Where FDEs Work

Where the openings are

Start with a correction. The FDE job market is not mainly the labs and Palantir. It is mostly companies you have not heard of, selling one product into enterprises that cannot install it alone.

The best census of postings comes from Plank, which on 12 September 2026 counted 1,333 live FDE posts at 565 companies 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, working from about 1,000 posts in an analysis published ten months earlier, 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, so do not combine them. They agree on the shape: the typical FDE opening is at a growth-stage company, not a lab.

Counting by title distorts the picture in both directions, and a count made for this book shows how.

Growth figures are left out on purpose. They measure different windows, several echo one Indeed series, and Appendix D lists them unblended; the main echo is set out in How to Use This Book. For choosing an employer they don't help anyway: a market can double while the job you are offered is still the wrong one.

Four employer types, by what they buy with your time

Origins and Economics sorted employers into three economic models. This chapter asks what each does to your week, your pay and your next job. It also adds a fourth type the economics chapter left out, the application startup, because 15.1 shows it is where most openings are.

The test: what is the employer buying when it pays for an hour of your time?

• An AI lab (OpenAI, Anthropic, Mistral, Cohere) is buying two things: production use at a few accounts large enough to matter, and evidence about where its models fail. The customer already pays for the model. Your hours are an investment in usage and in signal.
• An application startup (Sierra, Decagon, Harvey, Glean and hundreds of smaller firms) is buying deals and a working deployment of its own product. Your hours are a cost of sale today and, if the product absorbs what you build, a product investment tomorrow.
• A Palantir-style platform is buying account expansion. Pilots run at a loss by design, and your hours are justified when an account grows into a platform licence that runs without you.
• A deployment company or FDE consultancy (the labs' 2026 spin-outs, and the consultancies that have adopted the title) is selling your hours. Delivered outcomes are the product, and utilisation is the cost line.

Real companies mix these. Place an employer by where it leans, and check the lean with the sheet in 15.8. The types predict three things a title doesn't: who sets your priorities, what earns you a raise, and what you will be good at when you leave.

What the posts say the work is

Job posts are T4 evidence of how the work is done and primary evidence of what employers ask for. Read across the four types, the differences show up in what each post says success is.

Labs measure production and signal. OpenAI's San Francisco FDE post says: "You will measure success through production adoption, measurable workflow impact, and eval-driven feedback that changes product and model roadmaps" (OpenAI, FDE - SF, September 2026). Anthropic's Staff FDE post says "Customers bring you in to lead, not only to implement: you decide what should be built, and then make sure it gets built well" (Anthropic, Staff FDE, October 2026). The lab job is the closest to this book's definition: few accounts, deep ownership, and a duty to report back. Its risk is the one Origins and Economics names for labs: serving one large customer so well that nothing comes back to the product.

Application startups measure the deal. Decagon's Engineering Lead, Strategic Deployments will "own the technical build for 1–3 Strategic Deals end-to-end, operating like the founding CTO of each", find technical risks "and de-risk them before they threaten the deal", and turn "bespoke deal requirements into durable improvements to the core Decagon product" (Decagon, July 2026). Sierra's Flagship Deployment strategist is told to "drive revenue upside for Sierra, and work internally to manage down cost to serve over time, improving margins" (Sierra, September 2026). Glean's founding FDE post makes the product promise outright: "The best of what you build ships to every Glean customer in the world" (Glean, first published January 2026). The startup FDE sits closer to revenue than the lab FDE, and closer to the product: there is one product, and your customer's gap is its roadmap. The risk is that the deal wins every argument with it.

Platforms measure expansion, and hire young. Palantir's board is the only one in the sample with a full pipeline: internships, new-grad posts, a year-long placement, and specialised forward deployed roles in infrastructure, reliability, security and customer enablement, with Deployment Strategist posts, the non-coding partner role, listed alongside. Its FDSE post still describes responsibilities "similar to those of a startup CTO" (Palantir, FDSE - US Government). The platform job teaches the most and specialises earliest. Its risk: if the platform never absorbs the field work, the FDE is a consultant with stock.

Deployment companies measure delivery, and say what they fear. The OpenAI Deployment Company calls itself "the deployment arm of OpenAI", operating "at the intersection of cutting-edge AI, enterprise transformation, and scaled services delivery". Its founding FDE will "own both technical and business value outcomes end-to-end" and "ramp quickly toward management responsibilities". It wants candidates who "can make disciplined standardization-versus-customization decisions instead of defaulting to bespoke work" (OpenAI Deployment Company, FDE - US, June 2026). That is the most candid line in the sample: a company whose revenue is engineering capacity knows bespoke work is its natural drift, and is hiring people to resist it. A second post invites "an established team of two to five people" to apply together, and warns that "People who apply together may work on different deployments" (DeployCo, Apply with Your Team). That is how a services firm staffs: it hires capacity and allocates it.

Where you sit during the week also differs by type, and here the evidence conflicts.

The usable pattern, an inference from a handful of posts and one relayed interview: labs and deployment companies send you to the customer; startups keep you near the product team and send you out in bursts. If you can't travel half the time, that rules out more lab and deployment-company jobs than startup ones. Ask for the last quarter's actual travel, not the posted ceiling.

What pays: read the mix, not the midpoint

Posted base pay overlaps heavily across the four types. The differences that matter are in how the package is built and in what the width of the band hides.

Four rules turn bands like these into something you can use.

A wide band means the level is decided in the offer. A rule of thumb: when the top of a band is more than about 1.5 times the bottom, the employer is hiring across several levels with one post. Glean's post says so outright: "This range encompasses varying levels of the role, you will be assessed for your level during the interview process". Your number depends on the level they place you at. Ask which level the offer is at, what the next one requires, and who decides. The rubric in The Skill Map gives you the evidence to argue a level, not a number.

Variable pay tells you whose targets you share. On-target earnings mean part of your pay rides on something commercial. That can be fine for a domain-expert FDE who carries an account from pilot to daily use. First ask whether the posted range is base or on-target earnings; some posts don't say, and Anthropic's 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. Then ask what the variable part is paid on: bookings, renewals, usage or delivery. If it is bookings, the job is pre-sales, and Chapter 1's checklist will probably say the same.

Equity is a bet on the product loop. Startup stock pays only if the product scales past the custom work, which is what the field-to-product loop exists to produce. Before you value it, ask for one feature that started in the field. If no one can name one, the stock is a bet against your job's purpose.

Lower base can be the platform's design. Qureshi's account of below-market Palantir salaries recovered in stock is set out in Origins and Economics (T4, one alumnus, with the stock's later rise as a confound). The platform model asks you to wait for the account and the company to mature together; whether that pays depends on the company.

Upside and exits

First, the gap. Nobody has followed a cohort of FDEs to see where they went next: no attrition data, no tenure data, no study of exits by employer type. What follows is anecdote, employer claims and this book's inference, each labelled.

What the record shows. At the platform that invented the role, the best-documented exit was into the product. The Pragmatic Engineer reports that "Up until circa 2016, Palantir had more FDEs than it had 'normal' software engineers", and that after Foundry launched, "more FDEs went back to working as software engineers, working on Foundry and bringing their experience from the field" (T3). Qureshi argues that "former FDEs tend to be great founders", and adds that there are "usually more ex-Palantir founders than there are ex-Googlers in each YC batch" (T4; he gives no count, and an alumnus is not a neutral witness). Employers now recruit from that exit in reverse: Anthropic's and Glean's posts invite former founders to apply, Decagon's lead post names an "ex-technical founder" as one profile it wants, and DeployCo invites whole teams. Anthropic's Staff FDE post describes a senior individual path that leads "engagements delivered with partners" (T4 throughout).

The inference. Your exit follows what you owned, and the employer type decides what you can own. At a lab you own production at a few demanding customers and the evidence that changed a product or a model. That leads to product and research-adjacent roles, or to founding a company in a vertical you now know from the inside. At a startup you own deals and a product's gaps in one market, which leads to early-employee and founding roles in that market, or to running deployment at the next startup. At a platform you learn the most kinds of customer earliest, and the exit is into the platform's product or into founding. At a deployment company you own delivery across many customers, which leads to engagement management, delivery leadership and the consulting partner track. Leaving it for product work is harder unless your field work demonstrably fed one.

That last point is where the sceptics are right: the disappointed hires in What an FDE Is found "a typical IT services mindset" (Pragmatic Engineer, March 2026; T4), a job that has dropped the stages that make it an FDE job. For a career the cost is concrete: two years of services work under an FDE title prepares you for more services work. That can be a good career, but not the one most applicants think they are choosing. Levels, Burnout and What Comes Next takes the ladders and the exits further.

China: one title, very different jobs

The Chinese market makes this chapter's argument more starkly than the US one. What an FDE Is showed the same title filed under product, R&D and sales support. Origins and Economics showed a market led by in-house teams at the large platforms.

The titles already sort employers by type. Tencent posts 前线部署工程师; ByteDance posts "AI FDE工程师" in R&D, and one of its posts is titled "豆包AI大模型FDE", tied to its own model. Those are the in-house platform and lab end of the market. At the other end, an industrial software company in Tangshan advertises an "人工智能交付工程师（FDE）": an AI delivery engineer, which is the services end named in the title itself.

Two lessons carry over. First, annualise: Chinese offers quote monthly pay and a number of months, 15 or 16 at the large firms, so a lower monthly figure can be the larger offer. Second, read the spread within one title for what it is. The checkable figures mostly show seniority: they run from ¥20,000 a month for a junior FDE to ¥80,000 for a model company's FDE lead, and the large employers' posted bands overlap heavily (ByteDance ¥35,000–70,000, Ant ¥40,000–60,000, Zhipu's lead ¥60,000–80,000; all monthly, and Zhipu's months are not given). Only one figure points to a gap by employer type: the services-end Tangshan post at ¥8,000–16,000, roughly a tenth of a model company's lead pay by xdash's reckoning, and that post could not be checked. So treat "the employer type shows through in the pay" as plausible, not shown. The question to ask survives either way. As xdash puts it, ask whether a company "按结果收钱，还是按人头收钱": does it charge for results, or for heads? That is this chapter's question in one line.

The comparison grid

Each row is a lean, not a verdict.

Read the grid against your skill self-assessment. If your weakest domain is product, a deployment company will keep it weak unless its field work demonstrably reaches a product. If your weakest is customer, a startup forces it up fastest. Choose the type whose trap you are least exposed to and whose exit you want.

Before you sign: the question sheet

Ask these of your future lead, after the offer if not before. Accept only answers that name a customer, a person or a date. Chapter 2's five questions read the employer's economics; this sheet reads the specific job. For the trap check in section 4, count an answer as pointing to services if the threshold belongs to someone else, no field-to-product change can be named, the team is measured on utilisation or bookings, variable pay rides on bookings, or the only path up is management.

BEFORE YOU SIGN                                 date: ____________
Employer: ______________________   Title: _________________________
Type (lean): lab / app startup / platform / deployment co / mixed

1. THE WORK
   Last 3 engagements this team ran (customer, length, outcome):
   ________________________________________________________________
   Who agrees the success threshold with the customer?  __________
   One product or model change that started in the field: ________
   On site ____%   office ___ days/wk   travel ____%

2. THE MONEY
   Band posted: ________ - ________  base / OTE   level: ________
   What the next level needs, and who decides: ___________________
   Variable: none / bonus / OTE ___/___   paid on: _______________
   Equity: type _______   vesting _______   basis of value: _______
   Months paid per year (12 / 13 / 15 / 16): ___   annualised: ____

3. THE MEASURE AND THE EXIT
   Team measured on: adoption / outcomes / utilisation / bookings
   Where did the last two FDEs who left go?  ________ / __________
   Path above this level:  senior IC / management only / unclear

4. TRAP CHECK (answers that point to services, out of 5): ____
   Decision:  sign / negotiate ________________ / walk

Here is the sheet filled from what is public about one real offer type: a founding FDE at a lab-backed deployment company in the US, as posted in 2026. Everything marked ASK is not in the post.

BEFORE YOU SIGN                         date: offer stage, Oct 2026
Employer: lab-backed deployment co    Title: Founding FDE, US
Type (lean): deployment co (posts say "scaled services delivery")

1. THE WORK
   Last 3 engagements this team ran (customer, length, outcome):
   ASK. Company is months old; expect 0-2 that reached production
   Who agrees the success threshold with the customer?  ASK
   (post: FDE owns "technical and business value outcomes")
   One product or model change that started in the field: ASK
   (post promises field work "informs roadmap priorities")
   On site ASK   office 3 days/wk   travel up to 50%

2. THE MONEY
   Band posted: 170K - 400K USD  base? ASK       level: ASK
   Top is about 2.4x bottom: the level IS the negotiation
   What the next level needs, and who decides: ASK ("ramp quickly
   toward management": is there a senior IC path at all?)
   Variable: ASK   paid on: ASK (utilisation? delivery? renewals?)
   Equity: "offers equity"; ASK type, vesting, and valuation basis
   Months paid per year: 12              annualised: = base

3. THE MEASURE AND THE EXIT
   Team measured on: sibling post lists delivery reliability,
   customer trust, "operating leverage", field signal. ASK which
   one decides a promotion when they conflict
   Where did the last two FDEs who left go?  too new; ASK the lead
   about their own last employer instead
   Path above this level:  management stated; IC path unclear

4. TRAP CHECK (answers that point to services, out of 5): 2-4,
   depending on the ASKs
   Decision:  negotiate level first; sign only if a named
   field-to-product change exists or is promised in writing

Most answers are ASK. For a company months old that is normal, not a reason to walk, but get them from your future lead before you sign: the post alone can't separate a deployment company that feeds a product from one that sells capacity. And the one fully answered line, the band, is the least informative.

A reader who wants to manage early could reasonably take the other offer. The sheet doesn't choose; it makes the choice about the job rather than the title. If you are still in the loop, Getting Hired covers what it tests.