Skip to content

Part 11

Overview

Theories and patterns are powerful. But to actually build agents, you need frameworks and tools.

This section covers the concrete technologies for building, deploying, and orchestrating agent systems in 2025-2026.

Key Insight: Choose the right framework for your use case. No single framework is best for everything.


Chapter Statistics

Metric Value
Topic Files 20 comprehensive guides
Total Words 29,500+
Code Examples 150+ production-grade
Frameworks Covered 8 major frameworks
Protocols Covered 12 protocols & standards
Integration Patterns 30+ patterns
Comparison Tables 15+ matrices
Case Studies 20+ examples

Frameworks & Technologies Overview

Framework Type Best For Maturity Community
Claude Agents API SDK Production agents Prod Growing
MCP Protocol Infrastructure Tool integration Prod Active
LangGraph Workflow Complex agentic flows Mature Very Active
CrewAI Multi-Agent Team coordination Mature Active
AutoGen Multi-Agent Research & prototyping Mature Very Active
LangChain SDK All-in-one tooling Mature Very Active
Semantic Kernel SDK Enterprise integration Mature Growing
Haystack Framework Search & retrieval Mature Growing

-

Complete Chapter Organization

1. MCP Protocol (2,200 words)

01 Mcp Protocol

  • What is MCP and design philosophy
  • Request/response patterns
  • Tool definition and discovery
  • Resource management
  • Security model
  • Production deployment

2. MCP 2.0 Features (1,900 words)

02 Mcp 2.0 Features

  • Streaming improvements (video, audio, blobs)
  • Enhanced tool definitions
  • Resource protocol v2
  • Sampling methods
  • Performance enhancements

3. Skills System (2,100 words)

03 Skills System

  • Skill architecture overview
  • Registering and discovering skills
  • Skill composition patterns
  • Stateful vs stateless skills
  • Versioning strategies

4. Claude Agents API (2,300 words)

04 Claude Agents Api

  • Agent SDK overview
  • Creating agents programmatically
  • Deployment options
  • Tool and skill integration
  • State management

5. Subagents & Orchestration (2,200 words)

05 Subagents Orchestration

  • Subagent pattern (agent calling agents)
  • Parent-child relationships
  • Result aggregation
  • Error handling
  • Hierarchical agent trees

6. LangGraph Framework (2,100 words)

06 Langgraph Framework

  • StateGraph for state management
  • Tool calling patterns
  • Conditional routing
  • Cycles and loops
  • Streaming support

7. CrewAI Framework (1,900 words)

07 Crewai Framework

  • Role-based agent definition
  • Task assignment and execution
  • Tool integration
  • Memory and learning

8. AutoGen Framework (1,900 words)

08 Autogen Framework

  • Multi-agent conversation
  • Code execution
  • Tool use patterns
  • Research patterns

9. Additional Frameworks (1,500 words)

09 Additional Frameworks

  • LangChain, Semantic Kernel, Haystack
  • Quick reference guide

10. Framework Comparison (2,000 words)

10 Framework Comparison

  • Decision matrix by use case
  • Performance benchmarks
  • Cost analysis
  • Real-world selection examples

11. Structured Outputs (1,800 words)

11 Structured Outputs

  • Reliable function calling
  • JSON Schema validation
  • Claude vs OpenAI implementation
  • Hallucination prevention

12. Prompt Caching (1,600 words)

12 Prompt Caching

  • Cost optimization (90% savings)
  • Cache invalidation
  • Production patterns
  • Real ROI calculations

13. Batch Processing (1,500 words)

13 Batch Processing

  • Asynchronous bulk processing
  • 50% cost reduction
  • Use cases and examples
  • Error handling and retries

14. Vision & Multimodal (1,700 words)

14 Vision Multimodal

  • Image and video understanding
  • Visual tool calling
  • Document processing
  • Multimodal agent patterns

15. Extended Thinking & Reasoning (1,400 words)

15 Extended Thinking

  • OpenAI o1 deep reasoning
  • Cost model changes
  • Hybrid agent patterns
  • When to use o1 vs Claude

16. Agent Protocol (1,600 words)

16 Agent Protocol

  • Agent-to-agent communication (vs MCP)
  • Request/response patterns
  • Network-based coordination
  • Distributed agent systems

17. Function Calling Variations (1,500 words)

17 Function Calling Variations

  • Claude tool_use vs OpenAI function_calling
  • Schema differences
  • Migration paths
  • Provider-agnostic abstractions

18. Streaming Protocol (1,400 words)

18 Streaming Protocol

  • Real-time response streaming
  • Web UI integration
  • Implementation patterns
  • Error handling in streams

19. SDK Specifics (1,400 words)

19 Sdk Specifics

  • Anthropic SDK details
  • OpenAI SDK details
  • Async patterns
  • Error handling differences

20. Message Protocol (1,300 words)

20 Message Protocol

  • Multi-agent message formats
  • Message queuing
  • Request/reply patterns
  • Correlated messaging

-

Learning Paths

Path 1: Production Deployment (5 hours)

  1. Mcp Protocol
  2. Claude Api
  3. Subagents
  4. Comparison

Path 2: Workflow-Based Agents (4 hours)

  1. Langgraph
  2. Mcp 2.0
  3. Comparison

Path 3: Multi-Agent Systems (4 hours)

  1. Crewai
  2. Autogen
  3. Subagents

Quick Decision Tree

Single agent + Simple? → Claude Agents API Single agent + Complex workflow? → LangGraph Multiple agents + Coordinated? → CrewAI Multiple agents + Research? → AutoGen Need tool integration? → MCP Protocol Need reusable skills? → Skills System + Claude API


Key Insight

Choose the simplest framework that solves your problem.

Don't add complexity unless necessary.

-

Last Updated: August 9, 2026 Status: Complete chapter (10 files, 18,000+ words)