How to Get Into FDE from Zero? Career Path & Experience Sharing

The previous article said half of FDE roles are fake — so how do you enter the real half?

A哲Phil@Formulasearch·Aug 2 ArticleFDE is temporary outsourcing or real AI implementation? A quick-start guide for beginners and how to tell the differenceThe harder AI implementation gets, the more valuable FDE becomes?

This year’s hottest role in the AI world is FDE. OpenAI, Anthropic, and Google are hiring for it; in China, Tencent Cloud, Alibaba Cloud, NetEase, and ByteDance are all hiring too. On job platforms both domestically and internationally, postings for this role have multiplied several times over within a year…62311649K

In the previous article, I went through FDE job postings across four cities — Chengdu, Shanghai, Hangzhou, and Shenzhen — and concluded: half of these roles are doing real implementation work, while the other half are just “on-site & outsourcing” with a new name. I also gave three questions and three signals to help you tell the difference on the spot.

This article is about how to get in — what paths exist, what skills you need, how to prepare for interviews, and two similar things I’m currently working on myself. It ends by answering the question the previous article left open: is this role actually worth entering?

Let me start with one conclusion: the entry barrier is far lower than most people think, but don’t expect this to be easy money for light work. Many people never make it in not because they’re unqualified, but because they judge themselves against senior-level requirements. Those already in the field aren’t necessarily comfortable either — but overall, this is a role that really builds you up.

I. Three Paths into FDE

In order of recommendation: first check whether Path A is still available to you; whether or not you take Path A, you must take Path C

Path A: Internship and Entry-Level Positions — Lowest Barrier, Most Underrated

Across the four cities, there are eight FDE internship postings total, with daily pay of 120–350 RMB; big companies typically offer 250–350. The most lenient education requirement reads “no degree requirement,” and some even say “high school education acceptable.” Shanghai also has a dedicated management trainee track at 7–15K.

One internship posting lists requirements in quite specific terms — practically the entry standard for this path:

Have used large language models, not just for chat, but to actually solve a real problem with them

At least be able to write scripts in Python

Not afraid to get your hands dirty — try things yourself when you hit a problem, rather than waiting for someone to tell you what to do

Bonus points are listed as: having built AI-related projects yourself — coursework, personal projects, and hackathons all count.

Path C: Build Your First Real Delivery Yourself (🔺 The Main Path)

Whether or not you take the other two paths, you must take this one. Because interviewers ask what you’ve done, not what you’ve learned. Degrees, certificates, and courses can’t replace something that real people are actually using.

Step one: complete a fully runnable project

• The data is real

• AI is embedded in a workflow, not just used for conversation

• The results are actually delivered

• Deployed on a server, not just running on your local machine

• At least one person other than you is using it

• You can clearly explain how well it performs

The last two criteria are where most people fail.

Step two: find your first real client. Its purpose isn’t to make money — it’s to confront you with real uncertainty. Real clients can’t articulate their requirements, change their minds midway, and provide messy, unstructured data. These experiences cannot be learned on your own.

Where to find them: acquaintances from your previous industry; small organizations without technical teams, such as law firms, clinics, design studios, and independent creators; people in your communities who are complaining about tedious parts of their work.

Your first job can be low-priced or even free, but you must clearly define what will be delivered and how it will be accepted. After completion, you must write a retrospective.

Step three: distill your approach. By your third client, you’ll see patterns repeat — abstract them into templates. Once you reach this stage, you’ve distinguished yourself from people who just take orders one at a time; the latter put in just as much effort on their tenth job as on their first.

Step four: translate it into your résumé. Don’t write “responsible for implementation and deployment of the XX system.” Instead, write:

The client’s initial request was A. On-site diagnosis revealed the real bottleneck was B. I designed Solution C, spent X weeks getting a minimal end-to-end pipeline working to validate, and the metric improved from M to N. Along the way, I distilled D, which was reused across K subsequent projects.

“Diagnosed the real problem, built a minimal closed loop, had metrics, and created something reusable” — these four elements are the entire difference between an FDE résumé and an ordinary implementation résumé.

Path B: Policy Channels, Using Shanghai as an Example

Shanghai’s Human Resources and Social Security Bureau and the Municipal Commission of Economy and Informatization are running advanced FDE training programs, with about 50 people in the first cohort; winners of the associated competition can be directly certified as professional engineers, and Shanghai and Anhui have already established a joint certification committee to push for mutual recognition. Shanghai Open University launched the nation’s first FDE specialized course, and Shanghai Institute of Technology has established a new Pre-FDE pilot class.

The limitations are obvious: significant time investment, geographic restrictions, limited slots, and having credentials doesn’t guarantee a job. It’s a bonus on top of Path C, not a replacement for it.

II. One More Path: The AI Coding Direction

I’m giving this its own section because many people are already standing at the door without realizing it.

Some big-tech FDE roles revolve entirely around “helping client teams make good use of AI coding tools.” Hangzhou has a role paying 15–18K whose requirements literally state:

Proficiency in at least two AI coding tools, such as Codex, Cursor, Claude Code, Kiro, and similar.

If you’re already getting things done with these tools, that’s your credential — it’s just not on your résumé yet.

The advantages of this path are very tangible: no need to enter client intranets, no need to deal with infrastructure-level engineering, and you use the same tools you work with daily. You’re still doing the FDE thing — just with the scenario swapped from the client’s business to the client’s R&D workflow.

For people already using these tools to build things, this is the highest-conversion-rate entry point.

III. Skill Requirements: Layered by Salary

You don’t need to dig into the English terms in parentheses right now, but you’ll need them when searching for jobs and writing your résumé — this is how job postings are written.

What isn’t in the table, yet is the most valuable, comes down to two things: the ability to measure results, and the ability to ask the right questions. Job postings repeatedly mention “effect evaluation” and “tracking business metrics.” The most direct phrasing is “not satisfied with ‘delivered,’ but pursuing ‘targets met’”; another requirement says “able to understand the pain points clients don’t voice.” Only if you can measure results can you charge by outcomes; if you can’t, you can only charge by the hour — this is both a technical capability and a pricing capability.

The judgment of a team that’s been doing this for three years: the scarcest skill isn’t programming — it’s the ability to dive into an unfamiliar business and quickly understand how it works. Even the 00s-born French major on their team can pick it up. Technology gets you in the door; business understanding is the ceiling.

IV. What Interviews Cover, and What Employers Look For

In interview debriefs, interviewers spend almost the entire hour drilling into project details — testing whether the things you claim to have done can withstand scrutiny.

Preparation method: for every project you’ve done, interrogate yourself five levels deep — why did you design it this way, what other options existed, why didn’t you choose them, what went wrong after launch, and how did you fix it. The places where you can’t answer are exactly where interviewers will dig.

Image placeholder | Five levels of self-interrogation (see Guide Section C · Illustration E)

What employers are looking for. Reading between the lines of a dozen job postings, only three things keep recurring:

Can you turn something vague into something clear, can you take ownership all the way through, and can you communicate effectively with people who don’t understand technology.

They write out technical requirements at length, but those are filtering criteria; the three things above are the actual reasons for hiring.

Prepare a ten-minute verbal pitch. Pick a scenario you know well and practice until you can clearly explain who the client is, what the real problem is, which part you’d tackle first, how you’d measure results, what the risks are, and how humans and AI would divide the work. Being able to deliver this smoothly is more useful than memorizing ten technical terms.

V. What I’m Currently Exploring

I’m including this to make a point: you don’t need a job offer to start being an FDE — what you do is what determines whether you count as an FDE. Given my dual background as a designer who understands AI, I’ve been able to find some relatively unique scenarios.

1. Technical support for architects and artists, including creative work and portfolio websites

These clients almost never use specific technical terms — and when they do, you can only understand them if you’re a practitioner yourself.

An architect won’t say “I want a responsive portfolio website.” He’ll say: “I want a portfolio site… something that makes it easy for me to show my projects, works, and ideas to clients or peers, instead of sending a pile of drawings and WeChat article links.”

An artist won’t give you technical specs. She’ll say: “I want a clean atmosphere that expresses my thinking about nature and humanity — no, that’s not quite the texture I’m going for.”

The need in their head is a feeling, not a specification. My job is to translate that feeling into specifications.

The approach is exactly what I described earlier: don’t ask “what do you want?” — ask “what you’re seeing right now, what’s right, what’s wrong, where’s the gap?” Use samples to converge, not language. Producing a rough version for them to react to is more effective than three requirements meetings.

What this taught me: when a client can’t articulate their needs, it’s not their problem — it means you haven’t found the right way to ask.

2. SEO content optimization for businesses

I built my own workflow that produces content on schedule, reviews results regularly, and optimizes continuously.

This is already a complete FDE delivery: embedded in the client’s existing rhythm, using real data, starting with the minimal chain of “one piece of content from topic to publication to data collection,” the review process itself is measuring results, the workflow is an asset that can be reused for the next client, and because there’s data backing it up, I can negotiate pricing beyond hourly rates.

The most valuable thing here isn’t “I know SEO” — it’s turning something that depended on intuition into a workflow with metrics, review cycles, and the ability to be run by someone else.

Next time I take on a new client, after a brief conversation, this entire pipeline can start running again immediately.

VI. So, Who Is FDE Not Suitable For?

Job ads won’t mention these.

First, travel and on-site work are admission requirements, not bonus points. Job postings literally include phrases like “remote work not accepted,” “overtime expected,” and “willing to travel and be based on-site.” If you can’t accept this, there’s no point discussing anything else — FDE genuinely means going to battle with customers in person at their site.

Second, this role isn’t for people who want to build deep technical expertise. You’ll be adapting to other people’s environments long-term. Three years from now, you might be familiar with five outdated systems across five industries, but you’ll struggle to say which technical domain you’re an expert in.

Third, you may have zero authority at client companies. Everything gets done through influence. People who can’t handle that position will be miserable.

Two more subjective points, from a practitioner quoted directly:

• If you think this job makes money fast, you probably shouldn’t enter — it’s hard to understand human nature that way.

• People who only want a stable job shouldn’t force it either. FDE is ultimately a transitional role — the endgame of FDE is eliminating its own position.

Self-Check Before Sending Your Résumé

• Have I prepared answers to the three identification questions from the previous article?

• Does my portfolio include something “distilled that can be reused”?

• Can I name one achievement measured by outcomes rather than hours?

• Have I actually completed something with AI tools, or only used them for chat?

• Is “willing to travel” a genuine acceptance or just a verbal one?

VII. Is FDE Worth Entering?

FDE is an intermediate-layer role that will likely disappear — but not necessarily because AI matures

I originally thought the answer was simple: once the product matures, the gap narrows and headcount shrinks accordingly. The company that invented this role is itself the example — after the product got good, large numbers of FDEs moved into product engineering.

Later I came across a more realistic view: this model simply can’t scale in most Chinese companies, for three reasons, each more fundamental than the last.

First, FDEs are there to change processes — but in a company, only the boss can change processes. An outside hire with no administrative authority, no budget authority, and no personnel authority needs a sign-off at every step. The people who don’t sign aren’t necessarily opposing you — they just don’t want to back your success. If something goes wrong, the responsibility falls on them; if it succeeds, it may not even count toward their performance anyway.

Second, people who understand business and technology and know how to drive organizational change are scarce in the market. Someone with both skill sets isn’t going to accept an implementation-level salary to do the work of pre-sales, delivery, training, and operations support all at once. The compromise usually ends up being someone who understands only half of each side.

Third, and most fundamentally: what counts as success is never defined. Does “AI is running” count as success? Does “business metrics improved” count? Does “reduced headcount by a few people” count? If it wasn’t defined at project initiation — and that consensus itself is extremely hard to reach because every department wants different metrics — then FDE work eventually becomes something that no department’s performance can absorb. Nobody adopts the project, nobody maintains it, and three months later everything reverts to the starting point.

So the temporary nature of this role doesn’t just come from products getting better — it comes from projects dying.

There’s another layer of change moving even faster: today’s hardest technical barriers are precisely the ones tools will digest first. Deployment, tuning, environment adaptation — these can filter out half the candidates today, but they all have standard answers, and problems with standard answers are exactly what tools solve first. What remains is understanding the business and designing solutions, and there are no standard answers there.

What Do You Actually Gain from Being an FDE?

One practitioner’s phrasing — I think it’s the most accurate statement about FDE:

What it shares with To B sales is that you run through the entire process end to end; what’s different is that you get to be the client-side. If you can’t understand what “being the client-side” means, then it’s just outsourcing in a different wrapper.

The same person said the biggest gain from FDE isn’t money — it’s cognition, perspective, communication, judgment, and understanding of human nature. Management trainees used to spend six months in one rotation; now you can see a company’s entire business process, profit model, and organizational structure in two months.

He also gave a concrete exit path: find mid-tier companies in a specific industry, build an efficiency solution, then use that semi-customized solution to systematically win one similar company after another.

Summary

The previous article’s conclusion: half is temporary outsourcing, half is real AI implementation, and you have the means to tell the difference on the spot.

What this article adds is a second layer: even if you identify correctly and get into the right company, this job will most likely come to an end. Whether you have something in your hands when it ends is the real dividing line between outsourcing and real implementation.

After outsourcing, all you have is your hours — you leave empty-handed. Someone who genuinely did the work walks away with an understanding of an industry, a methodology of their own, and a network of people who’ve seen your delivered results. Both are roles that will disappear — but one resets you to zero when it’s gone, and the other upgrades you.

One more thing worth thinking about: the more popular FDE becomes, the more it signals that AI implementation capability is still lacking.

Companies fighting over talent, governments building training channels, salaries multiplying within a year — these look like indicators of industrial prosperity, but they’re actually a ledger of debt. The models are powerful, but they still can’t read the screenshots inside a law firm’s Word documents, can’t handle the offline server in a client’s machine room, and can’t judge which work should remain with humans. Behind every FDE role is something AI should have done well itself — but didn’t.

So it’s more like a thermometer, measuring how far AI still is from reality. The day AI can cross that gap on its own, this role will quietly disappear — not as a failure, but as a success.

The hype is high right now. Get in while the heat is still on, learn the industry landscape and firsthand experience, and turn them into your own assets.

🥳Thanks for reading this far. I’m A哲, former architect → AIGC designer & architect

Follow me @Formulasearch — I’ll keep sharing practical AI prompts, tool tutorials, and hands-on experience.

Appendix: Data Sources

Primary data: Approximately 55 active job postings manually searched across BOSS Zhipin for Chengdu, Shanghai, Hangzhou, and Shenzhen, plus 12 complete job descriptions; 3 complete FDE role descriptions and requirements from Tencent’s official recruitment website.

Policy information: Shanghai Municipal Bureau of Planning and Natural Resources, The Paper (澎湃新闻), Sina Finance.