State of the Role

Reading this snapshot

Nothing on this page is new. Each fact is copied from a chapter with its date, its source link and its tier exactly as the chapter gives them, and each links back to the section that uses it, where the caveats and the argument live. Where a chapter's box gives no tier or no link, none has been added here.

Two reading rules carry over from How to Use This Book. First, the tiers: T1 is measured, primary and independent; T2 is measured and primary but vendor- or self-reported; T3 is secondary reporting; T4 is anecdote, opinion, advice, a stated target or a job post. Second, job posts: a post is primary evidence of what an employer asks for and advertises, and T4 evidence of how the job is actually done. Every posted band below shows what an employer advertised on the date given, not what anyone is paid.

Postings change and close without notice, and the counts move by the hour. Unless a row says otherwise, "read 17 October 2026" means read from the employer's own job-board feed on that day. Check a posting again before you quote it. Figures from different samples and windows sit side by side here only so you can find them: do not average them, and never blend growth rates across windows.

Who hires, and how many

The census. Plank counted 1,333 live FDE posts at 565 companies on 12 September 2026 and linked each one. About two-thirds were in the United States, and 9% sat at the frontier labs and Palantir combined (Plank, FDE job market, September 2026; T2, a measured count published by a firm that sells FDE services). Bloomberry, from about 1,000 posts, found 58% of FDE roles at companies with 11 to 200 employees (Bloomberry, November 2025, updated January 2026; T2). The two samples cover different windows and use different rules; do not combine them. See Where FDEs Work, 15.1.

FDE-titled roles by employer. Roles with "forward deployed" in the title, out of all open roles, read from the employers' own job-board feeds on 17 October 2026 (T1 as a count of what was advertised; it says nothing about hires). Source: Where FDEs Work, 15.1, and for OpenAI's breakdown Levels, 18.1.

Plank's own caveat says the same as the overcount column: one posting can stand for many seats or for an evergreen pipeline.

Growth figures, unblended. Where FDEs Work leaves growth out on purpose and points here. Each figure below measures a different window, and the first two are one source counted many times (see the echo map). They are listed, not combined.

Seniority, team shape and size. These are the dated figures on who is hired and how teams are built.

Headcounts at the 2026 deployment companies are in D.6.

Pay in the US, and the aggregates

Posted ranges, US base unless marked, read 17 October 2026 unless a post date is given. They are posted base ranges, before equity, and show what employers advertise. Main sources: Where FDEs Work, 15.4, Levels, 18.1 and 18.4.

One line to read with the bands: Anthropic's posts carry a standard line that "For sales roles, the range provided is the role's On Target Earnings" without saying whether the FDE role is one (15.4).

The aggregates. Different samples and windows: do not average them. Source: 15.4.

Pay in Europe and Canada

Posted bands outside the US, in the currencies the posts use. Read 17 October 2026 unless a date is given. Source: Where FDEs Work, 15.4, and Levels, 18.1.

OpenAI also lists a Tokyo FDE manager post, with travel "up to 40%, although mainly local"; the chapters give no band for it (18.1).

Pay in China

Chinese offers quote monthly pay and a number of months, 15 or 16 at the large firms, so annualise before you compare. Main sources: Where FDEs Work, 15.6, Leading FDEs, 17.2 and Levels, 18.5.

xdash's ByteDance, Ant and Zhipu figures are identical to 中新经纬's, so they are an echo, not confirmation (D.11). The checkable figures mostly show seniority, and the large employers' posted bands overlap heavily; only the unchecked Tangshan post points to a gap by employer type, so "the employer type shows through in the pay" is plausible, not shown (15.6).

Employer types and the 2026 deployment companies

Where FDEs Work, 15.2 sorts employers into four types by what they buy with an FDE's time: AI labs (OpenAI, Anthropic, Mistral, Cohere); application startups (Sierra, Decagon, Harvey, Glean and hundreds of smaller firms); Palantir-style platforms; and deployment companies or FDE consultancies (the labs' 2026 spin-outs, and the consultancies that have adopted the title). The dated employer moves behind that sorting follow.

Pilot price points. Pear suggests a 20–50% discount on a prorated share of the expected annual contract (March 2026). Shimoni proposes a fee of about $10K for a 30-day AI pilot, and says it "eliminates 90% of the people who just want to 'see how this AI stuff works'" (July 2025; the 90% is his estimate, not a measurement). Heavybit gives no figure and speaks of a "token fee" (February 2020). None of these is a market rate (Before the Engagement, 4.4).

As of October 2026 no public evidence says whether the deployment companies' field work flows back into the labs' products, whether they report outcomes rather than headcount and funding, or whether their engineers own outcomes rather than supplying staff augmentation (2.5).

Time on site and travel, by employer. The figures disagree, and the Palantir ones may describe different eras (15.3, 18.4).

Titles and job families

The title no longer tells you which job you are looking at, and a posting's department is a better guide to whose targets you will be measured against (What an FDE Is, 1.3).

What the posts ask for

Every row is primary evidence of what one employer asks for and T4 evidence of how the work is done. Postings change or close; check them before you quote them. Bloomberry's frequencies across about 1,000 posts (T2): Python 66%, TypeScript 35%, AWS 32%, AI agents 35%, LLM experience 31%, RAG 12%, customer-facing ability 47%; "working directly with customers" 55% and integrating systems or APIs 32% among responsibilities (3.1).

Interview loops

No employer publishes its FDE loop, and nobody has published pass rates. What exists is candidate reports gathered by prep sites from small, self-selected samples, all T4. Accessed 17 October 2026; source: Getting Hired, 16.1.

All of it is perishable. Both Aced guides were updated in the month before access and change as reports arrive. Three gaps hold across every employer: no pass rates for any round, no account of how employers weigh rounds, and no evidence that a dedicated "customer role-play" round is standard anywhere (16.2). A five-round Anthropic loop with a client-simulation round and a failure rate is circulating in Chinese; it is a forming echo, and this book does not repeat the figure (D.11).

Prep material, dated: Paden Gayle's free FDE Handbook (Maven), about ten pages, promises "the four things you have to demonstrate to get hired" without naming them on its landing page (T4) (16.2).

What changed in 2025–26

The dated events behind the snapshot, in order. Posting publication dates are in D.8 and are not repeated here.

One product fact in the book carries no date: the voice engagement in 12.3 is reported as improving OpenAI's Realtime API (12.5).

Echo map

An echo is one self-reported figure repeated by several outlets until it looks independently confirmed (H.3). Each row traces one figure to its single origin. Count each once, at its origin, and never blend growth rates across windows.