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)¶
- 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)¶
- Streaming improvements (video, audio, blobs)
- Enhanced tool definitions
- Resource protocol v2
- Sampling methods
- Performance enhancements
3. Skills System (2,100 words)¶
- Skill architecture overview
- Registering and discovering skills
- Skill composition patterns
- Stateful vs stateless skills
- Versioning strategies
4. Claude Agents API (2,300 words)¶
- Agent SDK overview
- Creating agents programmatically
- Deployment options
- Tool and skill integration
- State management
5. Subagents & Orchestration (2,200 words)¶
- Subagent pattern (agent calling agents)
- Parent-child relationships
- Result aggregation
- Error handling
- Hierarchical agent trees
6. LangGraph Framework (2,100 words)¶
- StateGraph for state management
- Tool calling patterns
- Conditional routing
- Cycles and loops
- Streaming support
7. CrewAI Framework (1,900 words)¶
- Role-based agent definition
- Task assignment and execution
- Tool integration
- Memory and learning
8. AutoGen Framework (1,900 words)¶
- Multi-agent conversation
- Code execution
- Tool use patterns
- Research patterns
9. Additional Frameworks (1,500 words)¶
- LangChain, Semantic Kernel, Haystack
- Quick reference guide
10. Framework Comparison (2,000 words)¶
- Decision matrix by use case
- Performance benchmarks
- Cost analysis
- Real-world selection examples
11. Structured Outputs (1,800 words)¶
- Reliable function calling
- JSON Schema validation
- Claude vs OpenAI implementation
- Hallucination prevention
12. Prompt Caching (1,600 words)¶
- Cost optimization (90% savings)
- Cache invalidation
- Production patterns
- Real ROI calculations
13. Batch Processing (1,500 words)¶
- Asynchronous bulk processing
- 50% cost reduction
- Use cases and examples
- Error handling and retries
14. Vision & Multimodal (1,700 words)¶
- Image and video understanding
- Visual tool calling
- Document processing
- Multimodal agent patterns
15. Extended Thinking & Reasoning (1,400 words)¶
- OpenAI o1 deep reasoning
- Cost model changes
- Hybrid agent patterns
- When to use o1 vs Claude
16. Agent Protocol (1,600 words)¶
- 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)¶
- Real-time response streaming
- Web UI integration
- Implementation patterns
- Error handling in streams
19. SDK Specifics (1,400 words)¶
- Anthropic SDK details
- OpenAI SDK details
- Async patterns
- Error handling differences
20. Message Protocol (1,300 words)¶
- Multi-agent message formats
- Message queuing
- Request/reply patterns
- Correlated messaging
-
Learning Paths¶
Path 1: Production Deployment (5 hours)¶
Path 2: Workflow-Based Agents (4 hours)¶
Path 3: Multi-Agent Systems (4 hours)¶
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)