AI Agent Levels
Agents and Assistants
AI Agent or AI assistant is a generic word used to describe a system that interacts with an LLM. It takes the input from the user and then generates the output using LLM and the tools available.
ChatGPT, Gemini, etc.
AI agents are the most commonly used name for systems that interact with LLMs.
Both agent and assistant are terms used interchangeably. But there is a very subtle difference between them. To understand this, we need to consider the real world meanings of these two terms.
Always think of assistant as a personal assistant who has limited powers.
Whereas an agent/agency have more independent powers. We only ask an agent to do something on our behalf.
Assistant Messages
All messages generated from an LLM are grouped under the role 'assistant'. Under this, all CoT reasoning, tool calls and results, sub-agent calls are grouped.
Agentic AI
They're systems which takes a specific goal from the user and then performs a series of actions to achieve that goal.
Reasoning -> Planning -> Execution
All agentic AI systems are AI agents, but not all AI agents are agentic AI systems.
AI agent can be any system that interacts with an LLM.
AI agent is reactive whereas agentic AI is proactive.
Example - If we want to book a holiday, the agent will first reason about the flight booking process, then plan the steps to book the flight and then the hotels and then do them accordingly without any human intervention.
Agent autonomous levels
The below picture describes autonomy levels from lowest to highest.
