What is FDE
FDEs work on the client's site, moving "uncertain business problems" toward AI applications that are "verifiable, runnable and reusable."
What it determines is not "how much customization," but who absorbs the complexity and how it's captured.
Core workflow
An FDE's work revolves around four questions
Business modeling
How does the business actually work?
Solution and executor design
Where exactly are the problems, and how should work be done in the future?
Lightweight application validation
Does this new method actually work in practice?
Engineering · Productization · Operations
How do we run it in production, productize it and deliver business results?
Role comparison
How FDEs differ from similar roles
| Role | Main focus | Difference from FDE |
|---|---|---|
| Traditional implementation / delivery engineer | Deploys, configures and launches against fixed requirements | Requirements are usually set; an FDE must first define the problem |
| Business consultant | Diagnoses problems, designs blueprints and processes | Usually stops at the proposal; an FDE validates and delivers |
| Software engineer | Implements features to spec | Focuses on "building it right"; an FDE also judges "what to build and whether it's worth it" |
| Product manager | Abstracts common needs and plans products | Serves many clients; an FDE is embedded at one client's site and feeds lessons back into the product |
| Customer success manager | Drives adoption, renewals and expansion | Operations-oriented; an FDE is responsible for building the capability itself |
Core capabilities
Six core capabilities of an FDE
Business understanding
Industry priors, company profiling, SOP breakdown
Problem convergence
Carve out a topic with a clear value rationale
Human–AI division of labor
Every node: manual, automated, AI-assisted or AI-executed
Value validation
Metrics, baselines and success thresholds—let the evidence speak
AI engineering
Workflows, Agent + Skill, evaluation, production reliability
Codify and give back
Experience, Skills, data and failure cases become product capabilities
Delivery models & complexity
Don't classify by "how much customization"—look at who absorbs the complexity
They all look like "standard base + local extensions." The difference is who defines the base, who controls the extensions and who bears the cost of adaptation.
- Core goal
- Turn an uncertain business into a one-off system
- Does the client change?
- No change needed; adapt to the client as much as possible
- How it ends
- One-time acceptance, project ends
- Revenue source
- One-time project fee
Mode ① · Pure custom integration
All complexity sits at the project site
"Absorbed and digested by people on site"; People "do everything for the system"
Modes ②③ · Base + custom development / SAP
Complexity is shared in a structured way
The product absorbs part of the complexity + the methodology absorbs part; People "fill the gaps between product and reality"
Modes ④⑤ · Pure software/hardware / SaaS
Complexity is absorbed upfront by the product
"Absorbed and hardened upfront by the product"; People "configure, adapt and guide clients to use standard capabilities"
AI projects start in mode ①; the FDE's mission is to move them toward ②③⑤: on-site people absorb uncertainty → methodology captures the practice (SOP, 10-step method) → product capabilities harden it (Skills, evaluation, data).
Pure custom integrationIndustry background
Where FDE comes from, and why it's needed now
Origin: Palantir
Palantir internally calls FDEs "Delta." They work alongside product engineers, own the client's go-live outcomes and turn field experience into reusable assets.
Overseas: OpenAI / Anthropic
OpenAI's FDE team had about 39 people in 2025 with plans to expand; in May 2026, OpenAI and Anthropic each set up enterprise AI delivery joint ventures with investors, treating "deployment" as the biggest bottleneck beyond models.
China: demand surges
FDE job postings in China grew 21x year over year in H1 2026; ByteDance, Ant Group, Zhipu and others are publicly hiring FDEs, and Moonshot AI launched an FDE partner program in September 2026.
Sources: Palantir official blog, Business Insider, TechCrunch, Maimai and other public reports (as of October 2026).
FAQ
About FDE services
+How is FDE different from outsourced development?
Outsourcing delivers features against requirements; an FDE first defines the problem and sets a value baseline, then validates and delivers, owns the business outcome, and captures the practice as your company's own Skills and methods.
+How long do you need to be on site?
Assessment takes 1–2 weeks, a single use-case POC 4–6 weeks, and the full 90-day on-site engagement covers four stages from pilot preparation to deep integration. On-site and remote work can be mixed to suit the project's pace.
+Who do we need to provide?
One rollout lead (who can sign off on goals and budget), one seed user from each pilot department, and an IT lead to support accounts, permissions and system access.
+How is data security ensured?
Access is tiered from L1 Public to L4 Confidential: read-only first, dedicated integration accounts, and intercept-and-confirm for writes; dedicated cloud / on-premises deployment options are supported.
Tell us about your AI use cases
An FDE consultant will contact you within 1 business day and map out your 3 most worthwhile use cases for free.