Anthropic just shipped agent functionality. No more pure chat interface – Claude can now execute multi-step workflows autonomously. This means Claude can chain API calls, manipulate data structures, and handle complex task sequences without constant human prompting. Technically, it's extending their existing tool use framework with persistent state management and decision trees. For devs, this opens up automated code refactoring, data pipeline orchestration, and system integration tasks that previously needed human babysitting. The architecture likely uses their constitutional AI framework to keep agents bounded within safety constraints while maximizing task completion rates. Big shift from reactive assistant to proactive executor.