The Hottest Job in Silicon Valley Is Now in China: Grueling, Painful, and Always Putting Out Fires

FDE｜AI Implementation｜Three Cases｜Eight-Figure Deals

Produced by｜AI Nao

Intro

Another buzzword has crossed over from Silicon Valley to China: FDE, short for Forward Deployed Engineer.

The Valley narrative: a new breed of professional who combines technical chops, business acumen, and consulting skills — embedded on-site to shepherd companies through AI transformation, "the last mile of AI implementation."

FDE-related job postings in the U.S. have surged 800% year-over-year. Publicly available figures put senior FDE compensation at roughly 1.5–2 million RMB annually. OpenAI and Anthropic have teamed up with top PE firms and investment banks to launch FDE practices directly. Back home, tech giants like Volcano Engine, Tencent Cloud, and Alibaba Cloud are building out their own FDE teams or farming out FDE delivery to partners.

Like the Agent engineer of a year ago, FDE is being billed as "the sexiest engineering role of the AI era."

But the discourse so far lacks voices from the front lines. In China, what does someone actually doing this work on the ground think of the role? What have they been through? What do the results actually look like?

"FDEs in China are more like outsourced contractor labor. After the big tech companies sell the product, we're the ones doing the dirty work." That's Shen Yue speaking, the narrator of this edition of "AI Practitioners." One of China's earliest frontline FDEs and a former ByteDance employee, he started taking on FDE projects in early 2025. "Back then we weren't even called FDEs — we were 'AI consulting partners.'"

Contrary to the industry's hype around FDEs, his experience carries an almost "anti-climactic" drama. He's been embedded at four companies, private and state-owned alike, with project values ranging from nearly a million to over ten million RMB.

From this vantage point, he's observed the distinctive, uneven landscape of Chinese enterprises. "The power struggles inside an organization are always more complex than the technical solution. Where there are people, there's politics — what would AI know about that?"

What follows is Shen Yue's account, in his own words. Same role, FDE — China versus Silicon Valley. Two different worlds.

A Week as Firefighter FDE

Last week, I got dispatched to a large state-owned enterprise to put out a fire.

The project was an eight-figure deal signed in Q4 last year, with a commitment to deliver ten agents within six months. But nearly a year later, it still hadn't passed acceptance — severely paralyzed. The client made it clear: they were dissatisfied.

My firefighting mission: figure out "exactly what standard our agents need to meet" for the client to sign off.

Day one, I diagnosed the problem. The FDE contract had been signed by the SOE's top boss, but once inked, he handed it down to a subordinate with no guidance on which business scenarios to target. He'd simply bought into the AI concept — the enterprise needed to do this.

But the person assigned to execute couldn't articulate specific requirements either. SOE culture is risk-averse, so he told the FDE team to go figure it out themselves: find the needs, propose solutions, build them. So for the past year, the FDEs had buzzed around departments like headless flies, "getting lucky" with conversations here and there, ultimately forcing out seven agents. I pulled up the product — unusable, really. No clear stakeholder, no business value to speak of. No wonder the big boss wouldn't sign off.

Worse, the project lead originally appointed by this senior executive had, for reasons unknown, refused to manage us and instead set up his own AI transformation department within the company — essentially doing the same thing we were supposed to do.

So my counterpart became an IT manager reassigned on short notice. We met in a small conference room. He was direct: he was purely "taking the fall," saddled with extra work out of nowhere. His goal was to get me to close the project quickly without causing him trouble.

Privately, he suggested that since we both wanted this wrapped up, we could reframe the existing materials into "presentation logic" — demonstrate value to the boss with a demo, some before-and-after usage data, show comparative improvements.

It hit me: learning how to present should be recorded as an FDE's core competency.

So I asked if he could connect me with some business folks to get real data. He shut me down immediately — no, he couldn't disturb the business team. I'd have to figure it out myself.

Where was I supposed to find data? (Laughs.) I was suspended in a vacuum layer, unable to reach up or down.

So I set my goal: stop fantasizing about solving everything, and just list what could actually move forward this week.

If this were a success-story article, this is where the big reversal would happen: through my efforts, the client is moved, the proposal sails through, the project enters a new phase.

But reality doesn't work that way. After a busy week, the client's verdict: this demo still isn't very convincing, needs more thought.

Honestly, pretty demoralizing. It reminds me of the mobile internet era, when companies pursued "digital transformation" — paperless offices, buying systems, building products. The boss said to use them, but didn't specify how to evaluate; middle management didn't want to execute. In the end, the systems became "tools for filling out reports to leadership" or simply fizzled out.

The paralysis logic here is identical — organizational authority and responsibility were never clarified, nobody wanted to take ownership. SOEs, you know, everyone's afraid of making mistakes.

Outdone by a 60-Year-Old Principal's Three-Day Hack

I started doing FDE work in the summer of 2025, initially called "consulting partnership." Then the concept came from Silicon Valley: FDE. The Valley really knows how to coin terms (laughs).

Frankly, I feel like our work is still outsourced contractor labor — pre-sales pitching, sales closing, post-sales delivery. Just wrapped in an AI shell, same business model. Sometimes when we're bidding, the competitors at the same table are the same companies that used to do digital transformation for enterprises.

The only difference is, before, saying you did digital transformation couldn't even get you in front of the boss. Now, saying you're an FDE, you can at least get a conversation.

My first FDE client was a private education group with roughly 1 billion RMB in annual revenue. They'd bought Volcano Engine's agent development platform, and we started by teaching them to build on it. The company even ran a bootcamp, screening over 100 teachers and staff to participate — all complete technical novices.

But that bootcamp genuinely shocked me.

One woman in her twenties, in finance, hacked together a financial reconciliation agent on her own — an incredibly complex workflow that took me two hours to barely follow. But it ran flawlessly. Later I heard the company adopted her product and promoted her.

Another was an art teacher. He built a live-streaming customer service Q&A agent. His product documentation was better than most PMs'. Later I heard Geely poached him as an AI lead. I specifically asked him — any relevant background? He said no, he studied fine arts in college, was an art major.

AI will definitely unleash tremendous human potential. Eighty percent of product managers will be eliminated.

After the bootcamp came the FDE work. The company had a clear pain point: they frequently took over schools, and each takeover required a research report involving 4-5 principals on-site for at least two weeks. They wanted AI to improve efficiency.

Our team of three to four people worked for three months, ultimately delivering four agents covering intelligence gathering, data organization, report generation, and principal recommendation — the full chain, reducing overall time from two weeks to three days.

My first FDE project, I felt pretty accomplished when we finished. But after actual rollout, nobody used it.

The company required principals to digitize everything — recording, form-filling, uploading. But most people didn't have this habit, found it troublesome, leading to incomplete inputs and subpar output quality.

One principal in his sixties, who'd attended my bootcamp, whipped up a simplified agent in three days — just template generation, essentially partial AI-ification, with the rest still manually filled. He shared this agent with other principals, and everyone loved using it, found it much better.

Our professional team, hundreds of thousands of RMB, three months of work — outdone by an old principal's three-day hack. That was pretty discouraging (laughs).

Looking back, the biggest problem was "over-engineering."

Why did the old principal's product work? Because he didn't demand other principals change how they worked — zero barrier to entry. We were too fixated on "full-chain AI-ification," ignoring users' actual scenarios and habits.

This led me to reverse-engineer a detail: our initial research didn't talk to frontline principals but to the company's digital team — we were separated by a layer. This is also the FDE's biggest challenge: you might never actually meet the real stakeholders.

The client did pay in the end, but from their internal evaluation: the project was a failure.

I thought it was pretty much a failure too.

"Help Me Cut 3,000 Heads Directly"

Most private company bosses only care about one thing: I've spent several million — can you help me grow revenue or cut headcount? As for AI or not AI, honestly, they don't care much.

I remember a private home decoration company in Guangdong, hit hard by the real estate downturn in recent years, revenue consistently falling. The father didn't have the energy to reform, so the second generation took over — a post-85s woman, bold and aggressive, going all-in on AI. At our first meeting, she was very dominant, stated her demand directly: the company currently has 6,000-plus people, after FDE comes in, help me cut to 3,000.

Later I heard her husband also worked at the company, as HR head. The couple was pushing this together.

This project ran three months. It was brutal. I flew back and forth between Beijing and Guangdong four times — travel costs alone made the project unprofitable. Everyone thinks FDEs earn well; our real income isn't high. Many projects are pure loss-leader engagements (laughs).

Every progress review, I'd get chewed out. She was dissatisfied with pace, dissatisfied with value demonstration. Every time: I want to see value as in, how many people this agent can eliminate after going live. We kept trying to talk her down, lower expectations.

How could AI suddenly replace 3,000 people? We eventually deployed seven agents, mostly quality improvement, with only one sales agent that might drive revenue growth.

The bosses I see are all anxious, rushing to have AI replace people. But my understanding is, AI can currently replace tasks, not jobs.

Take legal roles: AI can take over "contract review," but that job also includes attending negotiations, organizing archives, handling emergencies — a whole series of tasks.

Only when AI can handle 80% or more of a role's core work can it fully replace a person. We're far from that.

Also, most enterprises I've engaged with haven't even laid a proper digital foundation. However powerful AI is, it can't land — AI actually demands extremely high organizational maturity from enterprises.

Doing FDE in China, I've summarized a few principles:

1. SOEs: it's about managing people; private enterprises: it's about getting things done
2. Best to have direct access to whoever signed the contract or whoever calls the shots
3. The enterprise needs some digital foundation
4. The boss needs some patience, not expecting results in three months

In the short term, FDE work definitely has value. But long-term, it's a transitional role.

On one hand, large enterprises will grow their own AI implementation capabilities internally. On the other, many FDEs in China are being handled by cloud vendor partners — too many layers in between, dooming some projects to failure from the start.

Frontline sales makes too many unrealistic promises to move product; we FDEs have to fill the holes, clean up the mess, do the dirty work. Nothing like Silicon Valley, where an FDE can tell a client directly: this won't work.

Oh, and I just got notified — next week I'm heading back to that large SOE to put out the fire again. Apparently there's been a turn of events.

The guy originally responsible for our project, the one who set up his own department? The big boss chewed him out and told him to clean up his mess. So after a full circle, we're back to square one. Let's see if we can push acceptance through this time (laughs).

But no matter how AI evolves, one thing won't change: SOE or private, the power struggles inside an organization are always more complex than the technology. Where there are people, there's politics — what would AI know about that? (Laughs)

Cover image｜Unsplash

Join Our Community