Career path
Forward Deployed Engineer
A forward deployed engineer (FDE) takes a customer's problem and builds software that works in the customer's real systems, not just in a demo. The job is part finding out what's actually wrong, part building the fix, and part staying responsible until it works. Use this path to see what the job involves, show you can do it, and size up a team before you join it.
Read these 12 guides in order
- What Is a Forward Deployed Engineer?
See what you'd be responsible for, and what another team handles.
- Forward Deployed Engineer vs Solutions Architect vs Consultant
See how the job differs from other customer-facing roles, like solutions architect and consultant.
- How to Transition Into an AI Career: Complete Guide
Match the experience you already have to a plan for switching.
- How to Learn AI from Scratch: Free Resources Guide
Fill in the technical skills you're missing.
- How to Become a Forward Deployed Engineer
Pick a portfolio project that shows you can get software working for a customer.
- Forward Deployed Engineer Interview Guide
Practice talking through how you'd find the real problem, design the fix and debug it.
- The Forward Deployed Engineering Playbook: Discovery to Production
Follow one customer project from the first conversation to software running in production.
- Forward Deployed Engineers for Enterprise AI: Why the Model Works
See what the job looks like when the customer is a large company with systems already in place.
- When Should You Hire a Forward Deployed Engineer?
Decide whether a problem actually needs an FDE.
- How to Build a Forward Deployed Engineering Team
Decide who on an FDE team owns what, and which work comes back every time.
- How to Measure FDE Teams: Metrics, ROI, and Unit Economics
Measure an FDE team by what changed for the customer, not just by what shipped.
- Why Forward Deployed Engineering Teams Fail
See how FDE teams fail, so you don't repeat the same mistakes.
Pay in employer job postings
Checked September 17, 2026. These are yearly ranges from 3 real job postings. They aren't market averages, the jobs aren't all at the same level, and no one is promised these numbers. Open the posting for the current range and the full package.
| Employer and role | Location | Posted range | What the range covers |
|---|---|---|---|
| OpenAI Forward Deployed Software Engineer | San Francisco | $185,000–$325,000 | Posting labels this compensation and lists equity separately. |
| Anthropic Forward Deployed Engineer | New York / San Francisco / Seattle | $280,000–$320,000 | Posting labels this annual salary; confirm the package and level with the employer. |
| Palantir Forward Deployed AI Engineer | New York | $135,000–$200,000 | Estimated annual salary; posting excludes potential bonuses, benefits and long-term incentives from this range. |
Who is hiring: Companies Hiring Forward-Deployed and GTM Engineers. Links straight to five employers. Compare who owns the build, how much travel there is, and what you're responsible for after launch, not just the pay.
What the job posts name
From 248 postings at 42 employers, read September 2026. The percentage is the share of those employers that name each one.
- Building AI agents52%
- Evals50%
- RAG26%
- Model Context Protocol (MCP)24%
- Prompt engineering19%
- Fine-tuning17%
The full Forward Deployed Engineer ranking, signature tools and new ones on the radar →
Monthly FDE hiring index
Checked 2026-09-17 (UTC). 65 matching job ads across five employers.
This counts public job ads at five employers. It isn't a count of open jobs, people hired or the whole FDE market. Two ads can describe similar work in different cities, and one ad can cover several openings. This is the first month counted, so there's no growth rate yet.
Scroll sideways to see the whole table.
| Employer | Matching ads | All postings read | Read at (UTC) |
|---|---|---|---|
| OpenAI | 17 | 817 | |
| Anthropic | 4 | 607 | |
| Baseten | 2 | 99 | |
| Clay | 0 | 58 | |
| Palantir | 42 | 313 |
Download every matching ad (CSV) · Raw data and method (JSON)
How the sample is counted
Method fde-title-sample-v1: an ad counts when its title starts with Forward Deployed Engineer, Forward Deployed Software Engineer or Forward Deployed AI Engineer, with or without hyphens. Each posting ID counts once per employer. Unlisted ads don't count, and neither do titles containing manager, management, director, head, strategist, sales, intern or internship. New-graduate roles count. Titles such as Forward Deployed Infrastructure Engineer and GTM Engineer don't match, but a specialty or location added after a matching title does. This is a rule about titles. It doesn't judge what each job involves or how senior it is. If a board can't be read, its count shows as unavailable, never zero, and no combined total is shown.
Each job board means something different by its date. Ashby gives the last time an ad was published, Greenhouse the first time (when it gives one), and Lever the time the ad was created. Created isn't proof of when an ad was published. Missing dates stay blank. The checked date is when the ad was read, not when the job opened. Every location an ad lists is kept, and nothing about seniority or the number of openings is guessed from them.
Ashby boards are read from their full-board v1 response. Greenhouse's stated total has to match the number of ads it returns. Lever is read page by page until a page comes back empty. The JSON keeps a fingerprint (hash) of each response, the counts and the request links. The boards are read one after another, not all at the same instant, so an ad that changes during a read can still affect the count.
The JSON lists every matching ad, and how many were left out and why. A closed or removed ad drops out of the next approved count, but that doesn't mean someone was hired. A change to a job's title, its location or how the board is set up can also change the count.
Each month a new count is collected and reviewed before it goes up. Until then, this table shows the last approved count.
Interview practice
Practice questions written for this site, not leaked questions or any employer’s real interview. Open one to see what a strong answer covers.
A customer asks for an AI assistant, but no one agrees what success means. What do you do first?
Before you pick a model, find out who will use it, how they do the work today, which mistake would be expensive, and what result everyone will accept as success.
The demo worked, but answers in production are unreliable. How do you investigate?
Collect real examples of bad answers. Look at what the assistant looked up and which tools it called, and sort the problems the system caused from the ones the model caused. Then turn those examples into a test set you rerun after every fix.
The system you have to connect to limits how often you can call it and sometimes goes down. How would you ship it?
Explain how a retry avoids doing the same work twice (idempotency), how many retries to allow, how to slow down when the other system is overloaded, how failures show up where people can see them, and a fallback that never loses the user's work.
A shortcut for one customer would ship faster but make the product messier. How do you decide?
Lay out the tradeoff, who owns the exception, when it expires or how it gets folded back in, and which part should become a feature every customer can use.
A deployed agent can change business records. Where should approval happen?
Point out the actions that can't be undone, limit what the agent's tools are allowed to do, show the exact change before it happens, and keep a record of who approved what and how to reverse it.
What would you show at the end of the project?
Something that works, the results of the checks the customer agreed to, the known limits, who runs it from now on, and documentation someone else can follow.
Other AI engineering roles
10-guide path
AI Engineer
Build useful software on top of AI models, then find out where it gets things wrong and fix it.
10-guide path
Agent Engineer
Build AI software that can take actions, with clear limits and a way to recover when something fails.
5-guide path
GTM Engineer
Fix the data and automate the handoffs a sales or marketing team depends on.
