Two Paradigms of Multi-Agent AI: Rust Parallel Agents vs Claude Code Agent Teams
Three multi-agent coordination positions, one codebase. A static Rust/Tokio fan-out assigns 20 agents at compile time with zero coordination overhead. A team.rs library implements the full Claude Code agent-teams model in pure Rust — TaskQueue, Mailbox, PlanGate, ShutdownToken — and the study pipeline now uses it to run a 2-step search→write flow with inter-worker messaging. Claude Code agent teams invert every assumption of static fan-out: dynamic task claiming, file-locked concurrency, full bidirectional messaging. The decision rule is one question: do your agents need to talk to each other? If no, tokio::spawn + Arc<T>. If yes: build team.rs, or use TeamCreate.
Multi-agent AI engineering has become a core discipline in production software development. The interesting question is no longer whether to build multi-agent systems. It is how — and specifically, which architectural pattern to reach for given the nature of the work. The clearest demonstration is that multiple fundamentally different paradigms live inside the same codebase.