Blog · AI
Forward deployed engineer (FDE): the role that puts AI into production

On this page
A forward deployed engineer works at the client's site to move AI from pilot to production. Where does this role come from, what does it do, and when do you need one?
In France, Google searches for "forward deployed engineer" grew ninefold in a year: 480 per month in January 2026, 4,400 in October (Google Ads volumes). The term comes from Palantir, OpenAI brought it back to the forefront, and many companies launching AI projects are discovering it. Here is what this role covers, why it is spreading, and what it says about how an AI project succeeds.
What is a forward deployed engineer?
A forward deployed engineer, or FDE, is a software engineer who works closely with the client. They do more than advise: they write code on the client's systems, put it into production, then feed what they learned back into their company's product.
Gergely Orosz, in The Pragmatic Engineer (August 2025), describes the role as an engineer who alternates between embedded work at the client and product development. The role blends three jobs: software engineering, commercial support, and platform work.
"Forward deployed" is a military term: it refers to troops deployed forward, in the field. Applied to software, it means the engineer goes where the problem is, instead of waiting for it to come up through a ticket.
Where the role comes from: Palantir
The role was born at Palantir in the early 2010s, under the internal name "Delta." At the time, Palantir worked for clients whose needs could not go through a standard product cycle: engineers had to be sent on site.
According to The Pragmatic Engineer, Palantir had more FDEs than product engineers until around 2016. With the launch of its Foundry platform, many of them then moved to the product.
Why everyone is talking about it in 2025-2026
AI model vendors have picked up the model. OpenAI now has a forward deployed engineering team, led by Colin Jarvis. The fund Andreessen Horowitz even called the FDE "the most sought-after job in startups" (June 2025): in its view, an AI startup that accepts lower margins early on, while it integrates its product at the client, builds a lasting advantage.
The underlying reason is simple: AI works in a demo, much less in a company's day-to-day.
- The MIT report "The GenAI Divide" (NANDA project, 2025) estimates that about 95% of generative AI projects in companies produce no measurable result. According to its authors, the cause is less the quality of the models than their poor integration with tools and ways of working.
- Gartner predicts that more than 40% of AI agent projects will be abandoned by the end of 2027, because of costs and poorly defined value (June 2025 forecast, reported by Reuters).
- At OpenAI, Colin Jarvis estimates that in about 80% of cases, what blocks is the deployment, not the model (interview with The Next Web, September 2026).
The FDE is the answer to that gap: someone who knows the technology and stays until the tool truly serves.
What an FDE does, concretely
- He starts from the business, not from AI. At OpenAI, every engagement starts with two days at the client's site. Leaders are asked to set AI aside and name the biggest levers in their business.
- He builds on real systems: the client's data, tools and access rights, not a separate mockup.
- He takes it to go-live and measures. At OpenAI, an FDE is judged on go-live and on the movement of a real metric, not on usage volume.
- He feeds back what he learns into the product or into his team's methods.
- He prepares his exit. "We should always be temporary," sums up Colin Jarvis. The goal is for the client's team to stand on its own.
FDE, consultant, solutions engineer, developer: what are the differences?
- Forward deployed engineer: delivers a tool in production and a self-sufficient team. They write code at the client's site and stay as long as it takes for the tool to be useful, then hand over.
- Consultant: delivers recommendations and a roadmap. They rarely write code, and their engagement ends with the report.
- Solutions engineer (pre-sales): delivers a demo or a target architecture. They mainly help close the sale.
- Software developer: delivers product features. They rarely work at the client's site and stay on the product.
The most debated boundary is the one with the consultant. Andreessen Horowitz itself notes that some service engagements are sometimes "rebranded" as forward deployed engineering. The difference comes down to one question: at the end of the engagement, is there a tool in production at the client's site, with people able to keep it running?
For the profile that designs generative AI applications themselves, see our article on the role of the GenAI engineer.
Does your company need an FDE?
The FDE model fits specific situations:
- your AI pilots work in demos but never make it to production;
- your data is scattered across business tools (ERP, CRM, inboxes, documents);
- your IT team has no time to integrate AI on top of its day-to-day work;
- the tool exists, but no one really uses it.
There are three ways to address this:
- Hire an FDE in-house. It is the most durable option, but the profile is rare: you need to know how to code, understand a business, and talk to teams.
- Go through your AI model provider. The major providers deploy FDEs, especially with their largest clients.
- Work with a team that operates this way: scope with the business, build on your systems, ship to production, then hand over.
How we apply this model at Scroll
We only advise on what we know how to build. Our AI projects follow the same logic as an FDE: the same people scope, build, ship to production, and hand over.
- Imalize, an industrial manufacturer: an AI agent reads every incoming email, identifies genuine sales inquiries and logs them in the ERP with their attachments, in under a minute per order. See the Imalize case study.
- Sistr, a temp staffing platform: job matching and a WhatsApp chatbot integrated into recruiters' tool. See the Sistr case study.
- Training "AI champions" : we trained the support team of a software vendor to build and publish its own AI agents, on its own use cases.
Our goal is the same as Colin Jarvis's: for your teams to no longer need us. We identify and train internal champions, we document what is built, and we organize the handover from the start.
Do you have an AI project that is not making it past go-live? See our approach to AI transformation or request a free scoping session within 48 hours.
FAQ
Frequently asked questions
What does "forward deployed engineer" mean?
Literally, "engineer deployed forward." The term comes from military vocabulary and refers to a software engineer who works directly at the client's site, on their systems, to bring a solution to go-live.
Is an FDE a consultant?
No. A consultant recommends; an FDE builds. They write code at the client's site, bring it to go-live, stay until the tool is actually in use, then hand it over to the internal team.
What is the difference between an FDE and an AI engineer?
An AI engineer designs and builds AI applications, often within a product team. An FDE may have the same skills, but applies them at the client's site, with a focus on go-live and handover.
Where does the FDE role come from?
From Palantir, in the early 2010s, where these engineers were called "Deltas" internally. The role spread in 2025 with generative AI vendors, including OpenAI.
Should you hire an FDE or work with an external team?
Hiring is more durable if you have several AI projects to run over time. For a first project, an external team working this way gets you to go-live faster, provided it trains your teams and organizes the handover.
Read next
Guides on the same topic
Explore the topic further with a selection of complementary resources.
Aug 06, 2026
GenAI Engineer: What is their role in a generative AI project?
The GenAI Engineer turns a generative AI idea into a reliable product. Discover their role, missions, and the skills needed to go from use case to production.
Nov 02, 2026
AI Act: the first harmonised standard is published, high risk delayed
EN 18286 was published in the Official Journal on 14 October 2026, and high-risk obligations are pushed to December 2027. What it changes for your AI projects.
Sep 22, 2026
Jev: the AI model that decides instead of writing
Launched in September 2026 by TypeSafe AI, Jev does not generate text: it returns typed, calibrated decisions. What it is, what it costs, and three use cases.