On May 5, 2026, Anthropic did something it has never done before: it shipped a vertical.
Not a model. Not a platform primitive. Not an API feature. A vertical — ten ready-to-run AI agent templates built specifically for financial services, packaged with connectors, skills, and subagent architectures for the grunt work that consumes junior analysts and back-office teams. It came bundled with Microsoft 365 add-ins that let Claude operate across Excel, PowerPoint, and Word in a shared context, and a Moody’s data partnership that gives agents access to verified financial data without requiring users to feed raw files.
Finance is famously conservative about software. The fact that Anthropic chose it as its first industry vertical says a lot about where enterprise AI adoption is actually happening — and where the next wave of agentic coding revenue is coming from.
What Shipped#
The ten agent templates cover both front office and back office:
Front office (deal work, client work)
- Pitch builder — turns earnings filings, comparables, and deal context into draft pitchbooks
- Meeting preparer — generates briefing packages from CRM data and public filings
- Earnings reviewer — flags material changes across quarterly reports
- Model builder — builds financial models from data feeds and audits formula dependencies
- Market researcher — aggregates sector intelligence into structured summaries
- Valuation reviewer — cross-checks model assumptions against live market data
Back office (operations, compliance)
- General ledger reconciler — matches entries across linked accounts
- Month-end closer — orchestrates the close sequence, flags anomalies, generates the management pack
- Statement auditor — verifies financial statement consistency and flags discrepancies
- KYC screener — runs name matches against sanctions lists, PEP databases, and adverse media
Each template is a reference architecture that packages three components: skills (Claude Code/Cowork plugin instructions plus domain knowledge), connectors (governed access to the data sources the task needs — Bloomberg, Moody’s, internal data warehouses), and subagents (additional Claude instances called for specialist subtasks, such as comparables selection or methodology verification). The templates deploy as plugins in Claude Cowork and Claude Code, or as a cookbook for Claude Managed Agents if you want to wire them into your own infrastructure.
The Benchmark That Matters Here#
Claude Opus 4.7 leads the Vals AI Finance Agent benchmark at 64.37%, ahead of GPT-5.5 at 59.96% and Gemini 3.1 Pro at 59.72%. The Vals benchmark is specifically designed to test finance-domain agentic tasks — not just question answering, but multi-step workflows that involve data retrieval, calculation, and output generation under domain-specific constraints.
That 4.4-point lead over GPT-5.5 on a benchmark this domain-specific is not a number to dismiss. Finance workflows involve high-stakes edge cases where reasoning quality compounds across steps. A model that’s broadly competitive on coding benchmarks might still fail the “does this forward schedule reconcile to the balance sheet?” check that a junior analyst catches on first review.
The Microsoft 365 Integration#
The bigger unlock is context continuity across the Office suite. Anthropic’s Microsoft 365 add-ins let Claude operate in Excel, PowerPoint, and Word with shared context across applications.
What this means in practice: an analyst builds a DCF model in Excel, flags the bear case, and a deck appears in PowerPoint that reflects the bear case numbers — without copying, re-explaining, or context-switching. Claude carries the financial model’s assumptions, data sources, and analytical conclusions across the entire workflow.
In Excel specifically: build models from public filings and data feeds, audit formula linkages across tabs and workbooks, run sensitivity tables. In PowerPoint: draft investor decks that auto-update when the underlying data changes. Outlook integration is coming.
This is the first Microsoft 365 integration that makes Claude a real workflow layer rather than a chat assistant bolted onto a document editor. The difference between those two things is the difference between a tool and a platform.
Why Finance First#
Anthropic did not choose finance by accident.
Finance has three properties that make it a natural first vertical for agentic AI:
High value per workflow. A junior analyst spending 40 hours on a pitchbook earns $150,000+ in loaded cost per year. Automating even 30% of that work has a measurable ROI that CFOs can put in a spreadsheet. Unlike developer productivity (where the ROI is real but harder to attribute), finance workflows have line-item costs that map directly to agent runtime.
Structured outputs. Financial documents — pitchbooks, credit memos, regulatory filings — have well-defined schemas. The grader in an Outcomes-powered agent can verify that a DCF output reconciles to its inputs. There is a ground truth, and the agent can check its own work against it. This is exactly what Claude Managed Agents’ Outcomes feature is built for.
Regulatory pressure. KYC, AML, and audit trail requirements in finance are not optional. Anthropic’s agent templates ship with documented dismissals, immutable audit logs, and governed data connectors — the compliance infrastructure that financial institutions need before they can put an agent in a production workflow. This is not an afterthought; it is a prerequisite for enterprise adoption in this sector.
The Moody’s data partnership matters here too. Agents that need live financial data have historically required custom data pipeline work to provision reliably. Moody’s partnership means that connection is pre-built, governed, and available out of the box — removing one of the major friction points for enterprise deployment.
What This Signals for the Broader Market#
Anthropic going vertical is a competitive move, not just a product addition.
Until now, enterprise AI deployments in finance have required significant professional services work — consulting firms, systems integrators, and in-house AI teams assembling custom agent architectures from primitives. Anthropic just commoditized the first layer of that stack. A bank that wants a KYC screener can now deploy a reference architecture in days instead of months.
That accelerates adoption and creates a new competitive pressure for the Cursors and Copilots of the world: if Anthropic is delivering packaged, domain-specific agentic workflows, an IDE-embedded assistant that helps developers write code is competing at a different level of abstraction. Cursor helps engineers build software. Anthropic’s finance agents help analysts close books and screen clients. The latter has a clearer ROI conversation at the C-suite level.
Claude Code is the tool that developers use to build these agentic workflows. The finance agents are the workflows that non-developers now deploy on top of Claude. These are two different wedges into the same enterprise budget, and Anthropic is now pushing on both simultaneously.
The Practical Question#
For financial services teams evaluating this: the templates are starting points, not finished products. Each ships with a documented architecture, but production deployment requires mapping connectors to your actual data sources (Bloomberg, FactSet, internal data warehouses) and customizing skills to your firm’s terminology and processes.
The Outcomes beta makes this significantly more robust than first-generation enterprise AI deployments. Rather than tuning prompts and hoping outputs are correct, you write a rubric: “The pitchbook must include a DCF with a sensitivity table, a comps table with at least five comparables, and a recommendation section with explicit supporting rationale.” The agent runs, the grader evaluates, and if the rubric isn’t satisfied, the agent tries again — in a separate context window, without the bias of its own prior reasoning.
That feedback loop is what turns these templates from impressive demos into production workflows.
Finance just became the first industry where Anthropic is shipping the whole stack: model, agent architecture, data connectors, office suite integration, and a self-verification loop. If this pattern holds, it won’t be the last.
Sources:
- Agents for financial services — Anthropic
- Anthropic deepens push into Wall Street with new AI agents — Fortune
- Anthropic rolls out AI agents to target financial services — TechRadar
- Anthropic Launches 10 Claude Agent Templates for Financial Services — how2shout
- Claude Managed Agents Outcomes — Claude Docs