AI Glossary Term

What does “Agentic Loop” mean?

Definition

An iterative cycle where an AI agent observes, plans, acts, and reflects until it completes a goal or hits a stop condition.

Browse the full AI GlossaryExplore all free AI guides

How Agentic Loop works

Diagram for Agentic Loop, a repeating loop: Observe the current state; then Plan the next step; then Act: use a tool; then Reflect: did it work?; then Stop when done or limit reached; then back to the start.Observe the current statePlan the next stepAct: use a toolReflect: did it work?Stop when done or limit reachedrepeat

Example

A coding agent runs a test, sees a failure, edits the code, and runs the test again. It repeats until the tests pass or it reaches a step limit and asks for help.

Why it matters

Loops let agents correct themselves. Without stop conditions they can run too long, cost too much, or repeat the same mistake.

Common questions

What ends the loop?

Success criteria, a step or cost limit, or a person's decision.

Agentic loop vs agentic workflow?

The loop is the repeating cycle at the core; a workflow is the wider multi-step process.

What can go wrong?

Getting stuck in repeats, drifting from the goal, or compounding small errors.

Related terms

Agentic WorkflowA multi-step process where an AI system plans, executes, checks results, and iterates toward a goal.AI AgentA software system that can observe, reason, and take actions to achieve a goal, often using tools and memory.Human-in-the-LoopA workflow where humans review, guide, or override AI outputs.Reinforcement LearningTraining by reward signals where an agent learns actions that maximize long-term return.Chain-of-ThoughtA reasoning style where an AI model decomposes a problem into intermediate steps.ClassificationA task where a model assigns an input to one or more predefined categories.

Learn more in our free guides

Self-Reflection in Agent LoopsAgentic Tool OrchestrationAgentic RAGHuman-in-the-Loop Agents

Test yourself

Self-Reflection in Agent Loops QuizAgentic Tool Orchestration QuizAgentic RAG Quiz

See also

MCP (Model Context Protocol)ReActKV CacheTree of ThoughtsSpeculative DecodingDPO (Direct Preference Optimization)Zero Data RetentionQLoRA