Anthropic's 10 finance agents: what does each one do?

This project is different: it wasn't released by an independent developer but by Anthropic, the company behind Claude. The repo is open and has over 41,000 stars.

Anthropic

Let's say it up front: this isn't built for ordinary investors. Its audience is banks, funds and their analysts. Still, it's interesting to see how they break down daily work. We read the docs but didn't run anything.

First: what it doesn't do

There's a warning at the top of the repo, and it shouldn't be skipped as a formality. The agents draft analyst work: models, memos, reconciliations. They don't make investment recommendations or execute transactions. Every output is left for a qualified professional to review.

That's the difference from the trader projects we covered earlier: the question "which stock should I buy" simply isn't here.

Research and modeling: closest to us

• Market Researcher. Give it a sector or theme and it prepares an industry overview, competitors, peer comparisons and a shortlist of ideas.
• Earnings Reviewer. Reads the earnings call and filings, updates the model and drafts a note.
• Model Builder. Builds valuation models such as DCF directly in Excel.

For an ordinary investor the interesting part is the Earnings Reviewer's logic: don't just read a report, ask "where did my earlier view change?" We have a separate article on what to watch during earnings season.

Coverage and advisory

• Pitch Agent. Builds a presentation from comparisons and past deals. That includes LBO math. An LBO means buying a company largely with debt, and because it relies on interest-bearing debt many scholars consider it impermissible. We mention it only to describe what the agent does.
• Meeting Prep Agent. Puts together a short briefing before each client meeting.

Fund admin: the invisible work

These four aren't glamorous, but this is where a lot of the time goes:

• Valuation Reviewer takes in fund packages and runs the valuation template.
• GL Reconciler finds breaks in the ledger and traces the cause.
• Month-End Closer prepares month-end entries and commentary.
• Statement Auditor checks investor statements before they go out.

The last one is the KYC Screener: it reads new client documents, checks them against the rules and flags gaps.

What else is in the box

The connectors plug into financial data platforms, and you should assume many of them need their own paid subscription. Some plugins add commands such as /comps, /dcf and /earnings.

Our takeaway

There's little here that an ordinary investor in Uzbekistan would use directly. The value is elsewhere: it shows how large finance firms hand work to AI. They aren't saying "predict the market". They're giving it boring, specific, checkable tasks and leaving the final signature with a person.

That approach alone is a good enough lesson for any investor who wants to use AI.

For information only, not investment advice. We're not affiliated with Anthropic.