Anthropic Launches Financial AI Agents and Self-Improving 'Dreams' Feature

Anthropic launched ten pre-built AI agent templates in New York designed to automate high-volume workflows for banks, insurers, and asset managers. Powered by Claude Opus 4.7, these agents handle tasks such as drafting credit memos, reviewing financial statements, and building pitch decks. To support this enterprise expansion, the company integrated Claude across Microsoft 365 applications and formed a $1.5 billion joint venture with Blackstone, Goldman Sachs, and Hellman & Friedman.

Strategic partnerships further extend the agents' reach, including an agreement with Moody's and Dun & Bradstreet for data integration. Additionally, Anthropic partnered with Fidelity National Information Services (FIS) to develop a specialized agent for anti-money laundering investigations. These moves come as the company reports annualized revenue of $30 billion and prepares for a potential initial public offering this year, despite a recent Pentagon designation of the firm as a supply chain risk.

Simultaneously, at the Code with Claude developers' conference in San Francisco, the company introduced 'Dreams,' a feature allowing agents to asynchronously review past sessions to identify patterns and curate memories. This is paired with an 'Outcomes' feature that enables agents to evaluate their work against rubrics and retry failed tasks. These tools aim to create self-improving agents capable of higher task completion rates in complex professional environments.