AI Engineering
Agent Architecture
Patterns for building AI agents that reason, act, and iterate autonomously. Not theoretical — these are the exact patterns running in your Claude Code framework: skills that execute, agents that plan, memory that persists, and guardrails that keep it safe.
2
Minutes
5
Concepts
+45
XP
1
The ReAct Loop

The foundation of tool-use agents. Claude Code runs this loop on every task you give it.

  ┌──────────────────────────────────────┐
  │           User Task / Goal           │
  └──────────────┬───────────────────────┘
                 ▼
  ┌──────────────────────────────────────┐
  │  REASON: Analyze state, plan next    │◄──────┐
  │  step based on goal + context        │       │
  └──────────────┬───────────────────────┘       │
                 ▼                                │
  ┌──────────────────────────────────────┐       │
  │  ACT: Call a tool (Read, Edit, Bash, │       │
  │  Grep, API call, etc.)               │       │
  └──────────────┬───────────────────────┘       │
                 ▼                                │
  ┌──────────────────────────────────────┐       │
  │  OBSERVE: Process tool result,       │───────┘
  │  update understanding                │
  └──────────────┬───────────────────────┘
                 ▼
  ┌──────────────────────────────────────┐
  │  DONE: Goal achieved → respond       │
  └──────────────────────────────────────┘