A forward deployed engineer (FDE) is a senior engineer who works inside one customer, on that customer's systems and data, and is judged on whether a business problem is solved in production, not on hours billed or a demo. Palantir created the role. AI companies now hire for it, because their software kept stalling at the customer's door.
Most of what is written about the role is for engineers who want the job: salaries, interviews, travel. This piece is for the other side of the table, the company that might end up with one in its office. Everything below comes from what the companies have published themselves: blog posts, job ads and interviews.
Where does the term forward deployed engineer come from?
From Palantir, which needed a name for engineers who did not sit with the product team. In a 2019 post explaining its engineering roles, it split them into "Devs", who build the core platform, and "Deltas", who work at the customer (Palantir). Its own shorthand is still the clearest definition: a Dev's focus is "one capability, many customers", a Delta's is "one customer, many capabilities".
The Delta was never meant to work alone. Palantir's job ads still file its forward deployed engineers under a team called Delta and its Deployment Strategists under one called Echo. The Delta builds. The Echo owns the relationship, the scope and the adoption, down to running training so the product is "used widely enough to have concrete impact" (Palantir). Hold on to that pair. Almost every company that copied the role copied the pair as well.
Why is everyone hiring forward deployed engineers now?
Because the hard part of enterprise AI has moved from the model to the customer. Job postings for the role rose more than 1,000% between January and August 2026 compared with the same months a year earlier, while the tech job market as a whole grew 13%, according to Lightcast data reported by Fortune.
The companies explain it in their own words. Wonderful, which raised $550 million at a $5 billion valuation in September (TechCrunch), wrote when it announced its previous round that "enterprise AI will not scale through technology alone" (Wonderful). Anthropic's job ad describes an engineer who "embeds directly with our most strategic customers" and ships "MCP servers, sub-agents, and agent skills" for them (Anthropic). Scale AI's ad locates the difficulty in systems that "reliably operate inside complex production environments" (Scale AI).
Read together, these sound like an admission. The vendors with the strongest models found that a model sold as a licence rarely reaches production by itself, so they started sending engineers with it.
What does a forward deployed engineer do all week?
Less model work than the title suggests. Gergely Orosz, who follows the role in The Pragmatic Engineer, estimates the split at roughly a quarter coding, half integration and plumbing, and a quarter meetings (The Pragmatic Engineer).
And the coding happens in an unusual place. At OpenAI, FDEs "write code directly on customer infrastructure", while solutions architects build proofs of concept "with anonymized or offline cuts of data" (The Pragmatic Engineer). A typical week covers:
- Reading the customer's data where it actually lives, with the access a real user has.
- Writing the connections to the systems the AI has to act in: the ERP, the CRM, the document store.
- Building an evaluation set from real cases, with the answers the business considers correct.
- Sitting with the people who will use the system, and changing it when they work around it.
- Reporting to a sponsor against a number agreed before the build.
Why on site? Colin Jarvis, who leads forward deployed engineering at OpenAI, gave the reason in the same interview: "often what the customer describes in scoping doesn't match the data/system reality on the ground". The only way to find the difference is to be where the data is.
How is a forward deployed engineer different from a consultant, a solutions engineer or an outsourced developer?
The four overlap in skills. They differ in when they work and what they answer for.
| Role | When they work | What they deliver | What they answer for |
|---|---|---|---|
| Consultant | At one point in time | A recommendation | The quality of the advice |
| Solutions engineer | Before the sale | A demo or a proof of concept | Winning the deal |
| Outsourced developer | While the contract runs | Code written to a specification | Tickets closed, hours billed |
| Forward deployed engineer | From scoping until the system runs and is handed over | A system running on the customer's data, used by its people | The business result in production |
Palantir drew the line against consulting itself. Its Deltas "measure success in terms of impact on the customer's goal", while consultants "generally create a one-time analysis, recommendation, or solution" (Palantir). The line against the solutions engineer is where the code runs: on a copy of the data before the sale, or on the customer's own systems after it.
The line against outsourcing matters most to a buyer, because on an invoice the two can look alike. An outsourced developer is told what to build and is paid for the time it takes. A forward deployed engineer is expected to find out what should be built, and to say so when the answer is "not this". The test for any offer that uses the title is short: who answers for the system three months after go-live, and which number is it supposed to move? If nobody can say, the title is new and the work is not.
Does a smaller company need a forward deployed engineer?
It needs the function more than the title. Anthropic's ad already said who gets its engineers: its most strategic customers. A company of 200 people is rarely on that list, and it has the same gap between a licence and a working system. The version that fits it is smaller: one engineer, part of the week, one use case, and someone who owns the result.
Why that fits a mid-sized firm better than more software, a large consultancy or a new hire is the argument of why AI projects fail without engineers on site. How Palantir, OpenAI and Wonderful run the model, and what a smaller company can borrow from each, is in the forward deployed engineering model.
At Sapio we do this as a sequence. It starts with the contact form on sapio.ro and a short call, after which I send a custom offer: AI consulting, or the Tech Audit, one morning on site and a report in three working days. Then comes a transformation roadmap, and only after that a senior engineer embedded about three days a week, who owns one use case until it runs in production.
"Forward deployed" sounds like jargon until you take it literally. Forward is where the plan meets the terrain. Every AI system is designed on a slide and runs on data the slide never showed, and the role exists for the distance between the two.
If you are working out who should own an AI system inside your company: vladtudor.com/consulting.