buyer's guide

Forward deployed engineering, explained for buyers

For the CIO, CTO or operations leader at a mid-market company weighing an embedded AI engineer against a consultancy, a vendor's deployment team or a hire of their own. What the role is, what it costs, how an engagement runs, and twelve questions to ask before you sign.

Updated November 202611 min read

01 · the role

what a forward deployed engineer is

A software engineer who works inside your company, on your data and your systems, until the AI system runs in production.

“turns research breakthroughs into production-grade systems”

OpenAI, describing its Forward Deployed Engineering team in a 2026 job posting

A forward deployed engineer (FDE) is a full-stack software engineer placed inside a client's team. The engineer scopes a use case with the people who own the process, builds on the client's real data and systems, ships a working system, and stays until it runs in production and the client's team can operate it. First Round Review, which published a hiring guide for the role in February 2026, describes the job as building the “last mile” of a product so it works in production, usually on site with the customer.

The day to day is engineering, not advisory work. An FDE writes and debugs production code, connects to legacy systems, models the client's workflows, and reports progress to the business owner of the problem rather than to a project office. The engineer is measured on adoption: a system in daily use and the number it moved, not a slide deck or a demo.

The role was devised at Palantir nearly two decades ago, for government agencies and large companies whose environments were too complex to support from a distance. For years the model was dismissed as consulting under another name. Palantir's results changed that reading, and in 2025 and 2026 the major AI vendors adopted the title.

The closest comparison is staff augmentation, and the difference matters. IDC's Jennifer Hamel told CIO Dive in August 2026 that embedding FDEs resembles consultancy staff augmentation, with one difference: the FDE role ties the business directly to building AI that can be deployed. A contractor fills a seat on a backlog your team already owns. An FDE finds the use case, owns the build and is accountable for the outcome.

What an FDE is not

  • A sales or solutions engineer

    Those roles demonstrate and configure a product during a sale. They do not write and own production code afterwards.

  • A consultant who delivers a report

    The deliverable of an FDE engagement is a running system, with the recommendations built into it.

  • A contractor from a body shop

    Staff augmentation fills a seat on your backlog. An FDE brings the backlog: the use cases, the architecture and the delivery.

  • A vendor's implementation team under a new name

    Gartner's Mukul Saha warned in October 2026 that some providers use the label for ordinary implementation work, at premium fees. The questions in section 06 tell the two apart.

02 · the market

why every vendor copied the model in 2026

Within a few months of each other, the largest AI companies built or bought forward deployed teams. The dates explain the shortage of engineers and the prices that follow.

Palantir's engineers worked inside intelligence agencies and defense organizations long before the title reached the rest of the industry. Monthly job listings for the role rose about 800% between January and September 2025, according to Financial Times figures cited by First Round Review. In 2026 the vendors moved from hiring engineers to building whole organizations around them.

  1. June 2025Andreessen Horowitz

    Partner Joe Schmidt publishes “Trading Margin for Moat”, the essay that called the forward deployed engineer the hottest job in startups. At the time, 22 of OpenAI's 311 open roles were forward deployed or solutions engineering positions.9

  2. December 2025Accenture and Anthropic

    A multi-year partnership trains about 30,000 Accenture professionals on Claude and creates “reinvention deployed engineers”, Accenture's name for FDEs placed inside client environments.7

  3. April 2026Google Cloud

    A $750 million commitment to its partner ecosystem for agentic AI, including Google forward deployed engineers embedded with the large systems integrators.63

  4. May 2026OpenAI

    OpenAI acquires Tomoro, a firm of roughly 150 forward deployed engineers and deployment specialists, as the founding acquisition of its Deployment Company, a services subsidiary launched at a $14 billion valuation.5

  5. May 2026Google Cloud

    CEO Thomas Kurian announces an AI-focused go-to-market organization and the hiring of additional forward deployed engineers. Google Cloud lists 59 roles in the United States, London, Paris and Hong Kong.311

  6. June 2026AWS

    $1 billion for a forward deployed engineering organization: pods of five or six engineers who build inside customer accounts in sprints of about 45 days, with the NFL, the NBA and Cox Automotive among the first customers.4

  7. August 2026Microsoft

    $2.5 billion for the Microsoft Frontier Company, which places 6,000 engineering experts in customer operations. Land O'Lakes and Unilever are early users.1

What Gartner expects

Gartner's Alex Coqueiro expects more than 85% of technology providers to have launched FDE programs as their main way of delivering AI by the end of 2026. The same firm expects about 70% of vendor-led FDE initiatives to be abandoned by 2028, for lack of internal skills on the client side and because of cost. Both predictions describe the vendor-run model: thousands of engineers deployed by a product company to install that company's product.12

The shortage

The supply is thin. Gartner's Mukul Saha put the number of active forward deployed engineers at about 2,000 in October 2026, against demand roughly four times that, with AWS and Microsoft alone trying to hire thousands. For a buyer, the shortage is the reason a vendor's team may not take a mid-market account, and the reason to check the seniority of whoever a provider proposes to send.12

03 · the options

four ways to buy forward deployed engineering

The same job title is sold four ways. The differences are in who the engineer works for, what the pricing rewards, and what remains when the engagement ends.

Who it suits

A vendor's deployment team

Large enterprises buying that vendor's platform at scale. OpenAI, Google Cloud, AWS and Microsoft staff these teams for their biggest accounts.

A global systems integrator

Companies with a multi-year transformation program and the budget for a large team.

A specialist partner

Mid-market companies that want a senior engineer on their own problems within weeks. FluentData's category: Google Cloud-native, from Miami and Lima.

Your own hire

Companies with several AI use cases a year and an engineering organization to hold the role.

Pricing shape

A vendor's deployment team

Bundled with platform consumption or an enterprise agreement. The engineering is priced to grow the platform spend.

A global systems integrator

Time and materials or a statement of work, with change orders when the scope moves.

A specialist partner

A fixed monthly retainer per engineer or squad. FluentData publishes three packages.

Your own hire

Salary, benefits and recruiting. OpenAI's 2026 posting for a deployment lead listed a base of $198,000 to $294,000.

Platform tie-in

A vendor's deployment team

Total. The engineer's job is to make that vendor's models or cloud your standard.

A global systems integrator

Partial. Most integrators partner with every major vendor and bring their own accelerators.

A specialist partner

Declared. FluentData builds on Google Cloud and says so; the code and the data stay in your account.

Your own hire

None. Your engineer chooses, and lives with the choice.

Time to a first working system

A vendor's deployment team

Set by the vendor's queue. AWS describes its sprints as about 45 days.

A global systems integrator

Months. Discovery, architecture and governance phases come first.

A specialist partner

FluentData commits to a working prototype on your data within 30 days.

Your own hire

Three to six months of recruiting before the work starts. Gartner counts about 2,000 active FDEs worldwide.

What stays when the engineer leaves

A vendor's deployment team

A deployment on the vendor's platform. Maintenance usually needs the vendor's services again.

A global systems integrator

Documentation and a transition plan. The people rotate.

A specialist partner

Code, pipelines and prompts in your repositories, and a trained team, if the contract says so. Ask.

Your own hire

Everything, for as long as the person stays.

Gartner's question: a premium fee forever?

A vendor's deployment team

Likely. The model is built to keep you on the platform and on the services.

A global systems integrator

Depends on the handover clause.

A specialist partner

Answerable. A fixed retainer with a notice period and an exit plan. FluentData's packages are monthly.

Your own hire

Not applicable, though the hiring premium is real.

FluentData is the third column, and this guide is written by the company that sells it. The other three columns are fair descriptions, and two of them are the better choice in the cases listed in section 07.

04 · cost

what forward deployed engineering costs

Few sellers publish a price. These are the public reference points, and a plain way to compare an offer.

The floor: an in-house hire

Salary postings set the floor for owning the capability. OpenAI's 2026 listing for a Technical Deployment Lead in San Francisco carried a base range of $198,000 to $294,000, before equity. Google Cloud's applied AI forward deployed engineering roles in New York and Atlanta listed $127,000 to $183,000 base in May 2026, per CIO Dive. Add benefits, recruiting and the months the seat stays empty.

Vendor teams and integrators

Neither publishes a rate card. Gartner estimated in October 2026 that FDE consulting fees alone could exceed $200,000 per quarter per use case, and that some providers charge premium fees for ordinary implementation work. A quote that cannot be tied to a use case and a deliverable is the warning sign.

A specialist partner: FluentData's published packages

FluentData publishes its FDE packages: an advisory engineer two days a week at $15,000 a month, a full-time embedded engineer at $28,000 a month, and an FDE-led squad of two to three builders with a program manager from $55,000 a month. Every package includes a Google Cloud-native AI stack, responsible AI governance and weekly stakeholder reviews.

See the packages

How to compare an offer

01

Price the first use case, not the engineer

Take the value the first production system is expected to move (hours, errors, revenue, cycle time) and set it against the monthly cost until it ships.

02

Count the handover

A system your team cannot run is a subscription. Gartner's question is whether you will pay a premium fee forever.

03

Check what the retainer includes

Governance, reviews, surge capacity and cloud costs are either inside the number or on top of it.

04

Compare equal seniority

A senior engineer two days a week often beats a junior one full time, because the hard part is choosing and scoping the use case.

05 · the first 90 days

how an engagement runs

Vendors describe the work in sprints of weeks, not phases of quarters. This is the shape to expect from any provider, and what FluentData commits to.

Weeks 1 to 3

Opportunity map

The engineer sits with the people who run the operation, reads the data and lists the AI use cases by value and effort. FluentData's engineers map every AI opportunity across your operations in the first three weeks.

Day 30

Working prototype

One use case, built on your real data, in front of its users. FluentData commits to a working prototype within 30 days. AWS describes its own sprints as about 45 days.

Days 30 to 90

Production

Access controls, monitoring, evaluation, error handling, cost controls and the integration into the system people already use. The prototype becomes a service someone is on call for.

Ongoing

Handover and the next use case

Your team learns to run and change the system while the engineer moves to the next item on the map. FluentData runs weekly executive reviews.

What you provide

  • Access to the data and the systems in the first week, with a named owner for each.
  • A business owner for the use case who can decide, not only advise.
  • A decision cadence: a weekly review that can approve, change or stop the work.
  • A place for the engineer: a seat in the team's channels and meetings, on site or remote.
  • An exit plan from day one: who runs the system after the engineer leaves.

What FluentData commits to

  • An opportunity map of your operations in the first three weeks.
  • A working prototype on your data within 30 days.
  • Weekly executive reviews, and a monthly AI opportunity review on the advisory package.
  • Delivery on Google Cloud's FedRAMP-authorized infrastructure with HIPAA-eligible services.
  • A fixed monthly retainer, with no per-project invoices.
Read the Forward Deployed Engineers page

06 · before you sign

twelve questions to ask a forward deployed engineering provider

Bring these to the first call. A good provider has answered them before.

  1. 01

    Who owns the code, the prompts, the evaluations and the models we fine-tune?

    A good answer You do, in your repositories and your cloud account, from the first commit. Anything else is a platform decision dressed as a service.

  2. 02

    Does the system run without your engineer?

    A good answer Yes, with documentation, monitoring and a named person on your side who can operate it. Gartner's test applies: if the answer is no, you are buying a subscription.

  3. 03

    Who exactly will be embedded, and how senior are they?

    A good answer A named engineer with production systems behind them, interviewed by you before the start. Not a bench to be assigned later.

  4. 04

    How many clients does that engineer serve at the same time?

    A good answer One, on a full-time package; a stated number of days a week on an advisory one. Shared engineers are shared attention.

  5. 05

    Is the price fixed or hourly, and what changes it?

    A good answer A fixed monthly amount per engineer or squad, with the triggers for a change written down: more people, more days, a new site.

  6. 06

    Which platforms and models does the work tie us to?

    A good answer A plain statement. FluentData builds on Google Cloud; a vendor's team builds on the vendor. Either is fine if you chose it.

  7. 07

    Where does our data live, and which compliance regimes does the work respect?

    A good answer In your own cloud account, with the controls you already use. The provider names the regimes it covers (HIPAA, SOC 2, FedRAMP, local data residency) and the ones it does not.

  8. 08

    How do you measure success?

    A good answer Production adoption and the business number the use case moves, reviewed weekly. Demos count for nothing.

  9. 09

    Can you name a client and a public result?

    A good answer A reference you can call, and a result published somewhere you can read it.

  10. 10

    What is the review cadence, and who attends?

    A good answer A weekly review with the business owner and the engineer, and a monthly review with the executive sponsor.

  11. 11

    What is the notice period?

    A good answer Thirty days is common for a monthly retainer. Anything that locks you in for a year is an enterprise agreement, not a retainer.

  12. 12

    Will our team be trained, and how?

    A good answer Pair work during the engagement, written runbooks, and a handover week before the engineer leaves.

07 · when to wait

when a forward deployed engineer is the wrong choice

An honest provider says no in these cases. FluentData does.

There is no usable data foundation yet

If the data lives in spreadsheets, exports and a system nobody trusts, the first months of an FDE engagement would be data engineering at an FDE rate. Do the data work first, as a scoped project.

You need a strategy or an audit, once

An FDE builds. A one-time assessment, a vendor selection or a board paper is a consulting engagement with a fixed deliverable.

Your AI team needs tooling, not people

If you already have engineers who ship, buy them the platform, the evaluation tooling and the training, and keep the headcount inside.

The budget cannot carry a monthly retainer

A retainer that stops after two months leaves a half-built system. A scoped project with a fixed price and a defined end fits better.

Regulated data cannot leave its system until a platform decision is made

Decide the platform and the controls first. An engineer embedded before that decision spends the retainer waiting for access.

08 · faq

frequently asked questions

Short answers to the questions buyers ask most.

A full-stack software engineer placed inside a client's team who scopes, builds and ships AI systems on the client's own data and systems, and stays until they run in production. The role was created at Palantir and was adopted across the AI industry in 2025 and 2026.

Talk to a forward deployed engineer

Twenty minutes with an engineer, not a sales deck. Bring one process, and we will tell you whether an FDE is the right way to fix it.

Talk to an FDESee the packages