Loading...
System Design and AI Engineering, by Roni Das. 980 interactive lessons.
Start LearningA forward deployed engineer (FDE) is a software engineer who works directly with one customer at a time. They sit with the customer's team, learn the real problem, and write the code that makes the company's product work inside that customer's systems. Palantir created the role. In October 2026, OpenAI, Anthropic, Databricks, Scale AI, Salesforce, Ramp and others are hiring for it, and most of the published US pay ranges we checked sit between $135,000 and $330,000 a year.
Preparing for an FDE interview? Our Forward Deployed Engineer interview pack has 100 questions on customer deep dives, vague problems, integrations, debugging on site and deploying AI, each with a full answer and a diagram.
Not sure you are ready? The free check is a short call where I tell you the two or three things to fix first for your target role.
Palantir created the role. Inside Palantir it is called a "Delta", and Palantir describes the job as "one customer, many capabilities". AI companies now hire for it because a model only earns money once it works inside a customer's own data and tools. On 22 October 2026 we counted the open postings with "Forward Deployed" in the title on each company's public job board: 236 at these nine companies alone.
Counted from the Greenhouse, Lever and Ashby job-board feeds on 22 October 2026. It includes internships, new-grad and manager roles, and each city a company lists as a separate posting. Salesforce also hires FDEs in many US cities; its board does not give a clean count, so it is left out.
“You will own discovery, technical scoping, system design, build, and production rollout, partnering directly with customer engineering and domain teams.”
“Contribute directly in the code when progress or clarity depends on it.”
“Travel frequently (25%) to customer sites to build in person with customers.”
“You don't need to have shipped agents to production.”
“Own delivery end-to-end: design, build, deploy, and maintain solutions at customer sites, from prototype to stable production.”
“You have prior customer-facing technical delivery experience (consulting, professional services, or forward deployed engineering)”
The three jobs overlap, and titles change between companies. This is the usual split. Palantir's own line is that the role does "significantly more engineering and technical work than a consultant".
| Forward deployed engineer | Software engineer | Solutions engineer | |
|---|---|---|---|
| Who you work for, day to day | One customer at a time, often on site | Your own product team | The sales team, before the deal is signed |
| Main output | Working code in the customer's systems | Features in the product many customers share | Demos, answers and proofs of concept |
| Writes production code | Yes, every week | Yes, every week | Sometimes, mostly prototypes |
| Talks to customers | Every day, from engineers to executives | Rarely | Every day, mostly before the sale |
| Travel in the postings we read | 20% to 50% | Usually none | Varies |
| Hardest part | Turning a vague problem into a plan, then shipping it | Deep technical problems in one area | Winning trust before the customer has paid |
Compared with an AI engineer: an AI engineer usually builds AI features inside one product. An AI-focused FDE builds the same kind of thing (RAG, agents, evals) but for one customer, with that customer's data and rules. See the AI engineer roadmap for that path.
You write real code, mostly Python, TypeScript or Java, plus SQL. The postings ask you to build and ship, not to advise. Ramp says it still checks coding, debugging, data structures and system design.
Most of the work is connecting things: a customer's database, their CRM (customer records, such as Salesforce), their files and their login system. You need APIs, authentication, webhooks (one system calling another when something happens), and safe retries.
Customers rarely give you a clear spec. You ask questions, find the real goal, and split it into parts you can build this week. Palantir tests exactly this in its decomposition interview.
At AI companies the job is putting models to work for one customer: RAG (answering from the customer's own documents), agents (models that call tools), and evals (repeatable tests that show if the answers are good enough).
You explain trade-offs to engineers in the morning and to an executive in the afternoon. Palantir lists engaging with stakeholders "from technical teams to executives" as part of the job.
When three customers need the same fix, it should become part of the product. Ramp calls scoping and generalization "the art of the role".
This plan is for someone who can already program. The times are rough guides for about 10 hours a week, not promises: go faster where you already have the skill. Each step lists free lessons you can read now (no signup) and the deeper lessons in our courses.
Weeks 1 and 2
Every FDE task starts with reading someone else's data and calling someone else's system.
Prove it: Pull data from a public API into a small database, and make the import safe to run twice.
Week 3
Customer systems sit behind single sign-on and tokens. You cannot build anything until you can get in safely.
Prove it: Connect a small app to a service that uses OAuth, and receive its webhooks without losing one.
Week 4
Half of a deployment is getting the right data to the right place, on time, without copies going missing.
Prove it: Keep a copy of a database table in sync with a second store, and show it survives a restart.
Week 5
OpenAI's posting lists system design as part of the job. You will sketch designs for customers, so the basics must be fast.
Prove it: Draw the design of your week 4 project, and add a cache where it helps.
Weeks 6 and 7
At AI companies, most FDE projects put a model to work on a customer's own documents.
Prove it: Build question answering over 100 real documents, and count how often search finds the right one.
Week 8
Customers now ask for models that act: update a record, file a ticket, send a report.
Prove it: Give an agent two tools, log every step, and measure how often it finishes the task.
Week 9
A customer will ask "how do you know it is good enough?". An eval is your answer, with a number.
Prove it: Write 50 test questions for your week 7 project, change one thing, and show the score before and after.
Week 10
You are handling another company's private data. One leak ends the project.
Prove it: Hide an instruction in a document and check your project does not follow it.
Week 11
The job ends at a working rollout, not a demo.
Prove it: Add cost and time logging to your project, and a fallback when the model provider fails.
Week 12
The loop tests coding, design and how you handle a vague customer problem out loud.
Prove it: Do two timed mock rounds: one coding, one decomposition. Write down what you would change.
Each company runs its own loop, and most do not publish it. Below is only what the companies themselves have written. We left out forum posts and paid guides we could not check.
Palantir (students and early talent)
HackerRank coding challenge, 30-minute recruiter chat, 30 to 45 minute technical interview, two one-hour onsite interviews (learning and decomposition), then a hiring manager meeting.
Ramp
Checks the fundamentals (coding, debugging, data structures, system design) without asking for perfection, and looks hard for drive, which Ramp calls the best predictor of real-world performance.
Glean
The posting says the process includes "a brief AI-focused exercise or discussion".
OpenAI
The posting asks for 5+ years of engineering or technical deployment work that includes customer-facing work, and says travel up to 50% is required.
Decomposition means splitting a big, vague problem into small parts you can build and test. Here is the shape of a good answer to a made-up prompt of the kind these rounds use.
A hospital group says: "Our nurses waste hours on paperwork. Use AI to fix it."
One worked answer is not enough for the loop.
Each question has why it is asked, a full answer, what the interviewer listens for, the follow-ups, the common mistakes and a diagram (127 diagrams in all). A PDF you keep, junior to staff level.
India: UPI, cards or netbanking through Razorpay. Elsewhere: card or PayPal. Download at once from My packs.
Every range below is copied from a live job posting on 22 October 2026. US law in several states makes companies publish a pay range, which is why these are US roles. Read the last column: some ranges are base salary only, some include expected bonus (OTE, on-target earnings), and stock is usually extra.
| Company and title | Where | Published range | What it covers |
|---|---|---|---|
| OpenAI, Forward Deployed Engineer (FDE) | San Francisco | $185K to $300K, offers equity | Listed as salary; equity is separate |
| OpenAI, Forward Deployed Software Engineer | San Francisco | $185K to $325K, offers equity | Listed as salary; equity is separate |
| Anthropic, Staff Forward Deployed Engineer | New York, San Francisco, Seattle | $280,000 to $320,000 a year | Labelled annual salary (the page says sales roles show on-target earnings) |
| Palantir, Forward Deployed Software Engineer | New York | $135,000 to $200,000 a year | Salary only; stock units and sign-on bonus can be extra |
| Palantir, Forward Deployed Software Engineer, New Grad | New York | $135,000 to $145,000 a year | Salary only; stock units and sign-on bonus can be extra |
| Ramp, Software Engineer, Forward Deployed | San Francisco | $189K to $330K, offers equity | Listed as salary; equity is separate |
| Scale AI, Forward Deployed Software Engineer, Public Sector | Several US cities | $184,000 to $292,560 | Base salary for new hires; equity is separate |
| Salesforce, Forward Deployed Engineer (Mid/Senior) | San Francisco and other US cities | $88,970 to $287,910 a year ($97,860 to $316,750 in California and New York) | Base salary only |
| Databricks, Sr. Forward Deployed Engineer | San Francisco and other US cities | $182,000 to $250,208 | Base salary, or on-target earnings if the role has commission |
| Glean, Founding Forward Deployed Engineer | Mountain View, New York, or remote in the US | $160,000 to $270,000 a year | Standard range; may add variable pay and equity |
| Harvey, Forward Deployed Finance Engineer, Private Credit | New York | OTE $254.6K to $350K, offers equity | On-target earnings: base plus expected variable pay |
Levels.fyi collects pay that engineers report themselves. These numbers are total yearly pay (salary, stock and bonus), read from its public pages on 22 October 2026.
Forward deployed engineers, all companies, United States
Median $220,000 a year (middle half: $172,000 to $292,000)
Total yearly pay, 160 reports
Palantir forward deployed engineers, United States
$201K to $590K a year, median $233K
Total yearly pay, 227 reports
Forward deployed engineers, India
Median ₹17,56,635 a year
Only 11 reports, so treat it as a rough hint
India: Databricks had live FDE roles in Bengaluru and remote in India on 22 October 2026, but Indian postings do not publish pay, and only 11 people had reported an Indian FDE salary on Levels.fyi. We will not guess beyond that.
Weeks 6 to 11 of this roadmap are our AI Engineering course: 210 lessons on RAG, agents, evals, security and serving, many built around a real experiment with real numbers. 10 lessons are free to read first, with no card needed. One payment, lifetime access.
Free 15-min Interview Readiness Check
A short call. I tell you the 2 or 3 things to fix first for your target role. It costs nothing, and it is booked on Topmate.
Also useful: system design roadmap, RAG interview questions, AI engineer interview questions and live mock interviews.
