The "last mile" pain point of AI implementation has fueled demand…

NBD reporter Ye Xiaodan, NBD editor Bi Luming

In 2026, the FDE (Forward Deployed Engineer) has become a hot new role.

The term FDE comes from Palantir (a Silicon Valley tech company), where Forward Deployed Engineers go directly into customer sites to write code and modify processes. Palantir divides this team into two types: Echo (responsible for going on-site, managing customer relationships, and translating requirements) and Delta (responsible for turning requirements into prototypes, deployment, and iteration). The former works closely with the customer's business to identify problems, while the latter handles engineering implementation.

The FDE is also seen as the core role in the "last mile" of AI (artificial intelligence) implementation. Platform data shows that over the past two years, the number of FDE positions has surged 42-fold. On social media, many netizens call the FDE "the hottest job in the AI circle," with some FDE positions offering annual salaries as high as one million yuan.

In 2026, OpenAI invested $4 billion to establish a Deployment Company; Anthropic partnered with Blackstone to set up a joint venture, upgrading the FDE from a functional role to a strategic business unit. Domestic tech giants such as Alibaba, ByteDance, and Tencent are also stepping up FDE recruitment.

During the "2026 Yunqi Conference," Huo Jia, Vice President of Alibaba Cloud Intelligence Group, told media including the Daily Economic News (NBD) that for the long-term commercialization of FDEs, he favors two models: pay-for-results and long-term service contracts for key customers.

Why does the AI era urgently need FDEs?

What pain points exist in AI commercialization?

Huo Jia noted that AI implementation currently faces three major pain points: first, talent who understand AI and can integrate it with business is extremely scarce; second, AI implementation still lacks real, reusable best-practice cases; third, traditional IT project-style management (plan—execute—plan) cannot keep up with the technical pace of models iterating monthly or even weekly.

The surge in demand for AI implementation has brought FDEs into the spotlight. It is understood that the FDE role was first created by Palantir in 2010 to serve government and defense clients.

LinkedIn's 2026 Workforce Report shows that global FDE job postings grew 42-fold from 2023 to 2025. On social media, users new to FDE roles have shared that what makes the job novel is the chance to meet people from all kinds of industries—retail, manufacturing, e-commerce, biopharmaceuticals—and even department heads or the bosses themselves. These leaders are willing to open up about the various problems in how their companies operate, hoping to get them solved and reach an ideal state of operation.

Huaxi Securities analysts believe that the FDE is a cutting-edge concept proposed by the AI industry. In terms of its specific business activities, the FDE is a customer-facing software engineer, combining software development and systems integration, and requiring direct collaboration with customer staff and end users. Against the backdrop of continuously rising ARR for large models at home and abroad, rapidly exploding Token consumption, and fast-growing agent usage, the FDE, as one of the important vehicles for extending these models to downstream enterprise applications, is expected to grow rapidly alongside them.

Favoring two models to achieve long-term FDE commercialization

After four years of buildup and training, large AI models have finally reached the moment when their ability to drive real-world application truly gets tested.

Huo Jia regards 2026 as the first year of AI industrialization. He divides the market into four types of customers: the first is AI-native customers, including startup teams built around new model companies and even short-drama companies. The second is high-tech customers—internet, automotive, and smartphone companies—with strong in-house technical capabilities. The third is enterprise customers. The fourth is government customers.

In terms of scaled usage and Token consumption, the closer to the front, the higher the share. The first and second categories are demand-driven, while the third and fourth are value-driven. Customers in the first two categories know what they want and can use it as soon as they buy Tokens. The latter two need to be proven worthy of use first.

"The demand for FDEs essentially stems from enterprises' uncertainty about AI applications." Huaxi Securities analysts believe that customers across all industries commonly face the same challenge—how to effectively use AI technology to achieve a return on investment. Enterprise customers often have legacy systems, data silos, and security and compliance barriers, and deploying a full system in a production environment is far more complex than the model itself. The customer's focus has long since shifted from "how high the model scores on benchmarks" to "how to use it to make money." In this context, what the FDE fills is precisely the gap between "what the product can do" and "what the customer needs."

A CITIC Securities research report states that the FDE, as an end-to-end delivery mechanism from customer scenario identification to production deployment, has reached a basic consensus among major industry players such as model vendors, CSPs (cloud service providers), and software vendors, and has become the main, pragmatic way to bridge the gap between foundation model capabilities and enterprise scenarios in the AI era.

In the interview, Huo Jia also emphasized the FDE's role as a "binding agent." He said Alibaba Cloud does not make money from FDEs. The core metric for evaluating the team is whether it can drive product adoption. If an FDE cannot become a binding agent and accelerator of product capability, it will be hard to sustain.

In his view, there is no essential difference between the working model and value positioning of FDEs overseas and in China; overseas model companies also do not intend to make money from FDEs—the goal is to get AI into use. Only when the flywheel of products and usage starts turning will the flywheel of underlying data and models pay off.

On commercialization, Huo Jia emphasizes "pay-for-results." He believes that in the traditional software industry, paying by features easily falls into a vicious cycle of homogenization and price wars, because the value of features is often hard to quantify. Pay-for-results, by contrast, requires service providers to have genuine delivery capability; only when AI actually delivers cost reduction and efficiency gains for customers does the service provider gain pricing power.

In addition, whether customers are willing to sign long-term service contracts is the second test. Huo Jia believes you should not try to make big money from the service itself, but rather think about what your moat is. If it is the product, use services to drive deep product usage; if it is the service, dare to sign long-term Service Contracts rather than short-term Project Contracts.

The era of large models is giving rise to FDEs, but Huo Jia also reminds the market to maintain a clear-headed understanding of FDEs. He advocates neither losing imagination out of fear of the unknown nor ignoring business fundamentals by blindly following trends.

"Value creation is an illusory term. If you cannot solve the customer's problem, no matter how much manpower and resources you have, the customer will not pay. Being able to tell customers 'what cannot be done' is itself a huge value."

Cover image source: NBD media library

NBD Headlines

A super-strong El Niño sounds the economic alarm—how to break the deadlock as extreme weather hits agriculture, energy, and other industries and supply chains?