Brand.Engineering

How to Become a Forward Deployed Engineer in 2026

The concrete path into forward deployed engineering: the skills FDE postings ask for, portfolio projects that prove them, what interviews test, and how to switch from SWE, solutions, or consulting.

Chris Riley

Chris Riley

@lcslates

To become a forward deployed engineer, you need three things: production engineering skills (usually Python plus TypeScript), hands-on experience shipping LLM systems with evals, and proof that you can work directly with a customer to get something into production. Most AI lab FDE roles want several years of experience, so the fastest route is usually a lateral move from software engineering, solutions engineering, or technical consulting, backed by a portfolio of real deployments.

This guide covers the forward deployed engineer skills employers list, portfolio projects that prove them, what the forward deployed engineer interview tests, and how to make the switch from your current role. If you're still deciding whether the job fits, start with what a forward deployed engineer actually does.

What FDE hiring managers ask for

The postings are consistent. Here's what two of the biggest FDE hirers list:

  • OpenAI (SF FDE posting): 5+ years of engineering or technical deployment experience including customer-facing work, production code across frontend and backend in Python or JavaScript, and experience building or deploying systems powered by LLMs.
  • Anthropic (London FDE posting): 4+ years in a technical, customer-facing role, or as a software engineer with consulting or product experience. Production LLM experience "including advanced prompt engineering, agent development, evaluation frameworks, and deployment at scale." Strong Python. Former technical founders are encouraged to apply.

Palantir, which created the role, has historically hired earlier. According to The Pragmatic Engineer, it hires FDEs with as little as a year of post-college experience, and Ramp hires some exceptional new grads.

The pattern: production engineering, LLM deployment, and customer-facing reps. You need all three.

Forward deployed engineer skills

Skill What good looks like How to build it
Production engineering You've shipped and maintained systems real people use Own a service end to end at work, or run a side project with real users
Python + TypeScript Comfortable across backend, APIs, and a simple frontend Build full-stack apps; both languages appear in the AI lab postings
LLM deployment Agents, tool use, retrieval, structured outputs Ship an agent that takes real actions, not a chat demo
Evals You can prove quality went up, with numbers Write an eval set before you change a prompt, then hill-climb on it
Integration work Auth, SSO, messy APIs, data pipelines Connect to tools businesses actually use: CRMs, ticketing, warehouses
Problem decomposition Turn a vague business problem into a buildable first version Practice scoping out loud (see the interview section)
Customer communication Run discovery, manage scope, explain trade-offs Volunteer for customer calls, incident reviews, pilot kickoffs

Two skills separate strong candidates. Evals, because OpenAI's FDE team, per The Pragmatic Engineer, builds evals with customer input and "hill-climbs" on them before delivery. And decomposition, which a Palantir FDE in Dev versus Delta calls "technical decomp": taking "a high-level problem" and breaking it down "to the lines of code you need to solve it."

Portfolio projects that prove you can do the job

A GitHub full of tutorials won't get you hired. FDE hiring managers want evidence that you can walk into a messy business, find the real problem, and ship. These projects mirror the job.

1. Deploy an agent for a real business

Find a small business, nonprofit, or a team at your current company with a painful manual workflow: intake forms, invoice matching, support triage, lead routing. Run a discovery session. Build an agent that handles it, connected to their actual tools. Get it into daily use.

Document it like an FDE would: the problem as stated vs the problem you found, the architecture, the eval set, the before-and-after numbers, and what broke in production. This one project answers most FDE interview questions on its own.

2. Build an MCP server for a tool a team depends on

Anthropic's FDE posting lists "MCP servers, sub-agents, and agent skills" as deliverables. Build an MCP server that exposes an internal or niche tool to an AI agent, handle auth properly, and write the docs a customer's IT team would need.

A practical variant: stand up marketing agents for a small business over MCP. Tools like Cuppa give an agent one MCP server for brand DNA, SEO articles, ad creative, and social, so you can focus on the deployment work: understanding the business, wiring it into their workflow, and measuring whether the output is usable.

3. Build an eval harness for someone else's LLM feature

Take an existing open-source LLM app or a feature at your company. Write a labeled eval set with the people who use it, measure baseline quality, improve it, and publish the results. This shows the skill that turns a pilot into a production deployment.

4. Write the field notes

FDEs feed what they learn back to product. For each project, write a short memo: what generalizes, what should become product, what should never be built again. Hiring managers read these more closely than code.

Our free Brand Engineer track ends in a portfolio project built the same way: a real system for a real business, documented end to end.

What the forward deployed engineer interview tests

Interview loops vary by company and change often, so verify the current process with your recruiter. Here's what's publicly documented.

Palantir publishes its own guidance. Its Getting Hired page says everyone starts with one or two phone interviews before an onsite. Its guide to navigating open-ended questions is the best prep document for the role anywhere: "articulate the alternatives and trade-offs, but don't forget to be pragmatic enough to be able to arrive at some concrete approach. Deliver a functioning idea first, then expand it afterwards."

Candidate reports compiled by The Forward Deployed describe a "decomposition" round where you get a vague real-world problem (for example, tracking every piece of equipment in a hospital that has never inventoried it) and are graded on how you structure it. Treat those as reported, not official.

OpenAI and Anthropic don't publish their FDE interview loops, and we won't guess. Their postings tell you what they're screening for:

  • Production coding across frontend and backend
  • LLM system design, agents, and evals
  • Customer discovery and stakeholder communication
  • Judgment under ambiguity ("make fast, sound decisions under pressure," in OpenAI's words)

How to prep

  1. Practice decomposition out loud. Take a vague prompt ("help a logistics company find where shipments get stuck"), name the user, state assumptions, break it into 3-5 sub-problems, pick the hard one, and propose a first version in under 15 minutes.
  2. Keep coding sharp. Every loop still checks that you can write working code.
  3. Prepare two deployment stories. One that went well with numbers, one that went badly and what you changed.
  4. Know the product cold. Build something with the company's API before the onsite.

How to get your first FDE role

From software engineering

You already have the hardest part. Close the customer gap: volunteer for customer escalations, pilot launches, and integration work. Ship an LLM feature with evals. Then apply internally or externally with a project like #1 above. Former founders and early startup engineers are a direct fit; Anthropic says so explicitly.

From solutions engineering or solutions architecture

You have the customer reps. Close the production gap. OpenAI's Solutions Architects, per The Pragmatic Engineer, "rarely write code on customers' infrastructure," and that is exactly what FDEs do. Take a prototype from a deal all the way to production yourself, and own it after launch. OpenAI's own Head of FDE started there as a Solutions Architect. Our FDE vs solutions engineer breakdown covers the differences in detail.

From technical consulting

You know how to run discovery and manage executives. The gap is usually depth of engineering and long-term ownership. Palantir FDEs distinguish themselves from consultants by building long-term solutions instead of one-time recommendations. Show that you've built and maintained something, not just designed it. Vertical expertise helps a lot: "Financial Services" is one of the most common tags on our FDE listings.

Where to apply

As of November 2026, our forward deployed engineer job board lists 44 roles at 20 companies. OpenAI (9), Cursor (6), Notion (5), and Anthropic (4) are hiring the most. The median posted base midpoint is about $243K; see the FDE salary guide for the full breakdown. Smaller AI companies like Gradial, Lovable, Ramp, and Snorkel AI (2 roles each) can be easier first stops than the labs.

FAQ

What skills does a forward deployed engineer need?

Production engineering (usually Python plus TypeScript or JavaScript), hands-on LLM deployment including agents and evals, integration work with real enterprise systems, and strong customer communication. Palantir adds "technical decomp," breaking a vague business problem into a buildable solution.

Can you become a forward deployed engineer without experience?

It's hard at AI labs, where OpenAI asks for 5+ years and Anthropic's London posting asks for 4+. Palantir and Ramp are exceptions that hire earlier, per The Pragmatic Engineer. Early-career engineers can build the case with real deployment projects and customer-facing work at their current job.

What is the forward deployed engineer interview like?

It depends on the company. Palantir publicly emphasizes open-ended problems where you weigh trade-offs and land on a concrete first version, and candidates report a dedicated decomposition round. AI labs don't publish their loops, but their postings stress production coding, LLM systems, and customer judgment.

Do I need a computer science degree to be a forward deployed engineer?

Not necessarily. Anthropic's postings list a bachelor's degree "or an equivalent combination of education, training, and/or experience." Demonstrated production work matters more than credentials.

How long does it take to become a forward deployed engineer?

For an experienced software or solutions engineer, a focused 3-6 months of LLM deployment projects and customer-facing work is a realistic target (our estimate, not a published benchmark). Starting from scratch, plan on several years of engineering experience first.

Related

Chris Riley

Chris Riley

Founder at Acme Studio. Building Cuppa and Bullseye.

@lcslates

New Brand Engineering roles, every week.

120 open roles at 41 companies right now. One email a week with what's new, verified against the original posting. Unsubscribe anytime.