Anthropic introduced Dreaming for Claude Managed Agents, a self-improvement loop where agents review past sessions, surface missed patterns, and store insights in long-term memory.

Two tools moved from research preview to public beta: one uses rubric-style outcomes to guide agents, and the other delegates work to multiple sub-agents.

Dreaming operates as a scheduled process that curates specific memories from sessions and memory stores for future use.

At a separate New York event, Anthropic introduced 10 financially focused AI agents, signaling finance as a large growth driver and enterprise revenue source.

Early adopters from legal, tech, and content teams report improved task success and workflow efficiency, underscoring the product’s emphasis on reliable, controllable AI and self-improving automation.

The company announced a major expansion of usage limits and new infrastructure deals to support demand, including a SpaceX data center contract for substantial GPU capacity and partnerships with Amazon, Google, Broadcom, Microsoft, and NVIDIA.

Claude Code-driven growth propelled an 80x annualized rate in Q1, creating significant compute capacity challenges and supply constraints.

Anthropic’s moves come amid industry-wide expectations that AI will disrupt legacy software and services, influencing investor sentiment and SaaS stock performance.

In consumer terms, Anthropic is increasing five-hour usage limits for Pro and Max subscribers to better align with demand for compute infrastructure.

Co-founder Jack Clark has suggested frontier AI models may autonomously train successors by 2028, signaling a broader push toward agentic, self-managing AI systems.

The feature aligns with Anthropic’s push to win enterprise business, following growing popularity of its AI coding agent and a recent enterprise-focused event.

Anthropic emphasizes branding that anthropomorphizes capabilities, continuing a pattern of naming and framing AI features in human-like terms.