Orchestrate subagents at scale with dynamic workflows (code.claude.com)

Claude has introduced dynamic workflows that enable orchestration of multiple subagents on a larger scale, significantly enhancing task execution efficiency. Unlike traditional subagents and skills, which operate sequentially under user prompts, workflows leverage a scripted approach where the coordination and execution of tasks are automated. This allows users to run complex tasks involving dozens to hundreds of agents simultaneously, while maintaining a responsive session, as background processes handle multiple phases of work without interrupting the user's input workflow. The implications of this upgrade are profound for the AI/ML community, streamlining processes like research and data synthesis. Users can now run the built-in '/deep-research' command to conduct comprehensive investigations, fetching and cross-checking sources automatically, thus producing a synthesized report with evaluated claims. Additionally, the ability for Claude to generate and save custom workflows based on user-defined criteria accelerates the automation of repetitive tasks, maximizing productivity and resource management while adhering to operational constraints such as concurrent agent limits and runtime isolation. This positions Claude as a powerful tool for researchers, developers, and businesses seeking to increase efficiency in AI-driven tasks.