A few days ago, I had a long chat with someone who had just finished interviewing for an FDE role at a major overseas tech company. The more we talked, the more I realized that many people's understanding of this new AI role goes wrong from the very first step. Being able to build agents doesn't mean you can be an FDE. Having a few demos under your belt doesn't mean you'll get an offer for this position either.
Today, let's break down a real interview and see what companies are actually looking for.
What Exactly Is an FDE?
FDE stands for Forward Deployment Engineer, usually translated into Chinese as 前沿部署工程师, with some also calling it 前线部署工程师.
Here's an analogy: a company spends a fortune on a full set of smart kitchen equipment, but no one knows how to prepare their own menu or control the heat, so meals still can't be served. The FDE is the specialist the manufacturer sends, stationed in your kitchen. The equipment is ready-made; their skill is cooking the dishes you order on your own stove.
Many people think of it as a new kind of programmer for the AI era, assuming it's just a technical role of assembling large model tools. The reality is far from that. This role is about making AI deliver results inside enterprises—half on the technical side, half on the business side.
This model was pioneered by Palantir, which sends engineers who know their technology best directly to customer sites, immersing them in the customer's business to connect general capabilities to specific production environments. OpenAI has now adopted this approach, with Forward Deployed Engineer openings listed on its website, hiring in both Singapore and San Francisco.
I read OpenAI's job description in full, and a few key passages are worth savoring on their own: Partner with the most important strategic customers to lead end-to-end deployment of frontier models in production environments. Own everything from requirements gathering, technical assessment, and system design through development and launch. Be based on-site at customers, spending half the time traveling.
The requirements are stated quite plainly: five or more years of engineering or technical deployment experience, including customer-facing work. Ability to write production-grade code in Python and JavaScript. Experience building systems based on large models, and an understanding of how model behavior affects product experience.
On compensation, the job posting shows a base salary between $160,000 and $280,000 for this role. Similar positions have also started appearing on domestic Chinese job platforms, with monthly salaries ranging from 35K to 55K. According to some media statistics, the number of related positions has grown more than fortyfold in two years, earning it the title of the hottest new role in the AI world.
Why Does This Role Exist?
So why did this role emerge? Because today every major tech company has its own large models, AI platforms, and agent development tools—all ready-made.
But once enterprise customers get these capabilities, they run into a very practical problem: how exactly should AI be applied to their business? How should processes be changed? How should data be connected? How should strategies be designed? What if results are poor after launch? A whole string of questions pops up.
And these problems can't be solved by a model engineer alone. The model team is responsible for whether the model itself is good; as for the customer's internal processes, data, permissions, and organizational structure, they can neither see nor change them.
So there's a missing link between technology and business. The FDE was created to fill exactly this gap.
It's Not the Same as Pre-Sales
Many people ask: is an FDE just an AI version of pre-sales?
The difference is actually huge. Pre-sales' main battlefield is before the contract is signed, and the core action is explaining the product clearly so the customer is willing to buy.
An FDE comes in after the customer has completed the purchase, digs into the business on-site, and builds and runs the solution. One only needs to know how to present the product; the other needs to understand technology and requirements, and even redesign the customer's business processes. OpenAI's job description specifically states that this role spends half the time traveling at customer sites, with engineers embedded directly in the customer's engineering and business teams.
FDE Interviews Really Test Three Abilities
This is the part I most want to explain thoroughly today. That person's takeaway after the interview was that people who only know technology basically don't make it to the later rounds.
First, AI application development skills.
Merely knowing how to call APIs is at best an entry ticket. The interviewer wants to hear you explain how to build RAG, how to tune retrieval when results are poor, how to design agents, how multiple agents collaborate, and how AI actually operates in real enterprise scenarios. Each of these questions can be probed several layers deep—you can't get by on memorized concepts.
Second, engineering capabilities.
This is the biggest weakness of many demo masters. How easy is it to get an agent running locally? A day or two following a tutorial will do it. The hard part is what comes after enterprise launch. What if the answer is wrong? What if task execution fails? How do you control costs? How do you manage data permissions? How do you monitor the system?
These things should look very familiar to QA folks. Online monitoring, exception fallback, permission boundaries, quality assessment—these are exactly what test engineers deal with every day. If you're serious about transitioning into this direction, a testing background might actually be an advantage.
Third, business communication skills.
This is the one programmers most easily overlook.
An FDE is not a role where you sit at a computer writing code all day—a lot of time is spent talking with customers. What are your business goals? Where in the current process is the most labor consumed? Which steps are suitable to hand over to AI, and which absolutely must not be? After these conversations, you translate business problems into AI solutions. This stage filters out a lot of people who only know technology—no matter how beautiful your code is, if you can't articulate business value, you won't pass.
If You've Built Agent Projects, Can You Transition to FDE?
You can try, but first adjust one expectation. There's a considerable distance between personal demos and enterprise-level AI projects.
What enterprise interviewers want to hear is which real business problem you solved. How much labor cost was reduced, how much processing efficiency improved, which process bottleneck stuck for half a year got unblocked. Presenting a project with these numbers is a completely different interview from demoing a feature list of an intelligent assistant.
Will FDEs Be Replaced by AI?
In the short term, I don't think so. The logic is simple: the stronger AI capabilities become, the more enterprises need someone to help them apply those capabilities to their business. Walking into customer sites, understanding processes, driving launches—models can't do these things.
But one type of person will definitely be eliminated: those who only know how to call models and stitch together workflows, unfamiliar with both business and engineering implementation. AI itself can do that part of the work. The FDEs who ultimately endure are those who understand engineering, understand business, and can push projects all the way to launch.
Want to Enter the AI Track? How to Prepare
Don't bet all your effort on learning a few more frameworks—frameworks change every six months, while capabilities are what you can carry with you. What's worth spending time polishing is still those three things: AI application development, enterprise project implementation, and solving real business problems. The latter two can only be practiced in real projects.
Now, back to the question in the title: what exactly is an FDE?
Simply put, an FDE is an on-site engineering delivery role based at the customer's location, responsible for integrating and implementing the capabilities of large model AI products at the customer's site, handling the entire process from requirements gathering and solution design to deployment and launch on-site, connecting AI capabilities into business processes, and ultimately being accountable for business outcomes.
Finally, a quick survey. If you're considering the AI product manager or FDE direction, what do you feel you're most lacking right now—technical skills, or enterprise project experience? Let's chat in the comments, and later I can write in detail about how to fill the specific gaps people mention.