Agentic workflow
Also known as: agentic automation, agent-driven workflow
Traditional automation follows rules: if X happens, do Y. Agentic workflows are different. The AI agent reads the situation, forms a plan, takes an action, checks the result, and decides what comes next. It can branch, retry, or change direction mid-task without a human re-prompting it.
The distinction matters in practice. A rule-based workflow that hits an unexpected response will break or stall. An agentic workflow can reason through the ambiguity, try a different tool, or ask for clarification. That flexibility is what makes it useful for real-world tasks that don't follow a clean, predictable path.
Builders encounter this term when setting up automations in tools like n8n, designing a multi-step agent in LangGraph, or describing what Claude Code is doing when it works through a coding task. The word 'agentic' signals that the workflow has some degree of autonomous decision-making baked in, not just a chain of hardcoded steps.