Published by: ChinFDE Version: v0.1 (draft for comment) Release date: 2026-11-13 License: CC BY 4.0
What This Standard Is
FDE (Forward Deployed Engineer) is an engineer embedded at a customer's site who turns AI capability into business results. Their work starts with understanding the customer and runs until the system is launched and kept in continuous use.
This standard answers one question: Which observable capabilities must a person have to be qualified for FDE work?
It serves three purposes:
- Learners use it to understand what to practice and what level they are currently at.
- Employers use it to describe role requirements, design interviews, and understand what a ChinFDE certificate represents.
- ChinFDE uses it to design simulated projects and scoring rules. Every score on the platform maps to a dimension in this standard.
This is v0.1, a draft for comment. We welcome suggestions from practicing FDEs, employers, and training peers. See the end of this document for how to send them.
Design Principles
Judge outputs, not résumés. Education, years of experience, and which frameworks someone has used are not part of the standard. The standard describes only behaviors and outputs that can be seen in a concrete delivery project.
AI tools are allowed by default. Real FDEs use AI every day to write code, look up information, and draft documents. So the standard does not test memorization or handwritten code. It tests judgment: which scenario to choose, which requirements to cut, how to prove something works, and what the fallback is when something goes wrong.
Business results come before technical complexity. Solving a real problem with the simplest approach scores higher than solving a non-existent problem with a complex architecture.
Security and compliance are the baseline, not bonus points. No matter how well a candidate does in any dimension, it cannot offset data leaks or out-of-bounds advice.
The Six Capability Dimensions
| Code | Dimension | Core question |
|---|---|---|
| D1 | Business insight and scenario judgment | Is what the customer asks for the same as what the business really needs? |
| D2 | Solution design and scope management | What do we build first, what do we explicitly not build, and which metrics prove it works? |
| D3 | Engineering implementation | Can models, knowledge, and business rules be assembled into a usable system? |
| D4 | Evaluation and quality assurance | How do we prove it holds up on problems it has never seen? |
| D5 | Launch, delivery, security and compliance | Can data leave the domain? Who is responsible when something goes wrong? How do we roll out gradually and roll back? |
| D6 | Communication, collaboration and value proof | Can the boss, the business team, and IT all be willing to move forward, and can the return on investment be worked out? |
D1 Business Insight and Scenario Judgment
Customers often bring a solution ("We need an AI customer service bot") rather than a problem. The FDE's first task is to trace it back to the underlying problem and judge which scenario is worth doing first.
Observable behaviors:
- Interview decision-makers, business owners, and IT leads separately, and recognize that their positions differ.
- Elicit quantified current-state facts: business volume, the makeup of problems, time spent on existing processes, and where errors occur.
- Elicit constraints: where the data is, whether it can leave the domain, whether systems have interfaces, budget, and timeline.
- Elicit the real success criteria instead of stopping at vague goals like "improve efficiency."
- Identify gaps between the customer's goals and their actual conditions, and point them out.
| Level | Performance |
|---|---|
| L1 Foundation | Given a project background, identifies the main pain points and key constraints through interviews, and explains how the customer's stated needs differ from their real needs. |
| L2 Professional | On projects with incomplete information and conflicting stakeholder accounts, conducts research independently and makes well-grounded choices among several candidate scenarios. |
| L3 Senior | Quickly builds business understanding in an unfamiliar industry, helps the customer see high-value scenarios they had not noticed, and influences the customer's project approval. |
D2 Solution Design and Scope Management
A good solution is first a list of trade-offs. Scope getting out of control is one of the most common causes of failure in AI projects.
Observable behaviors:
- State in one or two sentences what the first phase does, who it serves, and what problem it solves.
- Explicitly list what is not being done, with reasons.
- Success metrics are quantified, have a baseline, and have a measurement method.
- List key assumptions and risks, along with responses if an assumption fails.
- The solution matches the facts gathered in interviews, rather than applying a generic template.
| Level | Performance |
|---|---|
| L1 Foundation | Writes a first-phase solution with clear scope, a "not in scope" list, and quantified metrics. |
| L2 Professional | Designs a phased roadmap, makes trade-offs under budget, schedule, and system limits, and the solution holds up under scrutiny from all parties. |
| L3 Senior | Designs solutions for complex, multi-team, multi-system projects, and continuously manages scope changes throughout the project. |
D3 Engineering Implementation
Assembling models, knowledge, tools, and business rules into a system that works. This dimension does not require training models from scratch. It requires reliably combining existing capabilities.
Observable behaviors:
- Prompts have a clear structure: role, task, boundaries, output format, and what to do when uncertain.
- Knowledge sources are curated: outdated, conflicting, or sensitive content is excluded or explicitly handled.
- Fallbacks are designed for things the model cannot do or should not do, such as handing off to a human, collecting contact details, or refusing.
- Adversarial inputs are handled, such as prompt injection and attempts to extract internal information.
- L2 and above: Completes retrieval-augmented generation, tool calling, and integration with business systems, while controlling cost and latency.
| Level | Performance |
|---|---|
| L1 Foundation | Configures a question-answering or processing assistant based on given knowledge, with complete boundaries and fallback rules and sound knowledge selection. |
| L2 Professional | Builds a complete application independently, including retrieval, tool calling, and system integration; the code is maintainable and runs in the customer's environment. |
| L3 Senior | Designs multi-agent, multi-system architectures, solves performance, cost, and reliability problems, and builds reusable components. |
D4 Evaluation and Quality Assurance
"I tried a few questions and it seemed fine" is not evaluation. FDEs need to show how the system performs on inputs it has never seen, and know where it will fail.
Observable behaviors:
- Test cases cover normal questions, edge cases, questions outside the knowledge base, and adversarial inputs.
- Each test case has a clear pass criterion.
- When a case fails, the FDE can identify whether the cause is knowledge, prompting, or model capability, and make a targeted fix.
- Does not rely on trial and error to "pass" tests, but understands why failures occur.
- L2 and above: Builds automated eval pipelines and continuously monitors quality after launch.
| Level | Performance |
|---|---|
| L1 Foundation | The configured system passes most hidden test cases, and failed cases are attributed to reasonable causes. |
| L2 Professional | Designs their own eval sets and pass criteria, builds a repeatable eval pipeline, and uses data to support every change. |
| L3 Senior | Builds a long-term quality system for the customer, including production monitoring, spot-check mechanisms, and regression testing. |
D5 Launch, Delivery, Security and Compliance
There is a lot between a successful demo and a successful launch: data, permissions, networking, compliance, fallback, monitoring, and rollback.
Observable behaviors:
- Chooses a deployment model based on data sensitivity, and is clear about which data must not leave the customer's environment.
- Understands the basic requirements of personal information protection in China, and knows when to choose domestic models or local (on-prem) deployment.
- The launch plan includes a gradual-rollout scope, human fallback, monitoring metrics, and rollback conditions.
- Refuses requests that would create security or compliance risk, even when they come from the customer's decision-maker.
- Designs maintenance mechanisms for knowledge and systems: who updates them, how often, and how errors are reported back.
| Level | Performance |
|---|---|
| L1 Foundation | Writes a launch plan that includes gradual rollout, fallback, monitoring, and data handling, and recognizes and refuses obvious compliance risks. |
| L2 Professional | Completes deployment in the customer's real environment, handles constraints such as network isolation, permissions, and local deployment, and completes handover. |
| L3 Senior | Leads launches in high-compliance industries (finance, healthcare, government, etc.), and co-develops plans with the customer's security and legal teams. |
D6 Communication, Collaboration and Value Proof
An FDE's output must ultimately become a reason for the customer to keep investing. That requires getting people with different positions aligned, and working out the value clearly.
Observable behaviors:
- Tailors emphasis to each audience: results and risk for decision-makers, process and metrics for the business, security and upkeep for IT.
- When saying "no," offers an alternative and the reasoning behind it.
- Calculates return on investment using the customer's own data.
- Respects frontline staff's experience, treating them as sources of knowledge and co-builders of the system.
- Is honest: does not overstate results, does not hide risks, and does not make promises that cannot be kept to win a deal.
| Level | Performance |
|---|---|
| L1 Foundation | Builds basic trust in interviews, responds to unreasonable customer requests with clear reasoning, and does simple return-on-investment estimates. |
| L2 Professional | Manages expectations across multiple stakeholders, handles disagreements and changes during the project, and takes a pilot project through to renewal or wider rollout. |
| L3 Senior | Becomes an advisor trusted by the customer's senior leadership, influences the customer's AI strategy, and leads a team through delivery. |
Scoring Method
On the ChinFDE platform, each simulated project produces two kinds of evidence:
- Process and document evidence: Interview notes, solution documents, system configuration, and launch and retrospective documents. Examiners score each of the six dimensions from 0 to 4 against this standard.
- Objective test evidence: The pass rate of the learner's configured system on a hidden test set.
| Score | Meaning |
|---|---|
| 0 | Not shown: no relevant output, or output unrelated to the project |
| 1 | Emerging: an attempt was made, but key points are missing or judgment contains clear errors |
| 2 | Competent: identifies the main issues and the approach is basically workable, but lacks specificity or has gaps |
| 3 | Proficient: judgment is accurate, the approach is specific and executable, and the main risks are considered |
| 4 | Outstanding: a clear chain of reasoning and explicit trade-offs, approaching the level of a senior FDE |
The total score for a project is 100 points: the six dimension scores are converted to 70 points, and the hidden-test pass rate is converted to 30 points. A total of at least 70 points with no dimension scored 0 counts as a pass.
Three Certification Levels
| Level | Intended for | Assessment method | Status |
|---|---|---|---|
| L1 Foundation | Engineers, product managers, implementation consultants, and new graduates who want to enter the FDE field | Complete and pass all simulated projects. AI examiners score; a hidden test set is used for objective evaluation. | Open |
| L2 Professional | Practitioners who can independently own one customer project | A timed comprehensive project requiring a runnable system to be submitted, plus a defense under real-name verification. AI scoring plus human review. | In preparation |
| L3 Senior | Senior FDEs who can lead complex projects and manage teams | Submit real project cases, reviewed by a review panel. | In preparation |
A Note on L1
L1 is scored by AI examiners, with no human proctoring or identity verification. It certifies that the holder completed all simulated projects on the platform and that their outputs met the L1 requirements of this standard. It is suitable for showing learning achievement and foundational capability, and should not be treated as full proof of independent delivery ability. This is also why we set up L2 and L3 and introduced human review.
Nature of the Certificate
A ChinFDE certificate is a capability certification issued by ChinFDE based on this standard, a form of capability assessment conducted independently by an enterprise. It is not a national occupational qualification certificate, nor a vocational skill level certificate. Each certificate has a unique number, and anyone can verify its authenticity and the corresponding results on the ChinFDE website.
Versions and Feedback
- Current version: v0.1, released 2026-11-13, a draft for comment.
- Every revision of the standard keeps its earlier versions. Issued certificates state the version of the standard they are based on.
- Comments and suggested changes are welcome via GitHub.
This standard is released under CC BY 4.0. You may freely reproduce, cite, and adapt it, provided you credit "ChinFDE, FDE Capability Standard" and include a link.