What Is a Forward Deployed Engineer?

A forward deployed engineer (FDE) is a software engineer who works inside a customer's business instead of from the vendor's office. They learn how the operation actually runs, build the software in place against the customer's real systems, and stay on after launch so it keeps working. Palantir created the role in the early 2010s. In 2026 the largest AI companies started building whole firms around it, because the models are good enough and getting them to run inside a real business is now the hard part. This article covers where the role came from, why the AI labs want it now, what an FDE does day to day, and what the model looks like for a business too small for any of those labs to send one.

Where the role came from

Palantir created the role in the early 2010s and called it "Delta", according to Gergely Orosz's account in The Pragmatic Engineer. "Forward deployed" is military language for operating at the point of action rather than from a rear base. Palantir's customers were government agencies and large companies whose data and systems were too specific to build for from a distance, so the company put its own engineers on site. They wrote production code against the customer's real systems, sat in the customer's standups, and split their weeks between building for the customer and feeding what they learned back into Palantir's product.

That split is the point. An FDE is not a consultant who writes the plan and leaves, and not a contractor who builds to a spec and hands over. The engineer is in the room long enough to know which parts of the spec were wrong.

Why the AI labs want forward deployed engineers now

Two announcements in 2026 turned an unusual Palantir job title into an industry model.

On May 4, Anthropic, Blackstone, Hellman & Friedman and Goldman Sachs announced a standalone enterprise AI services firm, with Anthropic, Blackstone and Hellman & Friedman each committing $300 million. OpenAI's own venture, The Deployment Company, raised $4 billion from 19 investors the same day. TechCrunch described Anthropic's firm as adopting "the forward-deployed engineer (FDE) model popularized by Palantir." Goldman Sachs' announcement put it more plainly: the new company exists to give mid-sized companies in the partners' portfolios "access to forward-deployed engineers." We covered it in our May 14 roundup.

In June, OpenAI launched its Partner Network with $150 million behind consulting firms and systems integrators, in three tiers: Select, Advanced and Elite. For the most complex deployments it is piloting a Forward Deployed Experts program that pairs partner practitioners with OpenAI's own forward deployed engineering teams. That one was in our June 25 roundup.

The reason is the same in both cases. The model is no longer the bottleneck. Any business can buy access to a frontier model, and many have. What they cannot buy is someone who knows their quoting process, their job system and their staff well enough to make the model do something useful on a Tuesday. The labs have concluded that the only way to close that gap is to put engineers inside the business. That is a forward deployed engineer.

What a forward deployed engineer actually does

Strip the title away and the work is three things in a loop.

Learn the operation. The engineer sits with the people doing the work and maps how it actually moves, not how the process document says it moves: where a job starts, who touches it, where it waits, and where the same information is typed twice. This is where most of the value is found, because most businesses have never written down how they run. The knowledge lives in three people's heads and a group chat.

Build in place. Against the live systems, in front of the people who will use it. The first version ships in the first week or two so feedback shapes the build rather than arriving at handover. The engineer is not building a demo for a steering committee. They are building the thing the dispatcher will use at 7am.

Stay on. The first month of a system running for real reveals what the scoping call could not. The engineer stays to fix that, then picks the next bottleneck. The knowledge of how the system works stays with the team rather than walking out with a contractor.

Compare that with the three ways a business usually buys this. A consultant does the first step and writes a report. An agency does the second step to a spec and hands over. An in-house hire does all three but knows one stack, takes leave, and takes the knowledge with them when they go. The forward deployed engineer is the seat that does all three and stays.

Who actually gets one

Enterprise. The Anthropic venture is built for the mid-sized companies inside private equity portfolios. OpenAI's Forward Deployed Experts program is for its partners' most complex enterprise deployments. Palantir's forward deployed engineers have always gone to governments and the largest companies.

If you run a trades business, a distributor, a dental practice or a textile manufacturer, nobody is sending you one. The model the labs have just decided is the answer for their largest customers has never been available to the business whose owner is still doing the quoting at 9pm. That is the gap Kursol works in.

What the model looks like at a small business

Kursol is a forward deployed engineering team for small and mid-sized businesses in the United States and Australia. We embed with the business, learn how it runs, build the AI and automation that takes the manual work off the team, and stay on so it keeps working. The business gets the in-house engineer and the consultant in one seat, without hiring either. Here is what that has meant in practice.

A field service business. Its crews work on site, away from the office. When we arrived, the business ran on a job system, an accounting system and a sprawl of group chats that never talked to each other. Crews wrote up their jobs from memory at the end of the day, and on busy days that did not happen at all.

The first thing we did was not code. We recommended the team move off the group chats and onto one shared workspace, because nothing useful can be built on a chat that is not searchable and cannot be programmed. Then we built a way for the crew to send notes and photos to the job from site, and an AI system that wrote the job up from those notes whether or not anyone had time to.

Months in, the crew changed how they worked. They left the workspace and started typing notes straight into the job system, and the capture we had built went quiet. The notes had not stopped. They had moved.

A build handed over at launch would have ended there. We went to where the notes were. Every evening the AI now reads the crew's notes on each open job, keeps an internal summary, and adds two lists to the invoice: the extra work done and the repairs recommended.

The crew's raw notes stay off customer invoices, as the office asked, and the office checks both lists before it bills. The lists went live the morning they said yes, with the office's guide updated the same day.

One of the managers had asked, months earlier, for a way to catch quoted work the crew never wrote down. That runs now too, on trial, so the office can ask the crew before invoicing whether something was done or just not written down. On one of the first jobs it read, it flagged quoted work still to be done and parts charged with no crew note behind them.

Each step came from being in the operation and seeing the next problem, not from a scoping call. We are still in their workspace, working through the owner's list.

Koton Kraft, a Sydney textile manufacturer. The director sells face to face, so the sale runs from an iPad catalogue of the company's designs, and one press turns an order into an invoice. Invoices that took up to an hour, built twice, now take seconds, built once. Production sees an order the moment it is agreed, and the whole system runs on infrastructure the company owns.

None of this required a billion-dollar venture. It required an engineer who stayed long enough to see where the work really happened, and built there.

What to ask before you sign up for one

The title is about to be everywhere, so here is how to tell whether what you are being offered is the model or just the name.

• Will the engineer see the operation before scoping it? If the proposal arrives before anyone has watched a job move through the business, it is a spec, not a diagnosis.
• What ships in the first week? Working software in front of real people, or a deck.
• Who is in the channel after launch? The person who built it, or a ticket queue.
• What happens when it breaks at 7am? A response time in the contract, or a hope.
• Where does the knowledge live? In writing and across a team, or in one head that can resign.

Those five questions are also how we would ask you to judge us. Our engineering page covers how we build, and our pricing is public, including the tier that is priced against the developer you would otherwise hire. If you would rather start smaller, the free AI readiness assessment is a quick look at where the manual work in your business sits. And if you are weighing up AI for a smaller business more broadly, start with the wider picture before the engineering.

FAQ