China Market Sample

Knowledge Graph · Research Dossier

FDE at a Glance

Loading knowledge graph

01 / Role & Compensation

Check the methodology first, then the numbers

Domestic job postings usually quote monthly salary and months of pay, while overseas official sites mostly disclose annual base salary. Bonuses, equity, housing subsidies, and relocation support must be confirmed separately.

US official site sample

USD 110K-345K / year

OpenAI, Anthropic, Palantir annual base salary

Verification window

2026-08-09

Positions go offline; current compensation needs to be re-checked

02 / Core Judgment

To judge whether a job is an FDE role, look at only two things.

Whether the customer gets production results, and whether the team leaves behind reusable capability.

Customer customization

The project can be delivered, but the team still relies on starting from scratch next time.

Project delivery

Experience stays with individuals and retrospectives; other teams can hardly use it directly.

FDE closed loop

On-site evidence flows into Skills, connectors, eval sets, or product capabilities.

Scalable FDE

Assets shorten the cycle for the next customer and reduce duplicated work.

03 / Delivery Pipeline

A demo is only the starting point of risk exposure

Production value comes from a complete pipeline. Every stage has exit evidence; you can't advance just by counting documents.

  • Scenario discovery: confirm users, tasks, baselines, system boundaries, and business owners. Output: scenario map
  • Runnable validation: run the critical path with real problems and actively hunt for failure points. Output: runnable prototype
  • Controlled pilot: limit users and risk scope, establish evals and rollback conditions. Output: eval set
  • Production launch: complete permissions, auditing, monitoring, human takeover, and operational responsibility. Output: production plan
  • Organizational adoption: write the new capability into roles, processes, training, and operating rhythms. Output: adoption plan
  • Consolidation & reuse: distinguish customer differences from industry commonalities and turn them into maintainable assets. Output: asset package

05 / Value Assessment

Value requires evidence at four levels simultaneously

Call volume and model scores only explain part of the picture. An FDE must also prove user adoption, production reliability, and faster delivery next time.

Business results

Are tasks completed better

  • Task success rate
  • End-to-end process cycle time
  • Changes in quality, revenue, or risk

Organizational adoption

Do target users keep using it

  • Adoption rate and active frequency
  • Business volume covered by the new process

Engineering quality

Can the system withstand production

  • Failure rate and recovery time
  • Human takeover and regression evals

Reuse efficiency

Is the next delivery lighter

  • Asset reuse rate
  • Reduction in launch cycle time and work hours

Metric discipline

Every percentage needs a defined methodology

Specify the numerator, denominator, baseline, exclusions, time window, and data owner.

npx skills add Wishyouerehere9610/fde-insight \
  --skill fde-insight \
  --agent codex \
  --global \
  --yes \
  --full-depth

06 / Source Boundaries

Source boundaries

The initial knowledge model references the Tencent Research Institute's July 2026 report "FDE Model: Industry Observations and Practices". The original PDF is not distributed with the repository.

Recruiting, compensation, product capabilities, organizational structure, and policy information change over time. When answering these questions, re-verify current primary sources.