A comprehensive write-up from Addy Osmani's O'Reilly AI CodeCon talk covering patterns for coordinating multiple AI coding agents in real-world software workflows. Covers three main patterns: subagents (focused delegation with parallel execution), Agent Teams (Claude Code's experimental feature with shared task lists, peer
Table of contents
The moment we're inThe 8 levels of AI-assisted codingThe shift: conductor to orchestratorThe single-agent ceilingWhy multi-agent?Pattern 1: Subagents - focused delegationPro-tip: Hierarchical subagents - teams of teamsPattern 2: Agent Teams - true parallel executionAgent Teams: how the mechanics workAgent Teams: key takeawaysPro-tips for Agent Teams: reliabilityPattern 3: Orchestration at scaleTool spotlightsPro-tips for scalingQuality gates: trust but verifyThe bottleneck has shiftedDemo 4: Quality gates and self-improving agentsThe Ralph Loop and self-improving agentsThe discipline: what makes this workThe factory model5 patterns to start todaySort: