FDE is having a moment.
FDE stands for Forward Deployed Engineer. Today, it more often refers to people who go deep into customer sites and are responsible for integrating large models into business processes. They not only write code and build systems, but also communicate with bosses and employees, turning a company's vague "we want to use AI" into a product and workflow that actually runs.
In fact, as early as around 2010, the US tech company Palantir had already established and promoted the FDE role, but for a long time afterward it existed only at a handful of enterprise software companies. After large models took off, companies discovered that hooking up a model doesn't mean having a usable AI system. Legacy systems that won't connect, data scattered across WeChat and Excel, bosses who can't articulate their needs—all of this requires someone on-site to solve. Thus the FDE, which bridges AI products and enterprise business, was pushed into the spotlight.
Hiring demand is also rising. Data from the US job platform Indeed shows that FDE-related job postings grew from 643 in April 2025 to 5,330 in April 2026, up 729% year over year. In China, similar demand is emerging, mainly from two types of players: one is model companies serving governments, enterprises, and large clients, where FDEs handle after-sales support and delivery of models; the other is individuals or small teams providing customized services to small and medium-sized businesses that can't use AI well.
Demand has taken off, but the business model hasn't taken shape yet. Some quote per project, some charge consulting or subscription fees, some take a cut of the cost savings, and some turn solutions into templates to sell repeatedly. What bosses understand more easily is "the annual salary of a digital employee," but how much it's actually worth to overhaul a workflow still has no standard answer.
Controversy has followed. Some call it the most valuable new profession of the AI era, and training institutions use slogans like "zero experience, switch careers in three months, million-yuan salary" to attract users; others believe it's just a new name for solutions architects, implementation consultants, and delivery engineers, or even consider it essentially an outsourcing job with no real substance.
We spoke with five FDE practitioners. Among them: a newcomer who, right after graduation, took on a project for a company with 100 million yuan in revenue; an intern earning 200 yuan a day and working 50 hours a week; as well as a freelancer, a product manager, and the founder of an FDE community.
Once they actually entered companies, they found that writing code was only part of the job; more often they had to be stationed on-site, attend meetings, train employees, and figure out messy data and closed-off legacy systems. The income wasn't as glamorous as the rumors either: some landed orders worth tens of thousands of yuan but had to handle customer acquisition, development, rework, and maintenance themselves; others traveled constantly and pulled all-nighters to deliver, while still earning a product manager's salary.
The five reached similar conclusions: there is demand for FDEs, but it's not yet a mature industry; companies want to use AI to boost efficiency but can't articulate their needs, so FDEs have to do a lot of communication work; income is uneven, and almost no one has actually received the rumored million-yuan annual salary.
01. The company's first dedicated FDE, involved in 80% of the work on a project
Feng Youyou | Post-2000s, Shenzhen, FDE at a content marketing tech company
At the end of June this year, I joined a company doing B2B content marketing. The position I interviewed for was AI product manager, but within a week of joining, I was sent to work on-site out of town and started doing FDE work.
We are an AI-native small team with no rigid division of labor—development and frontend roles are split out separately. I'm the company's first dedicated FDE and also the person who interacts with clients the most.
A traditional AI product manager mainly creates prototypes, writes documents, then coordinates with developers and handles acceptance, only participating in the middle stages of a project. An FDE, however, has to follow a project from start to finish: digging out requirements, going on-site, sorting out the business, building demos, driving implementation, and finally distilling the experience into a reusable framework. All told, I'm involved in about 80% of the work on a project.
My daily work now is 60% communication, 40% technical, with most of my time spent traveling; my longest on-site stint at a client lasted nearly a month. A typical on-site rhythm: in the morning, meetings with company leadership to uncover real business pain points; in the afternoon, organizing the client's scattered requests and using various AI tools to quickly build process demos; at night, staying up late to refine the solution, then the next day closing the loop with the client for verification, adjusting round after round until the AI solution can actually land.
Image source / pexels
I use all kinds of tools—Codex, Kimi, Copilot—which help me organize materials, build project frameworks, and write documents. A prototype that used to take days of developer support can now be done in half a day. But AI can only do execution; product architecture, business trade-offs, negotiating with clients, and final implementation decisions absolutely cannot be replaced.
This year I did indeed feel that FDE positions suddenly multiplied, but the salaries are not as exaggerated as the internet rumors suggest. I'm still earning a product manager's salary. But at this stage, what I value more is the opportunity to work directly with clients, lead projects, and be accountable for results—experiences that are hard to get elsewhere.
This role is still in an exploratory phase, and each company's understanding and requirements differ. For example, some companies require FDEs to handle both frontend and backend development, but at our company, I'm only responsible for requirements gathering, prototype building, and implementation follow-up, with dedicated developers supporting the backend. B2B content marketing clients have simple decision chains—leaders can directly sign off and push implementation forward; if it were aimed at massive numbers of C-end users, with scattered user demands, balancing cost and experience would be far harder.
What I'm most anxious about is that I do too many things across too wide a range, but without enough depth. My communication skills keep improving, but my technical judgment may not be keeping up. I spend every day solving clients' problems, with little time to think about how to distill reusable product capabilities.
AI industry buzzwords iterate extremely fast—prompt engineers, agent engineers, wave after wave—but I believe that even if this role is no longer called FDE in the future, the capabilities it builds—communication, judgment, implementation—are transferable. One piece of advice for those wanting to enter the field: don't get carried away by the "high salary" label; first validate at low cost whether you're suited for it, for example by building a product from start to finish and talking to real users. If you're not averse to high uncertainty, heavy communication, and hands-on work, then give it a try.
In the AI era, it's hard to say whether the FDE role will always exist, but people who can solve problems on the front lines will always have a place.
02. After becoming an FDE, I start each day by studying how others work
@82LSF | 30 years old, Guangzhou, freelancer
I studied computer science as an undergraduate, and after graduation I worked on financial risk-control software at a listed company in Guangdong, mainly serving clients like the Hong Kong Stock Exchange.
Last year I started freelancing, taking on software development and AI projects. Last month, through a friend's introduction, I went to a traditional restaurant chain company as an FDE.
Strictly speaking, I'm more like an OPC (one-person company). My thinking is very practical: first serve this company well to build up case studies, then develop more clients. So far, I've accumulated more than 10 industry-generic skills, which is my biggest gain.
On my first day, I joined the company's WeChat Work groups. Instead of rushing to write code, I first followed operations, marketing, delivery, and store managers to see how they did their daily work, getting a clear picture of restaurant industry employees' daily workflows. This step is crucial. Very often, clients can only tell you they want to "use AI to boost efficiency," but they don't know which part of the process AI should actually be applied to. An FDE needs to enter the business processes personally, observe what forms employees fill out every day, what reports they write, what data they organize, and then identify the tasks that are highly repetitive and don't require complex judgment.
After getting a clear picture, I found that most of this company's work still relied on manually filling out forms, making reports, and writing contracts; at most they used Doubao, and all office processes were extremely dependent on desktop office software.
I started using WorkBuddy to develop dedicated skills for each business line and each department, simplifying daily chores and handing repetitive, low-value labor over to AI—for example, daily reports that automatically pull each store's revenue data, attendance anomaly sheets, and inventory price-comparison tables.
These tasks used to be filled in manually, requiring people to sit at a computer for half an hour to an hour a day; now they're basically done automatically by AI, saving each person at least half an hour a day. Over a month, that adds up to roughly one person's worth of working hours.
Now that skills are up and running, employees don't even need to sit at a computer. Just send a natural-language instruction from WeChat on mobile, and most repetitive daily work gets done.
Image source / AI-generated
Many people ask what capabilities an FDE actually needs.
My takeaway is, first, you need a systematic understanding of the underlying technology, such as grasping the boundaries of APIs and the MCP protocol. Especially in enterprises, the business boundaries different roles can access must be clear—some data can be viewed, some can only be queried but not written, and some can't even be seen.
Second, you need to understand the industry you serve, which is a bit like what consulting firms do. You have to first understand how the client operates before you know where AI can be applied.
Finally, and most importantly, communication. Compared to technical skills, an FDE's communication ability may matter even more. If you keep failing to understand the client's needs, or can't solve problems quickly, they'll lose patience.
But having these capabilities doesn't mean a project will definitely succeed. In my observation, there are far too many projects where an FDE is sent into a company, can't deliver in two or three months, and the whole thing falls through.
As for the future, I'm not anxious. Looking further ahead, I believe this field still needs at least three to five years—or even longer—of accumulation before a mature ecosystem of products, methodologies, and services can truly take shape.
03. Interns earn 200 yuan a day; at my busiest, I worked nearly 50 hours a week
Zhewei | Post-2000s FDE intern in Beijing
One month into the job, my most direct impression of being an FDE is that what's said online and what actually happens are two completely different things. It's a profession with a stark contrast between "ideal" and "reality."
Many people's impression of this role stops at "million-yuan salary," "60% communication, 40% technical," or "just use AI to boost efficiency and earn money lying down." But on the front lines, the AI automation that outsiders hype up so much often turns out to be a lot of tedious "dirty work."
As an FDE intern, my daily pay is 200 yuan. At my busiest, I worked nearly 50 hours a week, with three days a week on-site at clients or on business trips. The cost-performance ratio is really low.
In the morning, I usually go to the client's site and spend a few hours in meetings first, confirming what the requirements are and arranging how to deliver afterward. The rest of the time is all spent writing code. If the deadline is tight, I might code for seven or eight hours straight, and even before bed, I'll leave a super-long task for my AI agent to keep running overnight, then review the results the next day.
Image source / pexels
Of course, what's rushed out overnight isn't high quality. But now everyone feels that with AI, development should progress by the day. In reality, after code is generated, it still needs review and testing—these are the most time-consuming steps. Plus, AI changes fast, so there are many concepts and new things to learn. I often feel like I'm not working a job, but more like a "research laborer."
Before entering the field, I thought an FDE's job was to sort out the client's pain points, toss them to AI, and just do a bit of fine-tuning. Only after actually starting work did I realize that clients often have a pile of legacy systems, and you first have to figure out how they run and how open they are; if the client's data is messy, you have to clean it first. But clients don't understand—they think they just want an AI feature, so why all the groundwork?
Once, a client wanted to build a Feishu AI assistant internally. The hard part of this task wasn't installation—it was connecting the Feishu AI assistant, the program on our server, and Feishu's online spreadsheets. For these three things to communicate with each other, an intermediary service had to be added.
To enable batch file modification and style adjustments, we also integrated an external tool called "Miaoda." But as the tools multiplied, troubleshooting became more and more troublesome. Once something went wrong, I had to check at least three places: Did the intermediary service compute something wrong? Was there a problem with the spreadsheet data? Or was the AI assistant's configuration not set up properly?
We originally wanted to use tools to save effort, but ended up increasing our workload by integrating too many of them. After all the hassle, we gave up on that approach and went back to the old way—building the AI assistant on our own server and using other tools for debugging, which turned out to be much simpler.
I don't think FDE is what outsiders say—just a Solutions Architect or delivery engineer with a trendy new name. It's actually about merging many traditional roles into one, powered by AI. What used to be divided among several people is now all piled onto one person, who must not only understand code but also know the business details and risks better than the client.
Some say FDE is a transitional role, and that in two or three years, once AI solutions are standardized, so many people won't be needed. I see it differently. As long as enterprise business processes keep changing, and as long as there are new scenarios that need people to understand and implement, FDE won't disappear. It's just that the barrier to becoming an FDE will keep getting lower, because AI tools will keep getting stronger. But the lower the barrier, the scarcer those who can untangle complex business needs and make the call on technical solutions become.
In the long run, I might transition to being an independent developer, running a one-person company. Working for others means too many trivial matters and no rhythm of my own. When it comes to solving problems, I'd rather do it my own way.
04. An FDE is an "all-rounder," and even I haven't maxed out every stat
XIAO | 23 years old, Hangzhou, entrepreneur
I got into the FDE field more by accident. From June until now, it's been less than two months.
I did my undergrad at BUPT and my master's at Nanyang Technological University. I first saw a blogger sharing about FDE online, and I felt I was quite suited to be an FDE—my communication skills aren't bad, and I have my own understanding of AI. Later I found out he was a senior from my same program, and now he's become my co-founder.
Under his influence, I started trying to use AI to transform enterprise workflows. He helped me shoot a video about it, and not long after it went out, my first client came knocking. A "second-generation factory owner" was interning at a sales company his family worked with, and he introduced me to the boss. The client wanted a customized CRM system. We first chatted online about the intent to cooperate and our understanding of the industry. That weekend, I rushed to Wenzhou and showed the boss everything I had done before, piece by piece. The boss valued that we young people had technical skills, and we had a pleasant discussion. That very Sunday, we reached a verbal agreement.
The clients I work with now mostly have annual revenue above 100 million yuan. One of them, a supply chain company with annual revenue of about 200 million yuan, wanted a photo-recognition system that would automatically import the recognized information into their ERP.
It sounds simple, but doing it is a completely different matter. What truly tests an FDE isn't whether a model can recognize an image, but whether you can break down a company's chaotic, vague requirements into workflows that AI can execute. The client's data may be scattered across employees' WeChat, the ERP has no API, and the boss can't even articulate what he actually needs. I have to go on-site first to understand the business, train employees, build a knowledge base, and then judge which steps are suitable to hand over to AI.
Image source / pexels
Once, I noticed an employee spending every day repeatedly organizing Excel spreadsheets. I spent half an hour figuring out where the data came from, what steps it went through, and what form it needed to take in the end, and that very afternoon I had AI take over that work. An FDE's ability often shows exactly here: not just writing code, but quickly understanding the business and then turning the problem into an executable solution.
AI executes fast, but the early-stage business immersion can't be skipped. Even within the same industry, getting a solution to 30% standardization is already quite good; the rest has to be customized again based on each client's needs. What I'm selling isn't an off-the-shelf product, but a service that continuously delivers results.
My income is far less glamorous than the legends suggest. Customer acquisition, on-site work, delivery—I have to do it all myself, busy from morning to night every day. It's more exhausting than joining a big tech company. Online they say this field in the US can pay a million dollars a year, but among the people I know, not a single one has gotten that number. Abroad, people may be willing to pay for the process; in China, they only pay for results. If the product doesn't actually run, no amount of work done beforehand matters.
That said, this doesn't mean FDEs can't make money. The industry is still at a very early stage, with no established standards and a market full of mixed players. People with real capability who can actually deliver projects can indeed earn income that matches their abilities, even higher than comparable technical roles in the internet industry. But high income comes from the combination of technical, business, communication, and delivery skills—not from carrying the FDE title.
In my view, an FDE is an "all-around player": they understand technology and business, communicate well, and must also act fast and iterate fast. 99% of people can't do this, and I myself haven't maxed out every stat.
FDEs may just be a transitional vehicle for AI entering every industry. Even if the role disappears, I believe it will be among the last software roles to go. But if you're only drawn by the high pay, don't come; if you truly want to do it, start by using AI to boost your own productivity.
05. After touring Beijing, Shanghai, Shenzhen, and Hangzhou, I found that what FDEs lack most is trust
Lawted | Founder of the nationwide Ha7ch FDE community for 26-year-olds
I started working as an FDE in April this year.
At the time, a friend asked if I'd be interested in helping a freight forwarding company install OpenClaw. When I got there, I discovered the boss could barely use Doubao. The company's documentation clerks switched back and forth between WeChat, Excel, and PDFs every day, reading documents and entering information, with a huge amount of work done entirely by hand.
I felt these processes could be handed over to AI, so I took on my first project, hoping to help them replace the whole system.
After running back and forth between the boss and employees for a month, I gradually built out the processes. Two months later, the system was done. After uploading Excel, PDF, and other files, the AI could automatically parse them, create orders, and fill in forms, and employees only needed to do the final check.
Image source / pexels
But even though the product was built, the project fell through. The boss asked me: is the system stable, who takes responsibility when something goes wrong, can you provide 24-hour maintenance? As a solo OPC, I had no way to guarantee a system would run stably long-term.
Although the project didn't pan out, that was when the FDE concept started taking off. I posted my lessons learned online, sharing how FDEs work and what problems arise during on-site deployments, and gained a following. Later, I organized four closed-door meetups in Shenzhen, Shanghai, Hangzhou, and Beijing, meeting many people working on enterprise AI adoption.
After touring the four cities, I found that FDE is currently just a profession with "business but no industry." Across the four cities, there were small e-commerce jobs worth a few hundred or a few thousand yuan, as well as enterprise transformations worth hundreds of thousands or even millions. But big projects require assembling a team, and the project fees also have to cover customer acquisition, development, on-site work, rework, and maintenance costs—the money actually earned isn't as much as it looks.
Winning a project is only the beginning. Clients say they want AI but can't articulate their needs; enterprise data is messy and sensitive information is hard to integrate; employees fear being replaced, and veteran staff and business departments push back; a single task may seem to work, but once chained into a complete workflow, accuracy, model costs, and payment cycles can spiral out of control.
But the hardest part for independent FDEs is getting enterprises to trust you. Among those attending the closed-door meetups, sixty to seventy percent got their first project through relatives, friends, or accumulated connections. FDEs from big tech companies have their company and products as backing and can go straight into client sites, while independent FDEs have none of that and struggle to earn trust. For traditional business owners, handing business data to a stranger is inherently a high-risk move. Independent FDEs can only prove themselves through case studies and delivery after delivery. People without industry connections who are unwilling to speak publicly and just quietly write code will find it very hard to get that first ticket in.
To solve this problem, I started organizing hackathons, having participants solve enterprise problems within 48 hours. I brought in companies with AI transformation needs and screened people who wanted to become FDEs. The companies covered travel expenses, and participants came on-site to sort out requirements and build a demo within 48 hours. If the boss approved, the two sides continued to cooperate; if not, the project ended there.
Once, we went to a law firm to help them build a system. Their interns normally had to process five to six hundred scanned documents and organize them into over a thousand rows of Excel, which took nearly half a week. The participants used AI to complete the recognition and summarization, with some content even more accurate than manual entry. After the 48 hours ended, the company decided to sign a first-phase agreement to deploy the tool stably.
So, in my view, FDEs are more like "patches for AGI"—while AI still can't enter a company on its own, understand the environment, and start working, humans connect it into real business processes. Only by truly delivering a finished product whose results enterprises can see directly can you truly retain clients.
I think the barrier to entry for FDEs isn't as high as imagined—if you understand technology and are willing to go into enterprises to understand their needs, you can try. But the bar for actually delivering projects is very high: you have to make the boss comfortable handing over their business and data, and you have to stay around to take responsibility when the system breaks.
*At the interviewee's request, "Zhe Wei" in this article is a pseudonym.
This article comes from the WeChat official account "Dingjiao" (ID: dingjiaoone), authors: Li Mengran, Wang Hanxing, Wang Lu, Chen Dan, Lei Jing, editor: Lei Jing, published by 36Kr with authorization.