Short, citation-ready definitions for forward deployed engineering vocabulary.
An engineer embedded in a client organization to ship AI systems in production and transfer capability.
Forward Deployed Engineers sit inside the customer's business, learn domain context, build governed agentic workflows, integrate with enterprise systems, and carry projects from pilot to production. The role was pioneered in defense and intelligence technology and is now critical for enterprise AI adoption.
An engineer focused on evaluation, guardrails, and continuous testing of production AI agents.
FDEEs design eval harnesses, monitor drift, enforce policy guardrails, and keep agents accurate and compliant after go-live. As agents move to production, eval engineering becomes as important as build engineering.
A small deployed unit of FDEs and FDEEs trained together and embedded as one team.
Pods combine builders and eval engineers who share methods and stack fluency. Companies hire pods when they need production outcomes faster than traditional hiring or consulting allows.
Designing and governing the information layer agents use to reason and act.
Context engineering unifies data from source systems: via MCP, retrieval, knowledge graphs, and policy filters: into a governed layer agents can trust. It is a core FDE skill for enterprise deployments.
An open standard for connecting AI applications to external data sources and tools.
MCP enables agents to access enterprise systems through standardized connectors. FDEs implement MCP integrations inside client perimeters as part of context unification.
Automated tests that measure agent quality, safety, and regression over time.
Eval harnesses run continuously in production or staging to catch accuracy drift, policy violations, and performance regressions. FDEEs own harness design and evolution.
Teaching client teams to operate AI systems as FDEs build them.
Capability transfer is intentional knowledge sharing: runbooks, pairing, documentation, and joint ownership: so clients are not permanently dependent on external engineers.
The organizational and technical chasm between a successful demo and governed production.
Most enterprise AI pilots fail to reach production because teams lack embedded engineers who understand domain context, integration, security, and operational evals. FDEs exist to close this gap.
Coordinating multiple AI agents, tools, and workflows toward a business outcome.
Agent orchestration defines how agents hand off work, call tools, respect guardrails, and recover from failures. FDEs design orchestration inside client environments where latency, permissions, and audit requirements matter.
Policy and technical controls that constrain what production agents may do.
Guardrails block unsafe outputs, enforce compliance rules, and route edge cases to humans. FDEEs implement and test guardrails as part of continuous eval discipline, especially in regulated industries.
Monitoring when production agent behavior or accuracy degrades over time.
Drift detection compares live agent performance against baselines: catching model updates, data changes, and emerging failure modes before users or auditors do. Core FDEE responsibility after go-live.
An AI agent running in a governed client environment with real data and accountability.
Production agents differ from demos: they operate under permission models, logging, eval harnesses, and stakeholder scrutiny. FDEs ship production agents; FDEEs keep them reliable.
Working inside a client's organization, systems, and security boundary.
Embedding means operating with client data, stakeholders, and infrastructure (on-site or in-VPC) rather than from a vendor sandbox. It is the defining context of forward deployed engineering.
A measurable drop in agent quality detected by an eval harness.
Eval regressions trigger investigation: model changes, context staleness, or new edge cases. FDEEs treat regressions like production incidents, with triage, fixes, and updated eval coverage.
Grounding agent responses in retrieved documents and data from authorized sources.
In enterprise FDE work, RAG must be permission-aware, auditable, and fresh. FDEs implement retrieval as part of context engineering, not as a generic vector-db tutorial.
Letting agents invoke APIs, databases, and tools to complete tasks.
Tool use connects agents to enterprise systems. FDEs wire tools with auth, logging, and failure handling; FDEEs eval tool-selection accuracy and unsafe invocation patterns.
The backlog of AI demos that never reached governed production.
Most enterprises accumulate pilots that stalled on integration, security, or ownership gaps. FDE pods exist to move selected pilots out of the graveyard with embed accountability.
Cohort 02's final assessment scored by deployment partners, not internal exams.
Capstones are judged on one question: would this survive inside a real organization? Partner graders mirror the security and platform stakeholders graduates will face on embed.
Running agents inside a client's private cloud perimeter.
VPC deployment keeps data and keys under client control. FDEs routinely ship inside customer VPCs or on-prem: a core requirement for enterprise and regulated engagements.
FDE Factory's current open intake, 40 seats, 12 weeks, FDE and FDEE tracks.
Cohort 02 is the only cohort accepting applications today. It covers discovery, context engineering, agent build, and partner-graded production capstone across both FDE and FDEE certification paths.