# Ai Architecture And System Design
All Ai Architecture And System Design notes →1. Architectural Thinking
Introduces the pattern-catalog framing for the rest of this part: how to evaluate an agentic architecture pattern against determinism, cost, latency, and blast-radius trade-offs rather than picking the newest framework default.
2. Planner–Executor Pattern
Formalizes the planner–executor pattern — a planning component that decomposes a goal into a full upfront plan, and a separate executor that carries out each step — with applicability criteria, concrete failure modes, and how it composes with the rest of the pattern catalog.
3. Supervisor Pattern
Formalizes the supervisor pattern introduced in Multi-Agent Systems as a reusable architectural pattern, covering when a central supervisor agent outperforms peer-to-peer coordination and where it becomes a bottleneck.
4. Orchestrator–Worker Pattern
Covers the orchestrator-worker pattern for fan-out/fan-in task decomposition, including worker failure isolation, partial-result aggregation, and when it is preferable to a supervisor-style hierarchy.
5. Router Pattern
The canonical treatment of the router pattern -- classifying an incoming request and dispatching it to exactly one specialized handler, tool, or sub-agent -- covering the three real ways to build the classification step, confidence-based fallback, and the structural line that separates a router from a supervisor.
6. Blackboard Pattern
Covers the blackboard architecture - a shared, structured workspace multiple specialist agents read and write to opportunistically - and where it beats explicit message-passing for loosely-coupled multi-agent collaboration.
7. Event-Driven Pattern
Covers building agent systems on an event bus rather than direct request/response calls, including event schema design, at-least-once delivery handling, and idempotent agent reactions to replayed events.
8. Memory-Centric Pattern
Covers architectures where long-term and episodic memory (not the planner) is the primary coordination substrate for agent behavior, including memory write policies and staleness/consistency trade-offs.
9. Human Approval Pattern
Covers designing human-in-the-loop checkpoints for high-risk agent actions - approval gates, timeout/escalation policy, and how to keep the pattern from becoming a rubber-stamp bottleneck.
10. Agent Mesh Pattern
Covers a decentralized mesh of peer agents that discover and negotiate with each other directly, contrasted with the centralized orchestrator/supervisor patterns earlier in this catalog, and the discovery/trust problems a mesh introduces.
11. Pattern Selection Framework
Closes the catalog with a decision framework - a scorecard across coordination overhead, failure isolation, latency, and observability - for choosing among the patterns covered in this part for a given problem shape.