AI Glossary

Plain-language definitions of AI terms, concepts, and technologies — explained for every professional.

A

Agent Handoff

The moment one AI agent passes work, context, or responsibility to another agent in a multi-agent workflow.

Agent Memory

Agent memory is the broader system for deciding what an AI agent should remember, retrieve, update, and forget across interactions.

Agent2Agent (A2A) Protocol

The Agent2Agent protocol is a standard for agent-to-agent communication, delegation, and result exchange across systems, especially when one agent needs another agent to do part of the work.

Agentic AI

Agentic AI is the broader pattern of AI systems that can pursue goals with limited prompting between steps by combining models with tools, context, memory, and execution loops.

Agentic Loop

The iterative cycle where an AI agent calls a tool, inspects the result, and decides whether to call another tool or return a final answer based on the stop_reason field.

Agentic Workflow

A multi-step AI workflow where the system decides what to do next, uses tools, and adjusts based on results instead of returning one static answer.

AI Agent

An AI agent is a software system that can pursue a goal across multiple steps by choosing actions, using tools, observing results, and continuing until it reaches a stopping point.

AI Alignment

The field of research focused on ensuring AI systems behave in ways that are helpful, harmless, and aligned with human values and intentions — the challenge of making AI do what we actually want.

AI Benchmark

A standardized test or evaluation used to measure and compare the performance of different AI models — helping professionals understand which model is best for specific tasks.

AI Copilot

An AI assistant embedded directly in your workflow tools that provides real-time suggestions, completions, and help as you work — like having an expert looking over your shoulder.

AI Eval

AI evals are structured tests used to measure whether an AI model or workflow is accurate, reliable, and useful enough for a real task.

AI Governance

The frameworks, policies, and processes organizations use to manage AI systems responsibly — covering risk assessment, compliance, ethics, and accountability for AI decisions.

AI Guardrails

AI guardrails are the policies, validators, permissions, and workflow controls used to keep an AI system inside acceptable behavior.

AI Hallucination

AI hallucinations are confident-sounding outputs that are not properly grounded in evidence, current context, or reliable retrieval.

AI Memory

Information an AI system stores or reuses across steps or sessions so it can keep context, recall facts, and behave more consistently over time.

AI Orchestration

The process of coordinating multiple AI models, tools, and data sources into a unified workflow — connecting the pieces so they work together to accomplish complex tasks.

AI Overviews

Google Search results that use generative AI to summarize information and answer some queries directly on the results page.

AI Safety

The broad field dedicated to ensuring AI systems are developed and deployed in ways that minimize risks and harms — covering everything from preventing misuse to ensuring reliable behavior.

Anomaly Detection

A machine learning technique for finding unusual patterns, outliers, or rare events that differ from normal behavior.

API (Application Programming Interface)

A standardized way for software applications to communicate with each other — the mechanism that lets your apps connect to AI services like ChatGPT, Claude, and other models.

Attention Dilution

Quality degradation that occurs when a model processes too many items, instructions, or context elements at once.

Attention Mechanism

The core innovation inside transformer models that allows AI to understand which parts of the input are most relevant to each other — like a spotlight that highlights the most important words.

B

C

Chain-of-Thought Prompting

A prompting technique that asks AI to show its reasoning step by step before giving a final answer, dramatically improving accuracy on complex problems.

Chatbot

An AI-powered conversational interface that can answer questions, complete tasks, and hold human-like conversations — the most common way people interact with AI today.

Chunking

The process of splitting a document into smaller pieces so an AI system can store, search, and retrieve the most relevant parts.

Classification Model

A machine learning model that predicts which category an input belongs to, such as spam vs. not spam or approved vs. rejected.

Claude Certified Architect

A professional certification from Anthropic that validates expertise in designing, building, and deploying production systems with Claude.

CLAUDE.md

A configuration file that provides persistent instructions, project context, and conventions to Claude Code across all sessions in a repository.

Computer Audition

The branch of AI that analyzes sound and audio signals, enabling systems to detect speech, music, events, voices, and patterns in recorded or live audio.

Computer Vision

The field of AI that enables computers to interpret and understand visual information from images and videos, powering everything from facial recognition to medical imaging.

Context Caching

A broader caching strategy that reuses valuable AI context such as prompts, documents, summaries, or memory so systems do not rebuild the same working state repeatedly.

Context Engineering

Context engineering is the work of deciding what the model sees, in what order, and with what supporting structure so the system behaves more reliably and usefully.

Context Window

The maximum amount of text an AI model can process in a single conversation — measured in tokens, it determines how much information the model can "see" and remember at once.

Context Window Management

Strategies for managing the finite token limit in conversations with a language model to maintain quality and avoid truncation.

D

E

F

G

H

I

J

K

L

M

N

O

P

Q

R

RAG

Retrieval-augmented generation is a pattern where an AI system gets relevant external knowledge at runtime instead of relying only on model training.

RAG (Retrieval-Augmented Generation)

A technique that combines AI text generation with real-time information retrieval from your own documents, producing more accurate, up-to-date, and source-grounded responses.

Reasoning Model

A reasoning model is an AI model optimized to spend more inference-time effort working through ambiguous or multi-step problems instead of returning the fastest plausible answer.

Reasoning Tokens

Tokens spent on an AI model’s internal reasoning process before it produces the final visible answer, often affecting quality, speed, and cost.

Recommendation System

An AI system that predicts and suggests items a user might like based on past behavior, preferences, and patterns — the technology behind Netflix suggestions, Amazon product recommendations, and Spotify playlists.

Red Teaming

A testing practice where people deliberately try to break an AI system, expose weaknesses, and uncover unsafe or unreliable behavior before real users do.

Regression Model

A machine learning model that predicts a continuous numeric value, such as price, demand, risk score, or time to completion.

Reinforcement Learning

A type of machine learning where an AI agent learns by trial and error, receiving rewards for good actions and penalties for bad ones — the method behind game-playing AI and robotics.

Reranking

A retrieval step that takes an initial set of search results and reorders them so the most relevant items appear first.

Retrieval Pipeline

The full sequence of steps an AI system uses to find, rank, and prepare information before generating an answer.

S

Semantic Caching

Semantic caching is a strategy that serves a previously computed answer when a new request is close enough in meaning to an earlier one, reducing cost and latency without requiring an exact prompt match.

Semantic Search

A search method that finds results based on meaning, not just matching keywords, so users can find relevant information even when they use different words.

Sentiment Analysis

An AI technique that automatically determines the emotional tone of text — positive, negative, or neutral — used for analyzing customer feedback, social media, and brand perception at scale.

Silent Error Suppression

An anti-pattern where tool implementations catch exceptions and return default values instead of surfacing the error to the model.

Speech-to-Text (STT)

AI technology that converts spoken audio into written text — powering meeting transcription, voice assistants, and accessibility features.

stop_reason

The field in a Claude API response that indicates why the model stopped generating, used to control agentic loop flow.

Structured Output

The ability to instruct an AI model to return data in a specific, machine-readable format like JSON, XML, or a defined schema — critical for building reliable AI-powered applications.

Structured Outputs

Structured outputs are AI responses constrained to a defined schema so downstream systems can validate and use them more reliably.

Subagent Isolation

The design principle that subagents in a multi-agent system operate with their own context and do not share memory with the coordinator or other subagents.

Supervised Learning

A type of machine learning where the model learns from labeled examples — like a student learning from an answer key — to make predictions on new data.

Synthetic Data

Artificially generated data that is created to resemble real data and used for training, testing, or protecting privacy.

Synthetic Environments for Robotics

Synthetic environments for robotics are simulated worlds used to train, test, and validate robotic systems before they operate in the physical world.

Synthetic Media

Media such as images, audio, video, or text that is generated or heavily transformed by AI systems rather than captured directly from the real world.

System Prompt

A high-priority instruction that sets the model’s role, behavior, tone, and rules before the user’s actual request is processed.

T

Temperature (AI Parameter)

A setting that controls how creative or predictable an AI model's responses are — lower temperature means more focused and consistent, higher temperature means more creative and varied.

Test-Time Compute

Test-time compute is the extra inference-time work a model does while answering a prompt, often to improve performance on harder problems at the cost of more latency or spend.

Text-to-Image AI

AI technology that generates images from written descriptions — type what you want to see, and the AI creates it. Powers tools like DALL-E, Midjourney, and Stable Diffusion.

Text-to-Speech (TTS)

AI technology that converts written text into natural-sounding spoken audio — enabling voice narration, accessibility features, and AI-generated podcasts from any text.

Throughput

A measure of how much work a system can handle over a period of time, such as requests per second or documents processed per hour.

Tokenization

The process of breaking text into smaller units (tokens) that an AI model can process — the first step in how language models read and understand your prompts.

Tokens and Context Windows

Tokens are the chunks an AI system processes, and the context window is the total working space available for instructions, history, and retrieved information.

Tool Description Routing

The mechanism by which Claude selects which tool to call based on the natural-language description provided in the tool definition.

Tool Use

The ability of an AI system to call external tools like search, calculators, databases, or APIs instead of relying only on its own generated text.

Tool Use and Function Calling

Tool use and function calling let AI systems call external tools, APIs, and actions instead of only responding in plain text.

tool_choice

An API parameter that controls whether Claude must use a specific tool, can choose any tool, or must respond without tools.

Transfer Learning

A technique where an AI model trained on one task is adapted to perform a different but related task — dramatically reducing the time, data, and cost needed to build useful AI.

Transformer Architecture

The neural network architecture behind modern AI models like ChatGPT and Claude — it processes text by understanding relationships between all words simultaneously rather than one at a time.

U

V

W

Z