Anthropic builds a vertical-agent factory

Anthropic's ten finance agents reuse the architecture that turned Claude Code into a $2.5 billion business

// Share
Anthropic builds a vertical-agent factory

A NEW genre of corporate event has taken hold in technology this year. The set piece runs roughly the same way: a frontier-AI lab unveils ten or so "agents" for some new vertical, the relevant incumbents lose double digits in a single trading session, and the lab's chief executive suggests — gently, on stage, in a tone of mild surprise — that this is only the beginning. On Tuesday in New York, Anthropic ran the play in finance: shares of FactSet, the financial-data provider, fell 8.1%; Morningstar erased its earlier gains to fall more than 3%; selling pressure spread across S&P Global and Moody's. Ten new Claude agents — pitch builder, model builder, valuation reviewer, KYC screener, month-end closer, statement auditor and the rest — were the proximate cause.

The wider apparatus around them is the more telling part. Claude can now operate inside Microsoft Excel, PowerPoint, Word and Outlook through native add-ins. Data flows in from FactSet, S&P Capital IQ, MSCI, PitchBook, Morningstar, LSEG, Moody's and Dun & Bradstreet through governed connectors. A new joint venture with Blackstone, the world's largest alternative-asset manager, Hellman & Friedman, a private-equity firm, and Goldman Sachs Group will help deploy the suite to other Wall Street buyers. "Finance is a great blueprint for the rest of knowledge work," Nicholas Lin, Anthropic's head of product for financial services, told reporters. The technology, he added, is "just a few months behind" coding, where Anthropic has "seen massive acceleration."

Cut from the same code

The remark passed quickly, but it is the most important sentence Anthropic has uttered in months. The trouble with reading Tuesday's launch as a finance announcement is that almost nothing technical in it is new. The architecture of the ten agents — skills (markdown instructions and domain knowledge), connectors (governed access to data) and subagents (smaller Claude models called for sub-tasks like comparables selection or methodology checks) — is identical, line for line, to the architecture Anthropic shipped for Claude Code in May 2025.

// Members only

This article is for Vector members. Start a 7-day free trial to keep reading.

Start your free trial

// The Daily

Get Vector in your inbox.

A free morning briefing on the AI revolution. Weekdays at 6am CT.