A forward-deployed engineer (FDE) is a software engineer who works inside a single client’s business — often on-site, writing code directly against that company’s own data and systems — to make a piece of software actually work there, rather than building one generic product for thousands of customers. The role has existed for over a decade, but in 2026 it has become the default hiring pattern at AI labs, because getting a large language model to work reliably inside a real company has turned out to be harder than building the model itself.
What the job actually involves
An FDE’s work blends three things that are usually split across separate roles: writing production code, doing the requirements-gathering a consultant would normally handle, and feeding what they learn back into the core product. Instead of shipping one system to many customers, an FDE spends a large share of their time — often half or more — embedded with a single customer, adapting the software to that customer’s specific data, workflows and edge cases. The company that popularized the role has described the difference as “one customer, many capabilities,” instead of “one capability, many customers.”
That makes it a different job from a traditional software engineer, who mostly builds for an abstract user base, and from a management consultant, who advises but rarely ships the code. A forward-deployed engineer does both — an embedded builder who also owns a piece of the client relationship.
Where the idea came from
Palantir Technologies, the data-analytics company founded in 2003, is widely credited with pioneering the model. One of the earliest documented examples dates to 2009, when JPMorgan Chase had roughly 120 Palantir engineers embedded on-site to help deploy the bank’s software for detecting insider threats. By 2016, Palantir employed more forward-deployed engineers than conventional software engineers — for years, that embedded, customer-specific engineering work was effectively the company’s real product, more than any packaged software license. Other firms used embedded engineers before Palantir gave the role its name, but it’s Palantir’s version, and its results, that AI companies are now explicitly copying with the forward-deployed model.
Why AI companies are copying the model now
Training a capable model turns out to be the easier half of the problem. Making it useful inside a specific hospital, bank or logistics company means connecting it to messy internal databases, enforcing security and compliance rules, and redesigning workflows around it — work a general-purpose chatbot can’t do by itself. That gap between an AI demo and a working deployment is exactly what forward-deployed engineers are hired to close, and it’s why enterprise AI adopters increasingly hire for the role directly, rather than expect it to come bundled with a model subscription.
Demand shows it: postings for “forward-deployed engineer” on Indeed rose more than 700% between April 2025 and April 2026, and Anthropic, OpenAI, Google Cloud and Palantir itself are all now competing for engineers who can do deep technical work and sit across the table from a client in the same day.
The highest-profile bet on this model is Ode, a roughly $1.5 billion venture that Anthropic launched together with Blackstone, Goldman Sachs and several other investors, built around a team of about 100 engineers who embed inside client companies rather than serving every customer through Claude alone. Ode is not the first company to build a business around forward-deployed engineering, and it won’t be the last — but its size shows how seriously AI’s biggest backers are now taking the “last mile” problem: turning a capable model into software that actually runs a business.
In the news
Anthropic and Blackstone’s Ode is the clearest sign yet that forward-deployed engineering — not model quality alone — is what enterprise AI customers are willing to pay a premium for.