Career guide · 2026

How to become a Forward Deployed Engineer

FDE is not a traditional SWE role and the path in is not the traditional SWE interview. This guide covers what backgrounds actually land the role, what the loop looks like at Palantir, OpenAI, Anthropic, and Scale AI, and what you can build now to make your application credible.

Who actually gets hired

Four backgrounds recur in FDE hiring. The most common is a full-stack or backend SWE with customer-facing experience — someone who has spent time in a startup, agency, or consulting adjacent role where they held the customer relationship while writing production code. The second is an applied ML engineer who has taken models into production against real customer data and knows how to explain the failure modes to non-technical stakeholders.

The third is the founder-shaped generalist — someone who has built and shipped a product from scratch against real user constraints, where they were simultaneously engineer, product manager, and customer support. The fourth is the ex-consultant with genuine engineering depth: someone who came through McKinsey, Accenture Tech, or a Big-4 advisory practice and built a serious software portfolio alongside it. The common thread is ambiguity tolerance: FDEs land in customer environments with incomplete specs and ship something anyway.

The concrete skill stack

Across public FDE job postings on FDE Jobs List, the most consistently requested technical skills are: Python and TypeScript (Go is a plus, especially at Palantir), SQL and columnar formats (Parquet, Arrow, DuckDB), LLM product frameworks (LangChain, LlamaIndex, Model Context Protocol, vendor SDKs), and cloud infrastructure basics (AWS or GCP, containers, basic IaC). Domain expertise in government, defense, healthcare, or finance is a compounding multiplier for companies serving those verticals.

The non-technical skills matter as much: written communication that translates ambiguous requirements into a scoped technical spec, comfort running workshops and whiteboarding sessions with mixed technical/non-technical audiences, and — particularly at Palantir — genuine willingness to travel and work on-site at customer locations. An FDE who avoids the customer is a contradiction in terms.

What the interview loop actually tests

Standard SWE interviews test whether you can code. FDE interviews test whether you can code and hold a customer relationship. The loop adds at least one round that does not exist in standard SWE hiring: a customer scenario or case interview where you are handed an ambiguous problem ("a Fortune 100 defense contractor wants to deploy an LLM to their document workflow — design the integration") and must scope, defend, and course-correct in real time against pushback.

The coding rounds at Palantir and OpenAI tend to be context-rich — not pure LeetCode, but problems grounded in real data-engineering or integration scenarios. Systems design at FDE level is specifically about customer-deployed systems: data pipelines that handle messy real-world inputs, LLM pipelines with eval sets and fallback paths, deployment models that survive the customer's air-gapped or regulated environment. Behavioral rounds test customer-pushback situations: "Tell me about a time you pushed back on a customer" is a real Palantir interview question, and the wrong answer is "I never push back."

Portfolio moves that work

Three categories of portfolio work move the needle with FDE hiring managers. Open-source data integration projects — something that ingests messy real-world data (public APIs, flat files, event logs) and produces structured, reliable output — directly mirrors FDE daily work and gives interviewers something concrete to discuss. Hosted LLM workflow demos deployed against realistic datasets (not just a GitHub repo — a clickable URL with a real use case) demonstrate that you can ship, not just prototype.

The highest-signal portfolio item is a written case study: "Here was an ambiguous scope. Here is what I shipped. Here is every design call I made and why." This format directly answers the FDE hiring signal — can you think clearly under ambiguity, communicate that thinking to a non-technical audience, and own the outcome? Contributing to MCP servers, LangChain integrations, or Anthropic Claude tool-use projects also builds relevant signal with the AI-native FDE hirers. See live roles at FDE Jobs List.

Where to apply and how to enter

Palantir is the original FDE shop — both the Forward Deployed Software Engineer and the parallel Deployment Strategist track. Palantir recruits through campus and industry pipelines; the fastest entry is referral from an existing FDSE or through a Palantir recruitment event. OpenAI has scaled FDE hiring most aggressively in 2025–2026, posting roles across San Francisco, New York, London, Tokyo, Dublin, and Paris — apply directly through their ATS.

Anthropic uses Applied AI Architect and Forward Deployed Engineer interchangeably for the same role shape; their FDE track is smaller and primarily San Francisco-based. Scale AI runs an FDE track tied to its data labeling and evaluation platform, with more remote options than the others. Browse all live openings at FDE Jobs List and cross-reference with the FDE salary guide before negotiating.

Related resources

Frequently asked questions

What background do I need to land an FDE role?

The most common path is 2–5 years of production full-stack or backend engineering, ideally with at least one stretch of customer-facing or deployment work — agency projects, startup consulting, early-stage product work where you held the customer relationship. Most hirers (Palantir, OpenAI, Anthropic) treat 2+ years of shipped production software as the minimum; new-grad FDE roles exist but are rare and competitive. The profile they are looking for is a strong engineer who does not need to be protected from customers.

Do I need machine learning experience to become an FDE?

ML depth is not required at Palantir — their FDSE track has always been more data-engineering and integration than model work. At OpenAI, Anthropic, and Scale AI, LLM-product familiarity is increasingly expected: you should be comfortable calling model APIs, building prompt pipelines, designing eval sets, and explaining confidence calibration to non-technical stakeholders. Deep model research is not required. The practical bar is: could you build a RAG pipeline against a customer's document corpus and explain the failure modes to the customer's VP of Engineering?

How do I get past the FDE resume screen?

Recruiters filtering FDE resumes are looking for the words "deployed," "customer," "integration," and "shipped" — not "contributed to" or "worked on team that." Every bullet point on your resume should anchor to a customer outcome or a working production system, not a process. Cut any project that was internal only and never touched real users. If you have customer-facing work buried under generic SWE framing ("built microservices for X platform"), flip it: "deployed X integration for Y customer, reducing their data pipeline latency by 40%." The resume gate is fast — you have roughly 20 seconds.

What does the FDE interview loop actually look like?

Expect 4–6 stages: recruiter screen, hiring-manager call, technical screen (coding + system design), a customer scenario or case interview unique to FDE, virtual onsite (3–5 rounds: coding, systems design grounded in a customer use case, behavioral/customer simulation, bar-raiser), and a final/exec round. Palantir and OpenAI both weight the customer scenario heavily — a case where you are handed an ambiguous customer problem and must design + defend a solution in real time. The bar-raiser at Palantir specifically tests culture fit, not technical depth. Total loop length is typically 4–6 weeks.

Is the FDE role remote-friendly?

Mixed, and varies by company. Roughly a third of postings on FDE Jobs List are remote-global; the majority are hybrid in hub cities (San Francisco, New York, Washington DC, London, Tokyo, Paris). Palantir is the most travel-heavy — their original embedded model put engineers on-site at customer sites for weeks at a time, and that culture persists for government and defense accounts. OpenAI and Anthropic have softened the travel requirement for most commercial FDE roles. Filter live openings by location at FDE Jobs List.

What is the comp ramp for an FDE?

OpenAI publishes base salary bands in public ATS postings: current published ranges run from $185,000 to $325,000 base for Forward Deployed Software Engineer roles in San Francisco. Palantir, Anthropic, and Scale AI do not surface compensation in their ATS payloads. Levels.fyi self-reported data for senior FDEs at the same companies shows total comp (base + equity + bonus) in the $250,000–$550,000+ range at the IC5–IC6 level; entry-level FDEs at smaller companies cluster around $120,000–$160,000 base. Full breakdown at FDE salary guide (2026).

What portfolio work actually helps an FDE application?

Three categories move the needle. First, open-source data integration projects — something that ingests messy real-world data (public APIs, flat files, event logs) and produces structured output; this directly mirrors FDE daily work. Second, hosted demo deployments of LLM workflows against realistic datasets — not a GitHub repo, a clickable URL. Third, written case studies of "here was an ambiguous scope, here is what I shipped, and here is why I made each design call" — narrative format that shows how you think about ambiguous problems, not just that you can code.

Can I pivot to FDE from a standard SWE role with no customer experience?

Yes, but you need a bridge. Internal SWE experience alone does not answer the core FDE question: "Can this person hold the customer relationship while shipping code?" Build the bridge through: running technical workshops or lunch-and-learns for non-engineers at your current company, shipping a side project that real external users depend on, giving a conference talk, or doing contract work where you own the client relationship. Even one concrete example of "I worked directly with a non-technical stakeholder to scope and ship a solution they depend on" is enough to get an interview — zero examples is the actual barrier.