Being On-Site Is Not What Makes a Forward Deployed Engineer

Vijay Kasibhatla | Friday, October 30, 2026

Being On-Site Is Not What Makes a Forward Deployed Engineer

Sitting at the customer's site does not make an engineer a forward deployed engineer. What does is a product team back home that regularly receives what the engineer finds.

One engineer, three descriptions of the job

Picture a renewal call. The customer's chief data officer asks what the engineer who has been in her building every week since March actually does.

The account executive says the engineer is a technical advisor on the account. The head of services says the engineer is extra hands building on the customer's systems. The engineer says the work goes back to the product team.

Nobody disputed one thing: the engineer was in her building. That does not tell you which of the three descriptions fits.

Employers use the title forward deployed engineer (FDE) for a software engineer who builds inside a customer's own systems instead of at the vendor's office. The title says where the engineer works. It does not say whether anything the engineer learns there goes anywhere.

What could go anywhere is a finding: something the engineer learns at one customer that other customers will need too.

Here is an invented example, for illustration only. An engineer spends three weeks at a hospital network because its appointment data arrives in a format the software cannot read. She writes a converter and the rollout moves on. Suppose the next ten customers send the same format.

Now follow the finding. If it reaches a named product team on a regular schedule, the team builds the converter into the product and the next customer never hits the problem. I call that arrangement the return path. If it does not, the converter stays in one customer's environment and the next engineer writes it again.

On this example's numbers, each of the next ten customers costs an engineer about three weeks without a return path. With one, only the first customer did. Here is the distinction this piece makes. With a return path, the engineer is an FDE. Without one, the engineer is what I will call a delivery engineer: skilled, useful, and doing a different job. That distinction is my definition, not an industry standard.

If you take one thing from this piece: before you post an FDE job, name the product team that receives the findings.

Identical on site. Different after the third box. Two lanes show a forward deployed engineer workflow. In Lane A the work stays with one customer. In Lane B a return path sends findings to a named product team so the capability ships to other customers. Identical on site. Different after the third box. WHAT THE CUSTOMER SEES: THE SAME THING LANE A EMBEDDED DELIVERY ENGINEER Customer need arrives on site Engineer builds in their environment Ships for this customer Work stays here no receiving team, no schedule the next customer may start over LANE B FDE WITH A RETURN PATH Customer need arrives on site Engineer builds in their environment Ships for this customer Return path finding goes to a named product team, on a schedule Capability ships to other customers the next customer need can start from what shipped Author's illustration of the idea.

Both lanes look the same on site. They differ only after the work ships for the first customer. Author's illustration of the idea.

Job ads show the same title and cannot show the return path

The five postings come from Palantir, OpenAI (two), Anthropic, and Databricks. Their duties look much alike. All five say the engineer works embedded with customers. Four say the engineer sends findings back to product or engineering: OpenAI's two, Anthropic's, and Databricks's. Palantir's does not. None of the five states a schedule for it.

Overlapping duties, different bars Five live forward deployed engineer postings compared on stated years of experience and four duties. Palantir, New York asks for 1+ years and has no sentence about sending findings back to product. OpenAI Tokyo asks for 5+, OpenAI San Francisco 7+, Anthropic Munich 8+, and Databricks London states no minimum. Overlapping duties, different bars Five live forward deployed engineer postings, read on their employers' own systems on 10/18/2026 STATED YEARS OF EXPERIENCE Embeds with customers Writes production code Owns delivery Sends findings back to product Palantir's posting has no such sentence Palantir, New York Forward Deployed Software Engineer 1+ OpenAI, Tokyo Forward Deployed Engineer 5+ OpenAI, San Francisco Forward Deployed Software Engineer 7+ Anthropic, Munich Forward Deployed Engineer 8+ Databricks, London Forward Deployed Engineer no minimum stated Stated in the posting's own words Partly stated (no sentence says production code or owning delivery) No sentence found Five postings at four employers; OpenAI is counted twice (San Francisco and Tokyo). OpenAI also lists a second San Francisco posting (5+ years) and Databricks lists other London postings; I did not chart them. Years are as listed and sit under different headings (required, thrive, good fit), so they are stated expectations, not hard filters. Duty coding is mine. Palantir's posting shows no usable date; the other four were published 2025 to 2026.

Four of five postings say the engineer sends findings back, and the years run from none to 8+ (Palantir 1+, OpenAI San Francisco 7+, Anthropic 8+, OpenAI Tokyo 5+, Databricks none). Source: five postings read on 10/18/2026 on their employers' own systems. These five are a read, not a sample. Years sit under different headings (required, thrive, good fit), so they are stated expectations, not filters. Titles are Forward Deployed Engineer or Forward Deployed Software Engineer; I treat them as one family. Duty coding is mine and the rule is printed on the chart.

Four postings state the intent, which an ad can carry. No ad can show whether a product team is waiting at the other end, so an ad cannot tell you which job you are hiring.

At Palantir, field findings reached the people who build the product

Palantir's own posting has no findings-back sentence, yet accounts of its earlier years describe findings reaching the product side.

A 2019 Palantir post describes two engineering roles. Devs, the product engineers, build the product: their focus, it says, is "one capability, many customers." Deltas, the company's Forward Deployed Software Engineers, work at customers: theirs is "one customer, many capabilities." Devs sit in Product Development. Deltas sit in a different division, Business Development.

So the people closest to the customer sit apart from the people who build the product. Two accounts describe how the work crossed over.

Nabeel Qureshi, who joined Palantir in 2015, writes that FDEs were typically expected to work onsite three to four days a week, and that product engineers then turn what the FDEs build into product. Gergely Orosz, a newsletter writer, reports that after Palantir launched Foundry, its data platform, more FDEs went back to engineering to work on it, bringing field experience with them. Neither account describes a schedule, so they show part of a return path, not a full one.

Where the Deltas sat did not settle it. They sat outside Product, and findings still reached it.

Marc Andrusko, an investor at a16z, warns about the other half: FDEs in a separate professional-services unit lose the link back to the product team and drift toward a services business. That is a warning, not outcome data. Placement alone does not decide it. The link does.

My reading of what this costs: with a return path, the second customer starts from what the first one needed. Without one, the second starts over, and you are running a services business, one that sells engineers' time customer by customer.

Two tests sort the three descriptions

Back to the renewal call. Two tests sort the three descriptions.

Test one: does the engineer write production code, the code that runs the customer's live systems rather than a demo, in the customer's environment? First Round Review says the FDE is "still very much an engineer who writes and debugs production code," unlike a traditional solutions consultant or sales engineer. Orosz says OpenAI's solutions architects are closer to advisors who usually build proofs of concept, small demos, on anonymized or offline data, while its FDEs "write code directly on customer infrastructure."

Test two: do the engineer's findings reach the product team on a regular schedule? That is the return path, and test two is my addition. The sources I read say what other roles do not do. They do not say whether those roles send findings back.

Two tests that sort solutions architects, delivery engineers, and forward deployed engineers
Role Test one: writes production code in the customer's environment Test two: has a return path
Solutions architect (at OpenAI), sales engineer, or solutions consultant Typically no Not addressed
Delivery engineer Yes No
FDE Yes Yes

"Not addressed" means the sources I read do not say. The last two rows are this piece's definitions. The same engineer is one or the other depending on test two.

If each description is accurate, the account executive described an advisor, which typically fails test one. The head of services described a delivery engineer, who passes test one and fails test two. The engineer described an FDE, but only if a product team really receives the findings.

Three different jobs, one building, and quite possibly one title.

A return path is a staffing decision, not a slogan

A return path is easy to claim and easy to leave unstaffed. In my view, which none of the sources I read specifies, it takes three things.

First, a product team with capacity. Someone on the product side has to have receiving findings written into their job, with time set aside for it. A team that receives findings as a favor stops receiving them the first quarter the roadmap gets tight.

Second, a standing time to look. A finding that sits in a shared channel is not on a return path. A review that is on the calendar, with the same people each time, is. How often it meets is yours to choose, and I make no claim about the right interval.

Third, a rule for what becomes product work. Without one, every finding is either urgent or ignored. A simple rule, for example that a finding becomes product work once a second customer hits it, gives the engineer an answer and gives the product team a reason to say no to the rest.

Each of the three is small to set up and easy to let lapse. That is why a job ad cannot carry them, and why the leader who is hiring has to.

Do not post an FDE job until you can name the product team that receives the findings

If you lead engineering, product, or services and have been asked to hire FDEs, answer these four questions first. Questions 01 and 02 are the two tests. Questions 03 and 04 tell you whether the return path is working. 

01 — Will the engineer write production code in the customer's environment? If not, you are hiring an advisor, which is a different job.

02 — Where do field findings land, and who receives them? Name the product team, how often it reviews them, and what it did with the last one.

03 — How many days pass between a field finding and a product decision about it? Decide what is too slow before you look (30 days, say), then measure it once on a real finding.

04 — What would tell you this has turned into a services business? For example, findings pile up and nobody reviews them. Write the sign down before it happens.

If the answer to 02 is nobody, post it as a delivery engineer job, or stand up the product team first.

If you are the candidate, ask question 02 in the interview. Which product team received the last finding an engineer in this role sent back, and what did it do with it? A specific answer is worth more than any line in the ad. A vague answer is information too.

An FDE is not an engineer who sits with the customer. It is an engineer with a return path. The return path is yours to build.

The candidate cannot bring it.

The next piece in the series, The AI Deployment Gap (article 2 out of 6), asks why the title spread through 2025 and 2026, and why so many AI deployments stall between a working demo and a working customer.

Which of your open FDE jobs has a named product team on the receiving end? I would like to hear where the answer was nobody.

Sources

Job postings, all read on the employers' own applicant systems on 10/18/2026. I searched all five for words about schedule (cadence, weekly, monthly, quarterly, regular, recurring) and found none. OpenAI also lists a second San Francisco posting that asks for 5+ years, and Databricks lists other London postings; I did not chart them. Palantir's posting shows no usable posting date.

  • Palantir, Forward Deployed Software Engineer, New York
  • OpenAI, Forward Deployed Software Engineer, San Francisco (published 11/15/2025)
  • OpenAI, Forward Deployed Engineer, Tokyo (published 09/16/2025)
  • Anthropic, Forward Deployed Engineer, Munich (first published 08/17/2026)
  • Databricks, Forward Deployed Engineer, London (first published 10/08/2026) 

Essays, posts, and reporting

  • Palantir, Dev versus Delta: Demystifying engineering roles at Palantir (byline 04/08/2019; page metadata shows 11/29/2023) •
  • Nabeel S. Qureshi, Reflections on Palantir (10/15/2024)
  • Gergely Orosz, What are Forward Deployed Engineers, and why are they so in demand?, The Pragmatic Engineer (08/12/2025). The Foundry-era account of FDEs returning to engineering is a single-source account
  • Marc Andrusko, The Palantirization of everything, Andreessen Horowitz (01/16/2026)
  • First Round Review, So You Want to Hire a Forward Deployed Engineer (published 02/24/2026, updated 10/02/2026)
  • Gregor Ojstersek, Inside OpenAI's Forward Deployed Engineer Role, Engineering Leadership (08/15/2026) 
Contribute to the Ascendient Learning blog

Contribute to the Ascendient Learning blog

Whether you are an IT expert or want to talk about a particular IT training niche or industry topic, we invite you to share your knowledge through the ExitCertified blog. Submitting informative, well-written content will increase the likelihood of getting published. Read on to find out the guidelines that will give you the best chance.

News