FDE Handbook
What is a Forward Deployed Eval Engineer?
Production agents don't fail once; they drift. FDEEs are the engineers trained to catch it.
In short: A Forward Deployed Eval Engineer (FDEE) is an engineer who embeds with production AI systems to build eval harnesses, guardrails, drift detection, and continuous testing, keeping agents accurate, safe, and compliant after go-live.
Last updated July 2026
Why FDEEs exist
Shipping an agent is not the finish line. Models change, data drifts, policies evolve, and edge cases accumulate. Without embedded eval discipline, production agents silently degrade until an executive or regulator notices.
FDEEs pair with FDEs in pods: FDEs build and integrate; FDEEs prove the system still works, and keep proving it.
Core FDEE responsibilities
- Design eval harnesses tied to business outcomes, not vanity metrics
- Implement guardrails and policy controls for regulated environments
- Monitor drift, regression, and failure modes in production
- Produce compliance-ready reporting for security and legal stakeholders
- Iterate evals as the product and model stack evolve
FDEE vs ML engineer vs QA
ML engineers often optimize offline metrics on fixed datasets. Traditional QA rarely covers non-deterministic agent behavior at scale. FDEEs operate in the middle: continuous, production-grounded evaluation of systems that change every week.
A day in the life of an FDEE
Morning: review overnight eval regressions and drift alerts. Midday: pair with an FDE on a new edge case from production. Afternoon: update guardrails after a policy change from legal. FDEEs live in the gap between 'shipped' and 'still safe six weeks later.'
The work is less about one perfect benchmark and more about sustainable measurement: harnesses that stakeholders trust when models, data, and regulations shift.
How to become an FDEE
Track B in Cohort 02 certifies Forward Deployed Eval Engineers through the same 12-week embed method as FDEs, with capstone emphasis on eval harnesses, guardrails, and compliance reporting.
Apply to Cohort 02 and indicate FDEE track preference. Strong engineering fundamentals and comfort with ambiguity matter more than a traditional QA background.
Frequently asked questions
Does every pod need an FDEE?
For governed production in regulated or high-stakes environments, yes. Build-only pods risk go-live without sustainable eval discipline.
Can an FDE become an FDEE?
Some engineers cross-train on both tracks. Cohort 02 offers distinct certification paths for FDE and FDEE.
What metrics do FDEEs track?
Business-outcome metrics: task success, policy violations, escalation rates, latency under load, not vanity accuracy on static sets. Harnesses tie evals to what executives and regulators care about.
When should we hire an FDEE?
Before go-live in regulated or high-stakes workflows, not after the first incident. FDEEs should shape eval design while FDEs build, not bolt on testing later.
How does Cohort 02 train FDEEs?
Cohort 02 Track B shares weeks 1–10 with FDEs on embed and context work, then emphasizes eval harness design, guardrails, drift monitoring, and partner-graded compliance capstone in weeks 11–12.