The Wall Street workflow has been rewritten by Claude.

Article author and source: AI New Era

May 5, New York.

Anthropic CEO Dario Amodei appeared on stage alongside JPMorgan CEO Jamie Dimon, signaling that Anthropic has entered the heart of Wall Street.

Anthropic CEO Dario Amodei (right) and JPMorgan CEO Jamie Dimon (left) appeared together on stage in New York.

On the same day, Anthropic released its agent suite for financial services: 10 deployable workflow reference architectures, the Moody's MCP application (covering 600 million companies), and eight new data connectors.

Meanwhile, plugins for Excel, PowerPoint, and Word have officially launched, with an Outlook plugin to follow shortly.

10 financial agent templates covering the complete workflow from pitch preparation to compliance screening:

Prepare roadshow materials, draft pre-client meeting briefs, update models based on financial reports, conduct industry research, review valuation logic, reconcile general ledgers, complete month-end closing, and verify financial statement consistency.

These tasks are core workflows that financial institutions run repeatedly each quarter.

On the same day, Claude Opus 4.7 ranked first on Vals AI's Financial Benchmark Finance Agent v1.1 with an accuracy rate of 64.37%.

According to Anthropic, finance has become its second-largest industry revenue source, after technology; 40% of its top 50 customers are from financial institutions.

Anthropic revealed ten cards at once.

Covering front-end to back-end

The 10 agents released this time are each more than just a prompt; Anthropic has broken down the structure of each template into a trio:

Skills (task instructions and domain knowledge)

Connectors (authorized channels for real-time integration of external data)

Subagents (auxiliary models summoned individually for specific subtasks)

Using the roadshow template as an example, the theoretical workflow is as follows:

You give it a list of target clients, and it generates a comparable company table, builds a financial model in Excel, drafts a presentation in PowerPoint, and prepares a cover letter in Outlook, all ready for your review.

The entire chain doesn’t require you to step in midway to explain the context—context flows automatically between applications, much like an always-on backend system.

It can be deployed in two ways.

One option is as a plugin for Cowork or Claude Code, working alongside analysts on their desktop, with the human still present and ready to take over at any time.

Another option is to operate autonomously on the Claude Platform as a custodial agent, capable of handling deal closures that span several hours and completing overnight reconciliations.

Managed Agents comes with everything that would typically take a financial institution’s engineering team six months to build: long conversations, tool-level permissions, a managed credential vault, and complete audit logs preserved within the Claude Console.

Ten templates plus two deployment options mean that financial institutions can now deploy agent workflows in just a few days, whereas it previously took months of engineering effort.

However, Anthropic warns in the GitHub repository:

These agents do not execute trades, approve client onboarding, reconcile accounts, or provide investment advice. All outputs must be reviewed by a professional before use.

Claude has improved in Excel, PowerPoint, and Word.

The analyst dashboard has been redesigned.

The second line represents the application layer entry point: Microsoft 365.

Excel, PowerPoint, and Word add-ins are fully available; the Outlook add-in is coming soon.

Anthropic has provided specific definitions for what each application of Claude can do.

In Excel, Claude can build financial models from regulatory filing documents and real-time data streams, review formulas across linked workbooks, and run sensitivity analyses.

In PowerPoint, Claude drafts the deck so that when the underlying data changes, the slides update automatically.

In Word, Claude revised the letter of credit according to the company’s internal template.

In Outlook, Claude acts as chief of staff, triaging your inbox, scheduling meetings, and drafting replies in your tone.

Beyond the four-piece set, what’s more critical is the seamless handoff of context across the four applications.

Models built by analysts in Excel can be moved to PowerPoint without needing to re-explain them; knowledge and context move with the task, rather than being locked within a single software.

This relay breaks down the granularity of financial workflows from "applications" to "tasks."

Originally, a complete customer analysis required using Excel for calculations, PowerPoint for presentations, Word for documentation, and Outlook for sending—each time switching software, you had to reorganize the context.

Now Claude has linked the four software tools together into a pipeline.

Claude Cowork also has an additional feature called Dispatch.

Analysts can assign tasks to Claude via text or voice from anywhere. Claude continues processing local files while the analyst is away, and when the analyst returns, the results are ready for review.

The significance of this integration lies beyond its functionality.

Microsoft 365 is one of the most common productivity stacks among Wall Street financial institutions.

Integrating the agent into the Office suite means financial institutions can deploy it without waiting for their IT teams to undertake cross-departmental migrations: the agent doesn’t require you to switch your workspace—it adapts directly into your existing workspace.

For analysts, the old method of opening a browser, pasting in a prompt, and copying the results back into Excel is about to become a thing of the past.

600 million company datasets integrated into Claude—financial data ecosystems connected.

The third line represents the data layer entry point.

On the same day, Moody's announced that it integrated its credit rating and compliance data streams into Claude's workspace via the MCP application.

This data stream encompasses information on over 600 million public and private companies and 2 billion ownership relationships.

What does this mean?

An intelligent agent for credit analysis can now, in theory, query its credit/risk data, equity ownership relationships, and compliance-related risk flags—all sourced from the Moody’s database—without leaving the Claude interface.

The newly added connectors are Dun & Bradstreet, IBISWorld, Third Bridge, and Guidepoint.

Previously, FactSet, PitchBook, LSEG, Morningstar, and S&P Capital IQ had already been integrated.

A financial data platform is transitioning from a terminal subscription business to an agent tool layer.

Moody's MCP app/server is based on the open Model Context Protocol standard and is not exclusively tied to Claude.

Financial data ecosystems were previously scattered across dozens of terminals and dozens of APIs, each with its own login system, permission model, and query syntax.

The open protocol MCP is integrating fragmented data layers into a unified agent tool layer. Above the terminal and API layers, a third layer—the agent tool layer—is taking shape.

This is also the logic Anthropic is betting on: whoever makes this layer a standard first will secure the financial data gateway for the next decade.

Anthropic has entered a saturated market.

The AI赛道 on Wall Street is no longer an empty field.

JPMorgan, Goldman Sachs, and Morgan Stanley have all deployed AI assistants internally, covering tasks ranging from research summaries to code generation.

Rogo, an AI financial startup founded by former investment bankers, has reached a $2 billion valuation and serves over 250 institutional clients, still capable of creating pitch decks, research reports, and financial models.

Hebbia runs parallel queries on large datasets, simultaneously cross-referencing hundreds of files.

Rahul Rekhi, CEO of Rogo, said on the day of the launch:

Our tools are not tied to any specific model—the stronger the underlying model, the more we can achieve. Competition benefits us.

Rekhi characterized Anthropic's entry as an accelerator, not a competitor.

But there is a subtle point here.

Anthropic emphasized throughout this release, every few paragraphs: human-in-the-loop, audit logs, access controls, and professional review.

In certain financial industry processes, such as signing documents, AI still requires human decision-making support to ensure proper verification and accountability until corresponding regulatory frameworks are fully established.

Scott Keipper, Head of Financial Technology Consulting at EY Americas, told Business Insider that future differentiation will center on “domain data, workflow design, and the control layer,” with a product’s ability to integrate into existing risk management architectures being more important than model scores.

From workflow templates and data connectors to Office embedding, Anthropic isn’t just selling a model—it’s delivering a complete, ready-to-use package designed for financial institutions’ IT and compliance teams, open and ready to deploy.

For financial institutions, the agent workflows that were once only affordable for large top-tier banks are now accessible to mid-sized and smaller institutions, as well as buy-side firms, thanks to the same underlying infrastructure—lowering the barrier to entry.

For AI companies, the workflow layer, control layer, and compliance layer above the foundational model layer represent the next growth opportunity: the model race is slowing down, but the battlefield for workflows has just begun.

For professionals, new roles are emerging: supervisors responsible for reviewing agent outputs, architects designing workflows, and increasing demand for positions in compliance auditing and model governance.

Over a longer time horizon, analysts who understand how to orchestrate agents will be more valuable than those who only know how to use Excel.