AI Agent Glossary
94 terms explained in simple language. Every concept behind modern AI agent systems.
A10 terms
Agent (AI Agent)
An AI system that plans, decides, and acts autonomously toward goals, rather than simply responding ...
Agent Orchestration
Coordinating multiple agents, tools, and workflows into a cohesive, goal-directed system.
Agent SDK
A software development kit for building production AI agents (e.g., Claude Agent SDK — the same engi...
Agent Team
A group of specialized agents that communicate directly with each other and divide work collaborativ...
Agent-Compatible Digital Experience
A website or application designed to serve both human visitors and AI agents simultaneously.
Agentic System
A collection of agents, tools, and workflows operating together to achieve an objective.
Agentic Workflow
An AI-driven process where agents make decisions, take actions, and coordinate tasks with minimal hu...
Audit Trail
An immutable record of all agent actions and decisions, essential for compliance and accountability.
Auto-Compaction
Automatic summarization of older context when the context window fills up, preserving recent informa...
AutoResearch (Karpathy Method)
An autonomous optimization loop: propose change, experiment, measure, keep or revert, repeat.
B1 terms
C9 terms
Cache (LLM Cache)
Storing previous AI responses for reuse. Saves costs and speeds up repeated queries.
Chain of Thought
Prompting technique that encourages the AI to reason step-by-step before answering.
Claude Code
Anthropic's agentic coding tool that runs in the terminal, understands codebases, and executes tasks...
Computer Use
An agent operating a user interface (browser, desktop apps) by clicking, typing, and navigating — us...
Context Engineering
The practice of shaping and managing information fed to an AI model for optimal results.
Context Rot
Degradation of model accuracy as the context window fills with too much information.
Context Window
The AI model's "working memory" — how much information it can process at once (e.g., 1 million token...
Cookieless Analytics
Visitor tracking without cookies or third-party scripts. Privacy-first measurement.
Cost Tracking
Monitoring every AI call: model used, tokens in/out, cost, cache status. Essential for profitable op...
D3 terms
Data Extraction
Extracting structured information from unstructured data like text, emails, or web pages.
Dual-Layer Architecture
Page structure where every page serves both humans (visual UI) and agents (structured data, APIs).
Dynamic Tool Selection
An agent's ability to choose which tools to use based on the task at hand.
E3 terms
F4 terms
Fallback
An alternative approach when the primary method fails (e.g., Ollama fails, Claude API takes over).
Few-Shot Learning
A model's ability to learn new tasks from just a few examples without retraining.
Fine-Tuning
Adjusting a pre-trained model on specialized data to improve performance for specific tasks.
Function Calling (Tool Calling)
The model's ability to trigger external tools: searches, databases, APIs, code execution.
G3 terms
GEO (Generative Engine Optimization)
Optimizing content for AI-powered search engines like Google AI Overviews and Perplexity.
Governance
Policies, controls, and processes that ensure responsible agent behavior in production.
Guardrails
Rules and constraints that prevent an agent from taking harmful or unauthorized actions.
H6 terms
Hallucination
AI-generated information that sounds plausible but is factually incorrect.
Handoff
When one agent dynamically delegates a task to another agent with more appropriate expertise.
Heartbeat
A scheduled check-in where an agent wakes, reviews its tasks, acts, and reports its status.
Hooks
Scripts that fire automatically at lifecycle events for deterministic control over AI behavior.
Human-in-the-Loop (HITL)
A human reviews and approves specific agent actions before they execute.
Human-on-the-Loop
A human supervises agents and intervenes only when thresholds are crossed.
I4 terms
Idempotent
An operation that produces the same result regardless of how many times it's executed.
Inference
The process of an AI model generating a response or prediction from input data.
ISO 27001
International information security standard defining requirements for an information security manage...
Iteration
A repeating process where an agent improves its output round by round.
J1 terms
K3 terms
L6 terms
Latency
The time it takes for an agent to respond or act.
Living Lab / Case Study
A website or system that runs on the same AI agent technology it promotes — proof by demonstration.
LLM (Large Language Model)
A large language model like Claude, GPT, or Gemini. The "brain" that understands and generates langu...
LLM-as-Judge
Using a second AI model to evaluate the quality of a primary agent's output.
llms.txt
A text file at a website's root that tells AI agents what the site contains and offers.
Local Model
An AI model running on your own server (e.g., via Ollama), keeping data private and costs near zero.
M5 terms
MCP (Model Context Protocol)
An open standard for connecting AI tools to external services. "USB-C for AI."
Memory (Agent)
Mechanisms that allow an agent to retain and recall information across sessions.
Memory, Long-Term
Persistent storage of facts, preferences, and historical interactions across sessions.
Memory, Short-Term
Context retained during a single session.
Multi-Agent System
An architecture where multiple specialized agents work together toward a shared goal.
N2 terms
O3 terms
P7 terms
Paperclip
An open-source orchestration platform that models company structures for AI agents: org charts, budg...
Planner
The agent component that breaks down high-level goals into executable steps.
Plugin
An extension that adds capabilities to an AI tool (skills, MCP servers, hooks).
PRD (Product Requirements Document)
A product spec. In agentic development, the PRD is the "program" that the agent executes.
Prompt
An instruction or question given to an AI model. Quality determines output quality.
Prompt Injection
An attack where external content hijacks an agent into acting against its instructions.
pSEO (Programmatic SEO)
Search engine optimization through programmatically generated pages at scale.
Q1 terms
R6 terms
RAG (Retrieval-Augmented Generation)
A method where AI retrieves relevant information from external sources before generating a response.
Ralph Loop
An autonomous development loop: work, evaluate, fix, repeat until done. Git serves as memory.
Rate Limiting
Throttling agent action speed to manage costs and prevent abuse.
ReAct Pattern
Agent action model: Reason, Act, Observe, Repeat.
Remote Control
Continuing a local AI coding session from your phone or any browser.
RPA (Robotic Process Automation)
Traditional rule-based automation. Agentic systems differ by their ability to adapt.
S8 terms
Sandbox
An isolated environment where an agent can execute code without affecting production.
Scalability
A system's ability to maintain performance as workload increases.
Schema.org
Structured data vocabulary used by search engines and agents to understand web content.
Self-Healing
A system's ability to detect and fix errors without human intervention.
Semantic HTML
HTML that describes content meaning, not just appearance. Foundation of agent-readable pages.
Skill
A reusable instruction set that teaches an agent a specific workflow.
Structured Data
Machine-readable information embedded in web pages that agents can parse.
Subagent
A specialized agent under a main agent, handling a specific task and reporting back.
T6 terms
Task Decomposition
Breaking a complex task into smaller, manageable subtasks.
Temperature
Controls how creative or deterministic AI responses are (low = predictable, high = creative).
Token
The unit of text an AI model processes. Approximately 3/4 of a word.
Token Budget
A limit on how many tokens an agent may use for a task.
Tool Use
An agent's ability to call external tools: searches, databases, APIs, code execution.
Traceability
Following an agent's decision chain from input to action to outcome.