Forward Deployed Engineer (FDE) Playbook
Interactive
FDE Simulator
Step through realistic customer scenarios and get scored on technical correctness, architecture, customer understanding and production readiness.
Where do enterprises deploy AI, and what does it really take?
The use case is rarely the hard part. The implementation requirements are. These are the constraints FDEs design around in each industry.
| Industry | High-value use cases | Implementation requirements | Practise it |
|---|---|---|---|
| Financial services | Policy and procedure assistants, KYC document review, invoice and procurement agents, analyst research copilots | Document-level access control, citations on every answer, immutable audit trail, model risk management sign-off, data residency | Policy assistant scenario |
| Healthcare | Triage support, clinical documentation, prior-authorisation agents, patient messaging drafts | PHI handling under HIPAA, conservative safety thresholds, clinician approval on every patient-facing output, no autonomous diagnosis | Triage assistant scenario |
| Legal | Research assistants, contract review, matter intake, clause extraction | Verifiable citations, privilege protection, matter-level isolation, no fabricated authority | Legal research scenario |
| Manufacturing | Technician knowledge assistants, maintenance agents, quality-incident triage | Offline or low-connectivity operation, voice input, refusal on safety-critical procedures without a source | Manufacturing scenario |
| Retail & e-commerce | Customer support agents, returns and refunds, merchandising and catalogue agents | Hard limits on refund authority, escalation triggers, PCI scope kept out of the model, latency under a few seconds | Support agent scenario |
| SaaS & IT operations | Always-on ops agents, ticket triage, on-call summarisation, internal developer agents | A separate identity per agent, least-privilege tokens, approval before destructive actions, SIEM logging | SaaS ops agent scenario |
| Public sector | Caseworker assistants, benefits and permit intake, records search | Sovereign or on-prem hosting, accessibility, explainability for decisions, strict retention rules | Permit intake scenario |
How should an organisation embed AI engineering?
Three operating models work in practice. Most enterprises move from left to right as the number of production use cases grows. Pair this with the 90-day enterprise AI adoption playbook.
One team owns model access, the eval harness, guardrails, observability and shared tools (often MCP servers). Business units request use cases.
Watch out: Becomes a queue. Domain knowledge stays outside the team, so adoption stalls.
The hub owns the platform and standards. Spokes are AI engineers embedded in business units who build on the platform.
Watch out: Standards drift if the hub has no review gate or shared evals.
Small pods (an FDE, a domain lead, and a platform liaison) sit with the business or the customer and own a use case from discovery to production.
Watch out: Needs a strong platform underneath, otherwise every pod rebuilds the same plumbing.
How can an enterprise train its engineers into FDEs?
An 18-week cohort plan built entirely on Learn AI Engineering. Each phase maps to a roadmap or the simulator, and ends with a checkpoint managers can verify.
Weeks 1–4
AI Engineering foundations
Every engineer can ship an LLM feature with structured outputs, RAG and a basic eval.
AI Engineer roadmapWeeks 5–8
Agents, tools and MCP
Engineers build tool-using agents with stopping conditions, guardrails and MCP servers.
Agentic AI roadmapWeeks 9–12
Forward deployment
Engineers run discovery, design under constraints, and deliver a customer POC.
FDE roadmapWeeks 13–14
Simulated engagements
Each engineer completes the FDE simulator scenarios and is scored on architecture, customer understanding and production readiness.
FDE simulatorWeeks 15–18
Capstone on a real internal use case
Pods ship one production use case with evals, an audit trail and an owner. This is the proof the programme worked.
FDE learning path
Running this for a team? The AI adoption guide covers how to plan, sequence and measure the rollout.
FDE scenarios
Latest on FDEs and always-on agents
Forward Deployed Engineer FAQ
What does a Forward Deployed Engineer do?
An FDE takes an AI system from a customer's problem to a working production deployment. Day to day that means technical discovery with stakeholders, solution architecture under the customer's security and data constraints, rapid prototyping, integration with their systems and identity provider, evaluation, and handing over a system the customer can operate.
How is an FDE different from a solutions engineer or consultant?
A solutions engineer mostly works pre-sale and demonstrates the product. A consultant mostly advises. An FDE writes and ships production code inside the customer's environment and is accountable for the system working after launch.
What skills does an FDE need in 2026?
Strong software engineering, AI engineering (RAG, tool calling, agents, evals), plus the newer deployment skills that always-on agents demand: giving each agent its own identity and scoped credentials, designing approval rules, audit logging, and cost control. Communication with non-technical stakeholders matters as much as any of these.
How long does it take to become a Forward Deployed Engineer?
For an experienced software engineer, roughly 4 to 6 months of focused practice. The FDE roadmap on Learn AI Engineering is 18 weeks and assumes you already write production software.
Why are FDEs in demand now?
Model capability is no longer the bottleneck for most enterprises. Getting AI to work safely with real data, real permissions and real processes is. Always-on agents such as OpenAI Dots, Meta Muse and xAI Grok Bot make this sharper, because an agent that acts around the clock needs careful identity, approval and audit design before it is trusted.
Can an enterprise train its own engineers to become FDEs?
Yes. The most reliable approach is a cohort programme: shared foundations, then agents, then forward deployment, then simulated engagements, then a capstone on a real internal use case. The 18-week plan on this page maps each phase to roadmaps, labs and the FDE simulator so managers can track progress.
How should an organisation structure its AI engineering team?
Most start with a central platform team, move to a hub-and-spoke centre of excellence as use cases multiply, and add embedded FDE pods for high-value deployments. The platform team owns models, evals, guardrails and shared tools, and the FDEs own outcomes with the business.