Patterns: Agent Design, Multi-Agent, and Anti-Patterns¶
Recurring architectural patterns for building, coordinating, and debugging AI coding agents — plus the anti-patterns that derail them.
Architecture, coordination, and failure-mode patterns organized into three areas: single-agent design, multi-agent collaboration, and common anti-patterns to recognize early.
This page is the entry point for those three sections plus two selection aids below. For content grouped by cross-cutting theme, use the Concept Map; for the core disciplines, start at Foundations; for standards and surrounding context, start at Reference.
Agent Design¶
Architecture, delegation, memory, control, reliability, and harness patterns for building effective agents. Organized into core design, memory and state, control and orchestration, reliability, and harness engineering.
Multi-Agent¶
Coordination patterns for systems where multiple agents collaborate — handoff protocols, fan-out synthesis, orchestrator-worker topologies, and ensemble approaches.
Anti-Patterns¶
Common failure modes in AI-assisted development — from abstraction bloat and context poisoning to the yes-man agent and prompt tinkering. Recognizing these early prevents costly rework.
Pattern Selection Map¶
A trade-off matrix comparing 14 patterns from this site across token cost, latency, frontier-model dependency, blast radius, verification cost, and task class — to help you pick the cheapest pattern that solves your problem.
Agentic Pattern Vocabulary Crosswalk¶
A navigation aid mapping four external agentic-pattern vocabularies — Ng's four, Anthropic's five, Gulli's twenty-one, CoALA's three axes — onto this site's canonical pages, with per-row scope-difference notes.