How to Become a Forward Deployed Engineer: What 100 Job Postings Ask For

Marina Wyss, an AI and machine learning career coach who says she has coached more than 400 people moving into AI roles, read 100 forward deployed engineer (FDE) job postings to work out what the job is, who gets hired and how to get there. The role started at Palantir more than a decade ago; in the last year OpenAI and other AI companies have started hiring for it too. Her figures below come from that sample of postings.

Disclosure: the video includes a sponsored segment for DataCamp. We have left it out of this summary.

This article is our own summary, with short quotes, from the video's automatic transcript. Watch the full video above for everything that was said.

The short version

• What the job is. A company buys an AI product or wants to build on a model, and the hard part is making it work inside its systems and team. The vendor sends an engineer to work directly with that customer and build it, often from the customer's office. It is AI engineering for one client instead of one product for many.
• The pay. OpenAI pays up to $300,000 a year for the role, she says. Base salaries typically start between $175,000 and $195,000, with stock and bonuses on top, and frontier labs pay more. She counts more than 1,000 open positions.
• Who gets hired. Median experience required is 5 years; about a third of postings accept 3 or fewer and only about 12% are truly entry level. 68% never mention a degree. Close to half of current FDEs were software engineers and about a fifth were solutions engineers or architects.
• Half building, half customers. "Most people in my experience are only good at one of those things." Customer-facing experience appears in 71% of postings, end-to-end ownership in 72%.
• The skill most people skip is evals. Evaluations appear in 53% of postings and are named at Anthropic, Snowflake, Databricks and OpenAI. An FDE has to prove to the customer that the system works.
• How she'd get qualified. Build one end-to-end project (a production app, then an LLM feature, then an agent, then evals) and run one customer loop inside your current company.
• The usual way in is a lateral move. Many current FDEs moved over internally from solutions architect or solutions engineer roles; she says OpenAI's head of FDE came up that way.

What the postings ask for

Software engineering

• Production experience (87 of 100 postings). Her test: work counts as production "if there were users who could potentially break things and you would have to fix it." An internal tool counts; a notebook you run for yourself does not.
• APIs, integrations and system design (about 70% each). FDEs design inside a customer's existing architecture, so it is about making something work, not textbook design. Databases and data modeling appear in about 31%, data pipelines in 27%, SQL in only 15%.
• Languages. Python in 63%, TypeScript or JavaScript in about half, and nearly 40% want full-stack or front-end work.
• Cloud and security. AWS, GCP or Azure in about 38%; security, auth or compliance in 43%. She would skip Kubernetes, Terraform and CI/CD until you need them.

AI engineering

• Building with LLMs (67%) and agents (57%), meaning systems where the model decides what to do next, calls tools and runs multi-step workflows. Using a coding agent like Claude Code is simply assumed.
• Frameworks matter less. LangChain, LangGraph and similar appear in 19% of postings, and about half of those list them as nice to have.
• Evals (53%), the one she would study most.
• RAG and prompt engineering (about 28% each) and AI safety (about 21%, growing as more customers are banks and healthcare).
• What to skip if you're starting out: classical ML (18%), fine-tuning (11%) and GPU or inference optimization (5%).

Working with customers

• Customer-facing experience (71%), from consulting, solutions engineering or close work with non-technical teams.
• Communicating with technical and non-technical people (68%), and discovery and scoping (68%), including respecting the customer's systems and politics.
• Stakeholder management (56%) and executive presence (about half).
• Comfort with ambiguity and high agency (65%); business acumen and product sense (about 54% each, usually as nice to have).

How she'd make herself qualified

1. Build a normal production app first: an API, a database and auth, deployed on a cloud with logging.
2. Add the AI layer to the same app: an LLM feature, then make it agentic with tool calls and multi-step tasks.
3. Wrap it in evals so you can show it works. "Evals are the most important part that people skip."
4. Run the customer loop at your current job: find a team with a problem, have a discovery conversation, write a short scoping doc, build it and demo it back. She argues it can be worth doing on your own time if it moves you toward the goal.

Paths into the role

• Apply directly if you have about five years of production experience, a strong project and a network.
• Move laterally at a company starting an FDE team. She names Databricks, Salesforce, Google Cloud and AWS, plus many AI companies.
• Join a smaller company. 58% of FDE roles are at companies with 11 to 200 employees, and titles vary; she saw one listed as "software engineer, agent."
• Build the background through consulting, systems integration or founding a startup. One company report she found said roughly one in six of its hires was a former founder.
• Early-career programs: Palantir's new grad FDE track and a route where you submit a recorded demo built on its platform, a Sierra new grad track, and an early-career deployed engineer role at LangChain.

What the interviews look like

An open-ended scoping problem (they watch whether you ask about the user and the goal first), a round in code you didn't write, a practical coding round instead of LeetCode, often a take-home or live build you defend as if presenting to a customer, and a deep dive into a past project to check that you really owned it.

Is it worth it?

45% of postings mention travel, usually 25 to 50% of the time, and most frontier-lab roles are hybrid. You talk to stressed, often non-technical customers, and some weeks look more like consulting than engineering. In exchange, she says the pay is among the best in the industry, the field is growing fast and you see how AI is used across many companies. "If you like building things, you like people, and you don't mind a bit of chaos, this is one of the best jobs in AI right now."

What it means for your business

1. Hire for both halves. If you are building an FDE or solutions team, test for customer work as hard as for engineering.
2. Ask candidates about evals. It is the clearest signal that someone can prove an AI system works.
3. Look inside first. Solutions engineers and engineers who already work with non-technical teams are the most common source of FDEs.
4. Use the customer loop internally. Discovery, a scoping doc, a build and a demo is a good way to deliver AI projects to your own teams.
5. Expect the role to look like consulting. Budget for travel and for time spent keeping stakeholders informed.

Video details

• Title: How I'd Become a Forward Deployed Engineer (Complete Roadmap)
• Creator: Marina Wyss (Marina Wyss - AI & Machine Learning)
• Published by: Marina Wyss on YouTube, November 2026
• Watch: youtube.com/watch?v=ONv6-VdLBv8
• Note: figures come from her review of 100 job postings as stated in the video's automatic transcript, lightly cleaned. Check the video for exact wording.

Want to put this to work?

We help owners turn AI trends into a plan, a product and a launch. Book a free call or see our services.