Anthropic Just Shipped 10 Wall Street Agents. The Frontier Lab Is Now a Vendor.

Anthropic held an invite-only event in New York this morning and used it to do something frontier labs almost never do: sell finished products instead of capability promises. Ten preconfigured Claude agents for banks, asset managers, and insurers shipped today, alongside full Microsoft 365 integration, a Moody's app embedded directly inside Claude, and a co-engineered Financial Crimes Agent built with FIS. This is the day after the $1.5 billion joint venture with Blackstone, Hellman & Friedman, and Goldman Sachs. Today is the day the JV got a catalog.

I've been waiting for one of the labs to make this jump. For two years the model companies have insisted they only sell intelligence and let partners build the workflows. That story just ended.

The Ten Agents, In Order of Pain They Remove

Anthropic is shipping ten templates, each a Claude Cowork plugin and a Claude Code plugin, plus a cookbook for Claude Managed Agents. They are organized into three buckets: research and client coverage; credit, risk, and compliance; and the controllers' office. Here is the lineup as Anthropic described it on stage and on the product page.

Notice the through line. Every single one of these is a job that an analyst, associate, or mid-level operations person spends 20 to 50 hours a week on. Anthropic did not pick differentiated, novel use cases. They picked the workloads where any bank CIO can compute the headcount math in their head before the demo finishes.

Microsoft 365 Just Got a New Default Agent

The second piece is the one that quietly reshapes the desktop. Anthropic shipped general availability of Claude add-ins for Excel, PowerPoint, Word, and Outlook today, with a single shared agent state across all four. That last detail is the unlock. Until now, Claude in Excel was a separate session from Claude in Word. Today, the agent that just rebuilt your DCF in Excel can drop the chart and three bullet points into the PowerPoint pitch the coverage banker needs by 4 p.m., then draft the Outlook email to the client with the attachment correctly named.

Microsoft has been quiet on whether this conflicts with Copilot. The honest answer is that it does. Copilot is the default agent inside Office. Claude is now the default agent on top of Office, with finance-specific reasoning and a dedicated Wall Street vertical behind it. The tension is going to surface in enterprise contracts within 90 days.

Moody's as a Native App, 600 Million Companies Inside Claude

The Moody's integration is the one analysts will quietly ignore for a week and then panic over. Moody's embedded its full platform, credit ratings, financials, ownership data, ESG signals, and corporate hierarchy for over 600 million companies, as a native Claude app. You ask Claude for a credit memo on a private mid-market borrower and it pulls structured Moody's data inside the same context window. No copy-paste from a separate terminal, no CSV export, no API gluework.

That is a real shot at Bloomberg and S&P Capital IQ. Both still own the workflow because the data sits behind their interface and the analyst has to be there to ask the question. When the analyst asks the question to Claude and Moody's answers from inside the same chat, the desktop terminal is no longer the gravity well.

FIS, Financial Crimes, and the Embedded FDE Model

Yesterday FIS announced its own partnership with Anthropic, and the first deliverable is a co-built Financial Crimes Agent. The pitch: AML alert and case investigations compressed from days to minutes, evidence assembled across a bank's core systems automatically, false positives reduced, and SAR narrative quality improved. BMO and Amalgamated Bank are the design partners. General availability is targeted for the second half of 2026.

The structural detail matters as much as the product. Anthropic embedded its Applied AI team and forward-deployed engineers inside FIS to co-design the agent and transfer enough knowledge that FIS can build and scale follow-on agents independently. Client data stays inside FIS-controlled infrastructure. This is the Palantir playbook adapted for an agent vendor: ship the first one with your engineers in the building, then hand them the toolkit.

The roadmap that FIS published includes credit decisioning, deposit retention, customer onboarding, and fraud prevention. Each one is a multi-hundred-million-dollar line in a typical Tier 1 bank's technology budget.

Opus 4.7 Is the Engine, and the Benchmark Backs It

All ten agents are tuned to Claude Opus 4.7. Anthropic pointed at the Vals AI Finance Agent benchmark to back the choice: Opus 4.7 leads at 64.37%, ahead of GPT-5.5 and Gemini 3.1 Pro on the same task suite. We are tracking those numbers on our benchmarks page and will reproduce them once the public test set lands. Until then, the relative ordering matters more than the absolute number.

The pricing for these agents is Claude Opus 4.7 token economics ($15 input, $75 output per million), which means a single research-and-pitch cycle on a mid-cap target probably runs between $40 and $120 in inference. That is rounding error against an associate's fully-loaded cost for the same hours, which is the comparison Anthropic wants you to make.

What Just Changed for the Industry

Three things, in order of how quickly they will hit a P&L.

First, Anthropic is no longer selling tokens. It is selling workflows. The pitch to a CFO is no longer "here is an API, integrate it," it is "here is a pre-built KYC screener, plug it into your case management system, your false positive rate drops 40% next quarter." That changes who at the buyer signs the contract and how big the contract is.

Second, the consulting and ops outsourcing markets just got a new competitor with a structurally lower cost base. The Blackstone, Hellman & Friedman, and Goldman Sachs JV announced yesterday is the distribution channel. Today's ten agents are the SKUs. A managed service that combines preconfigured agents with embedded Anthropic engineers is what Accenture and the Big Four have spent two decades building, and Anthropic just shipped a version one of it.

Third, every other frontier lab has to respond. OpenAI launched Workspace Agents for ChatGPT Enterprise three weeks ago, but those are general productivity templates, not vertical finance products. Mistral has Le Chat Work mode but no banking partner. Google has Gemini inside Workspace and a deep cloud relationship with several US banks, but no shipped agent catalog. The pressure to publish a comparable vertical lineup is now on every leadership team.

Our Take

The Anthropic 900 billion dollar valuation we wrote about yesterday was framed as a counter to OpenAI in the funding race. Today is what justifies it. The jump from $9 billion ARR to a reported $44 billion in five months is not a model story, it is a deals story. Wall Street pays for things that do work. Anthropic now sells things that do work. The mapping between those two facts is what makes the multiple defensible.

For TensorFeed users, the practical takeaway is short. If you are building anything in finance ops, look at the agent templates as a baseline and decide whether to build on top or rebuild. If you are running a frontier lab strategy, the vertical product play is no longer theoretical. And if you are an analyst staring down a pitchbook tonight, your job did not disappear, but the boring half of it just got automated by a vendor you can buy on a standard MSA.

We will be tracking the rollout cadence and the bank-by-bank deployments on the agents page. The Financial Crimes Agent goes GA in H2, and that is the launch worth watching, because if it delivers the alert-time compression Anthropic and FIS are claiming, every Tier 2 and Tier 3 bank in the US will be in procurement by Q1 2027.