Anthropic announces 'dreaming' feature for AI agents that reviews activity logs to improve performance, as AI companies increasingly name tools after human cognitive processes

Anthropic unveiled a 'dreaming' feature at its developer conference in San Francisco as part of its AI agent infrastructure. The feature sorts through transcripts of an agent's completed work to identify patterns and improve the agent's performance.

The 'dreaming' capability works alongside a 'memory' system to create what Anthropic calls 'a robust memory system for self-improving agents.' Memory allows agents to capture learnings during work, while dreaming refines that memory between sessions and pulls shared learnings across agents.

The naming practice reflects a broader trend: since 2022, AI company leaders have consistently named features after human cognitive processes—OpenAI released 'reasoning' models in 2024 that require 'thinking' time, and multiple startups describe chatbots as having 'memories' about users. A research paper in the AI & Ethics journal warns that anthropomorphism distorts moral judgments about AI and risks users overly trusting tools and projecting qualities onto them that aren't present.