Agent Fundamentals
Eight conceptual playbooks covering the core disciplines Anthropic identifies for building complete AI agents — identity, context, retrieval, memory, tools, harnesses, skills, and evaluation.
Already shipping production architectures? View the advanced playbooks
System Prompt and Identity Engineering
Defining who an agent is before deciding what it can do.
Context Engineering
Treating the context window as a finite, costly resource — not an inbox.
Retrieval-Augmented Generation
Letting an agent fetch knowledge on demand instead of carrying it all in the prompt.
Agent Memory Architecture
Giving an agent continuity across turns, sessions, and time.
Tool Use and Integrations
How an agent acts on the world, and why tool design is a context budget problem.
Harness Engineering
The infrastructure layer around the model that turns a capable LLM into a reliable production agent.
Agent Skills
Reusable, on-demand procedures that let an agent perform specialized tasks well without bloating its core context.
Evaluation, Safety, and Observability
How you find out an agent is actually working — and catch it when it isn't.