FDE Hub · Forward Deployed Engineering · Updated September 2026

The engineer who ships AI inside the customer.

Forward Deployed Engineers sit with the customer, own the problem end to end and turn a model, a platform or an agent into a working system people actually use. This hub explains the role as it exists in 2026, the skills it demands, what it pays, and a credible route into it.

What an FDE does →Explore FDE programs
>10×Growth in FDE job postings in 2025 versus 2024LinkedIn News · 2026
9KAI-created FDE / PM roles globally, 2023–2025LinkedIn Economic Graph · 2026
$206KMedian US total compensation reported for FDEsLevels.fyi · Sep 2026
Up to 50%Travel required in one current OpenAI forward-deployed roleOpenAI Careers · 2026
01 · The role

Half engineer, half product owner, on-site.

Palantir coined the title. OpenAI, Anthropic, Databricks, Scale, Salesforce and ServiceNow made it a mainstream hiring category in 2025–26.

Definition

An FDE is deployed forward — into the customer's environment — to make an AI product deliver a measurable outcome.

Unlike a consultant, the FDE writes production code. Unlike a core engineer, the FDE owns the customer relationship and the result. The role exists because frontier models and enterprise platforms are only valuable once someone wires them into real data, real permissions and real workflows.

  • Embeds with the customer for weeks or months, often on-site
  • Owns discovery, architecture, build, evaluation and launch
  • Feeds what breaks in the field back into the core product
  • Is measured on adoption and business value, not tickets closed
A week in the role

What the calendar actually looks like

  • Mon — Workshop with the customer's ops leads to lock the first agent use case and its success metric
  • Tue — Map data sources, permissions and integration points; write the architecture and risk memo
  • Wed — Build: agent, tools, retrieval, guardrails; pair with the customer's engineers
  • Thu — Run the eval suite against golden cases; fix failure modes; review with security
  • Fri — Ship to a pilot group, instrument adoption, write the field report for the product team

The mix shifts with each engagement, but every FDE week contains discovery, code, evaluation and a conversation about value.

FDE vs core engineer

Outcome over roadmap

A core engineer owns a component of the product for every customer. An FDE owns one customer's result with the whole product — and ships glue, configuration and custom code the roadmap will never contain.

FDE vs solutions engineer / consultant

Production code over slides

Solutions engineers demo and advise. FDEs stay after the sale, write and operate production systems, and are accountable when the pilot fails to reach adoption.

FDE vs applied AI engineer

Customer over model

Applied AI engineers go deep on models, evals and inference. FDEs use that depth, but spend as much time on data access, change management and stakeholders as on the model.

FDE vs forward deployed software engineer

AI-shaped problems

The FDSE title (Palantir lineage) covers general software delivery in the field. The 2026 FDE wave is specifically about agents, retrieval, evaluation and AI governance inside enterprises.

02 · Delivery loop

Five stages, one accountable owner.

The same loop applies whether the platform is a frontier model API, Agentforce or the ServiceNow AI Platform. Each stage produces an artefact an employer can inspect.

Discover

Sit with users, map the workflow, quantify the pain, pick the first use case that is valuable and feasible, and lock the metric that defines success.

ArtefactVision-lock memo · process map · success metric

Build

Agents, tools, retrieval and interfaces — with the customer's engineers pairing so the system survives your departure.

ArtefactWorking system in the customer's environment

Integrate

Connect to data, identity and systems of record through APIs, MCP servers and platform connectors, respecting permissions and residency.

ArtefactIntegration map · MCP / connector inventory

Prove

Golden datasets, LLM-as-judge and human review; red-team for injection and data leakage; document cost, latency and accuracy against the bar.

ArtefactEval report · risk register · security sign-off

Launch & adopt

Pilot, train champions, instrument usage, remove blockers, expand — then hand over runbooks and report value to the sponsor and the product team.

ArtefactAdoption dashboard · runbook · value report
03 · Skills matrix

Three layers, rising expectations.

Job descriptions from OpenAI, Anthropic, Databricks, Salesforce and ServiceNow in 2026 converge on the same stack. Use the matrix to locate yourself.

Capability
Entry FDE
Mid FDE
Senior / lead FDE
Engineering foundationPython or TypeScript, APIs, SQL, cloud
Ships clean, tested services; comfortable in one cloud and one database
Designs integrations across systems; owns CI/CD, observability and cost
Sets architecture for multi-team, multi-system deployments; reviews others' code
AI engineeringLLMs, RAG, agents, MCP
Builds a RAG pipeline and a tool-using agent with an SDK; understands context and cost
Designs multi-agent systems, MCP servers and memory; tunes retrieval and prompts from eval data
Chooses build-vs-buy across model providers and platforms; designs for governance and scale
Evaluation & safetyEvals, guardrails, red-teaming
Writes golden datasets and runs eval suites; applies standard guardrails
Designs LLM-as-judge and trajectory evals; red-teams and remediates
Owns the AI risk register with security and legal; maps to EU AI Act / NIST AI RMF
Platform depthSalesforce, ServiceNow, Databricks…
Certified on one platform's core and its agent studio
Designs data, identity and agent architecture on the platform; knows its limits
Advises on platform roadmap fit; escalates product gaps with evidence
Customer & discoveryWorkshops, stakeholders, value
Runs structured interviews; documents processes and requirements
Facilitates executive workshops; negotiates scope; writes the value case
Owns the account relationship; expands from pilot to programme
Adoption & operationsChange, runbooks, support
Trains users; writes runbooks; monitors the pilot
Designs the champion network and adoption metrics; runs incident response
Builds the customer's own AI capability so the engagement ends well
04 · Where FDEs deploy

Six domains, one anchor each.

Vendors and service firms organise forward deployed teams into specialised pods. Each pod maps to an anchor discipline you can build first — then add the FDE loop on top.

Pod 01 · AI & agents

GenAI & agentic systems

Agents, RAG, MCP tools, evaluation and guardrails on frontier models — the pod most FDE postings describe.

Hired by OpenAI, Anthropic, Databricks, Scale, Sierra, Glean, AI practices at SIsAnchor: AI Engineer → Pod 02 · Enterprise platforms

Salesforce & ServiceNow

Agentforce, Now Assist and AI Agent Studio inside a customer's existing platform — data, identity and governance already in place.

Hired by Salesforce, ServiceNow, their partner ecosystems, Indian GCCsAnchor: Salesforce or ServiceNow FDE → Pod 03 · Data

Data & analytics platforms

Lakehouses, pipelines, semantic layers and the retrieval-ready data every agent depends on; Fabric, Databricks, Snowflake, BigQuery.

Hired by Databricks, Snowflake, Microsoft partners, analytics consultanciesAnchor: Data Engineering & AI → Pod 04 · Cloud & platform

Cloud, Kubernetes & DevSecOps

Landing zones, CI/CD, Kubernetes, security hardening and AIOps — the platform an AI deployment runs on.

Hired by AWS ProServe, Microsoft, cloud partners, platform teams in GCCsAnchor: Multi Cloud DevOps & AI → Pod 05 · App modernisation

Applications & integration

APIs, microservices and AI-native interfaces that put agents in front of users; modernising the systems they must integrate with.

Hired by Product companies, SIs, enterprise IT modernisation programmesAnchor: FullStack & AI Agents → Pod 06 · Innovation

POCs, MVPs & rapid prototyping

Validate an AI use case in weeks — discovery, prototype, eval, decision — before the customer commits to a full build.

Hired by Innovation labs, AI CoEs, venture studios, vendor FDE teamsAnchor: Forward Deployed AI Engineer →
Capabilities FDE postings ask for across pods
Agent frameworks & MCPRAG & retrieval qualityLLM evaluation & guardrailsSolution architectureCloud architectureIntegration & APIsData engineeringDevSecOps & automationSecurity & complianceObservability & reliabilityPlatform engineeringRapid prototyping & POCsDiscovery & business advisoryChange enablement & training
05 · How an engagement is measured

FDEs are measured on outcomes, not just output.

FDE interviews often probe whether you can connect technical delivery with adoption, reliability, cost and business value. Know the five signals a strong engagement tracks.

Time to value

From plan to evidence

Track the time from kickoff to a useful system in the customer's environment. Reusable components and a narrow first use case reduce avoidable delay.

Interview: "How would you plan a credible first deployment?"
Adoption

Usage, not launch

Weekly active users, tasks completed by the agent, containment or deflection rate — measured against the baseline captured in discovery.

Interview: "The pilot shipped but nobody uses it. What do you do?"
Cost

Efficiency delivered

Hours removed, cost per task, tokens and infrastructure per outcome — with a budget the customer signed off.

Interview: "How do you keep agent costs inside budget at scale?"
Reliability

Security & stability

Eval pass rate on golden cases, incident count, permission boundaries respected, audit evidence produced for security and compliance.

Interview: "Walk me through your eval and red-team process."
Handover

Knowledge transfer

Pairing, live documentation and runbooks so the customer's team runs the system and dependency on you decreases every sprint.

Interview: "How do you make yourself unnecessary?"
Focused sprint

One narrow deployment

One use case from discovery to pilot, with scope and timing agreed for the customer context.

Embedded engagement

Several releases

An FDE or pod works alongside the customer through multiple releases, adoption cycles and operational handover.

Outcome-led

Shared measures

The customer and delivery team agree measurable goals such as velocity, cost, reliability, adoption or customer satisfaction.

06 · Market & pay

A growing role with wide variation.

Compensation and working patterns vary materially by employer, level and location. Use current job postings and role-specific market data, not a single headline number.

Verified US compensation snapshot

Current reported distribution

25th percentileUS FDE total compensation
$175K
MedianUS FDE total compensation
$206K
75th percentileUS FDE total compensation
$283K

Source: Levels.fyi, accessed 18 September 2026. Total compensation can include salary, stock and bonus. This US dataset is not a salary guide for India or a promise of earnings.

Demand signalLinkedIn reported that Indeed postings for forward-deployed engineers grew more than tenfold in 2025 compared with 2024.LinkedIn News · Mar 2026
What current postings ask forCustomer discovery, architecture, production software delivery, iteration in the customer's environment and actionable product feedback.OpenAI Careers · accessed Sep 2026
Travel and locationWorking patterns differ by employer. One current OpenAI role is hybrid and requires travel up to 50%, so candidates should check every posting.OpenAI Careers · accessed Sep 2026
India contextTitles and pay vary across platform partners, services firms, product companies and GCCs. Compare live postings for your city, experience and anchor platform rather than applying US figures.Career guidance, not a compensation promise
07 · Programs

Three FDE tracks. One delivery method.

Each Digital Edify FDE program teaches the same five-stage loop on a specific stack, with a customer-style capstone and a public verification URL employers can inspect.

Cross-platform · Frontier models12 sections

Forward Deployed AI Engineer

Python, FastAPI and LLM engineering; agents with the Claude Agent SDK, OpenAI Agents SDK and LangGraph; MCP, RAG, evals and adoption — deployed into a customer environment.

Claude Agent SDKLangGraphMCPRAGEvalsDiscovery
Explore the program →
Salesforce · Agentforce 36012 sections · 50 modules

Salesforce Forward Deployed AI Engineer

Agentforce agents grounded in Data 360, connected through AIforce and MCP, governed by the Trust Layer — from vision-lock to a customer's production org.

Agent BuilderAgent ScriptData 360Headless ToolkitTesting Center
Explore the program →
ServiceNow · AI Platform12 sections · 48 modules

ServiceNow Forward Deployed AI Engineer

Now Assist skills and AI agents in AI Agent Studio, connected through AI Agent Fabric, MCP and Workflow Data Fabric, launched under AI Control Tower with evaluation evidence.

AI Agent StudioSkill KitAgent FabricData FabricControl Tower
Explore the program →

These programs develop capability and portfolio evidence. They do not guarantee a title, interview, employer, salary or placement outcome.

08 · 90-day roadmap

From anchor skill to field-ready evidence.

Start from an existing strength — software, data, cloud or an enterprise platform — and close the loop stage by stage. Each phase ends with something public.

Days 1–30

Anchor and AI foundations

  • Solidify one language (Python or TypeScript), SQL and one cloud
  • Build a RAG pipeline and a tool-using agent with one SDK
  • Learn MCP: consume a server, then build one for a real API
  • Write your first golden dataset and eval loop
ShipA deployed agent service with an eval report in its README
Days 31–60

Platform depth and integration

  • Pick a platform track — frontier models, Salesforce or ServiceNow — and certify on its core
  • Integrate the agent with identity, permissions and a system of record
  • Add guardrails, tracing and cost budgets; red-team it
  • Run a discovery workshop with a real or simulated customer and write the vision-lock memo
ShipAn integrated agent in a customer-style environment with a risk register
Days 61–90

Launch, adopt, defend

  • Pilot with users; instrument adoption; iterate from field failures
  • Write the runbook, the value report and the product feedback memo
  • Publish the case study with a verification URL
  • Practise FDE interviews: discovery, architecture, evals, security, adoption
ShipA public case study employers can verify — and a rehearsed story for each loop stage
09 · Resources

Go deeper on each question.

Guides written for people deciding whether, and how, to move into forward deployed engineering.

Role guide

What is a Forward Deployed AI Engineer?

The role family, its origins and how it differs from adjacent titles.

Read →Checklist

FDE skills & portfolio checklist

Assess your evidence across engineering, AI, evaluation, platform and customer delivery.

Use the checklist →Roadmap

How to become an FDE in India

Experience-aware routes from software, data, cloud, AI or an enterprise platform.

Open the roadmap →Interview prep

FDE interview questions

Discovery, architecture, evaluation, security, operations and adoption scenarios with rubrics.

Practise →Comparison

FDE vs Applied AI Engineer

Model and application depth versus the customer deployment mandate.

Compare →Comparison

FDE vs Forward Deployed Software Engineer

Where software delivery depth overlaps with, and differs from, AI deployment.

Compare →
10 · FAQ

Straight answers.

Do I need to be a strong software engineer to become an FDE?

Production software ability is central to most FDE roles. Current postings commonly ask for programming, APIs, data and cloud experience. Platform-track FDEs can begin from deep Salesforce or ServiceNow experience, then add scripting, integration and evaluation skills.

Is FDE a junior role?

Many frontier-lab roles ask for prior software and customer-facing experience. Entry paths can also begin in platform delivery, implementation or applied AI roles before expanding into broader FDE responsibilities.

How much travel is involved?

It varies from remote or hybrid work to frequent on-site delivery during critical phases. One current OpenAI forward-deployed role requires travel up to 50%, so candidates should check each employer's posting.

Which platform track should I choose?

Follow your existing context. Salesforce or ServiceNow admins and developers should stay on their platform and add agent delivery. Software and data engineers usually fit the cross-platform FDE track. An advisor can map this in one call.

What does a hiring manager look for in a portfolio?

A deployed system, an eval report with real failure modes fixed, an integration into identity and a system of record, evidence of user adoption, and a clear write-up of the customer problem and the value delivered — with a verification URL.

How fast is an FDE expected to show results?

Timelines depend on scope, access, security review and the customer's environment. Interviewers often look for candidates who can define a narrow first outcome, expose dependencies early and plan measurable iterations.

What does knowledge transfer look like in practice?

It is built into every sprint, not left to the close-out: pairing with the customer's engineers, documenting as you build, running enablement sessions and leaving runbooks. A good FDE engagement ends with the customer able to operate and extend the system without you.

How is this different from an AI consultant or solutions engineer?

FDE roles typically combine customer discovery with hands-on production delivery and post-launch iteration. The exact boundary with consulting, solutions engineering and customer engineering varies by employer.

Not sure which FDE track fits your background?

An advisor can map your current skills to a starting point, explain the evidence each path expects and set a realistic timeline.