A forward deployed engineer, usually shortened to FDE, is a software engineer who works inside a customer’s business until an AI system is running there and people use it. The job sits between writing code and changing how a team works. On October 2, 2026, Anthropic committed $100 million to training 10,000 engineers in a version of the role, and called people with these skills the hardest talent to find.
Editorial note: Role descriptions come from job postings by Palantir, OpenAI and Anthropic and from Anthropic’s Frontier Academy announcement, all read October 3, 2026. Postings change; the patterns below are what they had in common that day.
- 01Engineer firstEvery posting asks for strong production code. The customer work is added to engineering, not a substitute for it.
- 02Measured by useThe job ends when people use the system, not when a demo works. OpenAI’s posting names production adoption first.
- 03Not a solutions engineerA solutions engineer helps sell. An FDE builds and owns the result after the contract is signed.
- 04You can grow oneThe traits are learnable. Train an engineer who already knows your business before hiring a stranger who knows models.
01 — DefinitionWhat the job is
Most AI projects stall in the gap between a working prototype and a system a team relies on. The prototype runs on clean sample data. The real version has to reach the company’s own systems, pass a security review, handle the cases nobody wrote down, and change how people spend their day. An FDE is the engineer assigned to close that gap, working on site or embedded with the team that will use the result.
The title is not unique to AI companies. Palantir hires Forward Deployed Software Engineers to solve customers’ data problems alongside them. Its posting compares the responsibilities to “those of a startup CTO”: small teams, little supervision, and ownership of a project from start to finish. AI labs now hire under the same title, and the fit is plain. A model is not useful until someone fits it into a specific business.
We cover the failure modes on the other side of that gap in our framework for why agent projects stall before production. An FDE is the staffing answer to most of them.
02 — The evidenceHow employers describe it
Four current descriptions show the shape of the role. Read side by side, they agree on more than they differ: real systems, customer contact, and a duty to turn one deployment into a pattern the next team can reuse.
| Employer | Title | What the description stresses |
|---|---|---|
| Palantir | Forward Deployed Software Engineer | Start from an open question a customer brings, design and build the data solution, own it end to end. Travel to client sites of 25–50% preferred. |
| OpenAI | Forward Deployed Engineer, by industry | Lead deployments in regulated sectors such as financial services; set launch criteria, run evaluations, and turn lessons into reusable reference designs. Travel up to 50%. |
| Anthropic | Forward Deployed Engineer | Work inside major customer accounts until AI applications run in production, and report what the field teaches back to product teams. Travel of 25–50% on the manager posting. |
| Anthropic Academy | Frontier Deployed Engineer | The same skills, held by a customer’s own engineer: take a Claude project from idea to production inside their company. |
The last row is the new part. In the first three postings the FDE is the vendor’s employee. Anthropic’s Frontier Academy residency trains engineers employed by customers and consultancies to do the same work for their own organisations. OpenAI’s financial services posting is the most detailed public description of the vendor-side job, down to on-call readiness and audit evidence for regulated launches.
A small team of high-agency people with the right skills, access to Claude, and a deep understanding of how their business runs can transform an entire company.Steve Corfield, Global Head of Business Development and Partnerships, Anthropic, October 2, 2026
03 — ComparisonHow it differs from nearby roles
The title overlaps with three jobs most companies already know. The difference is in what each one is accountable for when the work is done.
Forward deployed engineer
Writes and owns production code, gets it through security review, and stays until people use it.
Solutions engineer
Runs demos and proofs of concept before a contract. Usually hands over once the deal closes.
Management consultant
Frames the problem and recommends a plan. May not write or operate the system.
Software engineer
Delivers what the requirements say. Rarely sits with the people whose work changes.
04 — SkillsThe skills postings ask for
Across the postings, the requirements fall into five groups. None of them is a model-research skill. The engineering bar is the same as any senior role; the difference is how much of the job happens in front of people who are not engineers.
- Production engineering. Python or TypeScript at a level that survives code review, plus integrations, data flows and monitoring. OpenAI’s posting adds on-call readiness.
- Building with language models. Anthropic’s residency asks for a record of building with them. OpenAI asks for an understanding of how model behaviour, evaluation and guardrails affect user trust.
- Evaluation. Defining what “good enough to launch” means for one workflow, measuring it, and iterating against the measure.
- Problem framing. Turning a vague request into a scoped project with a measurable outcome. Palantir’s example is a customer asking why so many flights are delayed.
- Bringing people along. Explaining trade-offs to executives, compliance and the team that will use the system, and helping that team change how it works.
Published pay for the role is scarce. Anthropic’s posting for the first manager of its own FDE team, read October 3, lists an annual salary range of $320,000 to $405,000 and asks for 10 or more years of experience. That is a senior leadership role at an AI lab, not a guide to what an FDE costs a mid-size company.
05 — DecisionHire, train or bring one in
Few companies need a full FDE team. Most need one person, or one engagement, for the first system that matters. The route depends on how many AI projects are queued and whether anyone inside already knows the business well.
The vendor route has a catch. When a model company’s own FDE builds the system, the knowledge leaves with them, and the system is tuned to that vendor. We looked at that trade-off in renting enterprise agents versus building your own. Training an insider is slower at the start but keeps the skill in the building, which is the bet Anthropic’s residency makes.
06 — Practical implicationsSetting the first project up
Whichever route you take, the first project decides whether the role works. Pick one process with a clear owner, data the engineer can reach in week one, and a result that can be counted, such as hours saved or errors caught. Agree the security review path before any code is written, because that is where most pilots wait.
Then write the handover into the plan. An FDE engagement that ends without a named person who runs the system is a demo with extra steps. Our AI transformation work runs on this model: embedded build, a review record, and a handover to the team that keeps it.
Name the project and its owner before the engineer
Write down one process, its owner and the measure of success. Then decide whether an insider can learn the model work faster than an outsider can learn your business. For most companies, that answer picks the route.