Agentic AI
Agentic AI describes AI systems that act with a degree of autonomy to achieve goals: they plan, take actions through tools, adapt to results and work across multiple steps with limited human direction. The term covers single AI agents and systems where several agents cooperate, in contrast to AI that only responds to prompts.
Agentic AI is less a single technology than a design approach. It combines a capable model, often a reasoning model, with tools, memory, planning and feedback loops. Systems range from a single assistant that can browse the web and edit files to multi-agent setups in which one agent plans, others execute and another reviews the work.
Agentic AI vs generative AI: generative AI creates content in response to a prompt; agentic AI uses generative models to decide and act toward a goal. An image generator is generative; a system that researches a topic, writes a report, builds charts and emails it to stakeholders on a schedule is agentic. Most agentic systems are built on generative models.
Why it matters: vendors label many products "agentic", and the term is used loosely. When evaluating a claim, ask what the system can do without a new instruction, which tools and data it can access, where a human must approve, and how its actions are logged and reversed.
Limits: more autonomy means more ways to fail. Reliability over long tasks, security against prompt injection and clear accountability remain the main open problems.
Related terms
Learn it step by step
Sources
- A Survey on Large Language Model based Autonomous Agents. Wang et al., arXiv, August 2023
- ReAct: Synergizing Reasoning and Acting in Language Models. Yao et al., arXiv, October 2022
Topic: AI agents
Last updated