Bonus 1

Glossary of 100 Agentic AI Terms

Every term here is written for a working data professional. Short, plain English, no marketing.

Showing 118 of 118 terms
A
Access Control
Policies that decide what data an identity is allowed to see.
Action Control
Policies that decide what an agent is allowed to do, not just what it can see.
Action Gateway
An in-line service every agent action passes through for policy checks and logging.
Action Inventory
The structured list of every action an agent can take, with risk tier and owner.
Agent
A system that can answer, reason, plan, use tools, and take action — not just chat.
Agent Graph
A multi-agent topology where agents call each other freely. Powerful and easy to overuse.
Agentic AI
AI that acts on real systems, not just answers questions.
Approval Queue
A shared inbox where humans review agent actions before they execute.
Audit Trail
A structured record of what the agent did, why, and with what evidence.
B
Backpressure
Slowing the producer when the consumer cannot keep up.
Blast Radius
How many people, records, or systems an action affects if it goes wrong.
Business Memory
The layer that captures how your company defines its world, beyond just schemas.
Business Rule
A condition or exception the business enforces, like discount eligibility.
C
Caching
Reusing a previous answer to avoid recomputing it.
Calibrated Confidence
Confidence derived from system evidence, not from raw model probabilities.
Cascading Models
Trying a cheap model first and escalating to a larger one only when needed.
Catalog
An inventory of where data lives and who owns it.
Choice
The ability to swap models, tools, or vendors without rewriting your system.
Citation
A reference the agent shows so users can verify where an answer came from.
Compositional Tools
Small, reliable tools that agents combine to do bigger work.
Confidence
How sure the agent is about its answer, ideally surfaced honestly.
Context
The meaning, definitions, and relationships an agent needs to answer correctly.
Context Engineer
A data professional who owns meaning the way engineers own code.
Context Layer
The queryable layer above storage and compute that holds business meaning.
Context Window
The maximum amount of text a model can read in a single request.
Control
The set of policies that make agent behavior safe and predictable.
Cost
The total expense of running AI — tokens, tools, retries, and infrastructure.
Cost-Aware Architecture
A design that treats cost like latency: a first-class requirement.
D
Data Product
A piece of data offered with quality, ownership, and a clear contract.
Delta Lake
An open table format that adds reliability features on top of Parquet files.
Determinism
Same input, same output, every time. Hard for LLMs, important for audits.
Downgrade
A policy outcome that lets a smaller, safer version of the action proceed.
Drift
When model or data behavior changes over time, often silently.
E
Embedding
A numeric representation of text used for similarity search.
Eval Harness
A test suite that scores AI quality across many examples.
Eval-Driven Development
Building AI features by writing evaluations first, then improving the system.
Evaluation
Measuring whether the agent is giving correct, safe, useful answers.
F
Fine-Tuning
Adapting a base model to your domain by training it on your data.
G
Gateway
A single entry point that all AI calls go through for security, routing, and logging.
Glossary
A list of business terms with agreed, simple definitions.
Golden Set
A curated set of examples used as the truth for evaluation.
Governance
The set of rules that decide what is allowed in the data and AI stack.
Grounding
Anchoring an answer in real, citable data instead of model memory.
Guardrails
Code-enforced rules that block dangerous inputs or outputs.
H
Hallucination
When a model produces something that sounds confident but is not true.
Handoff Contract
The typed schema two agents agree on when passing work between them.
Human in the Loop
A design where humans review, approve, or take over agent actions.
I
Iceberg
An open table format widely adopted for lakehouses.
Idempotent
Safe to retry — the same call twice gives the same result.
Inference
Asking a model to produce an output, as opposed to training it.
Ingestion
Bringing raw data into your platform from source systems.
Interface Dividend
The compounding velocity gain a platform earns by routing change through stable interfaces.
K
Knowledge Graph
A structured map of concepts and how they relate, used to enrich context.
L
Lakehouse
A storage architecture that combines lake flexibility with warehouse reliability.
Lineage
The path a piece of data took from source to answer.
LLM
A large language model trained to generate human-like text.
M
MCP
Model Context Protocol — a standard way for agents to talk to tools and data.
Metric
A defined business number, like revenue or churn.
Metrics Layer
The place where metrics are defined once and calculated consistently.
Model Choice Matrix
A map of tasks to the right model size for each one.
Model Routing
Sending each request to the smallest model that can handle it well.
Multi-Agent
A design where multiple specialized agents work together.
O
Observability
The ability to see what your AI system is doing in production.
Ontology
A structured map of business concepts and how they relate.
Open Format
A data or model format anyone can read without vendor lock-in.
Open Interface
A standard API anyone can implement, reducing lock-in.
Orchestration
Coordinating multiple steps, tools, or agents into one workflow.
P
Parquet
An open columnar file format widely used in data platforms.
Permission
A rule about what an identity is allowed to see or do.
Pipeline Pattern
A multi-agent topology where agents run in a fixed sequence — the safest default.
Policy as Code
Writing governance rules as software the system enforces automatically.
Policy Engine
A service that evaluates policy-as-code at runtime.
Prompt
The instructions and context sent to a model in a single request.
Prompt Engineering
Designing prompts so models behave well for a given task.
Provenance
The chain of sources behind an agent's answer, surfaced so users can verify it.
R
RAG
Retrieval-Augmented Generation — letting the model read your data before answering.
Reasoning
A model thinking through steps before producing a final answer.
Reference Architecture
A shared pattern teams follow so each new project does not reinvent the wheel.
Replayable
A system where you can re-run history to debug or recover.
Retrieval
Looking up relevant content to give a model better context.
Retry
Re-running a failed model call. Quietly expensive at scale.
Reversibility
Whether an action can be undone cleanly, and how quickly.
Rollback
Turning off or reverting an agent quickly when something goes wrong.
Router
A cheap classifier that decides which model or tool handles a request.
S
Schema
The shape of a table — columns and types.
Semantic Cache
A cache keyed by meaning, not exact text, that reuses similar answers.
Semantic Layer
The layer that maps business meaning to underlying data.
Semantic Search
Finding content by meaning, not exact keywords.
Sensitivity Tag
A label that marks data by how sensitive it is, like PII or financial.
Side Effect
Something an action changes in the outside world, like sending an email.
Signal
A measurable indicator used to monitor quality, cost, or safety.
Source of Truth
The one place a fact is considered authoritative.
Stewardship
The ongoing care and ownership of a data asset.
Stream
Continuous data that arrives event by event rather than batch by batch.
Structured Retrieval
Looking up answers in tables, metrics, or APIs before falling back to text search.
Supervisor Pattern
A multi-agent topology where one agent routes work to specialists.
Synthetic Data
Generated data used for training or testing when real data is scarce.
T
Telemetry
Data emitted by a system that lets you observe its behavior.
Termination Condition
The explicit rule that ends an agent loop. Without one, loops burn money.
Throttling
Limiting the rate of requests to control cost or load.
Token
The unit a model reads or writes — roughly a few characters.
Token Budget
A target for how many tokens a request is allowed to use.
Tool Call
When the agent invokes an external tool, like a query or API.
Trace
The full record of a single agent run, end to end.
Trust Path
The sequence of checks an agent action passes before it executes.
Trust Signal
A visible cue (source, definition, confidence, limit) that helps users decide when to trust an agent.
Trusted Agent Architecture
A nine-step reference flow every production agent follows.
Trusted Source
A dataset agreed upon as authoritative for a given metric.
U
Unit Cost
Cost per unit of value delivered, like cost per resolved ticket.
Unity Catalog
Databricks' governance layer; one example of a managed catalog.
V
Validation
Checking that an input or output meets the rules before using it.
Vector Database
A store that lets you search by embedding similarity.
Vector Search
Finding similar content using embeddings.
Vendor Lock-In
A situation where switching providers is expensive or slow.
Versioning
Tracking changes to data, models, or definitions over time.
W
Warehouse
A structured store optimized for analytics queries.
Workflow
A defined sequence of steps a system runs to complete a task.
Z
Zero-Shot
Asking a model to do a task it was not explicitly trained for.