Note_Tech

All technological notes.

---

Project maintained by simonangel-fong Hosted on GitHub Pages — Theme by mattgraham

AI Agent & Agentic AI

Back

• AI Agent \& Agentic AI
  • AI Agent
    • Types of AI Agents
  • Agentic AI
    • How Agentic AI Works
  • AI Agent vs Agentic AI

AI Agent

• AI agent
  • an autonomous software program that uses artificial intelligence(Large Language Models (LLMs)) to achieve specific goals by reasoning, planning, and using external tools.
• Key Characteristics
  • Goal-driven
  • Can use tools (APIs, databases, CLI)
  • Semi-autonomous
  • Often powered by LLM + tools + memory

---

Types of AI Agents

• Reactive Agent
  • No memory
  • Rule-based
• Tool-Using Agent
  • Calls APIs/tools
  • Most common today
• Planning Agent
  • Breaks tasks into steps
  • Uses reasoning
• Multi-Agent System
  • Multiple agents collaborate
  • Each has a role

---

Agentic AI

• Agentic AI
  • the autonomous systems that act as agents to achieve specific goals with limited human supervision, moving beyond passive, reactive AI to proactive, goal-driven action.
  • These systems leverage Large Language Models (LLMs) to reason, plan, take independent actions, and adapt to new situations to complete complex, multi-step tasks rather than just generating content.
• Agentic AI is a paradigm where AI systems:
  • Plan tasks autonomously
  • Make decisions
  • Execute multi-step workflows
  • Iterate until goal is achieved
• Core Capabilities
  • Task decomposition
  • Decision-making
  • Tool usage
  • Memory/context retention
  • Self-correction

---

How Agentic AI Works

1. Goal
   • e.g., “Fix CI/CD failure”
2. Perception
   • Read logs, configs, metrics
3. Planning
   • Break into steps
4. Action
   • Call tools (API, CLI, DB)
5. Reflection
   • Evaluate result
6. Repeat
   • Until goal achieved

---

AI Agent vs Agentic AI

• AI Agent = one worker
• Agentic AI = system of workers

Aspect	AI Agent	Agentic AI
Scope	Single task	Multi-step system
Complexity	Low–Medium	High
Behavior	Reactive + some planning	Autonomous planning
Architecture	Single component	Multi-agent orchestration
Example	Log analyzer	Full CI/CD assistant

---