Anthropic keeps Claude Code Agent Teams experimental
Nine months after its launch as a research preview alongside Claude Opus 4.6, Claude Code's "Agent Teams" feature remains experimental and off by default. The orchestration tool allows a team lead to coordinate independent teammates with their own context windows using a shared task list and JSON-based mailboxes. The author examines its practical use cases in parallel review and adversarial debugging, while noting its token cost tradeoffs and persistent limitations around session resumption. Open article →