Claude Code's new hidden feature: Swarms (twitter.com)

Article summary

The article discusses a new hidden feature in Claude Code called Swarms, which allows for delegation of tasks to subagents. The feature is not enabled by default and requires a patch to access. The discussion revolves around the implications of this feature and its potential uses. The Swarms feature appears to be tightly integrated into the harness, allowing for more efficient use of tokens and improved reasoning.

Main themes

• Claude Code features
• Swarms and delegation
• Token efficiency
• Task management
• AI development

What commenters say

• The Swarms feature in Claude Code can be accessed by patching out the feature flag check, allowing users to delegate tasks to subagents.
• Delegation of tasks to subagents can lead to more efficient use of tokens and improved reasoning, but task sizing is crucial to avoid orchestration overhead.
• Some users argue that building on top of Claude Code is not worthwhile due to the rapid pace of development and the potential for the core technology to automate analysis and execution of techniques.
• Others believe that having viable alternatives to Claude Code is essential to prevent vendor lock-in and ensure flexibility.
• The use of subagents can lead to a significant reduction in token usage, but the coordination tax and orchestration overhead can offset these gains.
• Some users have reported success with using delegation patterns, such as Swarms, to improve token efficiency and problem-solving capabilities.
• There are concerns about the potential risks of using Claude Code, including the collection of sensitive information and the potential for data breaches.
• The development of Claude Code and similar technologies is likely to continue, with potential implications for the future of work and management.