Agents · Finance
Claude gets a finance degree: 10 purpose-built agents, full Microsoft 365 integration, and a wall of data partners
Anthropic just made the most vertical-specific agent push we have seen from a frontier AI lab. Ten ready-to-run agent templates covering the heaviest lifts in financial services — from building pitchbooks to reconciling ledgers. Microsoft Excel, PowerPoint, Word, and Outlook integration. New data connectors. And a partner roster that includes some of the biggest names in market data. This is not a demo. It is an ecosystem launch.
What shipped
On May 5, Anthropic released a package of capabilities designed to let financial firms put Claude on real work in "days, not months." There are three big pieces:
Ten agent templates. Each is a reference architecture that bundles skills (task-specific instructions and domain knowledge), connectors (governed data access), and subagents (additional Claude instances called upon for specific subtasks like comparables selection or methodology checks). The ten agents cover the full front-to-back office chain:
Research and client coverage — five agents handling the pre-meeting and post-filing grind:
- Pitch — creates target lists, runs comparables, and drafts pitchbooks for client meetings.
- Brief — assembles client and counterparty briefs ahead of calls, pulling from filings, transcripts, and deal history.
- Monitor — reads transcripts and filings, updates models, and flags thesis-relevant changes across coverage.
- Model — creates and maintains financial models from filings, data feeds, and analyst inputs.
- Credit — tracks sector and issuer developments, synthesizes news, filings, and broker research, and flags items for risk review.
Finance and operations — five agents for the close, compliance, and control work:
- QA — checks valuations against comparables, methodology, and the firm's review standards.
- General ledger reconciler — reconciles GL accounts and runs net asset value calculations against the books of record.
- Close — runs the close checklist, prepares journal entries, and produces close reports.
- Vouch — reviews financial statements for consistency, completeness, and audit-readiness.
- KYC — assembles entity files, reviews source documents, and packages escalations for compliance review.
Firms can adapt any template to their own modeling conventions, risk policies, and approval flows.
Microsoft 365 add-ins. Claude now works directly inside Excel, PowerPoint, and Word (generally available), with Outlook coming soon. The killer feature: context carries automatically between applications. A model built in Excel informs a deck drafted in PowerPoint without re-explaining anything. In Outlook, Claude triages the inbox, schedules meetings, and drafts responses in the user's voice. On mobile — via Claude for Mobile — analysts can assign tasks by text or voice and return to finished work ready for review.
An expanded partner ecosystem. AI agents are only as good as the data they access. Claude already connects to FactSet, S&P Capital IQ, MSCI, PitchBook, Morningstar, Chronograph, LSEG, and Daloopa. The new additions deepen the bench significantly:
- Dun & Bradstreet — the global standard for verified business identity, with its D-U-N-S Number and Commercial Graph now accessible inside Claude.
- Financial Modeling Prep — real-time quotes, fundamentals, statements, filings, and transcripts across equities, ETFs, crypto, forex, and commodities.
- Tegus — 100,000+ compliance-reviewed expert interview transcripts with verbatim excerpts linked to source.
- IBISWorld — industry-level revenue, financial ratios, risk scores, cost structures, and forecasts across thousands of sectors.
- Datasite — DealCentre data rooms for document search, diligence Q&A, and deal-activity tracking.
- Guidepoint — primary-source expert interviews on companies, sectors, and value chains.
- A property & casualty insurance data provider — underwriting, claims, and risk analysis data for insurers.
Separately, Moody's launched an MCP app that embeds proprietary credit ratings and data on more than 600 million public and private companies directly inside Claude — a clear signal that the Model Context Protocol is becoming the standard interface for enterprise AI tool integration.
Two deployment paths, same agents
There are two ways to use any of the ten agents:
As a plugin in Claude Cowork or Claude Code. The agent runs alongside the analyst on their desktop, using the software they already have open. Hand the Pitch agent a target list and you get back a comps model in Excel, a pitchbook drafted in PowerPoint, and a cover note ready in Outlook — without copying data between tools.
As a Managed Agent on the Claude Platform. The same template runs autonomously for work that spans a book of deals or a nightly schedule. Managed Agents bring enterprise infrastructure: long-running sessions that can work through a multi-hour deal close, per-tool permissions, managed credential vaults, and a full audit log where compliance and engineering teams can inspect every tool call and decision.
In both modes, users stay in the loop — reviewing, iterating on, and approving Claude's work before it goes to a client, gets filed, or is acted on. Anthropic is not pitching full autonomy here. It is pitching augmentation with auditability.
Built on Claude Opus 4.7, which now leads finance benchmarks
These capabilities pair best with Claude Opus 4.7 (released April 16), which Anthropic says is state-of-the-art on financial tasks and leads Vals AI's Finance Agent benchmark. The implication: these agents are not just prompt-wrapped wrappers. They are built on a model that has been evaluated specifically for financial reasoning.
Each agent template packages three things: skills (instructions and domain knowledge), connectors (governed access to specific data sources), and subagents (additional Claude models called upon for specific sub-tasks such as comparables selection or methodology checks). This is a more sophisticated architecture than simple retrieval-augmented generation — it is composable, multi-agent orchestration designed for regulated workflows.
Who is already using it — and what they are saying
Anthropic lined up an unusually deep bench of customer quotes for this announcement, spanning banking, asset management, and insurance:
FIS — a company that sits at the center of how money moves for thousands of financial institutions — is building an agent that compresses anti-money laundering investigations from days to minutes. "Anthropic was the clear choice," the company said, noting that credit decisioning, fraud prevention, and deposit retention agents are next. "FIS clients won't need to build this infrastructure themselves. It is already here."
Carlyle described Claude as "a core tool for delivering value across our firm from investing to operations to portfolio management," citing its coding capabilities and agentic reasoning.
Walleye Capital revealed that 100% of its 400-person hedge fund uses Claude Code. "This level of adoption reflects our AI-first mindset: we expect everyone to constantly rethink how they work, always asking 'How can AI help me do this?' — whether or not they are in a traditionally technical role."
Moelis & Company said Claude has transformed prep time into idea time, with "faster workflows, richer client insights, and new use cases we didn't anticipate."
Northwestern Mutual reported "significantly elevated levels of engineering excellence and meaningful improvements in productivity" since introducing personalized Claude assistants.
Why this matters
This is not just another model release. It is a vertical integration play with three strategic implications:
1. The enterprise AI agent race is moving from horizontal to vertical. General-purpose coding agents and chatbot-style interfaces are table stakes. Anthropic is now shipping purpose-built agents for a specific regulated industry with specific workflows, specific compliance requirements, and specific data partners. Expect every frontier lab to follow — and expect the next round of competition to be about who has the deepest vertical integrations, not who has the highest benchmark score.
2. Microsoft 365 is the new agent operating system. By embedding Claude inside Excel, PowerPoint, Word, and Outlook — with cross-application context — Anthropic is effectively treating the Office suite as the UI layer for financial agents. This sidesteps the "where do agents live?" question that has plagued enterprise AI adoption. The answer: they live where the work already happens.
3. The MCP ecosystem is becoming the enterprise standard. Moody's building an MCP app is significant. MCP (Anthropic's Model Context Protocol) is shaping up as the interface layer between AI agents and enterprise data — and third-party data providers are now building for it natively. If MCP becomes the default, the agent ecosystem starts to look like the app store ecosystem: platform owners capture routing, data providers build on top, and customers benefit from interoperability.
The open question
Anthropic is promising that firms can adapt these agents in "days, not months" and that users "stay firmly in the loop." But regulated financial workflows do not tolerate probabilistic outputs lightly. A pitchbook that hallucinates a comp or a KYC agent that misses a red flag is not a minor error — it is a compliance event. The audit log in Managed Agents is a strong start, but the real test will be how these agents perform under the scrutiny of risk officers who have spent careers building deterministic controls.
The FIS quote — "compresses AML investigations from days to minutes" — is the kind of outcome that makes CFOs pay attention. If the agents deliver that reliably, this is a landmark release. If they do not, it is a very expensive demo.
All ten agents, the Microsoft add-ins, and the new connectors are available now on all paid Claude plans. Managed Agents are in public beta. The financial services marketplace and webinar registration are live on Anthropic's site.