Glossary · agentic-workflow

Agentic workflow

Definition

A development pattern where an AI agent drives a multi-step task end-to-end (planning, file edits, terminal commands, test runs, and revisions) with the human approving privileged actions and steering high-level direction. Distinct from one-shot prompts: agentic workflows run for minutes to days and produce diffs, PRs, or deployed changes rather than chat replies.

Context

Agentic workflows generalize the loop that Cursor's Composer, Claude Code, Codex, and OpenCode all implement: the agent reads code, plans, executes, observes results, and iterates. The workflow's productivity is bounded by where the agent runs. On a laptop, the workflow stops when the laptop sleeps. On an agent workstation, the same workflow runs through to completion. The human checks in when an approval is needed, and the agent keeps moving otherwise.

Also known as

agent loopagent-driven workflowautonomous agent taskagent task loop

Related terms

See also