xdash/FDE-the-Guidance-Book-of-Forward-Deployed-Engineer: a free Chinese guide to the forward deployed engineer role

What the FDE Guidance Book actually is, and who it is written for

This repository is a book, not a library. The README describes it as a from-zero introduction to the forward deployed engineer role, built on the framework Fan Bing used in his earlier book on growth hacking. Everything ships as Markdown chapters at the repository root, plus a merged PDF named 前线部署工程师（FDE）v1.0.24.pdf. There is no build step, no package manifest and no source directory.

The audience is narrow and specific. The README says the goal is to help engineers considering a career change see the whole shape of the role, to help founders building enterprise AI avoid a detour, and to save anyone trying to understand how AI actually lands inside a company dozens of hours of searching. That is a research audience, not a runtime audience.

The framing comes from two numbers the preface puts side by side: an MIT NANDA lab report cited as saying 95 percent of enterprise generative AI projects produced no value that could be written into a financial statement, against a claimed sevenfold year-over-year increase in FDE job postings at OpenAI, Anthropic and more than a hundred YC companies. Whether you accept that pairing as causal or not, it is the premise the whole book rests on, and the preface states it plainly rather than hedging.

How the book is structured: one chapter per stage of a delivery

The table of contents is the architecture. After a preface, seven chapters walk a single delivery journey in order: 第1章 FDE 的崛起, 第2章 解决正确的问题, 第3章 赢得客户, 第4章 激活部署, 第5章 守住续约, 第6章 扩大收入, 第7章 规模化复制. Chapter 8 is a case collection, and a postscript covers professional ethics.

Three appendices carry the reference material: appendix A lists metrics an FDE should track, appendix B is a roster of people and teams, and appendix C is the full case index with sources. That last one is the part that distinguishes this from a blog series. The README states that every data point and case in the book has its source marked in appendix C, and the preface says the author tried to make each citation checkable and invites corrections through Issues.

The reading paths are also documented. The README points to fde4.ai for the web version, which it says stays in sync with the repository and is faster to reach from mainland China, to the Markdown files for chapter-by-chapter reading on GitHub, and to the root PDF for a single download. It also states that where the PDF and the Markdown disagree, the chapter Markdown is authoritative. That ordering matters if you are quoting the book anywhere.

One structural gap: the repository has a VERSION file at the root and the PDF filename carries v1.0.24, but the README gives no changelog and no release notes. There are no releases. You can see that a version exists, and you cannot see what changed between versions from the repository alone.

Reading it: no install, just open the chapters or the site

There is nothing to install. The README offers three reading routes and no setup instructions, so the honest tutorial is about picking one and reading in order. The first is the project site at fde4.ai, which the README says is mobile-friendly and faster from mainland China. The second is clicking through the contents table and reading each chapter on GitHub. The third is downloading the merged PDF from the repository root.

The most useful form if you want to search across the whole book is the Markdown, which the README lists as the authoritative source when it disagrees with the PDF. The chapter files are numbered from 00-自序.md through 12-附录C-全书案例索引与资料出处.md, alongside README.md, VERSION and the PDF. A sensible first session is the preface plus 01-第1章-FDE的崛起.md for the definition, then appendix C to see how the sourcing actually looks before you trust anything in the middle chapters.

Note that the files are UTF-8 Chinese. A terminal without CJK font support will render boxes rather than characters; opening the Markdown in any modern editor avoids that entirely. The README does not document any offline tooling, search index or build command, because none exists.

The sourcing is the selling point, and also the weak point

The preface lists what went into the research: podcast retrospectives from early Palantir executives, memoirs from former employees, venture capital industry analysis, the original MIT report, dozens of job postings, salary reports, forum complaints from practitioners, and accounts from the first practitioners in China. That is a reasonable spread for a self-published study, and naming the categories is more than most books in this space do.

The limitation follows directly from that method. This is a synthesis of secondary sources, and the preface says as much by calling it a study process. Nothing in the repository is a primary interview conducted by the author, and no methodology section explains how sources were weighted or how conflicts between them were resolved. Appendix C tells you where a claim came from; it does not tell you how strong that source is.

There is also a structural blind spot worth naming. The book is organised around the vendor-side FDE role as practised at Palantir, OpenAI, Anthropic, Harvey and Sierra. A reader at a company that buys AI rather than sells it will find the delivery chapters useful and the go-to-market chapters less so, because the perspective is consistently the vendor's. The README does not claim otherwise, but a buyer-side reader should know that going in.

Licence: free to read, not free to reuse

The README's copyright section is unusually explicit. Copyright belongs to the author, Fan Bing. The repository content is authorised by the author for free reading and non-commercial sharing, and any reposting must credit the source and the author. Commercial use, including but not limited to publishing, training and paid content adaptation, requires prior written permission from the author.

For an engineer, this is simple: read it, share the link, quote it with attribution. For anyone at a company that runs paid enablement, builds a course, or folds third-party material into a product, the README puts that behind a permission step. This is not legal advice, and the exact boundary of non-commercial sharing is a question for the author, whose contact details are in the README: WeChat ifanbing and xdash@duck.com.

The repository carries no separate LICENSE file, so the README text is the only licence statement you have. Treat the README as the terms.

Maintenance and what an upgrade actually costs you

The repository is not archived, and the last push was on 2026-08-14, roughly a month before this writing. That is recent enough that the project is being touched, though there are no releases and no published changelog, so the cadence is visible only through the commit history.

The upgrade cost is close to zero and that is the point. Chapters are Markdown files, and the README's rule that the chapter Markdown wins over the PDF means you should re-read the current chapters rather than rely on a PDF you downloaded earlier. The VERSION file gives you a number to compare.

The real cost is citation drift. If you quote a figure from chapter 1 into an internal document, a later revision can change it, and the README offers no diff summary. Pin the commit you read if the quote matters. The PDF filename embeds v1.0.24, so a mismatch between the PDF and the Markdown is expected over time rather than a sign of breakage.

Alternatives, and when this is the wrong book

The closest thing to an alternative is the vendor engineering blog ecosystem. Palantir, OpenAI and Anthropic publish posts about how their forward deployed or solutions teams work, and those are primary accounts written by the people doing the job. The difference in approach is stark: a vendor blog post is a first-hand narrative about one company's practice, while this book is a cross-company synthesis with a citation index. If you want depth on one organisation, read that organisation. If you want the shape of the role across many, this book is the better starting point, precisely because it is not selling a platform.

Another alternative is a general enterprise AI adoption book. Those tend to be written for the buyer and organised around procurement and change management. This book is organised around delivery stages and written from the vendor's side, so the two cover different ground rather than competing.

Where this is plainly the wrong tool: if you do not read Chinese. The entire repository, including the README, the chapters and the appendices, is in Chinese, and there is no translation in the repository. If you need code, sample architectures or anything executable, there is none, and you should look elsewhere. If you want a short answer to what an FDE is, a single chapter is a large commitment compared with a definition page.

Editorial conclusion

Adopt this if you are an engineer, founder or solutions architect weighing a move into customer-facing AI delivery and want a structured Chinese-language starting point with cited sources; the appendix C case index and the chapter-per-stage layout are the parts worth your time. Do not adopt it if you need runnable code, English text, or material you can reuse in paid training, since the README states that commercial use requires the author's written permission and that the repository is authorised for free reading and non-commercial sharing only. Before relying on any figure, open 12-附录C-全书案例索引与资料出处.md and check the citation for that specific claim, because the README itself says the data and cases are sourced there and invites corrections through Issues.

Frequently asked questions

Official sources

1. Issues
2. README
3. xdash/FDE-the-Guidance-Book-of-Forward-Deployed-Engineer on GitHub