A job title that sounds rather abstract is becoming one of the key hiring priorities for big tech companies in 2026, with serious money being thrown at it.

ByteDance is offering a monthly salary of 35,000 to 70,000 yuan with 15 months of pay per year, for an annual salary of up to 1.05 million yuan. Alibaba Cloud Intelligence is offering a monthly salary of 20,000 to 50,000 yuan with 16 months of pay per year.

These positions are for Forward Deployed Engineers (FDE). In recent months, from overseas leading large model companies like OpenAI and Anthropic to domestic giants like ByteDance, Alibaba, and Tencent, tech companies have been adding FDE roles one after another. Stories like "earning 100,000 yuan a month taking on projects solo" and "switching careers with zero experience to a million-yuan salary" keep popping up on social media, and tech professionals, fresh graduates, and traditional consulting practitioners are all flocking to this track.

But controversy has always accompanied it: Is this truly a brand-new golden profession born of the AI era, or just old wine in a new bottle — concept hype? Does its core value lie in the code, or in the business?

Not exactly a new profession

So what does an FDE actually do?

As model capabilities keep evolving, the real challenge lies in genuinely connecting the needs scattered in every corner of traditional enterprises with model capabilities. Bridging the gap between these two sides is what an FDE does.

"A large part of the buzz around the term FDE in 2026 is concept arbitrage." This judgment from Zaniel, a PhD student at Stanford and independent consultant, hits at the industry's most central controversy. In the eyes of many industry insiders, this role, packaged as a brand-new profession, is essentially the old Solutions Architect, implementation consultant, or delivery manager, just rebranded with a trendier name riding the AI wave.

Tracing back to its origins, FDE is not a concept that appeared out of nowhere. The role was first scaled up by the US company Palantir, whose core model was dispatching engineers to work on-site at government agencies and large enterprise clients, embedding standardized data analysis products into the clients' existing systems and workflows.

Jolie Ni, a Silicon Valley tech worker, witnessed the evolution of FDE firsthand for over six months before decisively going independent and founding an FDE service provider called Hconsult.ai. Her target clients are small and medium-sized enterprises with annual revenues of 5 to 10 million USD — clients who have the budget and the need, but can neither attract the large companies mentioned above to serve them nor afford to build their own FDE teams. Her startup has carved out a differentiated competitive position against those large companies.

Similar to the US, the domestic market has also formed a dual structure. On one end is the big tech FDE camp: FDE teams at large model vendors serve Fortune 500 clients, while FDEs at Feishu and DingTalk enter mid-sized and large enterprises in "sales + tech" pairs, doing customized deployments built on their own platforms.

Through the "HA7CH" community he organized, Lawted learned that a major company in Hangzhou has already begun implementing a system where "new hires start with three months as an FDE," and even directly dispatches product managers on-site to conduct requirements research and Demo development. FDE is evolving from a standalone role into a standard capability for TOB teams.

On the other end are the independent service providers Lawted calls "grassroots FDEs" — mostly solo studios or small teams deeply rooted in small and medium-sized enterprises in lower-tier markets. With high flexibility and low costs, they fill the gaps that big tech companies' services fail to cover.

The ambiguity of the role's definition is precisely the source of the conceptual controversy. Many people can't tell the difference between an FDE and outsourcing, an AI engineer, or an AI product manager — and real projects best illustrate the boundaries.

Zaniel once received a request from a company that explicitly wanted to launch an AI customer service system. Only after conducting in-depth research on-site did he discover that the company's real pain point was that customer data across multiple business systems was completely siloed — the low efficiency of customer service was merely a surface symptom.

If you followed the outsourcing logic and simply built the AI customer service system, you'd have a deliverable, but the core problem would remain unsolved. This is the key difference between an FDE and outsourcing: outsourcing is only responsible for clearly defined deliverables, with requirements defined by the client; an FDE must first see through the surface to find the problem truly worth solving.

The divide from AI engineers is equally clear. At the leading fast-moving consumer goods company where Ram works, AI engineer roles have existed for years, with the core job being to replicate mature algorithms and solutions from within the industry. But FDEs often face vague requests like "I want to use AI to cut costs and boost efficiency," and must break down such vague ideas into actionable technical paths.

In Ram's view, the core of FDE lies in the "Forward" — taking a step forward to immerse oneself in the front lines of the business, rather than sitting in the back office waiting for requirements to be passed along.

Jolie Ni explains that an AI product manager focuses deeply on a single product, continuously iterating and optimizing; but an FDE must solve problems across industries and scenarios, more like a "consultant who can get hands-on and deliver." They not only diagnose problems but also personally build the solutions.

The consensus is that the core capability of an FDE has never been writing code.

Lawted puts the ratio at "six parts communication, four parts technology," a conclusion validated in Zaniel's financial institution reporting project. He did report automation for a local financial institution, and the most time-consuming and valuable part wasn't writing automation scripts — it was spending two weeks reverse-engineering hundreds of business rules, never before documented, from the experience of veteran employees. Anyone who knows Python can write automation scripts, but the rules hidden in people's heads can only be extracted by squatting on the business site and digging them out one by one.

This also means that FDE was never some brand-new profession that appeared out of nowhere. AI tools have dramatically lowered the barrier to development — projects that once required a whole team can now be delivered by one person, instantly amplifying the economic value of this role, which is why it has stepped into the spotlight riding the wave of concept hype. Stripping away the "brand-new golden profession" filter, it is essentially the "last mile" of AI implementation — a translator connecting technology and business.

Where is the value?

Only by shedding the halo of the concept and stepping into real business scenarios does the value of an FDE truly become apparent.

In real-world scenarios, the most common demand is tangible cost reduction and efficiency gains — and this is the value point that small and medium-sized enterprises are most willing to pay for.

Jolie Ni once did lead-generation workflow automation for a Korean GPU computing power company. Previously, the company's employees had to manually compile information on professors from the top 500 QS-ranked universities, match academic conference schedules, and write personalized business development emails — even a skilled employee could send at most 10 effective emails a day.

After the automation solution was implemented, the system automatically fetched conference information and scholars' research updates via API, used large models to match relevant cases and generate customized content, and could stably output 200 to 500 emails a day — with no drop in reply rate despite the automation.

The same was true for the local financial institution Zaniel served: multiple business departments had to produce dozens of daily reports, all relying on manually exporting data from core systems and filling in Excel. After the automation solution was implemented, hundreds of implicit business rules were all codified into the system, and employees were completely freed from mechanical, repetitive labor.

Ultimately, the value of an FDE may go far beyond saving money — it also helps companies build long-term market competitiveness.

Corresponding to different value demands, the industry has evolved two mainstream payment models.

Jolie Ni uses a "project-based + monthly maintenance fee" combination: the main project is charged as a one-time package, while subsequent process iterations and daily operations are billed as a monthly service fee. This balances one-time income with long-term renewals and suits clients with clear requirements and long-term operational needs.

The other model is pay-for-performance.

Lawted explains that many independent FDEs settle lead-generation projects by the number of valid leads, cost-reduction projects by a share of the labor costs saved, AI customer service projects by the number of orders effectively handled — and if no results are delivered, they don't get paid. Take AI customer service, for example: it's not just about selling the system to the client and calling it done — they charge 0.5 yuan per order handled, and if not a single order is handled, the client pays nothing. This model greatly lowers the decision-making threshold for small and medium-sized enterprises, and deeply ties the FDE's earnings to the client's business outcomes.

The industry has long debated "how to balance standardization and customization" — the spread of AI may offer a new answer.

Lawted believes the viable approach is "standardize internally, customize externally." He doesn't deliver standardized products to clients, because standardized solutions can never achieve 100% fit; instead, he accumulates reusable functional components, development frameworks, and diagnostic methodologies internally. It's like designers having their own color palettes and motion libraries—every client's website is unique, but the underlying components are reused.

He gives an example: Feishu's FDEs went to factories to build safety helmet detection and smoke/fire recognition systems. These looked like brand-new custom solutions for clients, but the underlying technical components had already been reused across dozens of companies. After AI drove down the cost of custom development, accumulating standardized components internally while delivering customized solutions externally has become the industry's new consensus.

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

For example, the high cost of switching systems. Lawted once developed an AI+ERP system for a freight forwarding company in Shenzhen. All the features were complete, yet the system couldn't officially go live for a long time.

He explains that the core system of a freight forwarder is the lifeline of the entire company's operations. Switching systems requires all documents to be entered in parallel into both the old and new systems, so it can only be done during the off-season from February to May each year. The project happened to coincide with a surge in logistics orders after the easing of the Middle East conflict—everyone was busy with business, so the project had to be shelved until the next off-season.

After this lesson, he no longer starts by replacing the client's core system. Instead, he builds lightweight "digital employee" plugins that embed into existing workflows—lower risk, faster results, and far more acceptable to clients.

While FDEs can help enterprises adopt AI faster, this is not a master key, and it won't conjure a path to success out of thin air. The consensus among several practitioners is that the core value of an FDE has never been delivering a set of tools, but helping enterprises discover their cognitive blind spots—this is the essence of the "Forward/Frontier" in FDE.

Is this role just a transition?

The popularity of FDEs is attracting a large crowd of trend-chasers, and bubbles are growing along with it.

The first thing to flood the market is the FDE training boom. A large number of paid courses have emerged in China, claiming "zero experience to FDE in three months, million-yuan salary," attracting many practitioners hoping to ride the wave.

But practitioners deeply rooted in the industry disagree. The community Lawted founded has gathered thousands of FDE practitioners and has held offline salons in Shenzhen, Shanghai, Hangzhou, and other cities. He has always insisted on a non-profit positioning and does not offer paid training.

In his view, FDE work depends heavily on industry experience. Implementation methods from the logistics industry are completely inapplicable to the racing industry, and manufacturing experience has no value in finance—there is simply no universal curriculum. Every industry's pain points, processes, and rules are different; it's impossible to teach them all with a single set of courses. That's why holding offline meetups is a great opportunity for cross-industry exchange.

Jolie Ni also noted that Silicon Valley has almost no FDE training for the general public; certification training from large model vendors is only for internal employees of partner service providers, aimed at improving delivery capabilities.

More concerning than the training bubble is that AI companies focused on FDEs are sliding into the predicament of consulting firms.

Zaniel says many AI companies have essentially become consulting firms while raising money at software-company valuations. Large model outputs are probabilistic; every enterprise client needs a customized validation process plus ongoing maintenance and iteration, so labor costs simply can't be eliminated. In the end, FDE labor is booked as operating costs, pushing the company's gross margin down to the level of traditional consulting firms.

How long can the FDE survive as a standalone role? Several practitioners' judgments are highly consistent: most likely it is transitional, but the underlying capabilities will persist long-term.

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

In his view, FDEs are essentially the "special forces" of an enterprise's AI transformation period. When the organizational structure can't keep up with technological change, you first deploy a team to blaze the trail; once the road is paved, you naturally don't need so many trailblazers.

Zaniel believes that once people who understand the business learn to use AI tools, they can complete scenario implementation themselves, and the dedicated FDE role will naturally have fulfilled its mission.

But this doesn't mean the value of FDEs will disappear. On the contrary, the deeper changes they bring will continue to permeate the organizational fabric of enterprises.

Many people think AI's value to enterprises is merely cost reduction and efficiency gains, but Zaniel sees a more fundamental change: AI provides executives with an information channel that bypasses the hierarchy. When data inside a company travels from the front line upward, it gets filtered and polished at every layer of management; reporting good news and hiding bad isn't a moral failing—it's a rational choice under organizational incentives. But AI is not bound by this incentive structure and can generate neutral judgments directly from raw data.

From this perspective, the significance of FDEs goes far beyond implementing a few AI tools. They are the touchpoints through which AI penetrates traditional organizations—the more touchpoints, the faster AI transforms from a "tool" into "infrastructure." And the people who create those touchpoints, whatever their title, will remain scarce.

This article comes from the WeChat public account "豹变" (ID: baobiannews), author: 高泽, published by 36Kr with authorization.