## 1. Core Philosophy ### Simple, composable patterns over complex frameworks ### Start with simple prompts; add agentic complexity only when needed ### Match technical complexity to business value ## 2. Workflow Patterns ### Prompt chaining ### Routing (input → specialised handler) ### Parallelization ### Orchestrator-workers (dynamic decomposition) ### Evaluator-optimizer (generator + critic loop) ## 3. When to choose what ### Single-agent vs multi-agent vs workflow-based ### Sequential vs parallel vs evaluator-optimizer ### Decision framework: complexity ↔ business value ## 4. Implementation Guidance ### Context management ### Modular design ### Use of Skills ## 5. Companion Documents ### Effective Context Engineering for AI Agents ### Writing effective tools for AI agents ### Code execution with MCP ### References - https://www.anthropic.com/research/building-effective-agents - https://resources.anthropic.com/building-effective-ai-agents - https://github.com/anthropics/anthropic-cookbook/tree/main/patterns/agents