Anthropic Deploys 10 Finance Agents for Wall Street
Anthropic releases 10 ready-to-run AI agent templates for finance, now live at JPMorgan, Goldman Sachs, Citadel, AIG, and nine more major institutions.
6 min read
Ten production-ready AI agents. Thirteen confirmed bank deployments. A first-ever shared stage with Jamie Dimon. Anthropic's push into financial services moved from strategic to structural today.
The company released 10 agent templates organized across two tracks - research and client coverage on one side, finance and operations on the other. Not prompts to copy-paste. Deployable reference architectures with native connectors to the data terminals analysts already use, wired into Microsoft 365 starting today.
TL;DR
- 10 agent templates released for investment banking, asset management, and insurance
- Live at JPMorgan, Goldman Sachs, Citi, AIG, Visa, Citadel, Carlyle, Mizuho, BNY, and others
- Microsoft 365 integration covers Excel, PowerPoint, and Word - Outlook is next
- New connectors from Moody's (600M+ companies), Dun & Bradstreet, IBISWorld, and eight others
- Claude Opus 4.7 leads the Finance Agent benchmark at 64.4%
The 10 Agents
| Category | Agent | What It Automates |
|---|---|---|
| Research & Client Coverage | Pitch builder | Target lists, comps, pitchbook drafts |
| Research & Client Coverage | Meeting preparer | Client and counterparty briefs |
| Research & Client Coverage | Earnings reviewer | Transcript and filing analysis, model updates |
| Research & Client Coverage | Model builder | Financial models from filings and data feeds |
| Research & Client Coverage | Market researcher | Sector tracking, research synthesis |
| Finance & Operations | Valuation reviewer | Comps validation and standards checks |
| Finance & Operations | General ledger reconciler | Account reconciliation and NAV calculation |
| Finance & Operations | Month-end closer | Close checklist and journal entry preparation |
| Finance & Operations | Statement auditor | Consistency review and audit readiness |
| Finance & Operations | KYC screener | Entity file assembly and document review |
None of these workflows are new. Pitchbooks, KYC, month-end close - the task lists have been the same for decades. What's different is the delivery: these are architectures that a bank's engineering team can take to production without months of prompt engineering. They come with the connector infrastructure already built.
Who's Running Them Now
The Largest Institutions
JPMorgan Chase, Goldman Sachs, Citi, AIG, and Visa have moved Claude into production workflows. The tone of their leadership comments has shifted from pilot-speak to operational dependency.
Jamie Dimon, who shared a stage with Anthropic CEO Dario Amodei at a New York financial services event today - the first time the two executives have appeared together publicly - described asking Claude to research asset swaps and Treasury bid-ask spreads: "In 20 minutes, it created a huge dashboard, with all the backup, and all the research, and it was very accurate."
Goldman Sachs CIO Marco Argenti put the change plainly: "This is the first time that instead of buying infrastructure, you can actually buy intelligence."
AIG CEO Peter Zafino reported Claude scoring 88% as accurate as human experts on insurance claims analysis. If that number holds under independent scrutiny - and the methodology isn't published yet - it's the strongest accuracy claim in today's release.
JPMorgan Chase, Citadel, Goldman Sachs, and others now run Claude in production for investment research and back-office operations.
Source: unsplash.com
Tier Two: In Production, Growing Fast
Citadel, BNY, Carlyle, Mizuho, Travelers, Walleye Capital, and Hg have also confirmed production use. Walleye Capital reported 100% employee adoption across its investment team. Mizuho's managing director described the shift as: "Prep time has been transformed into idea time."
FIS CEO Stephanie Ferris outlined a specific operational result from its anti-money laundering integration: "Together we're building an agent that compresses AML investigations from days to minutes." AML reviews are notoriously time-intensive - analysts combing transactions against sanctions lists, PEP databases, and adverse media. Minutes, not hours, changes the compliance economics.
The Plumbing
Microsoft 365 Integration
Claude now works inside Excel, PowerPoint, and Word through native add-ins. Context transfers across applications automatically - start a financial model in Excel, move to PowerPoint for the client deck, and Claude carries the working context without re-explanation. Outlook integration is next.
The design choice here is continuity. The agent doesn't demand a context window restart when the analyst switches tools. It follows the work.
Data Connectors
Eight new connectors launched with this release: Dun & Bradstreet, Fiscal AI, Financial Modeling Prep, Guidepoint, IBISWorld, SS&C IntraLinks, Third Bridge, and Verisk. They join an existing stack of FactSet, S&P Capital IQ, MSCI, PitchBook, Morningstar, LSEG, and Daloopa.
The Moody's integration stands out. An MCP app giving Claude access to credit ratings and data on 600 million companies isn't a minor connector - it puts credit risk context directly into the agent's working memory for compliance, underwriting, and credit analysis workflows. Citadel described the Excel integration in direct terms: "Claude for Excel meets analysts where they live."
Anthropic's agent templates are designed for workflows already running on Bloomberg terminals and data feeds like FactSet, MSCI, and the new Moody's integration.
Source: unsplash.com
The Two-Track Play
Anthropic isn't running a single go-to-market. It's running two parallel strategies.
Track one is direct enterprise: give major institutions the reference architectures to deploy and run agents themselves. The 10 templates, the M365 integration, and the connector ecosystem all serve this market - the JPMorgans and Citadels that have engineering capacity to configure and maintain their own deployments.
Track two is a $1.5 billion joint venture with Blackstone, Hellman & Friedman, and Goldman Sachs, announced yesterday. That vehicle embeds Anthropic engineers directly inside mid-market, PE-owned companies in healthcare, manufacturing, financial services, retail, and real estate - the forward-deployed engineer model applied to AI at scale. The PE firms' portfolio companies are the first proving ground before the JV targets the broader mid-market.
JPMorgan CIO Lori Beer named the real constraint: "There's this capability overhang...it's the actual organization's ability to digest and absorb it that tends to be where the gap is." The JV addresses that constraint directly. When Anthropic engineers are embedded inside the operation, the capability overhang becomes an implementation roadmap.
What It Does Not Tell You
The Benchmark Is Self-Reported
Claude Opus 4.7 leads the Finance Agent benchmark at 64.4%. That benchmark was designed and published by Anthropic. Independent replication hasn't happened. 64.4% is a baseline to test against, not a ceiling to rely on.
The Accuracy Claim Needs Methodology
AIG's 88% accuracy figure on insurance claims analysis came from AIG's own deployment report. No audit methodology is published: which claim types, what confidence threshold, which human experts, what time period. Claims audits involve contested liability decisions and complex exclusions. Without methodology, 88% is a directional signal, not a verified number.
Every Connector Is a Switching Cost
FactSet, S&P Capital IQ, Moody's, Dun & Bradstreet, LSEG - each connector added makes Claude stickier with every data source integrated. That's good for retention. The inverse is also true: a bank that has rewired its Excel workflows, model-building pipelines, and data feeds around Claude's context architecture has materially increased its cost to switch. This isn't unique to Anthropic - OpenAI is running the same playbook. But institutions assessing which agent platform to standardize on should price in the lock-in architecture, not just the agent capabilities.
Anthropic has moved quickly since launching its managed agents platform - from SDK primitives to 10 sector-specific production templates with 13+ confirmed deployments in one announcement cycle. Dario Amodei described Anthropic's situation as one of "absolute radical uncertainty" about where growth is going: "The cone is even wider than I thought." The financial services vertical, with its long workflows, high-value data, and tolerance for per-seat pricing, is where that cone looks widest right now.
Sources:
