Big tech companies are offering annual salaries of millions to recruit talents. What on earth is FDE?

A rather abstract-sounding position is becoming one of the key focuses for large companies to invest in recruitment in 2026.

ByteDance offers a monthly salary ranging from 35,000 to 70,000 yuan, with a 15 - month annual salary, and the maximum annual salary is expected to reach 1.05 million yuan. Alibaba Cloud Intelligence provides a monthly salary of 20,000 - 50,000 yuan and a 16 - month annual salary plan.

These positions are for Forward Deployment Engineers (abbreviated as FDE). In recent months, from overseas leading large - model companies such as OpenAI and Anthropic to domestic large companies like ByteDance, Alibaba, and Tencent, technology companies have been adding FDE positions one after another. Stories like "earning 100,000 yuan a month by taking individual orders" and "switching to a million - yuan - annual - salary job from scratch" are emerging endlessly on social platforms. Technical professionals, fresh graduates, and traditional consulting practitioners are all flocking to this field.

However, controversies have always accompanied it: Is this really a brand - new golden career born in the AI era, or just a concept hype of "new wine in old bottles"? Does its core value lie in code or business?

Not a new profession

What exactly does an FDE do?

When the capabilities of models are constantly evolving, the real challenge lies in truly connecting the scattered needs of traditional enterprises with the capabilities of models. Bridging this gap is what FDEs do.

"A large part of the popularity of the term FDE in 2026 is due to concept arbitrage." The judgment of Zaniel, a doctoral student at Stanford University in the United States and an independent consultant, hits the most core controversy in the industry. In the view of many industry insiders, this position, which is packaged as a brand - new profession, is essentially the same as the past solution architects, implementation consultants, and delivery managers. It just has a trendier name riding on the wave of AI.

Tracing back to its origin, FDE is not a brand - new concept that emerged out of nowhere. This position was first popularized on a large scale by Palantir in the United States. Its core model is to dispatch engineers to the sites of government and large - enterprise customers to embed standardized data analysis products into the customers' existing systems and workflows.

Jolie Ni, a technology worker in Silicon Valley, witnessed the evolution of FDE for more than half a year on the front line and then boldly started her own business, establishing an independent FDE service provider called Hconsult.ai. Her target customers are small and medium - sized enterprises with an annual revenue of 5 million to 10 million US dollars. These customers have budgets and needs, but they can neither attract the above - mentioned large enterprises to provide services nor afford to build their own FDE teams. Her startup has formed a differentiated competitive situation with the above - mentioned enterprises.

Similar to the situation in the United States, a dual - pattern has also formed in the domestic market. On one end is the camp of large - company FDEs: The FDE teams of large - model manufacturers serve Fortune 500 customers. The FDEs of Feishu and DingTalk enter medium - and large - sized enterprises in the form of a "sales + technology" duo and carry out customized deployments based on their own platforms.

Through the "HA7CH" community he organized, Lawted learned that a large company in Hangzhou has started implementing a system where "newcomers first work as FDEs for three months". They even directly dispatch product managers to the site for demand research and Demo development. FDE is changing from an independent position to a standard ability of the TOB team.

On the other end are the independent service providers that Lawted calls "local FDEs". Most of them are single - person studios or small teams, deeply involved in small and medium - sized enterprises in the sinking market. They are highly flexible and low - cost, filling the gaps that large - company services cannot cover.

The ambiguity in the position definition is the source of the concept controversy. Many people can't tell the difference between FDEs, outsourcers, AI engineers, and AI product managers, and real projects can best illustrate the boundaries.

Zaniel once received a demand from an enterprise that clearly stated they wanted to launch an AI customer service system. After in - depth on - site research, he found that the real pain point of the enterprise was that the customer data of multiple business systems was completely unconnected, and the low efficiency of customer service was just a surface symptom.

If they directly developed an AI customer service system according to the outsourcing logic, they would have a deliverable, but the core problem would not be solved at all. This is the core difference between FDEs and outsourcers: Outsourcers are only responsible for the clearly defined deliverables, and the requirements are defined by the customers; FDEs need to penetrate the surface first and find the real problems worth solving.

The difference from AI engineers is also clear. In the leading fast - moving consumer goods enterprise where Ram works, the AI engineer position has existed for many years, and their core work is to replicate mature algorithms and solutions in the industry; however, FDEs often face vague demands like "I want to use AI to reduce costs and increase efficiency" and need to break down these vague ideas into implementable technical paths.

In Ram's view, the core of FDE lies in "Forward" - taking a step forward and immersing in the front - line business instead of waiting for demands in the background.

Jolie Ni explained that AI product managers focus on a single product and continuously iterate and optimize it; however, FDEs need to solve problems across industries and scenarios, and are more like "consultants who can implement solutions". They not only need to diagnose problems but also implement the solutions themselves.

There is a consensus among everyone that the core ability of an FDE has never been writing code.

Lawted gives a ratio of "60% communication, 40% technology". This conclusion was confirmed in Zaniel's financial institution report project. He automated the report generation for a local financial institution. The most time - consuming and valuable part was not writing the automation script, but spending two weeks extracting hundreds of business rules that had never been documented from the experience of old employees. Any person who knows Python can write the automation script, but those rules hidden in people's minds can only be obtained by staying at the business site and carefully extracting them one by one.

This also means that FDE is not a brand - new profession that emerged out of nowhere. AI tools have significantly lowered the development threshold. Projects that used to require a team to complete can now be implemented by a single person. The economic value of this role has been instantly magnified, and it has come to the forefront with the concept boom. Stripping off the filter of the "brand - new golden career", it is essentially the "last mile" of AI implementation, a translator connecting technology and business.

Where is the value?

Stripping off the halo of the concept and entering the real business scenario, the value of FDE is truly revealed.

In real scenarios, the most common demand is to achieve real cost reduction and efficiency improvement, which is also the value point that small and medium - sized enterprises are most willing to pay for.

Jolie Ni once automated the customer acquisition workflow for a South Korean GPU computing power company. Previously, the company's employees had to manually organize the information of professors from the top 500 universities in the QS ranking, match the schedules of academic conferences, and write personalized business development emails. A skilled employee could send at most 10 effective emails a day.

After implementing the automation solution, the system automatically grabs conference information and scholars' research trends through APIs, uses large models to match corresponding cases and generate customized content, and can stably output 200 to 500 emails a day. The response rate has not decreased due to automation.

The same is true for the local financial institution served by Zaniel: Multiple business departments need to generate dozens of daily reports every day, all relying on manual data extraction from the core system and filling in Excel. After the automation solution was implemented, hundreds of implicit business rules were all solidified into the system, and employees were completely freed from mechanical and repetitive labor.

In the final implementation, the value of FDE may go far beyond saving money. It also helps enterprises build long - term market competitiveness.

Corresponding to different value demands, two mainstream payment models have emerged in the industry.

Jolie Ni adopts a combination of "project - based system + monthly maintenance fee": The main project is charged in a one - time package, and subsequent process iterations and daily operations and maintenance are charged a service fee on a monthly basis. This model takes into account both one - time income and long - term renewals and is suitable for customers with clear demands and long - term operation and maintenance needs.

The other model is payment based on results.

Lawted introduced that many independent FDEs settle accounts based on the number of effective leads for customer acquisition projects, share the cost savings in manpower for cost - reduction projects, and charge based on the number of effectively processed orders for AI customer service projects. If they can't achieve results, they don't charge. For example, for AI customer service, it's not just about selling the system to the customer. They charge 0.5 yuan for each processed order, and if no orders are processed, the customer doesn't need to pay. This model greatly reduces the decision - making threshold for small and medium - sized enterprises and deeply binds the income of FDEs with the business results of customers.

The industry has long debated "how to balance standardization and customization", and the popularization of AI may provide a new answer.

Lawted believes that a feasible approach is "internal standardization and external customization". He doesn't deliver standardized products to customers because standardized solutions can never achieve 100% adaptation; however, he will internalize general functional components, development frameworks, and diagnostic methodologies. Just like designers have their own color - matching systems and animation effect libraries, each official website they create for customers is unique, but the underlying components are reused.

He gave an example: When Feishu's FDEs go to factories for safety helmet detection and fire and smoke recognition, it seems like a brand - new customized solution for the customers, but in fact, the underlying technical components have been reused in dozens of enterprises. After AI has reduced the cost of customized development, internalizing standardized components and delivering customized solutions externally has become a new consensus in the industry.

However, there are still many uncertainties on the actual implementation path.

For example, the high cost of system switching. Lawted once developed an AI + ERP system for a freight forwarding company in Shenzhen. All the functions were developed, but it couldn't be officially launched for a long time.

He explained that the core system of the freight forwarding company is the lifeline of the entire company. Switching the system requires parallel data entry in the old and new systems for all documents, and it can only be promoted during the off - season from February to May every year. The project happened to coincide with a surge in orders in the logistics industry after the easing of the Middle East conflict, and all employees were busy with business. So the project had to be postponed to the next off - season.

After this lesson, he no longer replaces the customer's core system at the beginning. Instead, he develops lightweight "digital employee" plug - ins that can be embedded in the existing workflow. This approach has low risks, quick results, and much higher customer acceptance.

Although FDE can help enterprises achieve AI transformation faster, it is not a universal key and will not create a successful path out of thin air. The consensus among several practitioners is that the most core value of FDE has never been to deliver a set of tools, but to help enterprises discover cognitive blind spots. This is also the core meaning of "Forward/Frontier" in FDE.

Is this position just a transition?

The popularity of FDE is attracting a large number of people chasing the trend, and bubbles are also emerging.

The first to flood the market is the FDE training boom. A large number of paid courses have emerged in China, claiming that "you can switch to an FDE position from scratch in three months and earn a million - yuan annual salary", attracting a large number of practitioners who want to ride on the trend.

However, practitioners who are deeply involved in the industry do not agree with this. The community founded by Lawted has gathered thousands of FDE practitioners. He has held offline salons in many places such as Shenzhen, Shanghai, and Hangzhou. He always adheres to a non - profit orientation and does not offer paid training.

In his view, FDE highly depends on industry experience. The implementation methods in the logistics industry are completely inapplicable in the racing industry, and the experience in the manufacturing industry has no value in the financial industry. There is simply no universal course system. The pain points, processes, and rules of each industry are different, and it is impossible to teach through a single set of courses. Therefore, holding offline meetings is a good opportunity for communication among different industries.

Jolie Ni also mentioned that there are almost no FDE training courses for the general public in Silicon Valley. The certification training of large - model manufacturers is only for the internal employees of cooperative service providers to improve their delivery capabilities.

What is more worthy of vigilance than the training bubble is that AI companies focusing on FDE are sliding into the dilemma of consulting companies.

Zaniel said that many AI companies have essentially become consulting companies, but they are seeking financing with the valuation of software companies. The output of large models is probabilistic. Each enterprise customer requires a customized verification process and continuous maintenance and iteration, and labor costs cannot be avoided. Eventually, the labor cost of FDEs will be included in the operating cost, pushing the company's gross profit margin down to the level of traditional consulting companies.

How long can FDE as an independent position last? The judgments of several practitioners are highly consistent: It is most likely a transitional position, but the underlying capabilities will remain in the long term.

Ram predicts that in two or three years, the AI implementation solutions for most industries will gradually take shape, and enterprises will return to the traditional model of purchasing mature solutions and will no longer need so many FDEs for exploratory work.

In his view, FDE itself is the "special forces" during the enterprise's AI transformation period. When the organizational structure cannot keep up with the technological changes, a team is first formed to pave the way. After the path is paved, naturally, there will be no need for so many path - paving people.

Zaniel believes that when people who understand business learn to use AI tools, they can implement scenarios by themselves, and the specialized FDE position will naturally complete its mission.

However, this does not mean that the value of FDE will disappear. On the contrary, the in - depth changes it brings will continue to penetrate into the organizational fabric of enterprises.

Many people think that the value of AI to enterprises is only cost reduction and efficiency improvement, but Zaniel sees a more fundamental change: AI provides enterprise executives with an information channel that bypasses the hierarchical system. The data within the enterprise is transmitted from the front line upwards, and each layer of management will process it. Not reporting bad news is not a moral issue but a rational choice under organizational incentives. However, AI is not restricted by this incentive structure and can directly generate neutral judgments based on the original data.

From this perspective, the significance of FDE goes far beyond implementing a few AI tools. It is the contact point for AI to penetrate into traditional organizations. The more contact points there are, the faster AI will change from a "tool" to an "infrastructure". And those who create these contact points, no matter what their names are, will always be in short supply.

This article is from the WeChat official account "Leopard Change" (ID: baobiannews), written by Gao Ze and published by 36Kr with authorization.