Silicon Valley's Hottest Job Lands in China: Are 35-Year-Old Programmers Entering a Golden Age?

35-year-old programmers are entering a somewhat unexpected golden age.

At China's big internet companies, 35 often means a countdown: slower promotions, higher costs, energy no longer assumed to be abundant, and technical experience that doesn't always command a higher premium. But the AI era has suddenly changed the situation for some people.

They no longer just sit at their desks waiting for requirements, writing code, and fixing bugs. Instead, they're sent to customer sites, clocking in, attending meetings, and breaking down workflows alongside business staff—translating the boss's vague request of "can we use AI to boost efficiency" into an AI agent.

This role has a name that came from Silicon Valley: FDE (Forward Deployed Engineer).

In the US, it has become one of the hottest jobs in the AI world. OpenAI and Anthropic have both bet on the FDE model almost simultaneously, establishing multi-billion-dollar enterprise AI services companies. LinkedIn reports show that from 2023 to 2025, demand for FDE-related roles grew 42-fold.

This trend has also reached China. On recruitment platforms, ByteDance's "Doubao AI Large Model FDE" pays a monthly salary of 35,000 to 70,000 yuan with 15 months' pay; Ant Digital's B2B FDE pays 40,000 to 60,000 yuan monthly with 15 months' pay; Zhipu Huazhang's FDE lead position pays 60,000 to 80,000 yuan monthly.

For the past two years, the protagonist of the AI industry has been the model. But now the story is returning to people.

People becoming valuable again is more of a signal: the commercialization of major AI companies is shifting from "selling products" to "selling outcomes." Those senior programmers once surrounded by age anxiety—if they understand code, understand business, and are willing to travel—now stand at the gateway of enterprise AI deployment.

01

Customers Want Outcomes:

FDEs Are Like "Foremen" Bringing Elite Graduates into Enterprises

Companies today certainly don't lack enthusiasm for AI. But when it comes to implementation, everyone gets stuck. The boss only says "use AI to boost efficiency," with no clear requirements.

At this point, AI is like a batch of elite university graduates—smart, but unable to hit the ground running once inside a company. The FDE is the person who onboards it.

As the human interface of the AI era, the FDE must place the digital employee into the enterprise, teach it to recognize the boss, understand the business, follow permissions, avoid talking nonsense, and actually get work done.

Moreover, now that AI agents are understood as "digital employees," companies have less patience. If you were a company boss, would you give a new employee a 3-month training period? You'd simply assume it should get up to speed faster. Ideally, it runs in two weeks, shows results in four, and can be replicated next quarter.

This is also why this role is hard to hire for.

A-Ji, a programmer who just moved from an R&D role at a big tech company to an FDE position in March this year, told us that FDEs need to be "all-rounders." Technically, they must understand data, systems, and code; on the business side, they must understand on-site problems; organizationally, they must break through departmental silos; and in delivery, they must be accountable for results. "People with strong technical skills often prefer to do R&D and are unwilling to travel long-term. Business-oriented people, on the other hand, may not be able to handle the model, data, and system issues in AI projects."

Meanwhile, Rolling AI, another rising AI enterprise services company, is looking for young people with natural business sense. For example, someone from a small business family who grew up immersed in it; or someone self-made with no connections but an extremely keen natural curiosity and a love of observing things on the front lines.

What matters for an FDE isn't necessarily age or seniority, but rather self-driven problem-solving ability and sharpness of thinking. Young talent is hard to find, but 35-year-old big tech R&D engineers who understand human nature, can communicate, and can architect systems are plentiful. In the past, their awkward position was that they couldn't stay up late like young people, and they might not want to keep doing repetitive development. But after AI can handle 90% of the code, the abilities of R&D engineers with 5 to 10 years of experience—those that previously weren't "cool" enough—become more visible.

02

New Roles, Big Companies, Hot Money,

and a Group of Chinese Believers Crossing the River by Feeling Palantir

Starting last year, several business units at Alibaba, Baidu, and ByteDance began researching Palantir. Founded in 2003, the company initially served US intelligence and defense agencies, later entering commercial markets like manufacturing, finance, energy, and healthcare. Its description of the FDE: traditional R&D is "building one capability for many customers," while FDE is "enabling many capabilities for one customer."

In 2025, its market cap at one point approached $500 billion, with its stock price rising even more steeply than Nvidia's. As of the US intraday session on July 3, 2026, its forward P/E ratio was around 145x—far exceeding the premium an ordinary software company should command. The market is betting it will be the "enterprise AI operating system."

This company, now a godfather figure for China's major AI services companies, reported total revenue of $1.63 billion in Q1 2026, up 85% year-over-year. Most of it came from large US government contracts—for example, the US Army signed a $10 billion 10-year services agreement to procure its commercial software products; and the contract ceiling for the Pentagon's Project Maven AI system was also raised to $1.3 billion in 2025.

Palantir also has a growth mechanism that Chinese AI service providers have studied repeatedly: AIP Bootcamp, a co-creation workshop with users. According to its website, customers can go from zero to a use case in 5 days. In the past, B2B sales required PPTs, bidding processes, POCs (proof of concept before development), and long-cycle pilots. A digital project's launch was measured in years. But now customers no longer believe in "what could be possible in the future"—they want to see "can it run next week."

New roles, big companies, hot money—a batch of Chinese companies have begun hoping to become "the next Palantir."

Four types of players have entered almost simultaneously:

① Cloud providers and data intelligence platforms. This is the category most similar to Palantir—and the most anxious. Alibaba Cloud, Lingyang, Baidu Intelligent Cloud, Volcano Engine, Tencent Cloud, and Huawei Cloud are certain about their investments in models but urgently need revenue. The AI commercialization of these big tech companies inevitably rests on cloud, model calls, data platforms, and agent platforms. Only after customers buy cloud, models, and platforms—and actually consume tokens—can these companies make money.

Alibaba is the most typical case. In the fourth quarter of fiscal year 2026, Alibaba Cloud's revenue was 41.626 billion yuan, with external revenue up 40% year-over-year, and AI-related products accounting for 30% of external revenue. Alibaba expects AI to become the main growth engine of its cloud business within a year, contributing over 50% of revenue. But customers won't burn tokens just for the sake of it—only when agents enter core workflows like sales, customer service, marketing, supply chain, finance, and store operations will call volumes become stable.

Baidu, whose old engine of advertising is sputtering, is even more anxious. In Q1 2026, total revenue fell 2% year-over-year to 32.1 billion yuan, while online marketing services revenue (i.e., the advertising business) fell 22% year-over-year to 12.6 billion yuan. But AI Cloud revenue grew 79% year-over-year to 8.8 billion yuan—it needs a new story to hedge against the decline of its old business.

ByteDance and Tencent are also entering the field. ByteDance's "Doubao AI Large Model FDE" role sits under Volcano Ark MaaS, requiring work with industry-leading customers, deep understanding of business scenarios, and completion of the full process from solution design to code implementation. Tencent Cloud is also recruiting an "AI Delivery Solutions Engineer (FDE track)" for the automotive industry, requiring the use of AI technologies such as large models, agents, multimodal systems, and knowledge graphs to design end-to-end solutions for automakers.

Huawei is taking a different path. It hasn't yet packaged this as a new Silicon Valley profession; it's more like a Chinese government-enterprise version of "industry corps + partner ecosystem + solutions experts." Huawei Cloud emphasizes industry-specific agents as the center, claiming it has helped over 30 industries and more than 2,600 enterprises deploy AI applications across 500+ niche scenarios; Huawei's government-enterprise business has invested over 5,000 "industry + AI" experts in training and enablement.

② Large model and AI agent platform companies, such as Zhipu AI, MiniMax, and 01.AI. They have models, APIs, and development platforms, but in the past were somewhat distant from enterprise business. However, MiniMax is already recruiting a "Large Model Pre-sales Solutions Expert," which is another variant of the FDE; 01.AI, led by Kai-Fu Lee himself, has recruited and built a strategic consulting team from top consulting firms, working with leading industry players to "find scenarios, tune models, and build applications," and claims to have full-chain service capability from strategic consulting to platform implementation;

③ Enterprise software and industry digitalization companies, which are already embedded in ERP, CRM, marketing SaaS, and supply chain systems, are the most likely to transform their existing implementation teams into AI-powered FDEs;

④ Transforming consulting firms and AI-native service companies, on the other hand, excel at going on-site, sorting out processes, and driving organizational change.

This is also the biggest difference between China and Silicon Valley.

When OpenAI and Anthropic bet on FDEs, their partners included Goldman Sachs, SoftBank, McKinsey, Bain, and other investment institutions and traditional consulting firms. AI companies have models, investment institutions have batches of portfolio companies, and consulting firms have transformation methodologies—the three parties work together to turn AI deployment into an industrialized production line.

But Chinese companies are rushing to do FDE not just to earn a bit of consulting fees, but to seize the entry point for enterprise AI.

Because once an enterprise connects its data, knowledge bases, permission systems, business processes, and AI agent platform to a particular tech giant's foundation, the migration cost becomes high. Whoever first helps an enterprise get AI running in its processes is more likely to become that enterprise's default supplier in the future.

The FDE is the person who turns a small entry point into a big foundation.

03

What the Chinese field needs is not standard answers,

but street smarts

What's truly interesting about enterprise AI deployment in China lies in the complexity of the environment: things change too fast, and much of the effective experience simply isn't in the enterprise's SOPs.

Take the chain retail industry, for example. In a chain supermarket, large-format yogurt sells poorly, while premium yogurt sells well. If headquarters only looks at the data, it might mistakenly conclude that the store is located in an upscale community. But only the store manager knows that there is another, larger supermarket nearby that sells cheaper budget yogurt but no premium yogurt—so he chooses differentiated competition. This is the "street smarts" that AI needs to learn.

Take the pharmaceutical industry, for example. In 2025, the operating revenue of China's pharmaceutical manufacturing enterprises above designated size fell 1.2% year-on-year; under the continued advancement of centralized procurement, medical insurance cost controls, anti-corruption efforts, and compliance regulation, the old ways of driving growth through price, channels, or gray areas are becoming increasingly difficult.

A large pharmaceutical company found that, despite the overall pressure, its Fujian sales region continued to grow rapidly. When executives went down to take a look, they found that this team wasn't "better at selling drugs," but better at creating professional value on the physician side: some doctors wanted to give lectures at universities, some valued departmental guidance, and the pharmaceutical sales reps became the matchmakers between doctors and universities, and between hospitals. By achieving self-interest through altruism, they then delivered commercial results.

A Ji's task was to use AI to replicate such excellent practices to more sales regions. These excellent examples need to be unearthed bottom-up by the company's top leadership, but more often than not, AI needs to first learn frontline wisdom.

The difficulty is that this can't be solved by writing a set of SOPs. Truly effective experience is often semi-tacit: Why is this doctor willing to participate? Where is the line in how a pharmaceutical rep should speak? Which actions are compliant, and which cross the line? All of this requires the FDE to go on-site and blend the jargon, processes, data, permissions, and organizational relationships together.

The FDE team A Ji was on spent 4 months officially launching an AI agent for pharmaceutical sales reps to use—which is already quite a long time. But one of those months was spent understanding business terminology and jargon.

04

Is the endgame of enterprise AI a wave of FDEs entering the field?

Will FDEs always be this expensive? Not necessarily.

OpenAI executives also admit that the FDE model is transitional: the goal is not to do the work for customers forever, but to help customers get through the early complex stage so that, ultimately, they can run on their own.

An FDE is like scaffolding. Without scaffolding, a building is hard to construct; but a truly completed building shouldn't keep its scaffolding up forever.

The difficulties enterprises face in using AI really come down to four things: enterprises don't know what they truly want; data can't be used directly; real processes aren't the processes written in the documentation; and an AI agent isn't finished at launch—launch is just the beginning. So the essence of the FDE's emergence is not "enterprises need more programmers," but that once enterprise AI enters real business, someone needs to act as the translator.

For 35-year-old programmers, this is of course not a salvation in the universal sense. Not everyone surrounded by age anxiety can land a million-yuan salary through FDE work. It simply allows people to turn all those years of meetings attended, pitfalls encountered, blame shouldered, and legacy systems maintained into a new kind of capability.

It sounds a bit darkly humorous.

But the AI era doesn't only reward the young, the cheap, and those who can pull all-nighters. The people who understand chaos best may really see their value rise first.

This article is from the WeChat official account "Yicai Business School", author: Yicai Business School, Wu Lingwei, published by 36Kr with authorization.