This blog post discusses the limitations of Claude Code subagents in terms of knowledge retention, highlighting that they spawn fresh instances each time and discard previous discoveries. It also suggests a method for enabling shared memory among these subagents.