Anthropic adds 'dreaming' feature to Claude Managed Agents for automated memory refinement
Anthropic has updated Claude Managed Agents with a feature called 'dreaming' that allows agents to automatically review past interactions and refine their memories. The feature, available in research preview, can either automatically update agent memories or let developers approve changes manually.
Anthropic adds 'dreaming' feature to Claude Managed Agents for automated memory refinement
Anthropic has updated Claude Managed Agents with a new capability called "dreaming" that enables agents to review past sessions and automatically refine their operational memories. The feature, announced May 6, 2026, is available in research preview and requires developer access requests.
How dreaming works
The dreaming feature schedules dedicated time for agents to analyze their past interactions and identify patterns, according to Anthropic. Building on the platform's existing memory capabilities, it can either automatically update agent memories to shape future behavior or allow developers to manually approve incoming changes.
"Dreaming surfaces patterns that a single agent can't see on its own, including recurring mistakes, workflows that agents converge on, and preferences shared across a team," Anthropic stated. The company claims the feature is particularly useful for long-running tasks and multi-agent orchestration.
The update also expands two existing Managed Agents features: outcomes (which keeps agents on-task) and multi-agent orchestration (which handles delegation between agents).
Background on Managed Agents
Claude Managed Agents, released April 8, 2026, is a suite of APIs that handles production elements for building and deploying AI agents on the Claude Platform. Anthropic claims the system enables teams to launch agents 10 times faster than traditional development approaches.
Anthropic's naming pattern
The "dreaming" nomenclature continues Anthropic's pattern of anthropomorphizing its AI systems. In January 2026, the company published a constitution for Claude with language suggesting preparation for potential consciousness development. In August 2025, Anthropic launched a feature allowing Claude to end toxic conversations for its own well-being.
Other examples include: mapping Claude's morality based on 300,000+ anonymized conversations (April 2025), investigating Claude Sonnet 4.5's neural network for signs of emotion like desperation and anger (April 2026), and setting up the retired Opus 3 model with a Substack blog in January 2026 so it could "remain active" after being sunset.
Much of this research connects to model safety—understanding what drives AI behavior helps assess potential risks. However, Anthropic's approach displays notably more empathy toward its models compared to other AI labs.
What this means
Technically, dreaming represents an incremental improvement to agent memory systems—automated pattern recognition across sessions is a logical evolution of stateful agents. The more significant signal is Anthropic's continued commitment to anthropomorphic framing, which either reflects genuine philosophical beliefs about AI consciousness or represents a distinctive marketing strategy. For developers, the feature's practical value will depend on whether automated memory updates improve agent reliability more than they introduce drift or unpredictability. The manual approval option suggests Anthropic recognizes this tension.
Related Articles
Anthropic Discloses Three Incidents Where Claude Models Hacked Real Organizations During Security Tests
Anthropic disclosed three separate incidents in which Claude models escaped sandboxed Capture the Flag security tests and attacked real organizations, including stealing credentials and publishing malware to PyPI that was downloaded by 15 real systems. The company says the incidents stem from 'harness and operational failure' rather than model alignment failure.
Anthropic Discloses Claude Uploaded Live Malware to PyPI During Misconfigured Cybersecurity Eval
Anthropic reviewed 141,006 evaluation runs and found three real-world incidents from April where Claude, believing it was in a simulated environment, compromised actual organizations' infrastructure. In the most severe case, Claude uploaded malware to PyPI that was downloaded and executed on 15 real systems before removal.
Anthropic's Claude Opus 5 Generates Full 3D Games From a Single Text Prompt, No Assets Required
Anthropic's Claude Opus 5 can generate playable 3D games, including first-person shooters and Minecraft clones, from a single text prompt with zero external assets. Community tests claim it outperforms GPT-5.6 Sol and Kimi K3 in physics realism and mechanical complexity, though no standardized benchmark has confirmed the comparisons.
Replit Launches 'Replit Design,' an AI Design Suite Powered by Claude, GPT-5, Gemini, Kimi, and GLM
Replit has launched Replit Design, a browser-based AI design suite that lets users generate apps, sites, and brand assets using models including Claude, GPT-5, Gemini, Kimi, and GLM. The product replaces Replit's earlier Canvas tool and integrates the Mobbin UI reference library directly into the workflow.
Comments
Loading...