo into an enterprise production system," he said.
Report
For advertising cooperation with Yicai, please click here
This content is original work by Yicai (China Business News), and the copyright belongs to Yicai. Without written authorization from Yicai, it may not be used in any way, including reproduction, excerpting, copying, or mirroring. Yicai reserves the right to pursue legal liability against infringers.
如需获得授权请联系第一财经版权部：
- 程程
相关阅读
AI安全正在从实验室内部的技术问题，转变为一个同时涉及产业竞争、资本市场等多维角度的问题。
排名前十的城市中“中美”两国各占三席
这种技术潜力能否转化为持续增长，取决于竞争政策、教育体系和劳动力市场制度能否同步跟上。
AI进化速递丨人形机器人百米竞速再次刷新纪录
第二届世界人形机器人运动会今天开幕，新增拔河、乒乓球等项目。
Yicai's Hottest