2026-05 — Claude for Finance and the 10 agent templates

Why this matters

On 2026-05-05 — one day before Code w/ Claude 2026 — Anthropic shipped Claude for Finance: 10 ready-to-run AI agent templates for financial services, plus a finance-tuned model release (Claude Opus 4.7), plus Microsoft 365 add-ins, plus an enterprise joint-venture announcement totalling $1.5B in committed capital.

For this notes set, Finance matters for the same architectural reason that Claude Design matters: it’s a vertical surface built on the same agentic stack as Claude Code, with the structure made textually explicit. Anthropic’s own announcement defines each agent template as a reference architecture packaging three things:

  1. Skills — instructions and domain knowledge for the task
  2. Connectors — governed access to the data the task runs on
  3. Subagents — additional Claude models called by the main agent for specific sub-tasks (e.g. comparables selection, methodology checks)

That’s exactly the agentic-engineering vocabulary the rest of this repo runs on: skills + connectors (the finance-vertical equivalent of MCP servers) + subagents (managed-agents-style dispatch). The finance launch is the cleanest worked example of “agentic surface = skills + connectors + subagents” composing into a production-grade vertical product.

The ten agents

Agent What it does
Pitch builder Generates pitchbooks (the Wall Street deliverable)
Meeting preparer Prepares for client meetings: collects relevant data, drafts talking points
Earnings reviewer Reads quarterly earnings reports and surfaces what matters
Model builder Builds financial models (DCF, etc.)
Market researcher Synthesises market intelligence
Valuation reviewer Reviews valuation work
General ledger reconciler GL reconciliation (back-office accounting)
Month-end closer Month-end close work
Statement auditor Reviews financial statements
KYC screener Know-Your-Customer screening

Spread across both front-office (pitch builder, meeting preparer, market researcher) and back-office (GL reconciler, month-end closer, statement auditor, KYC screener) — Anthropic explicitly targeted both sides of the finance workforce, not just the high-billable analyst-shaped roles.

What’s worth noting from the Tort Mario writeup: these aren’t demo agents. They ship as templates customers can fork, customise to their firm’s data + conventions, and run as production tooling. The market shape is template-and-customise, not out-of-the-box-product.

The Opus 4.7 finance benchmark

Anthropic released Claude Opus 4.7 alongside the finance launch and pinned its identity to a benchmark number: 64.37% on Vals AI’s Finance Agent benchmark, leading the industry. Two useful framings of that number:

Microsoft 365 integration

The other piece that pulls Claude further out of “chat app” territory: Claude now works across Microsoft Excel, PowerPoint, Word, and Outlook through Microsoft 365 add-ins. Once installed, context moves between applications automatically — work that starts in a spreadsheet continues in a deck without the analyst re-explaining.

The cross-app context propagation is the architectural piece. It matches the same continuity-of-context shape we saw in the Claude Design entry (design system read once, applied to every project) and the CLAUDE.md entry (project context loaded once per session). Persistent context across surfaces is the 2026 Anthropic motif — chat, code, design, and now finance all share the pattern.

The $1.5B enterprise joint venture

Per Fortune’s coverage:

This is the bet under the bet. The finance agent templates are a product launch; the $1.5B JV is the distribution commitment backing it. Goldman, Blackstone, and H&F aren’t just buying — they have skin in selling enterprise-grade Claude integrations into their portfolio companies.

For someone working with Claude Code, this matters indirectly: the platform’s commercial trajectory determines the platform’s investment trajectory. Anthropic that’s $1.5B-JV-shaped invests differently than Anthropic that’s only-API-revenue-shaped.

Connection to Claude Code

The finance vertical pulls into this Claude-Code-adjacent notes set because the agent templates are constructed from the same primitives Claude Code exposes. If you:

…you’d find your workflow already has the same pieces: skills (reusable instructions for recurring tasks), connectors (MCP servers / --add-dir / file references), subagents (Task tool dispatches for context-isolation). The finance agents are your workflow, formalised as deliverables. That’s the thing worth noticing.

Try this

  1. Read Anthropic’s finance-agents announcement for the architecture diagram. The skills/connectors/subagents decomposition is the cleanest single statement of the agentic- surface shape Anthropic is building toward across all verticals.
  2. Pick one of the 10 agent templates. Try to write the same thing as a Claude Code skill + plugin in your own project. You won’t ship a production-grade KYC screener this way (no regulated-data connectors), but you’ll feel the gap between “skill that helps a task” and “agent that runs a workflow.”
  3. Read the Vals AI Finance Agent benchmark if you care about evaluation discipline. Vertical-specific evals are how this generation of agentic products will get scored; understanding the eval shape is part of agentic-engineering literacy.

Sources