Anthropic launches 10 AI agents for finance

Anthropic has introduced ten ready-to-use AI agents for banks, funds, insurers and other financial companies. They can prepare client presentations, review credit, analyze compliance documents and help close the books at the end of the month.

The aim is to reduce setup time. Instead of building each system from scratch, teams can install these agents as plugins in Claude Cowork or Claude Code, or use them as managed agents from Anthropic's platform.

Ten agents for specific financial tasks

Each template combines specialized instructions, controlled data access and subagents that handle specific parts of the work. A company can adapt them to its own models, risk policies and approval processes.

The ten agents cover these functions:

• Pitch builder: prepares target lists, comparable-company analyses and drafts of sales presentations.
• Meeting preparer: gathers information about clients and counterparties before a call.
• Earnings reviewer: analyzes transcripts and financial documents, updates models and detects relevant changes.
• Model builder: creates and maintains financial models using reports, data sources and analyst input.
• Market researcher: tracks sectors and issuers, and flags information relevant to credit and risk.
• KYC screener: reviews client and company documents for identification and regulatory compliance processes.
• Valuation reviewer: checks valuations against comparable companies, methods and internal standards.
• General ledger reconciler: reconciles accounting records and calculates net asset value against official records.
• Month-end closer: manages month-end closing tasks, prepares journal entries and generates reports.
• Statement auditor: reviews financial statements for consistency, completeness and audit readiness.

One example shows how these tasks connect. An analyst can give the presentation agent a list of target companies and receive a comparable-company model in Excel, a draft presentation in PowerPoint and a client note in Outlook.

Claude connects Excel, PowerPoint and Word

Claude can also work directly with Excel, PowerPoint and Word through Microsoft 365 add-ins. The Outlook add-in is coming soon. Context is maintained across applications, so users do not have to explain a model again when turning it into a presentation or report.

In practice, Claude can:

• Create financial models and review formulas in Excel.
• Prepare presentations that update when figures change.
• Edit credit reports using a company's templates.
• Analyze email, organize meetings and draft replies in Outlook when the add-in becomes available.

In Claude Cowork, the Dispatch feature lets users assign tasks by text or voice. Claude can keep working with local files while an analyst is away from their desk and leave the result ready for review.

Financial data with controlled access

Agents need more than a language model: they need reliable data and clearly defined permissions. Anthropic is adding connectors so Claude can query sources that professionals already use in real time, without giving it indiscriminate access to all of a company's information.

The new connections include Dun & Bradstreet, Fiscal AI, Financial Modeling Prep, Guidepoint, IBISWorld, SS&C IntraLinks, Third Bridge and Verisk. They cover everything from company and market data to expert interviews, virtual deal rooms and insurance data.

Moody's has also launched an MCP application. MCP is a standard that allows external tools and data to be integrated into Claude. In this case, the application includes credit ratings and data on more than 600 million public and private companies.

The recommended model and the role of people

Anthropic says these functions work best with Claude Opus 4.7, which scored 64.37% on Vals AI's Finance Agent benchmark, a comparative test of financial tasks. This figure was reported by the company itself. It is not a guarantee that every agent will achieve that performance at every company or for every use case.

Managed agents can work for hours on lengthy operations, use tool-specific permissions and keep an auditable record of their decisions and calls. Even so, a person still has to review and approve the result before sending it to a client, filing it or making a decision based on it.

For you, the most relevant change is not that Claude will suddenly replace an analyst, accountant or compliance specialist. It is that tasks that once required copying data between systems, reviewing documents and preparing drafts can now run as supervised workflows. The agents are already available in Anthropic's financial marketplace for paid plans, while managed agents are offered in public beta. The Excel, PowerPoint and Word add-ins are generally available.

What remains to be seen is whether these templates maintain their accuracy when adapted to internal policies, incomplete data and regulated decisions. In finance, automating work only creates value if you can also show what data the agent used, what it did and who approved the result.