The role, named
IndependentForward Deployed Engineer
The industry finally has a name for what I have been doing since 1994. Palantir coined it in 2005. The AI labs are hiring for it now. The only part I do differently is that nobody owns my answer.
The boundary
The defining trait is end-to-end accountability. The person who maps the problem in week one is the person who gets the call when it breaks in month six.
The definition
The engineer who moves in.
A forward deployed engineer embeds with the customer, learns the customer's field, and ships production code against the customer's own data, inside the customer's own security and approval reality. The role is customer-facing and carries no sales quota. The core of the job is building.
The industry draws the boundary in one line: a consultant hands you a report, a forward deployed engineer hands you the running system. Palantir invented the role in 2005 for the CIA, the NSA and Army intelligence, because the consultants available to them could not write production code. OpenAI and Anthropic run the same model today under their own names.
Why now
The demand went up tenfold, and it went where I already live.
Postings for the role grew more than tenfold in eighteen months, and the demand concentrates in fintech and heavily regulated industries, where an engineer has to work inside somebody else's compliance reality instead of around it. That is the part most engineers want no part of. It has been my job for twenty years.
What they screen for
Four things, and they call the profile the engineer-diplomat.
Palantir screens forward deployed candidates on four components and turns away roughly 99 percent of applicants. Here is the list, in their language, and what I put against each one.
Engineering depth
Production-grade experience. Typed languages, distributed systems, data engineering.
Thirty-two years shipping production, with a U.S. patent in security architecture. Fortune 5, federal and state government, and 40+ university platforms. All of it live, none of it a prototype.
Domain curiosity
The willingness to master somebody else's field: compliance law, logistics, credit.
I learned federal disaster recovery funding well enough that HUD published my practices as its development model. Last quarter I learned distressed energy credit well enough to build the screen a fund now sources from.
Conversational range
Pulling the real requirement out of an analyst, an engineer and a VP in the same week.
Agency counsel, university procurement, a Fortune 5 compliance desk, and a fund's partners. Reading that room is not soft skills. It is the reason the code lands instead of sitting in a branch.
Operational autonomy
Walking into a customer site with no statement of work and starting to deliver.
I ship in the first week. Working software, not a discovery deck, because the fastest way to learn what someone actually needs is to hand them something and watch what they do with it.
The difference
A vendor's engineer arrives with the answer already chosen.
A forward deployed engineer at Palantir deploys Foundry. One at OpenAI deploys OpenAI. That is not a criticism, it is the shape of the job: the role exists to make the platform land, and the loyalty runs to the roadmap. You are going to get their product, and the interesting question of whether you needed it never comes up.
I do not carry a product. I pick the stack for the problem, build it on your accounts, and hand you the keys on day one. If you want me gone, you cut my access and keep every line of it.
How it runs
Embedded, on a term, with the dates written down.
On site when the work needs it
Embedded weeks in your building when that is what the problem takes, remote the rest of the time. Austin, and I travel.
Fractional, on a defined term
A set number of days a month against a term with an end date, billed monthly, renewed in writing or not at all.
Deliverables and dates first
What I owe you and when I owe it, agreed before anything starts, along with what your side owes me to hold the date.
Your accounts, day one
Repository, hosting, database, domain and keys registered to you. Nothing to migrate away from later.
Metered spend capped
AI and data usage priced per run against a ceiling you set. A model choice never turns into a surprise invoice.
Documented to be handed over
Written up so your team operates it without me. The federal precedent is on the record: HUD took my practices and published them.
I have worked this way since 1994. The market named it in 2005 and started paying for it last year.
If you have a problem that needs somebody inside your building who can read the domain and write the production code, that is the whole job description, and it is the one I have had for thirty-two years.
Sources
- Forward Deployed EngineerWikipedia
- Palantir's forward deployed engineering playbookPerspective AI
- What is a forward deployed software engineerParaform
- How demand grew 10x in 18 monthsParaform
- OpenAI's forward deployed engineerParaform