Plank Research · Explainer
What is a forward-deployed engineer?
Definition
A forward-deployed engineer (FDE) is a software engineer who embeds inside a customer’s organization to make a product actually work in their environment. They write production code against the customer’s real data and systems, and they own whether it delivers a result, not just whether it ships.
The role began at Palantir. It spread across the AI industry in 2026, once it became clear that the hard part of enterprise AI is deployment, not the model. Here is the real mental model, in a few minutes.
The distinction that matters
Most engineering roles are measured on what they build. A forward-deployed engineer is measured on what happens after. The distinction sounds small and is not: it changes who you hire, how you sell, and what “done” means. A product engineer ships a feature and moves on, and success is that the feature works as specified. An FDE ships into one customer’s live operation and is accountable for whether that customer gets the outcome they were promised. If the software works but nobody adopts it, the FDE has failed.
The whole role reduces to three things held at once. Take away any one and you have a different, weaker job.
Builds
Writes production code in the customer's systems. Not demos, not decks, not recommendations.
Embeds
Works inside the customer's environment, data, and politics, against inputs no demo has to survive.
Owns
Accountable for the outcome in production, not for shipping a feature to spec.
Why the role has to exist
You might reasonably ask why this is a job at all. If a product is good, shouldn’t the customer just use it? The answer is the most important idea in the category, and it is counterintuitive.
For complex software, and especially for AI, most of the value is not in the software. It is in the integration with a specific business’s data, workflows, and tacit rules, and that integration is human-shaped.
Every organization runs on knowledge that exists only in people’s heads: which database fields are actually trustworthy, why underwriting really rejects certain applications, what the true approval chain is when the documented one is ignored. A model can be superhuman at a task and still useless until someone encodes that context around it. You cannot document your way across this gap, because the customer usually cannot articulate the rules themselves until an engineer is sitting next to them asking why the number is wrong. Foundation Capital calls the missing layer a company’s “context graph”: the reasoning behind its decisions, which was never captured as data. That is why deployment is a person and not a manual.
Where it came from
The role and the name come from Palantir, around 2006. The company found that selling data-integration software to intelligence and defense customers did not work: the software was powerful and the customers could not operationalize it. So Palantir stopped shipping software and started shipping engineers, sending them into agencies to build the workflows directly. Shyam Sankar, an early employee, is generally credited with turning this into a repeatable model. For roughly its first decade Palantir employed more forward-deployed engineers, internally called “Deltas,” than product engineers.
Two structural details from that origin still define the category.
The pairing
A Delta (the engineer) works alongside an Echo (a deployment strategist who owns adoption, politics, and executive alignment). The technical last mile and the organizational last mile are different problems, staffed separately.
The flywheel
Patterns Deltas discover inside customers feed back into the core product, so each deployment makes the next one cheaper. Consulting sells the same hours twice. Done right, FDE work compounds into product.
What it is not
Because the role sits between engineering, sales, and consulting, it is constantly confused with its neighbors. The differences are sharp once you look.
| Sales / solutions engineer | Supports the sale through demos, POCs, and pre-sales architecture. Their job ends when the customer says yes. An FDE's begins there. |
| Management consultant | Delivers analysis, strategy, and recommendations. An FDE delivers running code and owns the result, not the slide deck. |
| Professional services / SI | Implements to a fixed spec on a statement of work. An FDE is open-ended, embedded, accountable for the outcome, and feeds learnings back to the product. |
| Developer relations / customer engineer | Scales shallow across many users through docs and content. An FDE goes deep inside one customer's production system. |
The through-line: an FDE builds, embeds, and owns the outcome. Each neighboring role has one or two of those. Only the FDE has all three, and the combination is the point.
Why it is suddenly everywhere
Forward-deployed engineering was a Palantir peculiarity for fifteen years. In 2026 it became a category, for one reason: AI made the deployment gap impossible to ignore. The models got good faster than companies could absorb them, the bottleneck moved from model capability to human-shaped integration, and the industry bought the model wholesale.
- $9B+
- committed to forward-deployed ventures by OpenAI, Anthropic, Microsoft, and Amazon in 2026
- 95%
- of enterprise GenAI pilots showed no measurable P&L return (MIT NANDA)
- $185K
- median forward-deployed comp; past $250K at the frontier labs
- 85%
- of open FDE roles are at startups and scale-ups, not the labs
- 3%
- of open FDE roles are early-career: it is a senior job
- $1.5M
- Palantir revenue per employee, with almost no traditional sales force
Within a year, OpenAI, Anthropic, Microsoft, and Amazon each stood up a forward-deployed venture explicitly copying Palantir: OpenAI’s Deployment Company, Anthropic’s Ode, Microsoft’s Frontier, and AWS’s FDE unit. The less obvious fact, from the actual hiring data, is that this is not primarily a big-lab phenomenon: roughly 85% of open forward-deployed roles are at startups and scale-ups, not the labs that get the headlines.
The case against it
An honest account has to include why smart people distrust the model.
Incentives
When the FDE works for the vendor whose product they install, they are paid, in effect, to deepen your dependence on that vendor. Andrew Ng put it plainly: letting a lab's FDEs wire your processes to their stack “significantly reduces optionality.” They help you and entrench a single supplier at the same time.
What gets left behind
Deployment is only the start of an AI system's life, roughly 20% of its lifetime cost; the rest is operation and change. If the FDE ships a working system but leaves no documentation, no patterns, and no in-house capability, you have not bought a solution. You have rented a dependency, and the meter keeps running.
It does not scale like software
The model is bound by people, so Gartner and others expect many enterprises to walk away from FDE-heavy engagements on cost. The counterargument is the flywheel: the providers for whom this works turn each deployment into reusable product. Whether a given provider does that, or simply bills the hours, is the question worth asking before you hire one.
When you actually need one
The right answer when
The product is genuinely hard to integrate, the customer’s real data and workflows are messy enough that no demo predicts production, and the stakes are high enough that a failure to adopt is expensive. Most enterprise AI clears all three bars.
The wrong answer when
The product is self-serve, well-documented, and low-integration. Sending an engineer to sit with a customer who could have succeeded from the docs is just expensive.
And the person who does it well is a specific type: senior enough to make architecture decisions alone, comfortable working in someone else’s codebase and someone else’s building, and willing to own an outcome they only partly control. It is one of the few engineering roles where the hardest problems are as often organizational as technical, and where being right about the code is not enough if nobody uses it.
Go deeper
- The State of FDE as a Service, 2026 · the market, the $9B of lab ventures, and the deployment gap in depth.
- The FDE job market, mapped · a census of open roles, filterable by vertical, stage, region, and pay.
- The FDE race: key player profiles · how Palantir, OpenAI, Anthropic, Microsoft, AWS, and Distyl run it.
Sources: Palantir filings and executive interviews (Shyam Sankar); MIT NANDA, The GenAI Divide(2025); S&P Global Market Intelligence (2025); Andrew Ng, The Batch (2026); Foundation Capital, Context Graphs (2026); and the Plank FDE job census (n=1,206, July 2026). Figures indicative; verify before quoting.