Anthropic ships 10 finance agents

Finance AI has mostly been sold as a clever assistant. Anthropic is now trying to sell it as labor. On May 5, the company rolled out 10 ready-made agent templates for banks, insurers, and other financial firms — not for brainstorming, but for concrete jobs like building pitchbooks, screening KYC files, writing credit memos, and helping close the books. The bet is simple: Wall Street does not really need a free-ranging bot. It needs software that can do expensive, repetitive work inside rules-heavy systems. (anthropic.com) ### What actually shipped? Anthropic’s release is a bundle, not a single model upgrade. It includes 10 agent templates for financial-services tasks, packaged for Claude Cowork and Claude Code, plus cookbook-style setups for Claude Managed Agents so firms can deploy them faster. Anthropic framed these as “ready-to-run” patterns — basically prebuilt workflows wit(anthropic.com)r. (anthropic.com) ### Why target finance first? Because finance is full of high-cost drudge work with clear formats and high willingness to pay. Pitchbooks, diligence memos, sanctions checks, reconciliation, and internal reporting all eat analyst time, but they also follow repeatable steps. That makes them a better fit for bounded automation than open-ended consumer chat. Reuter(anthropic.com)vices, while Bloomberg and Business Insider both described it as a direct move for Wall Street workloads. (money.usnews.com) ### Why do the Microsoft hooks matter? Because a lot of the work already lives in Microsoft software. Anthropic said Claude now works more deeply across Excel, PowerPoint, and Outlook. That sounds boring, but it is the whole point — if an agent can read the spreadsheet, dra(money.usnews.com) They want the existing stack to do more. (bloomberg.com) ### What about the data side? This is where the pitch gets more serious. Anthropic paired the agents with connectors and partner data, including Moody’s and Dun & Bradstreet. One outside report said Moody’s data covers more than 600 million public and private companies, which gives Claude som(bloomberg.com)ible” and “came from an approved data source” is huge. (bloomberg.com) ### So are these autonomous agents? Not in the sci-fi sense. Anthropic’s own framing is much narrower. Each template is built around governed access, specific tools, and subagents for limited tasks. That matters because banks and insurers need audit trails, permissions, and predictable behavior. The product is less “AI employee” and more “workflow software with a very flexible reasoning layer.” (anthropic.com) ### Why mention FIS and financial crime? Because Anthropic is also trying to prove this can plug into real banking infrastructure, not just analyst desktops. FIS said on May 4 that it is working with Anthropic starting with a Financial Crimes AI Agent, with a roadmap that stretches into onboarding, fraud prevention, and credit decisioning. That gives Anthropic (anthropic.com)ch bigger prize. (fisglobal.com) ### What is the real competitive angle? Anthropic is moving away from “our model is smart” and toward “our system fits your industry.” That is where enterprise AI is heading. The winning product may not be the most general model. It may be the one that comes pre-wrapped in controls, connectors, and domain data that legal, compliance, and IT teams will actually approve. (money.usnews.com) ### Bottom line? Anthropic is trying to turn Claude into finance software, not just an assistant. If that works, the first jobs AI changes on Wall Street will be the repetitive, document-heavy ones that firms already know how to measure. (money.usnews.com)