A forward deployed engineer (FDE) is a software engineer who works inside one customer’s environment, writing production code that connects a vendor’s product to that customer’s systems and gets it live. In the US the median FDE earns USD 220,000 a year in total compensation (Levels.fyi’s US salary data, 159 submissions, 19 October 2026). With employer benefits and payroll taxes, an employer pays about USD 258,000 a year, and about USD 302,000 in the first year once a recruiter’s fee is paid. Paraform’s FDE hiring guide puts that fee at 25% of base salary. Those figures are before management time and equipment. Fully loaded, the first year costs USD 328,572 (the table below shows each line). In the UK the Levels.fyi median is GBP 144,377 from 31 submissions. Employer National Insurance and the minimum pension take that to about GBP 161,000 a year, and to GBP 188,600 in the first year with the same fee.

Our guide to production-ready AI agents covers what the engineer must get done before each go-live. This one covers where the cost figures come from, what the alternatives charge and a worked example.

Where the forward deployed engineer role comes from

Palantir created the role in the early 2010s and called its forward deployed software engineers Deltas. Until about 2016 it employed more of them than ordinary software engineers (The Pragmatic Engineer, 12 August 2025). In Palantir’s Dev versus Delta post (8 April 2019), a product engineer builds one capability for many customers, while a Delta works on “one customer, many capabilities”. When Foundry launched in 2016, more FDEs moved into product engineering and took their field lessons with them. That loop back to the product is what Gartner expects most FDE engagements to miss through 2028.

Who hires forward deployed engineers in 2026

In public job feeds on 19 October 2026, Databricks had 100 open roles with “forward deployed” in the title, Palantir 77, OpenAI 26 and Anthropic 6. Titles vary (Decagon hires Agent Deployment Engineers, while LangChain and Cognition hire Deployed Engineers), so title counts understate demand.

The model labs now sell the work too. OpenAI launched the OpenAI Deployment Company on 11 May 2026 with more than USD 4 billion of initial investment, and agreed to buy Tomoro, a consultancy that would bring about 150 FDEs and deployment specialists. On 4 May 2026, Anthropic, Blackstone, Hellman & Friedman and Goldman Sachs announced an enterprise AI services firm to build Claude into mid-sized companies’ operations, with Anthropic engineering staff embedded. Google Cloud is hiring FDEs for a new AI organisation in its go-to-market team (The Pragmatic Engineer, 24 May 2026).

Forward deployed engineer salary in the US and UK

United States

Around the Levels.fyi FDE salary median of USD 220,000, total pay runs from USD 175,000 at the 25th percentile to USD 300,000 at the 75th. The separate medians for base, equity and bonus are USD 175,000, USD 37,500 a year and USD 3,250, which don’t sum to the total. The median moves as salaries are added. Paraform’s own hiring data puts base pay higher, at USD 150,000 to 217,000 with a midpoint of USD 183,000.

New York requires businesses with four or more employees to list pay ranges in job ads (New York State Department of Labor), and every US posting below shows one. Open postings on 19 October 2026:

Employer and role Location Range What the range covers
Palantir, Forward Deployed Software Engineer New York USD 135,000–200,000 Salary, excluding sign-on bonus, benefits and long-term incentives
Decagon, Agent Deployment Engineer San Francisco USD 175,000–230,000 Salary, plus equity
Databricks, Sr. Forward Deployed Engineer United States USD 182,000–250,208 Base salary (20% travel)
OpenAI, Forward Deployed Engineer (FDE) San Francisco USD 185,000–300,000 Salary, plus equity and bonus (up to 50% travel)
Anthropic, Staff Forward Deployed Engineer New York, San Francisco, Seattle USD 280,000–320,000 Annual salary
Anthropic, Forward Deployed Engineer London GBP 225,000–255,000 Annual salary (25–50% travel)
Decagon, Agent Deployment Engineer London GBP 110,000–150,000 Salary, plus equity

United Kingdom

UK data is thinner. Levels.fyi’s UK salary data gives a median of GBP 144,377 in total compensation from 31 submissions, with a median base of GBP 109,160 (converted from dollars, so it moves with the exchange rate). Of 42 UK deployment postings in the feeds we read, five showed pay, all at Anthropic or Decagon. ITJobsWatch has no FDE category, but for UK AI engineers it reports a median advertised salary of GBP 87,500 (six months to 17 October 2026, 717 ads) and a median contract rate of GBP 600 a day (536 ads). The sources disagree because they measure different markets, so the cost model uses both.

The fully loaded cost of an in-house FDE

Fully loaded cost is everything an employer pays for one engineer. Per year:

Line US: Levels.fyi FDE median UK: Levels.fyi FDE median UK: AI engineer job-ad median
Total compensation USD 220,000 GBP 144,377 GBP 87,500 (salary only)
Employer benefits and payroll taxes USD 38,324 GBP 16,945 GBP 13,696
Recruiter’s fee, 25% of base (first year only) USD 43,750 GBP 27,290 GBP 21,875
People management, 10% (our assumption) USD 25,832 GBP 16,132 GBP 10,120
Laptop over three years (our assumption) USD 666 GBP 500 GBP 500
First year USD 328,572 GBP 205,244 GBP 133,691
Later years, without the fee USD 284,822 GBP 177,954 GBP 111,816
  • US employer costs: in June 2026, insurance, retirement and legally required benefits cost private employers USD 13.03 an hour for professional workers, on top of USD 60.66 in wages, paid leave and supplemental pay (BLS employer costs, table 4). That is 21.5%, applied here to cash pay of USD 178,250 (median base plus bonus). Employer Social Security (6.2% up to USD 184,500 in 2026) and Medicare (1.45%) alone come to USD 13,636 (Social Security Administration).
  • UK employer costs: National Insurance at 15% above GBP 5,000 a year (HMRC rates for 2026 to 2027) is GBP 15,624 on a GBP 109,160 base, and the minimum pension of 3% of earnings between GBP 6,240 and GBP 50,270 (GOV.UK) is GBP 1,321. That is a statutory floor. Eligible employers can offset up to GBP 10,500 of Employment Allowance.
  • Fee and time to hire: Paraform charges 25% of first-year base salary when a hire is made and says to budget 8 to 12 weeks for an FDE search. A.Team, which sells an alternative to hiring, puts a permanent hire at 6 to 9 months. We found no UK fee norm we could verify, so the UK columns assume 25%.
  • People management (one-to-ones, reviews, the next hire) is work a contractor or provider absorbs. Change the 10% to suit your team. The laptop is the USD 1,999 price of a 14-inch MacBook Pro with 16GB in Apple’s US store.
  • Ramp-up: any engineer new to your product, hired or rented, spends part of the first month learning it, so we leave it out of every row. A provider that rotates engineers makes you pay it again.
  • Not included: travel (the postings above ask for 20% to 50%), payroll taxes on equity and US state taxes.

Cost model for hiring, contracting or outsourcing

This table compares like with like. Every row is per engineer per month, at fully loaded cost for in-house rows and at the all-in price a buyer pays for the rest. UK figures are converted at GBP 1 = USD 1.3335, from the European Central Bank’s reference rates for 16 October 2026.

Option Per month, first year Per month, later Basis
(a) In-house FDE, US USD 27,381 USD 23,735 Fully loaded cost, Levels.fyi median
(b) In-house FDE, UK USD 22,808 USD 19,776 Fully loaded cost, Levels.fyi median
(b) In-house AI engineer, UK USD 14,857 USD 12,426 Fully loaded cost, ITJobsWatch median
(c) Contractor through a US agency USD 22,400–38,400 Same AI/ML hourly bill rates of USD 140–240 (KORE1, a staffing firm, updated 22 August 2026) for 160 hours
(c) Contractor, UK USD 16,002 Same GBP 600 median day rate for 20 days, before any agency margin
(d) Talent network (A.Team) About USD 25,000 Same A.Team’s figure for a senior engineer, fully loaded
(d) White-label engineer, Volans Aquilae USD 12,000 Same Our price, about USD 75 an hour spread over the same 160 hours, with a minimum of one month

The 160 hours and 20 days are our assumptions. KORE1’s bill rate includes the agency’s employer costs and margin. In the UK, settle IR35 first. When the off-payroll rules apply, the deemed employer must deduct tax and pay employer National Insurance (GOV.UK guidance), so ask who bears that cost. No row includes travel, ramp-up or your own time directing the work.

A US hire costs about twice our price per month after the first year. In the UK it depends on the salary. At the Levels.fyi median a hire costs USD 19,776 a month, and at the job-ad median USD 12,426, within 4% of our price, with the knowledge kept in your company.

Worked example of three customer go-lives in a quarter

We assume three go-lives land in one quarter, each needing about one engineer-month of integrations, evals on the customer’s data and security-review evidence, run back to back by one engineer.

Option Cost for the quarter Timing
Hire in the US now Up to USD 67,485 An 8–12 week search leaves one month at most (the fee plus that month)
Hire in the UK now (Levels.fyi median) Up to USD 56,168 Same search (GBP 27,290 fee plus a month)
Contractor through a US agency USD 67,200–115,200 Depends on who is free
Contractor, UK USD 48,006 Depends on who is free (IR35 decision first)
Talent network (A.Team) About USD 75,000 Profiles in three to five business days, says A.Team
Volans Aquilae USD 36,000 Our price for three engineer-months

A hire made today can’t cover this quarter, and it then costs USD 284,822 a year in the US, which is the right trade if go-lives keep coming. Over a first year, that is USD 328,572 against about USD 300,000 through A.Team, USD 268,800 to 460,800 through a US agency and USD 144,000 at our price. Rented capacity can double for a month when go-lives overlap, but one hire can’t.

Risks and trade-offs of each option

  • In-house: slowest to start and a fixed cost if go-lives stop, but the knowledge stays and the engineer can travel, as the postings above require.
  • Contractor: no fee and easy to stop, but US agency rates are the highest monthly cost here, the knowledge leaves with the contractor and code ownership must be assigned in writing.
  • Talent network or FDE as a service: fast and monthly. Check who employs the engineer, whose brand your customer sees and who owns the work.
  • White-label, our model: the lowest monthly price here, engineers under your brand and cover for your customers’ hours in any time zone, because we staff around the clock. We work remotely, and we have no clients or case studies yet, so you’d be buying capacity you have to test.

Gartner’s prediction and FDE washing

On 29 September 2026, Gartner predicted that by 2028, 70% of enterprises will abandon agentic AI built by vendor forward-deployed engineering, because costs rise and they can’t change the systems on their own. It also expects less than 20% of FDE engagements through 2028 to turn recurring customer needs into product capabilities, and warns of “FDE washing”, meaning consulting sold under the FDE label. A white-label provider, ours included, sits close to that line. We can build what one customer needs, but only your product team can turn it into product. Gartner advises putting deliverables, knowledge transfer, intellectual property rights and exit duties in the contract before signing. If you outsource, put them in both contracts, with the provider and with your customer.

When outsourcing is the wrong choice

  • Deployments are how you learn what to build. Palantir’s FDEs carried field lessons into Foundry, while a provider’s engineer passes them on second-hand.
  • Your customer needs someone on site, or staff with a security clearance.
  • Your customer’s security review rules out access to its data from outside its country.
  • You have go-lives every month for a year or more and can wait for a search. At the UK job-ad median, a hire costs about our price after the first year.

What to ask any FDE-as-a-service provider

  1. Which named engineers will do the work, can you approve each one, and will the same person stay for your next customer?
  2. From where will they work and reach your customer’s data? Your customer’s AI vendor security questionnaire will ask.
  3. Does your customer see your brand, with your email domain and tickets, or the provider’s?
  4. Who owns the code, prompts, evals and documents, and what does the handover contain?
  5. How will they test before go-live, and what evidence will your customer’s security team get? Our page on enterprise customer onboarding for AI vendors lists what we would show them.
  6. What are the minimum term, the notice period and next month’s price?
  7. Which comparable deployments have they done, and can you call those customers? A new provider, us included, should say it has none.
  8. Will they write down the product gaps they worked around at each customer?

Our view

Hire forward deployed engineers when deployments are how you find out what to build, or when go-lives arrive every month with no end in sight. That knowledge belongs in your company, and at UK market rates a hire costs about what we charge once the fee is paid. Rent capacity when go-lives come in bursts, when dates are signed before a search could finish, or when customers need hours your team doesn’t work. Either way, keep one employee who owns each customer, write the exit into every contract and judge a provider by whether your team can run the deployment without it. And do your first enterprise go-live with your own engineers, because it shows you what the product lacks.

What the cost model leaves out

  • Pay data is partial. Levels.fyi is self-reported (159 US and 31 UK submissions) and leans towards employers whose staff use it. Posting ranges are base pay, not offers.
  • The BLS ratio is an average for professional workers. Health insurance doesn’t rise with pay, so it may overstate costs at a USD 175,000 salary, while leaving out equity taxes, state taxes and travel understates them.
  • The 10% management, 160 hours, 20 days, one engineer-month per go-live and three-year laptop are our assumptions. Change them and the totals move.
  • It compares prices, not people, so ask to see each named engineer’s work.
  • Medians and exchange rates move weekly, and we will check these figures again by 30 April 2027.
  • We have no clients yet, so nothing here is evidence of our results.

If you’d rather we did this

Our engineers work under your brand on customer go-lives (integrations, evals on each customer’s data, evidence for the security review) for USD 12,000 per engineer per month, minimum one month. See white-label forward deployed engineers for AI vendors.

Sources

All pages opened on 19 October 2026.