Agentic Systems Design
Engineering agentic systems · Volumes 1–3
The broad discipline of engineering the system around a capable model — the environment an agent acts in and the context it reasons over, then the tools it uses, how those are orchestrated, and how the whole is evaluated and operated. Delivered foundation-first across three volumes.
The volumes (Design v1.0 — all complete)
- Volume 1 — Environment & Context — the environment an agent acts in + the context it reasons over.
- Volume 2 — Tools & Orchestration — tool & MCP design, tool minimization, sub-agents, multi-agent, build-vs-buy a harness.
- Volume 3 — Evaluation & Operations — evals & validation, observability, cost, oversight / HITL, security.
Design v1.0 = Volumes 1–3 (28 chapters, all complete). A later problem-first applied volume (v2.0) will re-traverse the material through real problems. (Prompting lives in the Handbook and the Architect's Reference, not a Design volume.)
Chapters
Volume 1 · Environment & Context
- Chapter 1: Agent = Model + Harness
- Chapter 2: Beyond Autocomplete: The Environment & Context Discipline
- Chapter 3: Repo & Doc Design for Agents
- Chapter 4: The Instruction Layer: CLAUDE.md & AGENTS.md
- Chapter 5: Skills & Progressive Disclosure
- Chapter 6: Guardrails, Permissions & Reversibility
- Chapter 7: Environments at Scale: Large Codebases & Monorepos
- Chapter 8: Context Rot: Why Windows Degrade
- Chapter 9: Context Assembly: Engineering the Window
- Chapter 10: Memory: Persisting Context Across Sessions
- Chapter 11: Designing the Whole: Environment + Context as One System
Volume 2 · Tools & Orchestration
- Chapter 12: Beyond One Agent, One Tool
- Chapter 13: Build vs. Buy: Choosing a Harness
- Chapter 14: Tool Minimization: Subtract First
- Chapter 15: MCP: Designing External Capability
- Chapter 16: Shaping Input — The Prompting Craft
- Chapter 17: Shaping Output — Structured & Reliable
- Chapter 18: Sub-Agents: The Context-Isolation Primitive
- Chapter 19: Multi-Agent: Coordinating Many
- Chapter 20: Composing Tools & Orchestration: The Two Axes as One System
Volume 3 · Evaluation & Operations
- Chapter 21: Measuring & Operating Agents: The Discipline
- Chapter 22: Evaluating a Prompt: The Four-Step Loop
- Chapter 23: Evaluating an Agent: Harnesses, Suites & the Judge
- Chapter 24: Observability: Seeing What the Agent Did
- Chapter 25: Cost: The Economics of Running Agents
- Chapter 26: Human-in-the-Loop: Keeping a Human in Control
- Chapter 27: Security: The Adversarial-Input Layer
- Chapter 28: Operating the Whole: Eval + Ops as One Loop
Other pages
- Chapter listing (auto-generated)
- Search (Pagefind)
- References