Picture a software company in the early 2010s with a strange problem: its software was powerful, its customers badly needed it, and almost none of them could actually use it.

The company was Palantir. Its customers were government agencies and large enterprises sitting on huge, messy, disconnected data, often in secure environments that no outside tool could simply plug into. Shipping a product and waiting for customers to figure it out wasn't going to work.

So Palantir did something unusual. Instead of keeping its engineers at headquarters, it sent them out to live with the customer.

The engineers called "Delta"

Internally, these engineers were known as Deltas, officially Forward Deployed Software Engineers. Their job was simple to describe and hard to do: sit with the customer, understand the real problem, and write whatever code it took to make the software work in that customer's world.

Palantir described the difference neatly: a normal product engineer serves many customers with one capability. A forward deployed engineer serves one customer with many capabilities.

It wasn't a side team. According to The Pragmatic Engineer, until around 2016 Palantir had more forward deployed engineers than "normal" software engineers. Some worked in air-gapped environments with no internet. One former FDE described working on the final assembly line at Airbus.

For years, the model looked like a Palantir quirk. Then generative AI arrived.

History repeats itself

Fast-forward to the AI boom, and AI labs hit exactly the same wall Palantir did.

The models were astonishing in demos. But a model on its own doesn't reduce support tickets, process insurance claims or speed up a factory. It only creates value once someone connects it to a company's documents, systems, permissions and rules, and proves it works well enough to trust.

So the AI labs borrowed the playbook:

  • OpenAI built a forward deployed engineering team that started with a handful of engineers and expanded to cities including San Francisco, New York, London, Paris, Munich, Dublin, Tokyo and Singapore.
  • Anthropic said it planned to grow its applied AI team, which includes forward deployed engineers, fivefold.
  • Cohere and a long list of AI startups followed with FDE roles of their own.

The Financial Times reported that job listings for forward deployed engineers rose by more than 800% between January and September 2025. A job title that once belonged to one company had become one of the fastest-growing roles in tech.

What the job looks like today

The modern AI FDE does what the Deltas did, with a new toolkit. A typical engagement moves through four stages:

  1. Embed: spend time with the people doing the work and learn how things really happen.
  2. Scope: turn "we want AI" into one concrete, measurable problem.
  3. Prototype: build something rough in days and put it in front of real users.
  4. Production: make it secure, tested and reliable enough to keep running after you leave.

Along the way, an AI FDE builds retrieval (RAG) so the assistant answers from the client's own documents, agents that take actions only after human approval, secure connections into business systems, and evaluations that prove the system is good enough to launch.

And then comes the part most engineers underestimate: explaining all of it to a business leader in plain language, including where the AI is still uncertain.

Why this role is worth paying attention to

Most engineering roles ask you to be good at one thing. The FDE role asks for an unusual mix: solid coding, comfort with messy real-world data, product judgement, and the ability to earn a client's trust. That combination is rare, which is exactly why demand has outpaced supply.

The good news is that it's learnable. Most of the toolkit is open source, and the customer-facing skills come from practice, not credentials.

Where to start

If you're an FDE aspirant, check out Introduction to Forward Deployed Engineer (FDE). Instead of teaching tools in isolation, it follows a new FDE named Sam through one complete client engagement, building an AI assistant step by step with LangChain, LangGraph, MCP, LangSmith and FastAPI. Along the way you make the calls yourself in branching scenarios: whether to automate refunds, whether to launch after evaluation scores drop, and how to present results to a business leader. No paid API keys are needed to follow along.

The role Palantir invented to solve its own problem has become one of the best entry points into applied AI. The next Delta might be you.

Sources: The Pragmatic Engineer: What are Forward Deployed Engineers? · eWeek: OpenAI, Anthropic, Cohere step up FDE hiring · PYMNTS: FDEs emerge as one of AI's fastest-growing jobs

— Team SkillMint