Anthropic adds self-improving “Dreaming” to Claude agents

Dreaming adds self-improvement directly into Claude’s agent behavior, which is a different model than static task execution. For video workflows, that matters because agent systems are increasingly being used to automate operational work, and this feature points to a more autonomous class of tooling. The article does not name competitors or deployment plans, so the key signal is how Anthropic frames and productizes this capability. Watch for whether Anthropic provides concrete examples of Dreaming in Claude, beyond the feature announcement itself.