Silicon Valley's hottest new job has drawn the kill line for careers in the AI era

If a traditional bank wants to use AI to automate the work of its veteran analysts, what should the first step be?

A company has already purchased AI services and built its own AI system, but reports that it's completely unusable—is the problem with the AI model, the data, permissions, or a wrong choice of workflow?

Management wants to let an AI agent directly modify the existing production system, but the security team firmly objects—whose side should the engineer take?

These questions come from an Anthropic FDE interview guide posted on Reddit, which fully describes what a candidate should best prepare in order to land the Anthropic role known as FDE.

What is an FDE, and how did this recently famous role in Silicon Valley suddenly become a hot item on AI companies' recruiting pages?

Forward Deployed Engineer is the full name of FDE, often translated as Forward Deployed Engineer or Forward Deployed Engineer. These engineers must both write production code and understand AI agents, evals, and security; they must both go on-site with customers and decide which problems are worth solving.

OpenAI has listed Forward Deployed Engineering as a separate team, hiring in San Francisco, Seattle, Tokyo, Seoul, Singapore, and elsewhere; Anthropic calls its new members "founding FDEs."

Major Chinese companies like ByteDance and Tencent also have related FDE positions. Tencent posted three Forward Deployed Engineer roles in its social recruiting, all requiring three to five or more years of work experience, under the product job category.

ByteDance simply calls it AI FDE Engineer, without using a translation like "Forward Deployed Engineer," placing it under the R&D job category, with openings ranging from Feishu and Volcano Engine to data platforms.

There are even FDE internship positions, but ByteDance's FDE interns fall under the "Sales - Sales Support" category.

We keep hearing that AI will only take away human jobs—so what exactly is this FDE role that has risen because of AI, and can I apply for it too?

Up 42x since 2023

According to LinkedIn's January 2026 Global Workforce Report, Forward-Deployed Engineer has grown 42x since 2023, while AI Engineer grew 13x over the same period.

According to data from the recruiting platform Indeed, in April 2025 there were only 643 FDE postings site-wide; by April 2026, that number had become 5,330—more than 7x in one year. The Wall Street Journal directly called it "the hottest job in tech."

It sounds trendy, but this role has nearly 20 years of history. It originated at an American company dealing with customers whose data required strict confidentiality and who couldn't articulate their specific needs—so the company simply "deployed" engineers on-site at the customer.

In their own words, the essence of the role is: an ordinary engineer uses one skill to serve all customers; an FDE uses all skills to serve one customer.

On the company's job posting page, they also write: "We pioneered this unique role, embedding brilliant engineers directly into customer teams to tackle their most pressing challenges."

Twenty years later, all kinds of large models are omnipotent in demos and in the hands of individual users, but they struggle badly once connected to enterprises' real systems. To truly bring AI applications to a larger market, the "last mile" seems to be the biggest reason companies are hiring FDEs.

So a complete FDE needs to participate in problem discovery, technical scoping, architecture design, development, launch, and user adoption. After a project is done, they must also judge which parts apply only to that one customer and which should become shared components, reference architectures, or core product capabilities.

This approach once sat somewhat awkwardly with the standard SaaS model. Software companies pursue build once, sell many times, while FDEs put expensive engineers back into every customer project. All of this became justified because AI made this investment in human labor worthwhile.

What does the FDE interview test?

From the JD listed in Tencent's recruiting, you can see the specific responsibilities, requirements, and bonus points.

Generally speaking, FDEs need a bachelor's degree or above in STEM fields such as computer science, software engineering, math, physics, or AI, plus 3+ years of engineering or toB delivery experience.

Most importantly, they must master the various AI application tech stacks and AI agent products, and be proficient in the AI world's professional jargon like RAG, prompts, model selection, harnesses, fine-tuning, Skills, MCP, multi-agent, and so on.

The specific work content is similar to the FDE job details described above: going on-site with customers, turning vague business requirements into AI systems that can go live, driving real customer adoption, and being accountable for delivery results and ROI.

One more thing: having hands-on experience building OpenClaw or other AI agent tool projects counts as a bonus. The OpenClaw guide we once wrote, from installation to uninstallation, can come in handy here.

Swipe up and down to see more content

The interview guide posted on Reddit goes further, dividing the Anthropic FDE interview process into roughly five rounds, each answering the same question: can you safely integrate Claude into any enterprise's business?

The HR screening already requires you to explain why you chose an on-site role instead of traditional R&D—your motivation outweighs everything.

Next comes the technical scenario interview, centered on deployment walkthroughs of toolchains like Claude + MCP. This round mainly examines how candidates manage the context window and handle unreliable ultra-long contexts.

Anthropic may directly give interviewees access to Claude, letting them answer while using it.

Since the title says engineer, the coding round is still indispensable. Beyond grinding lots of LeetCode algorithm problems, the guide recommends targeted practice, like building a Claude rate limiter, refactoring a retry queue, building a small connector with a mock database, and building a task scheduling system.

After the technical round comes the customer simulation round, rumored in the industry to be the most brutal stage. Some organizations estimate that about 60% of candidates who pass the technical round fall here.

The interviewers play non-technical executives and picky architects, and candidates cannot open a code editor—they must complete requirements discovery entirely through questioning.

Finally comes Anthropic's signature round, widely considered the hardest of all. It was also hotly debated on social media recently—Anthropic's CEO worried that the people he hires care only about money, not the mission.

So this round's questions dig into moral conflicts, pressure from superiors, and "how did you feel at the time." Foreign media reported that some interviewees are directly asked, "If the company's mission conflicts with making money, how would you choose?"

How would you feel if the company one day abandoned its AI initiatives for safety reasons—and that decision caused the company's stock price to fall to zero?

The Anthropic job candidate interviewed also asked foreign media not to disclose any of his information, fearing it would affect future job opportunities.

Anthropic is also hiring an "Insider Risk Investigator" responsible for monitoring and classifying external threats against employees

Setting aside Anthropic's pursuit of its mission, what the FDE role actually assesses draws from algorithm engineers, Prompt engineers, and pre-sales consultants alike—it combines product manager, Solutions Architect, full-stack engineer, and customer delivery lead into one position.

As for compensation, the FDE annual salary at Anthropic is $1.88M–$2.15M, at the high end of applied, customer-delivery engineering roles, with the ceiling just touching the starting point of senior software engineers and regular software engineers; research engineering roles in core models, training, and evals still pay more.

Talent in the AI Era

The talent of the future AI era won't be those who are best at writing Prompts, but those who understand industries and know how to integrate AI into the traditional workflows of different industries—they are the FDEs.

There are many similar comments online, especially under various posts explaining or training people on FDE.

Remember when Prompts first became popular—the most sought-after people were those who could write Prompts; when Harness became popular, the most sought-after were those who could build Harnesses; now it seems the most sought-after have become FDEs.

Prompt engineer, Harness engineer, FDE—job titles change round after round, yet what AI needs to break through becomes ever clearer: getting AI from demos into production environments, and being accountable for the final results.

A while ago, Andrew Ng shared on X the four AI skills individuals need to master in the AI era. He noted that this AI engineering skills map came from analyzing a large number of job postings, expert interviews, and surveys.

The four skills are building and deploying AI applications, software engineering fundamentals, using coding agents, and Shaping the Build—that is, understanding business and customer goals and participating in deciding what the product should do.

The work of an FDE covers almost all four of these major capabilities at once: software engineering fundamentals get systems into production, AI application skills handle model uncertainty, Shaping the Build clarifies customer goals, and Coding Agents speed up on-site trial and error.

As for the first skill, his recent article also pointed out that the essential difference between AI applications and traditional software lies in unpredictable outputs.

We don't know in advance what an LLM will spit out; building AI systems is a highly iterative process, and the hallmark of an AI expert is "being able to skillfully decide what to do next based on intermediate results."

So he broke this skill into six modules: large model fundamentals, using data to provide context for models, agent systems, eval-driven development, production operations, and machine learning fundamentals.

For ordinary people, the FDE, like this AI skills map, is like a capability manual for the AI era. We need to find a problem worth solving, use AI to turn it into a deployable system, and get real users willing to use it.

The AI industry is willing to pay a premium for FDEs precisely for these abilities: turning uncertainty into productivity amid other people's messes.

In the domestic job market, FDE may be hard to roll out as comprehensively as in Silicon Valley, but we may not need to become FDEs—rebuilding our own capabilities according to FDE standards is enough.

This article comes from the WeChat official account "APPSO", author: 发现明日产品的, published by 36Kr with authorization.