AI Agents · TRACK 05/05
Multi-Agent Systems & Computer Use
Teams of agents, subagents, and agents that drive a screen — safely.
// THE TRACK
01 · START HEREMulti-Agent SystemsLearn what changes when several agents share one job — the benefits, the overhead, and the cases where a single agent still wins.BEGINNERComputer Use AgentsUnderstand how AI agents that control real screens work, how leading systems compare, and the sandboxing practices that keep them safe.INTERMEDIATEOrchestrator-Worker PatternSee how a lead agent plans and delegates to parallel workers, and how this architecture compares with pipelines, swarms, and debate.INTERMEDIATESubagentsUnderstand subagents — short-lived helper agents launched for one subtask with their own clean context — and how they keep the main agent focused.INTERMEDIATEMulti-Agent Failure ModesLearn the failure modes that make multi-agent systems worse than a single agent — fragmented context, coordination overhead, and conflicting actions.INTERMEDIATEComputer-Use SafetyLearn the practical guardrails for letting an agent control a real screen safely — sandboxing, scoped permissions, and human-in-the-loop approval.INTERMEDIATEMulti-Agent Debate & VotingUnderstand how having multiple agents debate, critique, and vote can improve answer quality — and when the extra cost isn't worth it.INTERMEDIATEA2A ProtocolYou will understand what the A2A protocol standardizes, how agents from different vendors discover and talk to each other through it, and how it differs from a tool protocol like MCP.INTERMEDIATE