• Anthropic shipped ready-to-run Claude agent templates for finance teams covering pitch decks, valuation reviews, and month-end close

• Templates install as plugins in Cowork or Claude Code, or run as Managed Agents via the public cookbooks

• Three named workflows: pitch building, valuation review, and closing the books at month-end

• I tested the same template pattern on solo-studio finance jobs (quarterly invoicing, prepaid retainers, German USt-VA), and most need real adaptation before you trust them

Anthropic just dropped a set of Claude agent templates aimed at finance teams. Pitches, valuation reviews, month-end close. Three workflows that normally eat a junior banker's weekend, packaged as installable plugins. I read the announcement twice because the wording is sneaky: these are templates, not products, and where you run them changes the story completely.

What "Claude agent templates" actually means

Claude agent templates are pre-built workflows that show Claude how to handle a specific job, plus the tools, prompts, and structure it needs to finish it. You do not get a finished app. You get a starting point that already knows the steps.

Anthropic shipped the financial-services pack with three options for where to run them. You can install them as plugins inside Cowork (Anthropic's collaborative workspace product) or inside Claude Code (the terminal coding agent most readers here already use). Or you can grab the cookbook code and run them as Managed Agents in production, which means Anthropic hosts the runtime and you just call an endpoint.

That last path matters more than it sounds. Managed Agents handle the orchestration layer most teams rebuild from scratch: queueing, retries, memory, tool routing. If you already build with the Claude Agent SDK, the cookbook is a head start. For the deeper context on the SDK side, see I Built 3 Production Agents With the Claude Agent SDK in One Weekend.

The financial-services templates are the first vertical pack. The pattern is obvious: Anthropic is publishing reference workflows for industries where the agent-shape is well understood and the data is repetitive. Finance, then probably legal, then probably sales ops.

The three named workflows, unpacked

Pitches. Valuations. Month-end close. Every analyst has lived all three.

The pitch-building template. A pitch deck draft is mostly assembly: company background, market size, comparable transactions, recommended structure, key risks. Claude is good at structured assembly when it has the right inputs. The template likely wires Claude to a CRM or research source, drops the data into a deck skeleton, and outputs slides ready for a human to refine. The painful 60% of the deck gets done while the analyst sleeps. The 40% that needs judgment still needs judgment.

The valuation review template. Valuation work is forensic: pull comps, sanity-check the DCF model, flag where assumptions look optimistic, summarize for the deal team. Claude reads spreadsheets, follows formulas, and writes plain-English memos. The template probably hands Claude a model file plus market data, asks it to recompute key cells and flag anomalies, then writes the review memo. This is the workflow I would actually trust earliest, because the output is a memo a human reviews line by line. Claude is auditable here. If the memo says top-line grows 40% a year and the underlying assumption is a single bullish analyst note, you see that in plain text and you push back. Compare that to a black-box "AI valuation tool" that spits out a number. Memos win.

Month-end close. This one I find the most interesting because it is not glamorous. Reconciling accounts, matching invoices to payments, flagging variances, building the close package. Repetitive, rule-driven, deadline-sensitive. The exact shape Claude is good at when you give it the right tools. A template that closes the books in 4 hours instead of 4 days is not a moonshot. Mid-market accounting teams are already doing it manually with junior staff. Replacing that with a reviewed agent run is a real workflow change, not a demo.

Cowork plugin vs Claude Code plugin vs Managed Agent

Three install paths, three different audiences.

Cowork is the workspace product. Plugins there sit next to a finance team's actual collaboration surface. The CFO can run a valuation review from the same place they run everything else. No new tool to adopt, no terminal, no cookbook. This is the path enterprise will take.

Claude Code plugins are for developers who want to trigger the same workflow from the terminal. If you live in Claude Code (I do), an installed financial-services plugin gives you new slash commands. That is the right shape for solo operators and small teams who treat the terminal as the office.

Managed Agents via the cookbooks is the production path. You take the template, adapt the prompts, wire it into your stack, and run it through Anthropic's hosted runtime. This is for product teams building finance features into their own apps, not for finance teams using Claude internally. Background on the runtime: Claude Managed Agents Now Have Filesystem Memory shows what you actually get for free at that layer.

The reason Anthropic shipped all three install paths together is honest: the same workflow has a different shape depending on whether you are a CFO, a developer, or a vendor. Forcing one shape on all three would have killed adoption.

What a solo studio actually gets from this pattern

I run the books for a one-person studio. So my reaction to "ready-to-run finance agent templates" was direct: which of my actual finance jobs does this pattern fit, and which ones need adaptation?

Quarterly invoicing. This is a fit. I have recurring clients, predictable line items, EUR amounts, VAT rules I never want to think about. A template that pulls the project log, drafts invoices in the right format, and queues them for review would save a clean afternoon every quarter. The pitch-building template is the closest sibling because invoicing is structured assembly with fixed inputs.

Prepaid retainer tracking. Mostly a fit. When a client pays upfront for a 3-month engagement, that 3,000 EUR sits on the balance sheet as a liability until the work is delivered. Tracking it across multiple projects gets ugly fast. The valuation review template is the right shape: read the contract, read the time log, compute what is recognized, flag anything weird. The catch is the German chart of accounts. The cookbook will be US-GAAP shaped. Adaptation is real work, not a 5-minute prompt edit.

German USt-VA prep. This is where the honesty kicks in. Quarterly VAT filing in Germany has a specific form, specific deadlines, specific reverse-charge rules for EU clients. No US cookbook handles this out of the box. You can absolutely build a Claude agent for it, but you are starting from the SDK, not from the financial-services template. Treat the templates as inspiration for the agent shape, not as a turnkey solution for German tax.

Receipt processing. A fit, but the tooling matters more than the template. Claude reads PDFs and screenshots fine. The template pattern (read, classify, file, flag) maps directly. The real work is wiring it to wherever your receipts live. Mine sit in three folders, two inboxes, and a stubborn pile of paper. Any agent has to deal with that mess before the prompt even matters. For my routine, see Solo Studio Bookkeeping in 90 Minutes a Month: My Stack and Routine.

One more thing worth saying out loud. Templates are not models. The same Claude template runs differently depending on which model you point it at, how much context you give it, and which tools it can call. Anthropic shipping a template does not skip the part where you test it on your own data, your own edge cases, your own rounding rules. It just skips the cold start.

The honest summary: the template pattern is right. The specific cookbooks are US-shaped. Solo operators outside the US should read the cookbooks for the agent design, then build their own version. That is still a 10x speedup over starting from a blank prompt.

Bottom line

Anthropic shipped Claude agent templates for finance because finance is one of the few jobs where the shape of the work is predictable enough that a generic template actually saves time. Pitches, valuations, month-end close. Three workflows that map onto a real Friday afternoon. The three install paths (Cowork plugin, Claude Code plugin, Managed Agent via cookbook) each fit a different team. Pick the one closest to where you already work.

For solo studios, do not wait for a German-tax cookbook that may never come. Read the financial-services templates, copy the agent shape, swap the data sources for your own. The pattern transfers cleanly. The data layer never does.

If you want a hub for the rest of the Claude agent coverage on this blog, the Claude Code MOC collects the SDK, Managed Agents, and tooling pieces in one place. Start there, then come back here when Anthropic ships the next vertical pack.