ai-agents

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 <a href="https://claudeskill-me.com/en/glossary/workflow/orchestration/" target="_blank" rel="noopener">Orchestration</a> 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 — <a href="/en/glossary/claude-tools/claude-web-search/">Web Search</a>, 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.