What Is Forward Deployed Engineer (FDE)?

A forward deployed engineer (FDE) is a software engineer who works inside the customer’s environment to make a product work on the customer’s real systems. They are part engineer, part consultant, part product researcher. They do not hand over a spec and leave.

The role came from Palantir. It now shows up wherever AI products need heavy integration, which is most places that sell AI agents.

Where it came from

Nabeel Qureshi, a former Palantir engineer, described the role in his post Reflections on Palantir. FDEs were typically expected to go onsite at the customer’s offices and work from there three to four days a week. Day to day, he says, they learned the customer’s business processes in difficult industries, wrote code that got the job done fast (which usually meant technical debt and hacky workarounds), built trust with senior counterparts, did data integration by getting access to enterprise data and cleaning it, and designed software that solved the customer’s actual problems.

That mix of skills is the point. An FDE has to read a messy database, talk to an operations lead, and write the fix in the same week.

Why AI agents revived the role

An AI agent is not a drop-in tool. It needs access to internal systems, knowledge of business rules, and tuning against real cases.

Joe Schmidt of a16z made the case in Trading Margin for Moat (4 June 2025). He argues that enterprise AI products have an implementation requirement because they need deep integrations and context. He describes services staff, sometimes called FDEs, who operationalize models into real solutions, implement, integrate, and solve problems so agents can resolve requests on their own. He says hands-on support can let young companies accept larger contracts, and he frames the trade as giving up some margin for a stronger moat.

Gartner’s June 2025 prediction that over 40% of agentic AI projects would be canceled by the end of 2027, due to costs, unclear value, or weak risk controls, shows what happens without that hands-on help. Integration and value problems sink projects. An FDE works on both.

What an FDE does on an agent project

Maps the workflow. Sits with the people who do the work and writes down the real steps, not the official ones.
Gets data and access. Finds the systems, requests credentials, and handles the awkward formats.
Builds the integration. Connects the agent to the systems it needs, with narrow permissions.
Builds the evaluation set. Collects real cases and expected outcomes.
Tunes and fixes. Reads failures, adjusts prompts, tools, and rules.
Trains the users. Shows staff how to review and correct the agent.
Reports upstream. Feeds recurring needs back to the product team.

FDE versus similar roles

Role	Main focus	Typical difference from FDE
Solutions engineer	Pre-sales demos and scoping	Usually does not ship production code at the customer
Consultant	Advice and process	May not write or own production software
Support engineer	Fixing reported issues	Reactive, not building new capability
Software engineer	Product code for many customers	Not embedded in one customer’s systems

Trade-offs

The model has real costs.

Margin. People time lowers gross margin, which is why a16z framed it as trading margin for moat.
Scaling. Each customer needs engineer attention. Lessons must flow back into the product or the work never gets cheaper.
Custom drift. Every special request pulls the product toward a one-off.
Talent. The role needs rare mixed skills.

My recommendation for a buyer: ask any AI vendor who will actually do the integration work, what they will have access to, and who owns the result. If the answer is “your team, with our documentation,” plan for that cost.

A short example

Say a manufacturer buys an agent to triage supplier quality complaints. A forward deployed engineer spends two weeks on site. They learn that complaints arrive by email, a spreadsheet, and a phone log. They connect the agent to all three, find that part numbers follow three formats, write a normalizer, and collect 40 past complaints with correct routing. The agent goes live on one product line first. The engineer sits with the quality team for the first week and fixes misroutes as they appear.

Common mistakes

Treating the FDE as free labor for every request.
No path to productize repeated work.
Giving the engineer broad access with no review.
Measuring effort instead of customer outcomes.
Skipping the handover, so the customer cannot run it alone.

An AI agent is what the FDE gets working. Pilot purgatory is the stall this role aims to prevent. Outcome-based pricing can reward vendors who get integrations right. The venture studio model uses shared builders across companies. For a short-cycle approach, see how to ship AI pilots in one week.