The core operating pattern of AI Agents: perceive environment → decide → execute action → observe results → decide again, cycling until the task is complete. The key mechanism that upgrades AI from "one question, one answer" to "autonomously completing multi-step tasks."

A task process executed autonomously by AI Agents, involving multiple steps and tool calls. Unlike one-shot Q&A, Agentic Workflows let Claude continuously sense environmental state, plan next actions, execute operations (search, write files, call APIs), observe results, then decide the next step — until the task completes or a stopping condition is reached. The core architecture for upgrading Claude from 'assistant that answers questions' to 'executor that can proactively complete work.'

Agent Orchestration is an architecture in which one orchestrating agent breaks a complex goal into sub-tasks and delegates each to specialized worker agents, then integrates all outputs into a complete result. An analogy: you want a full market research report; the <a href="https://aiagent-bible.com/en/glossary/multi-agent/orchestrator/" target="_blank">Orchestrator</a> plans the steps, sends one agent to research competitors online, another to draft the analysis, and a third to review — while you only communicate the goal to the orchestrator.

An architecture where multiple AI Agents collaborate to complete complex tasks — different Agents handle different subtasks, coordinating through message passing. Compared to single Agents, multi-agent systems can handle tasks exceeding a single Agent's context limits, execute multiple independent subtasks in parallel, and improve overall quality through specialization. One of the current frontier directions in <a href="/en/glossary/core-concepts/ai-agent/">AI Agent</a> research and applications.

The mechanism that lets Claude proactively call external functions during conversation — web search, database queries, code execution, file system access. Developers define "tools" (name, description, parameter format) in API requests; Claude decides when to use which tool and with what parameters; results return and Claude continues generating a response. The core foundation of <a href="/en/glossary/core-concepts/ai-agent/">AI Agent</a> capability.