Everything public about forward deployed engineering

An eight-chapter Chinese book by the author of 《增长黑客》, running from why the role appeared through winning the account, activating a deployment, holding renewal, scaling revenue and copying the model. Fourteen Markdown chapters plus a PDF.

Good for

The clearest commercial framing of the role in Chinese: it treats a deployment as a revenue cycle, not a project.

Limitation

Prescriptive in places and predates the 2026 wave of agent-native delivery. At 5.4k stars it is the most-starred Chinese FDE repository, which also means it circulates widely in unattributed summaries.

Licence

No OSI licence. The author's terms allow free reading and non-commercial sharing; commercial use needs written permission.

A bilingual knowledge base that adds structure to the field rather than another narrative: a 3C capability model, a role map, a tool map, a reference library, and a 《FDE 实战蓝皮书》.

Good for

Anyone who needs to describe the role in competency language — hiring, training plans, or mapping an existing team onto it.

Limitation

The linked site (open-fde.com) redirects to openfde.net, which did not load when we checked. Read the repository, not the site.

Licence

Documentation CC BY-SA 4.0; site and tooling code MIT.

Roughly 200,000 characters on running a deployment on site, organised around a five-stage method: reconnaissance, definition, delivery, write-back, withdrawal. Written by practitioners with cases anonymised before publication.

Good for

The delivery stage specifically — what each phase produces, and the checkpoints that stop scope from drifting.

Limitation

Dense; better used as a reference during a deployment than read end to end. One of at least eight unrelated repositories called some variant of “fde-handbook”.

Licence

CC BY-SA 4.0 — reuse with attribution, indicate changes, share alike.

A practitioner's manual in eight parts and 32 chapters plus eight appendices (141 pages): the role, its capabilities, the technical base, enterprise AI demand, implementation, delivery, three full cases, and reusable assets. Available as a site, a PDF and a single Markdown file.

Good for

A second voice on delivery after the longer Chinese manuals — shorter, more concrete, and case-led.

Limitation

Small circulation and no open licence: read it in place, do not mirror it.

Licence

No open licence. Copyright stays with the author; commercial use needs to be arranged with them.

Four parts, sixteen chapters, written from the engineering side: finding the problem on site, then moving it into an AI-native production delivery, with a glossary, a tool index, a delivery evidence-package template and a fact-check table.

Good for

Engineers who want the delivery discipline expressed as artefacts — what a handover package actually contains.

Limitation

Terminology stays close to English in places, so it reads oddly if you expect a fully localised Chinese text.

Licence

No licence declared — default copyright applies. Read it; republishing needs permission.

Eleven chapters in Chinese and English covering the parts the other handbooks skip: enterprise data audits, modern data architecture, a secure landing zone (Shared VPC, private GKE), VPC Service Controls, enterprise RAG, LangGraph multi-agent systems, and LLM evaluation with guardrails.

Good for

The cloud-and-security half of a deployment, which is usually where a pilot stalls in a large enterprise.

Limitation

Non-commercial licence, and the cloud specifics assume Google Cloud.

Licence

Code MIT; content CC BY-NC-SA 4.0. Non-commercial reuse only.

A community list split by purpose: getting started, best practices, cases and reference implementations, tools, a member network, and a small library of hosted Chinese reports, including a 15-page preview of the 崔牛会 white paper.

Good for

Finding the source behind a claim — this is where several Chinese white papers actually live as PDFs.

Limitation

The list is only as current as its last edit (August 2026), and several links inside it are already dead.

Licence

CC BY 4.0.

The definition, a comparison against software and solutions engineering, a hiring list that includes a China and Asia section (with the local job titles the role hides behind), core skills, and ten Palantir-style interview questions.

Good for

The one English list that names the Chinese and Asian equivalents — 解决方案工程师, 技术实施工程师 — instead of assuming the title exists everywhere.

Limitation

Short, and the hiring list is a snapshot rather than a live feed.

Licence

MIT.

The technical half of the role collected in one place: ontology, agents and workflows, RAG, knowledge bases, inference optimisation (vLLM, SGLang), a four-week learning path and 171 concept lessons.

Good for

Filling the stack knowledge a delivery engineer is assumed to have but seldom has written down.

Limitation

Curated aggregation rather than original analysis, and 323 stars rather than the ~4k some pages claim.

Licence

No licence recognised by GitHub (NOASSERTION). Treat it as read-only.

Eleven topics — the role, LLMs, prompting, RAG, agents, evaluation, production, fine-tuning, system design, an FDE behavioural interview, and the interview loop — plus a question bank and recall cards.

Good for

The behavioural round specifically, which is where most technically strong candidates lose an FDE loop.

Limitation

Written for two adjacent titles at once (Applied AI and FDE), so not every chapter applies.

Licence

CC BY 4.0.

A three-stage curriculum — data engineering, cloud architecture on GCP, consulting judgement — with playbooks for multi-agent systems, evaluation and enterprise RAG, plus disconnected and tactical-edge deployment, an interview blackbook and reusable templates.

Good for

The most structured English route from a neighbouring role into this one, with templates you can reuse in a real engagement.

Limitation

It is a curriculum rather than a narrative: there is no single through-line about how a deployment actually runs.

Licence

MIT.

A 24-week syllabus (compressible to 20, extensible to 28) that moves from a role baseline through a vertical slice, governed RAG, a read-only MCP server, observability and handover, to a capstone combining enterprise RAG with an MCP assistant.

Good for

A self-study plan with checkpoints rather than a reading list — each week ends in something you can show.

Limitation

Rights are reserved, so it is a link to follow rather than material to quote.

Licence

No licence. The README states all rights are reserved and forbids whole-repository reposting and paid redistribution.

A single-file interactive site: a fifteen-lesson tutorial across five levels, a ten-stage game about the trade-off between decisions and trust, and a 30-question six-dimension self-assessment drawn as a radar chart. No build step, no dependencies.

Good for

The part of the role that is judgement under uncertainty — the trust game makes the trade-off concrete in a way prose does not.

Limitation

A teaching artefact rather than a reference; you will not go back to it mid-deployment.

Licence

No licence declared.

Notes and code for lectures 01–19, published alongside a CoderArmy YouTube playlist. The repository itself carries no syllabus or ordering beyond the lecture numbers.

Good for

Free video-led walkthroughs if you learn by watching someone build.

Limitation

Repository and videos are incomplete without each other; the licence is undeclared.

Licence

No licence declared — default copyright applies.

An interview-prep site built for this role specifically: 658 questions, 111 company guides, 180 concepts, 30 labs, 408 mock questions and 200 paper walkthroughs, plus its own job board.

Good for

Drilling for loops at the companies that actually run FDE interviews.

Limitation

Freemium — six months of access costs about US$25 (₹2,000), discounted from double that. The site states it is not affiliated with any of the companies it covers.

An eight-month, 32-week postgraduate programme in FDE and applied AI solutions, with a technical and a consulting track, labs, a capstone and an optional IIT certification, run out of Bengaluru.

Good for

Understanding how the role is being formalised into an education product outside the US — useful as a comparison for training design.

Limitation

Tuition is not published, and the placement claims come from the provider. Nobody here should treat it as a neutral benchmark.

The Chinese portal around the xdash book: online edition, case library, ecosystem map, a people directory, and a knowledge graph of seven capability domains and 48 capability points, each linking back to a chapter. Video course sold separately.

Good for

The knowledge graph is the most usable artefact here — it puts capability, maturity and customer journey on the same map.

Limitation

A single author's view of the ecosystem, and parts of it double as marketing for the paid course.

Licence

The book is free to read online; the paid course is separate. Commercial reuse of the text needs the author's permission.

A Chinese FDE community site: 100+ cases, 100+ knowledge articles, 31 open-source skills, courses, a jobs and projects board, and enterprise services (diagnostics, co-creation workshops, PoCs).

Good for

The largest collection of Chinese deployment cases in one place — and the skills library is an unusual, practical artefact.

Limitation

The operating company behind the brand is not disclosed, and prices are not listed for any service.

A free newsletter by a working FDE: an essay every Tuesday and a practitioner interview every Thursday (Palantir, Sierra, Ramp, Cresta, Baseten, HappyRobot), plus research analysing 2,000 practitioner profiles.

Good for

The most consistent English read on how the role is actually staffed at named companies.

Limitation

One author's network; the interview sample skews to companies already doing this well.

A market-intelligence site for the role: a job board (self-reported 1,198+ roles across 50+ companies), salary data (median base $186K; $210K in SF, $195K in NY; 62% remote-friendly), an annual top-25 voices list and a weekly brief.

Good for

Salary and distribution data, especially if you are negotiating and want a second data point next to the Aced list.

Limitation

Self-reported counts, and the author runs a company in the space — read the numbers as one source among several.

A job board organised by company (358 curated roles across 239 companies at last check), plus company pages and long-form articles on the role, pay and interviews. Newsletter is free.

Good for

Reading a company before you apply — the per-company pages say more than a single job listing does.

Limitation

The operator is not disclosed, and some widely-quoted figures ($385K median mid-level total comp, 729% year-on-year growth) only describe the top slice of the market.

Eighteen companies scored in three tiers on a published methodology (pay 30%, maturity 25%, career capital 25%, momentum 20%), alongside 4,900+ US openings, a median base near $190K and a $550K top-end total comp figure.

Good for

Choosing an employer rather than a job — the weighting is explicit, so you can disagree with it precisely.

Limitation

Data last verified 2026-09-02, and the “42× growth since 2023” headline is a very small base inflated by a short window.

Japan's only dedicated FDE job site (56 postings, of which 16 are FDE proper), covering solutions engineers, AI implementation, customer engineers and deployment strategists, with 19 salary and career articles and a ten-question recommender.

Good for

The Japanese market specifically — the non-FDE titles listed here are the local equivalents most English lists miss.

Limitation

Japanese-first; the English and Chinese switches are partial. Employer-side pricing is unpublished.

Announced 2026-05-11 in Boston: a majority-owned deployment company with more than US$4bn behind it, nineteen institutional partners, and the acquisition of Tomoro, which brings roughly 150 experienced forward deployed engineers and deployment specialists on day one.

Good for

The clearest signal that model vendors now consider deployment a standalone business, not a services line.

Limitation

OpenAI's own announcement returned 403 when we checked, so this links to the co-investor's release of the same announcement.

Announced 2026-10-02: US$100m to train 10,000 Frontier Deployed Engineers by the end of 2027, starting with a residency of a few days of instruction plus a twelve-week placement on a real Claude deployment in San Francisco, New York and London. Nomination only; first full certifications expected in early 2027.

Good for

The clearest public statement of what a vendor thinks the role should be — read the two-tier badge definition closely.

Limitation

These are targets, not results: as of publication nobody had been certified. Participants so far are firms large enough to be nominated.

A 25-page white paper released 2026-10-28 that frames the role as a service model: a five-step shared-responsibility method and four service layers, built on the gap between the share of enterprises using AI and the far smaller share seeing profit improvement from it.

Good for

The service-design view — how a deployment is packaged and who carries which risk at each layer.

Limitation

Published by an institute with its own interest in the service market, and the headline percentages are survey figures without a published sample.

A 27-page Chinese talent white paper that explains why the role appeared through eight classes of “last mile” problem, then proposes a value formula and a six-part capability set.

Good for

The demand-side explanation — the eight problem classes are the most reusable part of it.

Limitation

The online copy sits behind a login we could not pass, so we worked from the PDF; its salary and demand figures are not independently sourced.

A local HTML workbench for a single class teacher — rosters, seating, notices, parents' meeting notes — that runs from the teacher's own machine, plus a companion kit of agent prompts, data-mapping templates and acceptance sheets.

Good for

Seeing what a single-user, non-SaaS deployment looks like when the buyer is a teacher rather than an IT department: no accounts, no cloud sync, data lives in a local file.

Limitation

The workbench itself is HTML adjusted by prompting an agent, so the kit is really a delivery method for one person. Read the licence before redistributing anything.

Licence

A custom non-commercial licence: free for a teacher's own classroom use with attribution; commercial redistribution needs the author's written permission. Not MIT or Apache.

The company that defined the role still runs two families: Delta for forward deployed software, AI and infrastructure engineers (30+ open roles when we checked), and Echo for deployment strategists (18+). Postings run from New York and London to Seoul, Tokyo, Sydney and US government sites, including internships and new-grad tracks.

Good for

Reading the actual job descriptions side by side — the Delta/Echo split is the clearest public statement of what the two halves of the role are.

Limitation

Palantir's own careers site is JavaScript-rendered and its blog returns 403, so this links to the Lever board, which is the only machine-readable version.