# Agent Architecture

> Deconstruct the core components of AI agent architecture, including the LLM brain, memory systems, and tool orchestration layers, to identify critical security vulnerabilities.

Canonical: https://codelivly.com/rooms/agent-architecture
Last updated: 2026-10-13

| | |
|---|---|
| Level | Advanced |
| Category | Cybersecurity |
| Lessons | 3 |
| Estimated time | 30 min |
| XP | 100 |

This module breaks down the internal anatomy of modern AI agents. You will explore the distinct roles of the central LLM "brain," the different types of memory (ephemeral scratchpads vs. long-term vector stores), and the tool orchestration layer (e.g., LangChain, AutoGen). Understanding this architecture is crucial for mapping trust boundaries, identifying where state can be poisoned, and recognizing how attackers hijack the agent's reasoning loop.

## Lessons

1. Planner, Executor & Memory Components Explained (30 min)
2. How Agents Decide & Sequence Actions (30 min)
3. Where Agents Store and Recall State (30 min)

## Related

- [Learning path: AI Agent Fundamentals and Trust Boundaries](https://codelivly.com/learning-paths/ai-agent-fundamentals-trust-boundaries)
- [All rooms](https://codelivly.com/rooms)