A forward deployed engineer (FDE) is a software engineer whose job is to get a product working inside one customer's business, in production, with real data and real users, and then to make sure the next customer needs less help. The title has gone from a Palantir oddity to one of the most-posted roles in AI. This post covers what the job actually involves, how it differs from the sales-engineering roles it is often confused with, why AI companies now lead with it, and how we are hiring for it.
Where the role came from
Palantir created the role in the mid-2000s. Its Forward Deployed Software Engineers work directly with customers in small teams, start from open-ended questions such as "why are we delaying so many flights?", and own the work from the data plumbing to the custom application to the conversation with the customer's executives. Palantir pairs them with a sibling role, the Deployment Strategist, who owns the problem framing and the customer relationship (Palantir FDSE posting, Palantir Deployment Strategist posting).
For a long time that model looked expensive and hard to copy. Then large language models made a new kind of product common: one that is powerful in general but only valuable once it is wired into a specific company's data, tools and approvals. That is exactly the gap an FDE closes, and the role spread quickly. OpenAI, Anthropic, Scale AI and many startups now post FDE roles, and the Pragmatic Engineer has tracked the rise in demand.
What an FDE does, end to end
The postings from the companies that hire the most FDEs describe the same loop.
- Discover. Sit with the customer's team and find the decision or workflow where the product changes a result. The customer usually arrives with a solution in mind; the FDE's first job is to find the actual problem.
- Scope. Turn that into something buildable in weeks, with a measurable definition of done, and say no to the parts that do not move the number.
- Build. Write the code: data connectors, integrations, agents and tools, custom applications, often on both the customer's side and the vendor's. OpenAI describes its FDEs as leading discovery, technical scoping, system design, build and production rollout with strategic customers (OpenAI FDE posting).
- Prove it. Build the evaluations and monitoring that show the system works, and keeps working, on the customer's data. Scale AI's version of the role names agent runtimes, human-in-the-loop workflows, LLM-as-judge evaluation and guardrails (Scale AI posting).
- Ship and hand off. Take the system through the customer's security review and into production, then hand it to the people who will run it, with documentation they can follow.
- Feed it back. Turn what was custom into something reusable: a pattern, a reference integration, a product feature. Anthropic's FDE posting asks for exactly this, codifying repeatable deployment patterns and bringing findings back to product and engineering (Anthropic FDE, Applied AI posting).
The last step is the one that separates a good FDE program from expensive consulting. If every deployment is bespoke forever, the vendor has built a services firm. If each deployment leaves the product better, the next customer goes faster.
The customer owns the product
One idea runs through all of this. Say a client runs a product of their own, call it Product Y. The FDE is not there to sell them a different product or to rebuild Product Y. The FDE's job is to make the vendor's platform work inside Product Y, so that Product Y gets better for the client's own users. The client keeps owning their product, their roadmap and their customers. The FDE owns the technical outcome of the integration until it runs without them.
That framing explains the skills the role asks for. An FDE needs to be a strong engineer, because the integration is real code in someone else's production system. They also need judgment about someone else's business, because "working" means Product Y's numbers improved, not that the demo ran.
FDE vs. pre-sales and post-sales engineering
FDEs are often confused with two older customer-facing engineering roles. The difference is mostly about timing and ownership.
| Role | When | Owns | Measured by |
|---|---|---|---|
| Solutions engineer (pre-sales) | Before the contract | The technical case: discovery, solution design, demo or proof of concept, security questions | Deals won |
| Forward deployed engineer | From signature to production | The deployed system and its result in the customer's business | Production use and impact |
| Integration / implementation engineer (post-sales) | After go-live, ongoing | Healthy integrations, configuration, monitoring, training | Retention and account health |
| Deployment strategist | Alongside the FDE | The business case, success metrics, executive cadence | Value proven and expanded |
A pre-sales solutions engineer proves the product can solve the problem; their work usually ends when the deal closes, and pay often includes a commission. A post-sales integration engineer keeps a working deployment healthy as both sides change. The FDE sits in between and goes deeper than either: they write production code, they stay until it is used, and they carry what they learned back into the product. In practice the four roles work as one team, and a hand-off between them should never lose context.
Why AI companies lead with FDEs
AI-forward companies now present FDEs as part of how they go to market, not as an afterthought to sales. There are three reasons.
- The value is in the integration. A frontier model is general. Its value to one business depends on that business's data, tools and approvals. Whoever does the wiring decides whether the customer sees value.
- Adoption, not access, is the bottleneck. Most enterprises can already buy model access. What they lack is people who can take a use case through security review into production. On October 2 Anthropic announced a $100 million Claude Frontier Academy to train 10,000 "Frontier Deployed Engineers" by the end of 2027, with early cohorts from firms including Accenture, Deloitte, McKinsey and Morgan Stanley (Anthropic, program page).
- The field is the best product research. An engineer who lives in the customer's workflow sees the failures a vendor's own team misses. Those findings, fed back weekly, shape the roadmap.
Note the two names. A forward deployed engineer is sent out from the vendor to the customer. Anthropic's frontier deployed engineer often sits inside the adopting organization and is accountable for adoption there. The skills are nearly the same; the employer differs.
Skills and pay
The postings converge on a profile: several years of shipping production software, hands-on experience building and evaluating LLM or agent systems, strong Python and SQL, comfort working directly with customers and executives, clear writing, and willingness to travel (Anthropic and Palantir list roughly 25% and up). Pay is high because the profile is rare. Published base ranges include $135,000–$200,000 at Palantir, $180,000–$225,000 at Scale AI, and $200,000–$300,000 for Anthropic's Applied AI FDE (see the postings linked above). An analysis of about 900 FDE postings this year put the median base at $195,000 (Recruiting from Scratch).
The known failure modes are worth naming too: FDE work drifting into unowned consulting, scope that grows without limit, and field engineers treated as outsiders to the core team, so their findings never reach the product. A program that avoids those keeps FDEs inside engineering with commit rights, and gives every engagement a written definition of done.
How we are hiring
We are building this team now, with two openings in each of the four roles: Forward Deployed Engineer, Solutions Engineer (Pre-Sales), Integration Engineer (Post-Sales) and Deployment Strategist. Each card on our careers page lists the market-rate salary range, and all four are remote. If you are a company that wants forward deployed engineers to put AI to work in your own stack, our Forward Deployed AI Engineering page describes how an engagement runs.
Pay ranges and program details are as published by the linked postings and reports and may have changed since; we did not independently verify them.