FDE job postings grew 4.6x in a year: what a forward deployed engineer actually does at the client's site
Companies are hiring Forward Deployed Engineers more and more often: engineers who work at the client's site and write code for its specific tasks. According to TrueUp, there are about 1,500 such openings. Over the year they grew 4.6x, while all engineering positions grew only 2.5%. The author of the analysis bets that competition here is lower: not every developer wants to deal with a client's dissatisfaction. So for interviews, it's worth preparing a story about a deployment, not just the code.
AI-processed from Habr AI; edited by Hamidun News
While some developers wonder how much of their work AI will take over, the market is quietly growing demand for another profession. In a November review, Lenny Rachitsky, citing a TrueUp sample, mentions about 1,500 open Forward Deployed Engineer (FDE) positions. Over the year their number grew 4.6x, while across all engineering roles in the same sample the increase was only 2.5%. The author of the Habr analysis, a former programmer now working as an IT manager, is making this a personal career bet. We retell his arguments and separately note where this is opinion and where the figures can be checked.
What an FDE does at the client's site
An FDE works next to the people who will use the system. They sit at the same table as a factory engineer, see where that person loses time, and start writing code the same day. Then they come back and check whether the solution helped. This is a model Palantir used long before AI agents: one of its engineers recalled living in Toulouse for a year and spending four days a week at the Airbus plant. He brought work orders, data on missing parts, and the history of quality remarks during A350 assembly into a single interface.
A recent example is OpenAI. In mid-2025, about 15 of its FDEs worked with a large semiconductor company on at least ten use cases: generating tests to verify chip behavior, finding the causes of failures, and an architecture assistant. Within three to six months, the solutions reached production use. By OpenAI's own estimate, the time spent preparing targeted tests dropped by roughly 48%. When Codex started handling most of the work, the team moved the solutions onto a shared foundation and removed about 90% of its own code from each project. In one example, the first deployment took two months, while connecting a second team took one day. These are OpenAI's data, and it's worth discounting for marketing.
The practical lesson is that it's the code that gets thrown away. What remains is an understanding of which error the test must catch. Serval reached a similar conclusion: its FDEs added more than 60 integrations to the product, along with a mechanism through which users rate the agent's work, that is, they moved the repeated parts into the common product.
Who pays and for what
TrueUp tracks more than 9,000 tech companies, and in November they had over 65,000 open engineering positions. Against that backdrop, 1,500 FDE openings is a small niche. Some companies may call the same role something else, such as sales engineer. The author doesn't know how many resumes arrive per such opening, so "less competition" remains a hypothesis.
Three openings from the analysis show how differently one profession can be named:
- OpenAI, San Francisco: deploying LLMs at the client and production code for the client-side and server-side. Requires 5+ years of experience, three days a week in the office, up to 50% travel, salary $185–300K plus equity.
- Anthropic, Munich: Python applications in the client's infrastructure, connecting Claude, developing and evaluating agents. Requires 8+ years of experience, fluent English and native-level German, 25–50% travel.
- X5 Tech, Moscow: tasks for retail divisions, Python and React, CI/CD, releasing and supporting its own applications. Requires at least three years of Python. The listing is already archived.
The common threshold is one: write working code, embed it in someone else's system, and explain to the business the terms under which they can use it. What differs is who uses the AI. At Anthropic the engineer evaluates the model's behavior; at X5, AI is the developer's own working tool. That's why the author advises asking at interviews: what did the engineer write personally, who uses the result, and whom do they call when everything goes down.
How to prepare for the interview
An example task the author used for rehearsal: an assistant generates a test for a queue of four elements. A test that puts one element in and takes it out passes on both the correct and the faulty version. The right test puts in A, B, C, D, tries to write E, and checks two things: the write is rejected, and reading returns exactly A, B, C, D. The faulty version accepts E and returns B, C, D, E. The candidate is asked to show which check the error fails on and which line of the specification explains the expectation.
The second question is about economics. If generation has sped up but every test still has to be checked by the most expensive specialist, there is no gain. You need to measure the total time to an accepted test, including fixes.
The author identifies eight skills, from debugging and integrations to agent evaluation and ownership of launches. To prepare, he advises taking your own project and finding the places where you can't show your personal contribution. He has published his material on GitHub (AI Interview Atlas) and included a prompt so that an AI agent can run a 25-minute mock interview.
The limitations are obvious. This is a personal text with a career stake, not a market study. The TrueUp figures come from a single sample, and OpenAI's results are its own estimate. The check, however, is simple: go to the person for whom you built the last feature and see how they use it. If you're interested in digging through their old Excel files, an FDE role might suit you.
- Nov 21, 2026A swimming team captain handed ChatGPT three scheduled tasks and cut her routine work by a multiple
- Nov 21, 2026Anthropic Mathematician Claims Riemann Hypothesis Proof with Claude: 1.3 GB of Code
- Nov 21, 2026OpenAI GPT-6 Cyber Found About 35 Flaws in iPhone, Mac and Other Apple Devices
- Nov 21, 202622 Claude Agents Spent Six Weeks on Black Hole Problems — and Solved Two
Key Questions Answered
What an FDE does at the client's site?
An FDE works next to the people who will use the system. They sit at the same table as a factory engineer, see where that person loses time, and start writing code the same day. Then they come back and check whether the solution helped. This is a model Palantir used long before AI agents: one of its engineers recalled living in Toulouse for a year and spending four days a week at the Airbus plant.
Who pays and for what?
TrueUp tracks more than 9,000 tech companies, and in November they had over 65,000 open engineering positions. Against that backdrop, 1,500 FDE openings is a small niche. Some companies may call the same role something else, such as sales engineer. The author doesn't know how many resumes arrive per such opening, so "less competition" remains a hypothesis.
How to prepare for the interview?
An example task the author used for rehearsal: an assistant generates a test for a queue of four elements. A test that puts one element in and takes it out passes on both the correct and the faulty version. The right test puts in A, B, C, D, tries to write E, and checks two things: the write is rejected, and reading returns exactly A, B, C, D. The faulty version accepts E and returns B, C, D, E.
Ship AI Faster with Founders, Engineers & Weekly Architecture Reviews
Inside Vibe Club (hamidun.com/club) — weekly live masterminds with Zhemal Khamidun, Tech Radar, full LMS access, and 370+ production Claude Code skills.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.