With monthly salaries as high as 80,000 yuan being offered for new hires, a cohort of FDEs are still unable to secure any orders | Market Trend
"Career Bonus" (ID: ZhiChangHongLi)
Written by / Yutian Jiang Cover Image / AI Generated
Art Editor / Yutian Jiang Text Editor / Tong Chen
The Other Side of High Salary · 01
You Cannot Become a Qualified FDE Without Communication Skills · 02
Who Is Suitable to Enter This Industry? · 03
The Coexistence of Sense of Achievement and Anxiety · 04
When clients clock off from work, it is the exact time point for full-time FDE ZHAO Dongxu to switch his work content.
During the daytime, he follows the client's meetings, sitting in whatever empty meeting room is available: communicating requirements, submitting revised versions, collecting feedback, one meeting after another. In the evening, employees leave one after another, but the lights in the empty meeting rooms are still on. He stays behind to write programs, using code to automatically classify a large amount of data and label them — for example, a user message should be categorized as "price consultation", "complaint" or "refund". The rules and hierarchy of this classification are exactly the label system on which the client's AI middle platform relies to operate. The next morning, he appears in the next meeting with a new version. The system must be finalized within a few days.
ZHAO Dongxu has been engaged in back-end development for more than ten years, and has served as a technical lead and architect. Half a year ago, he left a large tech factory in Beijing and became an independent FDE.
FDE (Forward Deployed Engineer) refers to a compound engineer who is permanently stationed at the client's site, understands both technology and business, and implements AI capabilities into usable production systems. Nowadays, the popularity of this position is being driven up by several forces at the same time:
On October 2 this year, Anthropic pledged to invest 100 million US dollars to train 10,000 FDEs by the end of 2027. Earlier on June 30, AWS announced the investment of 1 billion US dollars to establish the FDE (AWS Forward Deployed Engineering organization) department, planning to allocate thousands of engineers in the initial stage. In August, China included FDE in the document of national ministries and commissions for the first time. The Ministry of Industry and Information Technology issued the Notice on Launching the Special Action for Cultivating Artificial Intelligence Application Service Providers, clearly encouraging service providers to build teams of forward deployed engineers.
Data from Maimai shows that the number of FDE-related job postings in the first half of 2026 has increased 21 times year on year, and one in every three companies recruiting FDEs is an AI-native enterprise.
According to comprehensive information, the salary of open FDE positions in leading large tech factories and AI companies is generally 20,000 to 80,000 RMB per month: ByteDance offers a salary of 35,000~70,000 RMB * 15 months, with a maximum annual salary of 1.05 million RMB; Alibaba Cloud Intelligence offers 20,000~50,000 RMB * 16 months; Ant Digital Technology offers 40,000~60,000 RMB per month for B-end positions; the FDE lead at Zhipu AI offers 60,000~80,000 RMB per month.
On the other side, when Ren Yuhang, an 8-year veteran of large tech factories who has left his job to run an FDE incubation platform, was asked "the most difficult problem to overcome in FDE's daily work", he told *Career Bonus*: "I feel the most direct pain is having no orders."
The Other Side of High Salary
"From the market perspective, the demand for AI implementation in all walks of life does exist," said Luke, an independent FDE practitioner based in Shenzhen. However, the high salary offered by large tech factories cannot represent the overall salary level of the FDE market. The salary of leading companies is significantly higher than that of small and medium-sized companies, and there are obvious gaps between different industries and regions, which even depend on the individual's delivery capability.
Last year, ZHAO Dongxu left a large tech factory in Beijing and became an FDE taking orders independently, undertaking projects for small and medium-sized enterprises. According to his observation, the prices offered by small and medium-sized enterprises are generally not high: the remuneration for some projects is less than 10,000 to 20,000 RMB, slightly better ones can be close to 100,000 RMB, and projects offering hundreds of thousands of RMB are already large projects.
FDE Wayne, who used to work in North America, told *Career Bonus* that 200,000 US dollars is the basic salary level in foreign countries; very few people in the domestic market can get a salary of more than 1 million RMB. He even believes that "offering 30,000 to 50,000 RMB per month to an FDE may be reluctant for enterprises, but this amount of money is actually an insult to the FDE".
It is precisely because the professional portrait is not clear that there is a large deviation in salary anchoring. One of the reasons is that the portrait of this profession itself is not clear — a large part of domestic FDEs are not the FDEs defined in advance in Silicon Valley. In this regard, Wayne's definition is: "If the main work is still making demos, explaining solutions, cooperating with sales, and writing bidding documents, then it cannot be called an FDE in the AI era."
In addition, to judge whether a person is an FDE, we can look at the assessment indicators. For example, OpenAI's assessment of FDE is evaluated based on actual adoption rate, workflow impact and measurable business results; AWS has also made it clear that the goal of FDE is to enable clients to have independent operation capabilities after project delivery.
At first glance, FDE is easily understood as a "technical product manager". In fact, the work process is far more complicated than that: complete solution communication and architecture design at the client's site, develop rapid proof of concept, then deploy and go online, and finally precipitate the on-site experience to feed back to the own product platform.
The FDE model originated from Palantir, an American big data company founded in 2003, which initially served large clients such as military and aerospace on site — deployment engineers stationed in the camp with toolboxes to solve problems. The company also has a highly complete data processing platform and mature ontology system. The service objects have rich payment returns, large order scales and small demand changes, so FDEs have the conditions to precipitate in vertical fields for many years.
The situation in China is different. Practitioners told *Career Bonus* that there is no domestic ontology modeling, nor a unified data processing platform to solve industry problems; the demand for domestic FDEs emerged after the development of AI Agent, which is mainly responsible for handling clients' AI demands.
There are essential differences in professional demands and organizational support between domestic and foreign FDEs, and the salary gap is a manifestation of this.
You Cannot Become a Qualified FDE Without Communication Skills
Jove is in charge of the FDE team of an AI company. When interviewing candidates, he will ask them to record a video introducing the AI Agent they have developed.
The purpose of the video introduction is to examine the candidate's expression and communication skills. Because this is almost the prerequisite guarantee for work progress, if someone is needed to translate every communication, the collaboration cost will be very high. Wayne believes that, "without good communication, if the preliminary research and judgment are wrong, the faster you do later, the faster you will make mistakes."
In the process of engineering promotion, information deviation will be amplified by the workflow. In ordinary projects, information deviation at most leads to rework after communication, but in AI projects, Agent performs chained tasks, and the deviation of each layer is superimposed continuously, and the final output version may have completely deviated from the direction.
When the Lingyang FDE team undertook the Yadea project, they invested nearly one month in the early stage to carry out cross-departmental interviews, conducting 78 interviews, with an average of more than 3 interviews per day, covering multiple departments such as customer service, stores, after-sales, product, quality and IT. In the process, handling an after-sales problem requires cross-operation of more than a dozen systems. They first sorted out the process, repeatedly listened to the hotline recordings, restored the work process together with customer service staff, and sorted out the problems handled by senior employees based on experience one by one. On this basis, they integrated the call handling, knowledge base and work order into the same workbench, opened up the data call of more than a dozen systems in the background, and finally introduced robots to undertake high-frequency repetitive problems.
Therefore, in the daily work of FDE, communication is a very important part.
Zeneth, who is engaged in AI venture capital in Hong Kong, shares the same feeling. She helps enterprises connect FDE implementation. The company receives a lot of demands every day, and enterprises hope to arrange FDE to be stationed on site, but when it comes to specific goals, "the employer actually can't clearly state the demand, sometimes there is only one sentence 'we need AI Agent'."
The on-site FDE of her team once undertook a semiconductor factory project. Due to strict data confidentiality requirements, remote operation is not allowed, so colleagues need to be stationed directly in the factory, build data pipelines, RAG architecture and Agent orchestration in the client's internal network, and continuously communicate with colleagues in the production link.
When Wayne undertook small and medium-sized enterprise projects, he initially thought that such communication could be processed in a standardized process. When he sat next to front-line employees to observe, he found that the actual operation was very complicated: sometimes there is no relevant information in the materials provided by the client, FDE needs to continuously communicate with various business employees to solve stuck points, and sometimes collect unwritten rules, after many modifications and debugging, the AI capability can run, and ensure that the design and decision-making meet the requirements of the employer.
"Who the reporting object is is very important," ZHAO Dongxu emphasized. If the docking person does not understand the business well and the demand docking is vague, the project will fall into repeated modifications. In his recent project, the label system alone was revised three times. Each revision requires reprocessing a large amount of data through the model, and the Token cost alone cost tens of thousands of RMB. This sum of money is usually borne independently by the FDE.
He and his peers in the industry community OpenFDE unanimously believe that independent FDEs prefer to take projects that can report directly to CTO or CEO, "otherwise you need to explain repeatedly, the information gap is relatively large, and the communication cost is very high".
Who Is Suitable to Enter This Industry?
"People who are too obsessed with technology or only complete tasks according to Tickets will tire both themselves and others as FDEs," Wayne said.
FDE is known as the "Hexagon Warrior". As the head of an FDE department, Jove believes that "let alone a hexagon, even a icosagon is not enough". Because FDE implementation needs to be adjusted according to the actual situation on site, the demands and processes of projects in different industries are different, and the accumulation of complete project experience is the key to entering this industry.
When recruiting, Jove basically does not recruit candidates with less than three years of work experience — experience represents accumulation, and companies rarely hire people who are new to the industry, as the risk is too high; he requires candidates to be able to do Agent Loop and eval. Years of experience are not the first criterion, he evaluates candidates based on their accumulated effective project experience, so job seekers need to come with AI Agent projects. If candidates show sufficient strong potential — outstanding learning ability, enthusiasm and communication ability, he will make an exception for consideration even if they do not have much customer-facing experience.
He also prefers people with consulting and entrepreneurial backgrounds, so he has recruited many employees from McKinsey. "They have rich experience, communication skills, and are good at promoting the process in a high-paced and long-term working state. Entrepreneurs know that every order is hard-won, and they can't wait for work to be assigned."
However, if the consulting background is too strong and the engineering implementation ability is weak, it is difficult to become an FDE. First, they lack the practical ability to build production-level systems; second, they are easy to become shallow senior implementers. There is a deviation in the work thinking between the two. The consulting habit takes the solution report as the delivery end point and tends to comply with client demands, while FDE requires promoting the system to go online and create business value, and dares to put forward different judgments in the absence of administrative authorization.
As for evaluating whether you are suitable for this industry, Wayne believes that you can look at work preferences: some people prefer to deal with uncertain problems, are willing to contact clients, can do it by themselves, and are willing to take responsibility for the results; some people hope to define the boundaries such as requirements, interfaces and acceptance standards clearly before starting to work. The former can try FDE, while the latter is more suitable for pure R&D.
AI Coding has lowered a large number of work thresholds, and non-professionals do have more possibilities to enter this industry.
But the threshold of engineering capability has not disappeared. Wayne believes that FDE can transfer from backgrounds such as business, implementation and consulting, but practitioners must have certain engineering capabilities, at least be able to understand systems, data, APIs, permissions, deployment and production problems. AI Coding can greatly improve coding efficiency, and writing code is just the simplest part — there are many practical difficulties when arriving at the client's site. For example, FDE has to face problems such as data cannot be imported, interfaces cannot be aligned, permissions are difficult to open, and model output is unstable, which are difficult to solve by generating a few pieces of code, which requires engineering judgment, debugging experience and understanding of the whole system behind.
In addition, FDE also needs to customize solutions according to specific situations. Some companies' label systems rely on large model generation, which must be confirmed repeatedly on site; some clients have strict requirements for data security, and the model must be deployed on local servers, with data not leaving the internal network.
ZHAO Dongxu provided an evaluation method for his own ability: find a real scenario, use AI Coding to complete the prototype, and then go through the whole process by yourself. If the closed loop of deployment, integration, evaluation and iteration can be completed completely, you are qualified to enter the industry; if you can't complete it, it means that what you lack is not the concept, but the engineering capability.
When AI manufacturers and large tech factories recruit FDEs, they mainly follow the standards of engineers. The requirements are production-level engineering capabilities, demand disassembly and on-site practice capabilities, and data permission and audit capabilities for highly regulated industries. The screening focuses on candidates with client site delivery experience and real project review records; those who have not run production and have poor client communication skills will be eliminated.
The second type of recruiters are major demand-side enterprises. FDEs need to judge the enterprise's demand gap, and then configure evidence according to the industry. The logic of different industries is completely different, for example, software focuses on integration and replacement cost, manufacturing focuses on parameter working conditions, and finance focuses on compliance boundaries and continuous operation. Therefore, people who are able to submit acceptance, use result indicators and deliver clearly will stay.
The Coexistence of Sense of Achievement and Anxiety
Luke, former solution architect of a large tech factory, is currently engaged in independent FDE in Shenzhen. He believes that the sense of achievement brought by following projects to the site, understanding different industries, solving difficult problems and delivering projects far exceeds the production in the office.
Ren Yuhang, who left his job to run an FDE platform, believes that FDE capability equals workplace premium. "With FDE capability, an individual's time efficiency and results will be better than others." For people who want to work independently, the attraction lies in vision and opportunities. "When docking with clients, you will understand their market demands and business, your vision is broader, and you have more opportunities."