Should you switch to FDE? A $173,816 median salary, and travel in 68% of roles
The salary is only half the story. The other half is how many days a month you get to spend at home, and where the role leads next.
The newspaper of the Forward Deployed Engineer
Editor, FDE Times
Editor of FDE Times, the publication on Forward Deployed Engineering: what the job is, who is hiring and how developers move into it.
The salary is only half the story. The other half is how many days a month you get to spend at home, and where the role leads next.
No system can keep the promise that it will never go down. This 2017 book shows you how to turn that promise into a number that you and your customer can both measure.
Extra weeks of study do not turn into years of experience. Aimed at the right hiring tier, though, they can build the evidence recruiters actually check.
At the same FDE level, base salary barely moves, but equity can differ by as much as $200K. That makes where you start the conversation matter.
Harrison Chase and Rotem Weiss's short course teaches the two things clients ask about first: where the agent stops and who gets to approve. Resuming after a failure gets no more than an introduction.
Evidence that you can handle ambiguous problems can start in your own company's support queue, before you send a single application.
A model with an excellent F1 score can still sit untouched in a notebook. For a forward deployed engineer, the real work starts when that model has to be connected to the customer's daily operations.
On a customer site, one wrong field in an API response is enough to push an agent into a bad decision. A Zod schema at the entry point catches that error before the agent ever reads the data.
Once the contract is signed, your demo meets the real world. If the system goes down at 2am, the person who gets out of bed is the FDE, not the SE.
Most FDE job ads ask for three to five years of experience. There is still a way in for newcomers, but getting through it means proving you have worked with a real customer, even on a small project.
At your old company, you decided the schema. At the customer's site, the schema was there before you arrived, and it does not care what you prefer.
Two years of writing production code gives you most of what you need. Many candidates still fail because of a backend habit: drawing the architecture before asking the customer what "done" means.
The company that created the FDE title once paid below market, yet it kept strong engineers for years and produced a stream of founders. The real risk in the job lies elsewhere.
Set aside the 145 roles from newly tracked careers boards, and the count of open Forward Deployed Engineer roles changed by -8 between 7 October and 11 October 2026.
Palantir asks for Python, Java or C++. Anthropic asks for strong Python and experience shipping production applications. Your consulting experience only counts when it comes with code you wrote yourself.
Two hiring datasets give two different remote shares. Five specific job ads all show the same thing: for a forward deployed engineer, "remote" rarely means you can skip the customer.
A survey of about 1,000 FDE job postings shows the role meets customers more than anything else, yet no posting assigns it revenue responsibility. That detail is the key to reading a job description correctly.
Once the contract is signed, one role starts to step back. For the other, the main work is only beginning.
On a plant floor, operations teams don't ask how clever your model is. They ask whether it can write to the PLC, and the answer has to be visible in the design itself.
With no network, one file missing from the USB stick means a drive back to the plant, and a broken update can leave the line running with nobody checking for defects.
A night-shift technician needs only one manual excerpt stripped of its model name, one superseded SOP or one error code the embeddings miss before they stop trusting the assistant.
The count reached 440 on 9 October 2026, but 131 of those roles came from 35 newly tracked careers boards, while 12 roles closed and 3 new ones appeared.
However good the anomaly detection model, it is useless if every alert becomes a maintenance order that nobody trusts.
In a factory, an FDE's first job is not to pick the best platform but to locate the problem: automation code, the equipment data model, or the data path at the edge.
Agents are good at making decisions but don't know how far a machine can be pushed. Digital twins understand the physics but usually just watch.
PostHog's public handbook does not explain the reasoning. Its pricing rule still makes it fairly clear what customers are paying for now that writing code has become cheap.
AI now makes every CV look polished, so FDE hiring managers want evidence that holds up when they dig into it.
The course launched in 2019 for a reason tied to job ads. It remains one of the cheapest ways to learn to read a client's JavaScript codebase.
All three coding assistants can be restricted. Their defaults differ, though, and the exceptions most likely to put an FDE in breach of a client's policy are buried in documentation few people read.
A client has just sent you a folder of exported data, and the next meeting is an hour away. You need a database that runs inside your script, with no server to set up.
When a client asks "can we try it on real data?", an FDE rarely has time to build infrastructure. Modal lets you skip most of that work.
Before you write your first evaluator, read your traces, note the first failure in each one, then count. A few focused sessions give you a priority list that no dashboard can generate for you.
In a scoping session, every process sounds neat. You only see the Excel file open on the second monitor, and the shortcuts no document records, when you sit next to the person who does the work.
Batch scripts rarely fail because they run slowly. They fail because nothing holds them back: there is no limit on concurrent requests, no pacing, no timeout and no end to the retries.
Geoffrey Moore wrote this marketing book in 1991. It still explains clearly why a deployment that delights the first customer can drive the second one away.
The demo went well, so why does the chatbot still give wrong answers? The creator of the Instructor library teaches you to answer that question with numbers, not guesswork.
A candidate who took OpenAI's take-home found that most of the grading seemed to turn on whether you could explain why each choice served the customer's need, not on the code alone.
Principal FDE, product manager, engineering manager and founder all draw on the same field experience with customers. Each route also demands one skill that day-to-day FDE work barely teaches.
Neither round scores how quickly you find the bug. Both score how you track it down, and how you learn something new while someone watches.
Second Talent says two Vietnamese engineers with the same technical profile can be paid 60% apart, mainly because of their ability to work with clients in English. The available salary figures show where that gap sits.
The median base salary in US forward deployed engineer job postings is about $190,000 a year. The $550,000 figure is the top of an estimate of total compensation at OpenAI and Anthropic. Most of the difference is equity, which pays out slowly and is hard to value.
A job title won't tell you whether you will write code or sit in meetings. Read the job description with two questions in mind: who owns the build, and who owns the outcome?
Clients rarely get angry over a bad number. They get angry when it surprises them, and when the person delivering it arrives with no options.
The SQL runs without errors and the number looks plausible. It is still wrong, because nobody told the model how your client calculates MRR.
At least one FDE job ad still uses the old name. The hard parts are the ones few people mention: where data is processed, how many RU/s Cosmos DB gets and who creates the private endpoints.
The book never mentions FDEs, but its idea of an enabling team describes much of what you do every day: help other teams past obstacles and point out the capabilities they are missing.
A script that runs on your laptop is not yet something you can hand over. It becomes a tool when the client team can type one command on their own machine and have it work.
Three sources report growth in the thousands of per cent. A curated index found only 224 open roles, though, and one author's manual classification puts about 40% of postings with this title outside the builder kind of FDE.