Enterprise AI Dictionary
Concise definitions with links into deeper Learn material.
Agent
A system that observes, reasons, and acts toward a goal under a runtime.
Coming SoonAgent Harness
Execution runtime that governs context, tools, state, and authorization around an LLM.
Agent Runtime
The deterministic layer that owns sequencing, permissions, and recovery.
Coming SoonAgent Loop
Observe → reason → act → evaluate until a stop condition.
Coming SoonAgentic Workflow
A workflow that uses an LLM for some decisions, not unbounded autonomy.
Coming SoonContext
The assembled workspace a model sees for a given turn.
Coming SoonContext Engineering
Assembling, bounding, and governing what the model is allowed to see.
Memory
Persisted information retrieved later — not the current prompt.
Coming SoonState
Durable workflow or session facts required to resume correctly.
Coming SoonTool
An external capability invoked with structured arguments.
Coming SoonCapability
A permissioned, discoverable action a runtime may grant.
Coming SoonRAG
Retrieval-augmented generation: fetch evidence, then generate.
Coming SoonReranker
A second-stage ranker that reorders retrieved candidates.
Coming SoonGrounding
Tying generation to retrieved or deterministic evidence.
Coming SoonMCP
A protocol for exposing tools and context to model clients.
Coming SoonEval
A measured check of quality, safety, or task success.
Coming SoonModel Gateway
Shared entry point for model providers, routing, and policy.
Coming SoonAI Gateway
Enterprise control plane in front of models, tools, and spend.
Coming SoonContinuous Authorization
Re-evaluate access at the execution boundary, not once per login.
Coming SoonWorkload Identity
Identity for services and runtimes, independent of a human session.
Coming SoonDelegation
Acting on behalf of another principal with a constrained grant.
Coming SoonTrust Boundary
A line where identity, data, or control changes hands.
Coming SoonDurable Execution
Work that survives crashes via checkpoints and replay.
Coming SoonHuman-in-the-Loop
A required human decision inside an otherwise automated path.
Coming SoonControl Plane
Policy, identity, routing, and configuration — not the model loop.
Coming SoonExecution Plane
Where tools actually run and side effects occur.
Coming SoonOutcome Coordination
Commitments, verification, and settlement instead of fragile tool choreography.
Semantic Search
Retrieval by meaning using embeddings rather than exact terms.
Coming SoonVector Search
Nearest-neighbour lookup over embedding space.
Model Routing
Sending a request to the right model for cost, quality, or policy.
Coming Soon