06 - Agentic AI
Level: Beginner to Advanced
Format: Prose + code + Q&A pairs tagged [Easy] [Medium] [Hard]
What You Will Learn
- What an AI agent actually is and how it differs from a chatbot or workflow
- How the agent loop works at the API level - what the LLM sees at every step
- How function calling works, how to design good tools, and how to handle failures
- The four memory types and how to manage context across long-horizon tasks
- Planning and reasoning strategies: ReAct, Plan-and-Execute, Tree of Thoughts, Reflexion
- The 8 architectural patterns and how to choose between them
- The 5 core design patterns used across virtually all production agents
- How multi-agent systems coordinate, communicate, and fail
- The four major agent frameworks (LangGraph, ADK, CrewAI, AutoGen) and when to use each
- Production system design: reliability, security, HITL, cost management
- Trajectory evaluation and observability - how to measure and debug agent behavior
- 40+ curated Q&A pairs covering all domains
Chapter Map
| # | File | Topic | Level |
|---|---|---|---|
| 1 | What Is an AI Agent | Agent vs chatbot, autonomy spectrum, core properties, key vocabulary | ★★☆ Beginner |
| 2 | The Agent Loop | The core loop mechanics, what the LLM sees, ReAct, context growth, termination | ★★☆ Beginner |
| 3 | Tool Use & Function Calling | Function calling API, schema design, tool chaining, parallel calls, security | ★★★ Intermediate |
| 4 | Memory & State | 4 memory types, working memory, checkpointing, semantic memory, context management | ★★★ Intermediate |
| 5 | Planning & Reasoning | CoT, ReAct, Plan-and-Execute, Tree of Thoughts, FLARE, Reflexion, replanning | ★★★ Intermediate |
| 6 | Architectural Patterns | 8 patterns, HITL architectures, feedback loops, hybrid patterns, selection guide | ★★★ Intermediate |
| 7 | Design Patterns | Tool-use, Reflection, Planning, Coordination, Routing - behavior within patterns | ★★★ Intermediate |
| 8 | Multi-Agent Systems | Coordination strategies, communication protocols, state, failure modes, resilience | ★★★ Advanced |
| 9 | Agent Frameworks | LangGraph, ADK, CrewAI, AutoGen - internals, code examples, comparison matrix | ★★★ Intermediate |
| 10 | Agentic System Design | Production architecture, reliability engineering, security, cost, scalability | ★★★★ Advanced |
| 11 | Evaluation & Observability | Trajectory eval, LLM-as-judge, metrics, observability stack, debugging | ★★★ Advanced |
| 12 | Q&A Review Bank | 40 questions across 8 domains - concepts through production engineering | ★★★ |
Recommended Learning Paths
Path A - Full Progression (beginner to interview-ready)
Read chapters in order: 1 → 2 → 3 → 4 → 5 → 6 → 7 → 8 → 9 → 10 → 11, then work through 12-Interview-QA-Bank.md. ~6–8 hours.
Path B - Accelerated (already understand LLM basics)
- What Is an AI Agent - vocabulary (30 min)
- The Agent Loop - the mechanics (45 min)
- Tool Use & Function Calling - building blocks (45 min)
- Architectural Patterns - structure (45 min)
- Agentic System Design - production (60 min)
- Q&A Review Bank - all 40 questions (90 min)
Path C - System Design Focus
- What Is an AI Agent - vocabulary
- The Agent Loop - understand what you're designing
- Agentic System Design - production architecture
- Architectural Patterns - structure options
- Multi-Agent Systems - coordination
Path D - Framework Focus (building something now)
- The Agent Loop - understand what frameworks wrap
- Tool Use & Function Calling - tool design
- Agent Frameworks - pick and use a framework
- Browse Code Labs - Agentic Systems
Path E - Weak spots only
- Don't know what an agent is → 01-Agentic-Concepts.md
- Don't understand how the loop works → 02-The-Agent-Loop.md
- Memory / state confusion → 04-Memory-and-State.md
- Planning strategies → 05-Planning-and-Reasoning.md
- Which framework to use → 09-Agent-Frameworks.md
- Production system design → 10-Agentic-System-Design.md
- Evaluation and debugging → 11-Evaluation-and-Observability.md
Relationship to Other Sections
| Section | What It Covers | Relationship to This Section |
|---|---|---|
| 05 - Agents | Framework-specific implementation (LangChain, LangGraph, CrewAI, ADK) | How to build - this section is how to design |
| [06 - Agentic AI] | System design, patterns, evaluation | System-level view - concepts through production |
| Code Labs - Architecture Patterns | Code implementations of the 8 architectural patterns | Executable versions of the patterns in chapter 6 |
| Code Labs - Agentic Systems | End-to-end system designs in 4 frameworks | Full implementations combining multiple concepts |
| System Designs | Production-grade system designs with GCP mapping | Applied versions of chapter 10 |