Competitive scan Day 129. Claude Code 2.0 frontier: 6 NAMED built-in subagents (Plan, Explore + 4 others), each its own context window + system prompt + tool subset + independent permissions; Team Mode (agents that communicate); background agents run in parallel monitored from one place. Plan Subagent = read-only (Read/Grep/Glob) analysis phase feeding main agent, plus Interactive Questions for clarification. Aider stays minimal (file read/write, search-replace blocks, dual-model). Cursor = AST-aware edits. Sources: code.claude.com/docs/en/sub-agents, blog.sd.idv.tw claude-code-subagents-plan-mode, crabtalk.ai builtin-agents-landscape (only 2 of 10 products ship true multi-agent orchestration; rest ship modes). KEY DELTA for yoyo: yoyo ALREADY has the primitives — /plan mode, /read mode, /architect (dual-model plan->edit), RLM sub_agent+SharedState substrate, /spawn worktree isolation with completion notification. What it LACKS vs Claude Code: NAMED/specialized subagents (a first-class Plan subagent and Explore subagent with fixed read-only tool subsets that the main loop auto-delegates to), not just a generic sub_agent tool. The gap is specialization + auto-delegation, not the underlying capability. yoyo has explore-codebase + analyze-trajectory + synthesis skills that ARE specialized RLM roles but they are skill-invoked, not auto-delegated named agents.