AI-Native Glossary

AI workflow

A bounded operational process where AI handles defined steps end-to-end with measurable KPIs.

AI-native

A delivery model where AI is the operating layer of the workflow, not a feature added on top.

Discovery sprint

A 2-week paid engagement that maps the workflow, baseline metrics, systems, and risk model.

Thin slice

A narrow, end-to-end production deployment that proves an AI workflow on real data and edge cases.

Agentic AI

AI systems that can plan, take multi-step actions, and use tools to complete tasks autonomously.

Autonomous agent

An AI agent that completes a defined task without per-step human input.

Chain of thought

Prompting the model to show intermediate reasoning steps before producing a final answer.

Embeddings

Numerical vectors that represent the meaning of a text, image, or other piece of content.

Function calling

Specific implementation of tool use where the model emits structured JSON calls to registered functions.

Hybrid search

Combination of keyword (BM25) and vector (embedding) retrieval, often re-ranked.

MCP (Model Context Protocol)

Open protocol for exposing tools, resources, and prompts to AI models in a standard way.

Multi-LLM architecture

Routing different tasks to different models based on cost, quality, latency, and capability tradeoffs.

RAG (Retrieval-Augmented Generation)

Generation grounded in retrieved source documents rather than the model's parametric memory alone.

ReAct

A reasoning pattern where the model alternates Thought → Action → Observation steps.

Semantic search

Search by meaning, not by keyword overlap.

Structured output

Constraining model output to a strict schema (JSON, regex, grammar) for reliable downstream parsing.

Tool use

An LLM's ability to call external functions, APIs, or services within a generation step.

Vector store

A database optimized for similarity search over embeddings.

Confidence score

A scalar that estimates how reliable a model's output is for a given input.

Evaluation harness

A test framework that scores model or prompt output against a labelled set of expected outputs.

Labelled test set

A frozen, hand-curated set of real input examples with expected outputs, used to score model behavior.

Prompt versioning

Treating prompts as code: stored, diffed, reviewed, and rolled back like any production artifact.

AI governance

Policies, processes, and controls that make an AI system auditable and accountable.

Audit log

Tamper-evident record of every model input, output, version, and reviewer action.

Grounding

Anchoring model output to verifiable source material to reduce hallucination.

Guardrails

Pre and post checks that filter unsafe, off-topic, or non-compliant model outputs.

Hallucination

Plausible but factually incorrect output generated by an LLM with no grounding.

Model card

Documentation describing a model's intended use, limitations, evaluation, and risks.

NIST AI RMF

U.S. NIST's voluntary framework for managing risks in AI systems across the lifecycle.

Prompt injection

An attack where user input manipulates the model into ignoring its system prompt or executing unintended actions.

Reviewer queue

A workflow where low-confidence or high-impact AI outputs route to a human for approval.

Context window

The maximum number of tokens a model can process in a single request.

Extended thinking

A model mode that performs longer internal reasoning before producing the answer.

Fine-tuning

Continuing the training of a base model on task-specific data to specialize behavior.

Foundation model

A large model pre-trained on broad data, then adapted to many downstream tasks.

Frontier model

The leading-edge foundation models with the highest reasoning, coding, and multimodal capabilities.

LLM (Large Language Model)

A transformer-based model trained on language data to predict and generate text.

LoRA

Low-Rank Adaptation: a parameter-efficient fine-tuning method that trains small adapters instead of full weights.

Multimodal

Models that accept and produce more than one type of content (text, image, audio, video).

Transformer

The neural network architecture that powers modern LLMs, based on self-attention.