Rohan Paul@rohanpaul_ai · 5月6日70Anthropic just shipped 10 finance-agent templates that turn Claude from a chat assistant into a controlled worker for pitchbooks, KYC reviews, month-end close, valuations, earnings analysis, and financial modeling.
The shift is from asking an LLM questions to giving it a repeatable finance job with skills, connectors, and subagents already packaged around the task.
Skills tell the model how the firm wants the work done, connectors give it governed access to data, and subagents split the job into smaller checks like comparables selection, methodology review, or source-document screening.
The Microsoft 365 add-ins is a strong feature because finance work rarely lives in one place: a model starts in Excel, becomes a deck in PowerPoint, gets explained in Word, and lands in Outlook.
Claude can now carry context across those apps, so the analyst does not need to restate the deal, company, assumptions, and prior work every time the format changes.
So its a Claude Managed Agents, where the same templates can run longer jobs with permission controls, credential vaults, and audit logs for regulated teams.
The most serious claim is performance: Anthropic says Claude Opus 4.7 leads Vals AI’s Finance Agent benchmark at 64.37%.
译Anthropic发布了10款面向金融服务的Claude智能体模板,使其从聊天助手转变为可执行募资书制作、KYC审查、月度结算、估值分析等标准化任务的工作流。其核心是通过预置“技能”规范操作,利用“连接器”安全访问数据,并通过“子智能体”拆分任务步骤。该功能支持在Microsoft 365应用间保持上下文连贯,无需重复输入。同时提供具备权限控制与审计日志的托管代理模式。据称,Claude Opus 4.7在Vals AI金融智能体基准测试中以64.37%领先。模板可安装于Cowork和Claude Code,或部署为生产级托管代理。