LEARNING MATERIALS / GLOSSARY

Bilingual glossary

Look up unfamiliar terms and abbreviations with plain-language explanations and examples. Apply them in the context of your task, permissions and evidence.

FDE: Forward Deployed Engineer: an engineering role working with customers to understand problems, build systems and deliver them into production. This course teaches delivery methods; coding and engineering skills need separate training.
Deployment Strategist: A deployment strategy role that investigates field work and connects business needs with data, product, adoption and value.
Echo / Delta: Palantir team labels. This course uses Echo to discuss business discovery and Delta for engineering delivery; other employers may use different roles and responsibilities.
System of record: The system holding the authoritative business state. To verify a submitted request, check the request record rather than relying on an AI response.
Human oversight: Review by people with enough information, time, skills and authority to correct or stop an action, rather than simply click approve.
Evaluation / Evals: Using tasks and test records to judge readiness for use. Check outcomes, failures, performance across task groups and domain experts’ judgments.
Adoption: People using the system in eligible everyday tasks and changing how they work. Account creation and login counts alone do not establish this.
Residual risk: Risk that remains after controls. An authorized owner must decide whether to accept it, address it further or stop.
Context engineering: Choosing and managing the instructions, tool descriptions, data and history an AI task receives, within the available context window.
Agent harness: The mechanism that repeatedly calls the model, routes requests to tools and manages execution.
Idempotency: Repeating the same request does not create an additional business action. For example, retrying after a timeout should not create two requests. This must be implemented and tested.
Handover: Transferring responsibility, access and operating procedures, then verifying that the receiving team can pause, handle problems and recover independently. Sending documents alone is not enough.