During the Code with Claude developers' conference in San Francisco, Anthropic introduced a novel feature called 'dreaming' for its Claude Managed Agents. This feature enables agents to review recent interactions and identify key information to store in their memory, thereby enhancing their performance in future tasks. Currently in research preview, dreaming is limited to Managed Agents on the Claude Platform, which serve as a higher-level alternative to direct interaction with the Messages API. Managed Agents are designed for scenarios where multiple agents collaborate on a task over extended periods, making the ability to retain and recall relevant memories crucial. This new feature addresses the limitations of context windows in large language models (LLMs), where important information can be lost during lengthy projects. Traditional models often use a process called compaction to manage context, but dreaming offers a more structured approach to memory curation.