Anthropic Just Released 10 AI Agents for Financial Services. Here Are 3 Things You Can Do Today to Get Started.

On May 5th, Anthropic released ten ready-to-run AI agent templates built specifically for financial services work. Pitchbook builders. KYC screeners. Month-end closers. Valuation reviewers. Each one ships as a plugin you can drop into the tools your team already uses, or as a managed deployment that runs in the background on Anthropic’s infrastructure.

This is the most concrete release we’ve seen for finance firms so far. The infrastructure that used to take months to stand up is available out of the box. Compliance, audit logging, scoped permissions, human approval flows: all built in.

If you run a small or mid-sized financial services firm, you don’t need a strategy offsite to act on this. You need three small moves this week. Here’s exactly what to do today.

First, what’s actually in the release

Ten agents, grouped by function.

On the deal and research side: a Pitch Builder that drafts pitchbooks and runs comps, a Meeting Preparer that assembles client briefs, an Earnings Reviewer that reads transcripts and updates models, a Model Builder, and a Market Researcher.

On the finance and operations side: a Valuation Reviewer, a General Ledger Reconciler, a Month-End Closer, a Statement Auditor, and a KYC Screener.

Each template bundles three things: domain knowledge (what a good pitchbook looks like, what KYC standards apply), governed connections to your data (FactSet, S&P Capital IQ, PitchBook, Morningstar, Moody’s, plus your own systems), and sub-models that handle specialized tasks like methodology checks.

Anthropic also brought Claude directly into Microsoft Excel, PowerPoint, Word, and (soon) Outlook through add-ins. Context travels between them. A model your analyst builds in Excel can move into a deck without re-explaining itself. For a firm whose work lives in those four apps, that’s significant on its own.

That’s the landscape. Now to what you do about it.

Thing 1: Install Claude for Excel today and give it to one analyst

The Excel add-in is generally available. Installation takes minutes, not weeks.

Pick one analyst on your team, ideally someone who already lives in Excel and is curious about AI. Have them spend an afternoon doing one of these things on a real workbook:

• Rebuild a comps model with Claude doing the data assembly
• Audit formulas across a linked workbook they already trust
• Run a sensitivity analysis on a model they built last quarter

You’re not deploying anything. You’re building informed intuition about where the tool helps and where it still misses. Every real decision from here depends on that intuition. Buying it from a vendor or a webinar doesn’t work the same way. You need a person on your team who has actually used it.

By Friday, that analyst should be able to tell you, in plain language, what they’d trust Claude with and what they wouldn’t.

Thing 2: Identify the one workflow that’s eating your most expensive hours

Open your calendar from last week. Walk through it task by task and ask one question: which work was structured, repetitive, and deadline-driven, but didn’t actually need senior judgment?

For most finance firms, the candidates cluster fast:

• Month-end close and reconciliations
• KYC document assembly and review
• Comparables pulls and pitchbook drafting
• Meeting prep and client brief assembly
• Statement reviews for audit-readiness
• Earnings transcript review and model updates

You’re not picking the most strategically interesting task. You’re picking the one that drains the most hours from your highest-paid people. That’s your candidate.

Write it down. One sentence. “We spend X hours per week on Y.” That sentence is what you bring to step three.

Thing 3: Map that workflow to one of the ten templates

Go to Anthropic’s financial services marketplace on GitHub (linked from their announcement). Open the README for the template that matches your candidate workflow. Read what the agent does, what data it needs, what subagents it calls, and what the human-in-the-loop checkpoints look like.

You’re answering three questions:

• Does this template do roughly what we'd want it to do? Not perfectly. Roughly.
• What data would it need to access, and can our compliance posture allow that access today?
• Who on our team would own the human approval steps?

If all three answers are clean, you have a real pilot candidate. If any answer is messy, that’s also useful: you’ve found the blocker that needs to be solved before AI can help with that workflow.

This step is half an hour of reading. It is the single highest-leverage thing you can do this week, because it converts a vague sense of “we should do something about AI” into a specific, named workflow with a specific, named template attached to it.

What “today” actually buys you

By the end of this week, you can have: an analyst who has hands-on experience with Claude for Excel, a named workflow that’s a real candidate for automation, and a specific template mapped to that workflow with a clear read on what’s blocking a pilot.

That’s not a transformation. That’s a starting position. It’s enough to make a real decision about what to pilot next quarter, and it costs you a few hours and zero contracts.

Start there.

ProxyClaw scopes and deploys managed AI agents for professional services and finance firms. If you want help mapping one of the new finance templates to a workflow at your firm, book a thirty-minute scoping call.