# Agentic Workflow A pattern where an AI model operates in a loop rather than producing a single response: it receives a task, breaks it into steps, executes each step (often using [[Tool Use]]), observes the results, adjusts its plan, and continues until the task is complete or it decides it needs human input. The model acts as an autonomous problem-solver rather than a one-shot text generator. ## The buzzword and the substance "Agentic" became the defining industry buzzword of the 2024–2025 AI cycle, attached to everything from email assistants to autonomous coding tools. Like most AI marketing vocabulary, it describes something real while dramatically overstating how broadly it applies. The real phenomenon: models operating in multi-step loops with tool access *do* accomplish things single-shot prompts cannot — debugging code across multiple files, researching topics across multiple sources, completing multi-step workflows. The marketing inflation: calling any chatbot with a plugin "agentic." The technical architecture typically follows one of several patterns: ReAct (interleaving reasoning and acting), Planning (decomposing tasks upfront), or Orchestration (a central model coordinating specialized sub-agents). All share the core loop structure: perceive → reason → act → observe → repeat. The governance challenge is that agentic workflows operate at a speed and scale that makes meaningful [[Human in the Loop]] difficult. A model that takes 50 actions to complete a task cannot realistically have each action reviewed by a human. The "agentic" shift is fundamentally a shift toward autonomy — which is precisely why the word carries both excitement and anxiety depending on who's using it. ## See Also - [[AI Agent]] — the entity that executes agentic workflows - [[Tool Use]] — the capability agentic workflows depend on - [[Human in the Loop]] — the oversight mechanism agentic workflows strain