What it is
Forward-deployed engineering (FDE) describes a staffing model in which a technology vendor sends its own engineers to work inside a customer’s organization — often on-site, sometimes for months — to build, customize, and debug software against that customer’s specific data, systems, and requirements. It is distinct from traditional enterprise software sales, where a vendor ships a configurable product and leaves implementation to the customer or to third-party integrators. In the FDE model, the vendor’s own engineers do that implementation work directly, writing production code against the client’s infrastructure rather than handing off a generic tool.
The term borrows its military flavor deliberately: a “forward-deployed” engineer operates at the point of need rather than from a vendor’s home office, the way forward-deployed military units operate near a front line rather than at a rear base. Palantir Technologies is widely credited with originating the role and the title; the company’s engineers were already using the “forward-deployed engineer” designation by 2009, and Palantir went on to build much of its early government and enterprise business around embedding engineers directly with clients such as intelligence agencies and large corporations, rather than selling off-the-shelf software.
The model re-entered the spotlight in 2025 and 2026 as AI companies adopted it to solve a specific problem: large language models perform impressively in demos but often fail to function reliably inside a customer’s actual workflows, data pipelines, and compliance constraints. OpenAI stood up a forward-deployed engineering function in early 2025, starting with a handful of engineers and growing to dozens across multiple cities; Anthropic, Google Cloud, AWS, and Microsoft have since announced comparable units, with AWS and Microsoft committing roughly $1 billion and $2.5 billion respectively to embedded-engineering programs in 2026. Job postings for the role grew roughly eightfold between January and September 2025, according to reporting cited in industry coverage.
Why it matters for AI governance and narratives
Forward-deployed engineering sits at the center of a framing contest over who is actually responsible for agentic AI’s reliability problem. Vendors present FDE teams as a service — evidence that they are serious about production-grade deployment rather than shipping unaccountable black boxes. Critics, including Gartner, read the same arrangement as a liability: it concentrates deep, undocumented knowledge of how an agent was actually built inside the vendor’s own staff, leaving the customer unable to maintain, audit, or modify the system once the engagement ends. That distinction matters for AI governance debates about accountability and auditability — a system that only its vendor’s engineers understand is harder for a customer, a regulator, or an independent auditor to scrutinize than one built on documented, transferable architecture.
The model also reframes a recurring narrative in AI coverage — that agentic AI pilots frequently fail to reach production — not as a model-capability problem but as a deployment and organizational-capability problem. Gartner’s research places vendor-built, FDE-delivered agentic systems at the center of that failure pattern, which shifts scrutiny from “are the models good enough” toward “who controls the integration layer, and what happens when the vendor leaves.”
Key facts and dates
Palantir is the originator of the role, documented under the “forward-deployed engineer” title by 2009; the company has described itself as organized substantially around FDEs for much of its history, with FDEs reportedly outnumbering core product engineers before the 2016 launch of its Foundry platform. OpenAI launched its own forward-deployed engineering function in early 2025 under Colin Jarvis, growing from roughly two engineers to more than ten across eight cities, according to reporting by the Pragmatic Engineer newsletter; other accounts describe further growth and a 2026 restructuring into a larger, separately-backed unit. Anthropic, Google Cloud, AWS, and Microsoft have all announced comparable programs in 2026.
On September 29–30, 2026, Gartner published research predicting that up to 70% of enterprises will abandon agentic AI systems built through vendor forward-deployed engineering by 2028, citing rising maintenance costs, vendor lock-in, and the fact that fewer than 20% of FDE engagements convert bespoke customer requirements into features incorporated into the vendor’s standard product. Gartner analyst Mukul Saha framed the risk as a matter of engagement structure — scope, incentives, governance, and exit terms — rather than a flaw in agentic AI itself. Gartner also warned of “FDE washing,”: ordinary consulting services relabeled as forward-deployed engineering and priced at a premium without the operational depth the term implies.
Where to learn more
- Gartner Predicts 70% of Enterprises Will Abandon Agentic AI Built by Vendor Forward-Deployed Engineering by 2028 — Gartner’s original press release and the source of the 70% prediction.
- Forward Deployed Engineer — Wikipedia — overview of the role’s origin at Palantir and its 2024–2026 spread across major tech companies.
- What are Forward Deployed Engineers, and why are they so in demand? — The Pragmatic Engineer — detailed reporting on Palantir’s history with the role and OpenAI’s 2025 buildout.
- Gartner Warns 70% of Vendor-Built AI Agent Projects Face Abandonment by 2028 — Techstrong.ai — secondary coverage with direct analyst quotes from the Gartner research.