AI automation, defined.

Agent loop

The iterative cycle an AI agent runs: think (decide next action) → act (call a tool) → observe (read the result) → think again. Continues until the task is complete or escalated.

AI agent

An AI system (typically built on a large language model) that takes actions in the world via tools, follows instructions to accomplish a goal, and can operate with limited or no human intervention per step.

Context window

The maximum amount of text an LLM can consider in a single request, measured in tokens. Modern frontier models handle 200K-2M tokens.

Human-in-the-loop

A design pattern where AI systems include explicit human review or approval gates at critical decision points, instead of running fully autonomously. Standard in production AI automation.

Integration layer

The architectural layer of an AI automation system that handles how agents read from + write to external systems (CRM, billing, data warehouse, comms). Modern stacks use a mix of native APIs, iPaaS tools (n8n, Make), and MCP servers.

iPaaS (Integration Platform as a Service)

A hosted platform that provides pre-built connectors between SaaS applications, enabling workflow automation without custom integration code. Examples: Zapier, Make, n8n, Workato, Tray.io.

LLM (Large Language Model)

A type of AI model trained on massive amounts of text data, capable of understanding and generating natural language at human-comparable quality for most tasks.

MCP (Model Context Protocol)

An open standard introduced by Anthropic in late 2024 for connecting AI agents to external tools and data sources via a uniform protocol — like a "USB-C for AI agents."

Orchestration layer

The architectural layer of an AI automation system that decides what runs when, what escalates to humans, what gets retried on failure, and how steps coordinate. Distinct from the agent itself.

Prompt engineering

The craft of designing the instructions, context, and constraints given to an LLM to produce reliable, high-quality outputs for a specific task.

RAG (Retrieval-Augmented Generation)

A pattern where an AI model retrieves relevant information from a knowledge base before generating a response, grounding its output in real source data rather than relying purely on what it learned during training.

RPA (Robotic Process Automation)

Automation software that emulates human use of a GUI — clicking buttons, filling forms, copying data across legacy systems that lack APIs. Increasingly being replaced by AI agents + direct API integration.

System prompt

The long, persistent instructions given to an LLM that define its role, behavior, constraints, and tone. Distinct from the per-turn user message, which is just the immediate request.

Tool use (function calling)

An LLM capability where the model can decide to call external functions (APIs, code, integrations) with structured arguments, then incorporate the result into its response.

Webhook

An HTTP callback that one system sends to another in real time when an event occurs. Foundational primitive for event-driven workflow automation.

Workflow automation

The use of software to execute a sequence of business steps without manual intervention. Earlier generations were rule-based (Zapier, Make); modern systems combine rule-based execution with AI reasoning at decision points.