DataTerminology.comNEW
Also known as: Agent-based AI, Autonomous AI
AI systems that autonomously plan, decide, and act to achieve goals with limited direct human oversight.
DESCRIPTION
Agentic AI integrates perception, reasoning, planning, and action selection to pursue objectives in dynamic environments. Recent LLM-driven agents use tools, memory, and multi-step planning to accomplish complex tasks. The approach can accelerate operations and enable continuous automation but raises concerns about reliability, safety, and control. Best practices include constrained tool access, human-in-the-loop checkpoints, and auditable decision logs. Evaluation focuses on task completion, robustness, and harmful behavior prevention. Agentic AI contrasts with assistive systems that only provide recommendations.
ALTERNATIVE DEFINITIONS
NVIDIA
Agentic AI refers to AI systems that can plan, reason and act to achieve goals with minimal human intervention.
Stresses planning and acting capabilities in real applications.
Microsoft Learn
AI agents are systems that autonomously perform tasks by interpreting goals, planning steps, and executing actions in an environment.
Frames agentic AI through practical deployment patterns and components.
RELATED TERMS
Software Agent Function Calling Planning Reinforcement Learning Large Language Model Autonomous Systems MLOps AI Safety
COMPETING TERMS
Assistive AI
RELATIONS
- Broader: Agent Software Agent
KNOWLEDGE AREAS
AI MLOps AI Governance Data Engineering
TARGET AUDIENCE
AI Engineer MLOps Engineer AI Product Manager Chief Technology Officer
EXTERNAL SOURCES
- Wikipedia - Software agent
- NVIDIA - Accessible overview of agentic AI
- Microsoft - Azure perspective on AI agents
- arXiv - Survey on LLM-based agents
TOP QUESTIONS
What distinguishes Agentic AI from assistive AI?
Agentic AI plans and acts toward goals autonomously, while assistive AI provides recommendations for humans to execute.
Category: Concepts
What are risks of Agentic AI in production?
Unintended actions, tool misuse, and escalation without oversight. Guardrails and human checkpoints are essential.
Category: Risks
How to evaluate Agentic AI systems?
Measure task completion rates, safety incidents, recovery from failures, and adherence to constraints.
Category: Testing
What components form an Agentic AI system?
Planner, memory, policy for tool-use, execution engine, monitoring, and safety mechanisms.
Category: Example
Where is Agentic AI useful?
Workflow automation, customer support, data operations, and autonomous monitoring/remediation scenarios.
Category: Use cases
How does Agentic AI relate to RL?
Both involve sequential decision-making; agentic systems may use planners or LLM reasoning rather than learned policies alone.
Category: Opinion
How to govern Agentic AI?
Define scope, enforce permissions, log decisions, and regularly audit outcomes against policy and ethics standards.
Category: Governance
What is a hot topic in Agentic AI?
Tool-augmented LLM agents with memory and multi-agent collaboration for complex tasks.
Category: Topic
How to start building Agentic AI?
Prototype with framework agents, restrict tools, add observation logging, and iterate with safety-in-mind.
Category: Getting started