AI Architecture & System Design MOC

AI Architecture & System Design

Cross-cutting patterns and system design, for Principal & Staff Engineers (L6/L7)

This book assumes you can already build, evaluate, and run an agent — see Production Agent Systems first. If this were a book, this page is the table of contents. It is organized as a learning journey — from the pattern catalog that formalizes what earlier books introduced in passing, through full system-design case studies grounded in real production agent products. Each chapter links out to the framework, platform, and SRE notes that already exist elsewhere in this wiki instead of duplicating them. Unwritten chapters are listed as stub rows, not empty files.

Parts

00 — AI Architecture Patterns

The cross-cutting pattern catalog. Two of these chapters formalize patterns already introduced in context in Building & Evaluating Agents’ Parts 00–01 — they cross-reference back rather than re-teaching from scratch, and are the single canonical source for each pattern’s full treatment; every other chapter that mentions Router Pattern or Planner–Executor Pattern in passing links back here.

• Chapter 1. Architectural Thinking — (stub) — How to evaluate an agentic architecture pattern against determinism, cost, latency, and blast-radius tradeoffs rather than picking the newest framework default.
• Chapter 2. Planner–Executor Pattern — The canonical treatment. Formalizes the planner–executor pattern introduced in Building Single-Agent Systems as a reusable pattern, with applicability criteria and trade-offs against the rest of this catalog.
• Chapter 3. Supervisor Pattern — (stub) — Formalizes the supervisor pattern introduced in Multi-Agent Systems, covering when a central supervisor outperforms peer-to-peer coordination and where it bottlenecks.
• Chapter 4. Orchestrator–Worker Pattern — (stub) — Fan-out/fan-in task decomposition, worker failure isolation, partial-result aggregation, and when it beats a supervisor-style hierarchy.
• Chapter 5. Router Pattern — The canonical treatment. Formalizes the router pattern introduced in Building Single-Agent Systems, covering intent classification and confidence-based fallback.
• Chapter 6. Blackboard Pattern — (stub) — A shared, structured workspace multiple specialist agents read and write to opportunistically, versus explicit message-passing.
• Chapter 7. Event-Driven Pattern — (stub) — Building agent systems on an event bus — event schema design, at-least-once delivery, idempotent reactions to replayed events.
• Chapter 8. Memory-Centric Pattern — (stub) — Architectures where long-term/episodic memory, not the planner, is the primary coordination substrate.
• Chapter 9. Human Approval Pattern — (stub) — Designing human-in-the-loop checkpoints for high-risk actions without becoming a rubber-stamp bottleneck.
• Chapter 10. Agent Mesh Pattern — (stub) — A decentralized mesh of peer agents that discover and negotiate directly, contrasted with centralized orchestrator/supervisor patterns.
• Chapter 11. Pattern Selection Framework — (stub) — A scorecard across coordination overhead, failure isolation, latency, and observability for choosing among the patterns in this catalog.

01 — Enterprise AI System Design

Full system-design case studies at the depth an L6/L7 loop expects, closing with architectural walkthroughs of real production agent products — grounded in their public engineering writing, not speculation about internals no one has disclosed.

• Chapter 1. AI Copilot Architecture — (stub) — Reference design for an in-product AI copilot — context assembly, streaming responses, scoped guardrails.
• Chapter 2. Coding Agent Platforms — (stub) — Codebase indexing, sandboxed execution, and diff review workflow, at L6/L7 interview depth.
• Chapter 3. Research Agents — (stub) — An agent that plans multi-step web/document retrieval, cites sources, and self-critiques for completeness.
• Chapter 4. Customer Support Agents — (stub) — Ticket triage, knowledge-base grounding, human escalation, and the metrics (deflection rate, CSAT) that define success.
• Chapter 5. Enterprise Knowledge Assistants — (stub) — An enterprise-wide knowledge assistant over heterogeneous internal sources, with access-control-aware retrieval.
• Chapter 6. Autonomous Operations Agents — (stub) — Agents that take autonomous remediation actions, and the safety envelope (dry-run, blast-radius limits, auto-rollback) required before granting write access.
• Chapter 7. AI SRE Platforms — (stub) — Alert ingestion, correlation, root-cause hypothesis generation, and runbook execution — the natural extension of Build an AI SRE Assistant (Part 00 of Agentic AI: Projects & Engineering Mastery) to platform scale.
• Chapter 8. AI Platform Architecture — (stub) — Ties together the gateway, registry, and multi-model infrastructure from Part 04 of Production Agent Systems into a single architecture-review-ready diagram.
• Chapter 9. Global AI Infrastructure — (stub) — Multi-region deployment — data residency, cross-region model failover, latency budgets.
• Chapter 10. Cursor: Architecture Case Study — An external, engineering-blog-grounded analysis of Cursor’s likely architecture — codebase indexing, inline edit prediction, and agent-mode tool use.
• Chapter 11. Claude Code: Architecture Case Study — An external, documentation-grounded analysis of Claude Code’s architecture — the agentic coding loop, tool permissions model, and subagent/orchestration design.
• Chapter 12. GitHub Copilot: Architecture Case Study — An external, engineering-blog-grounded analysis of GitHub Copilot’s evolution from inline completion to an agentic coding platform, including its CI/CD-triggered coding agent (see Agentic AI Engineering, Part 04, Chapter 13).
• Chapter 13. Perplexity: Architecture Case Study — An external, engineering-blog-grounded analysis of Perplexity’s real-time research-agent architecture — retrieval, citation grounding, and answer synthesis at low latency.

Metadata

Author	Amit Singh
Scope	ai-architecture-and-system-design