Anthropic open-sources 12 agents

Anthropic just made a pretty aggressive move into finance software. On May 5, it released a set of ready-to-run agent templates for financial-services work and published the underlying reference agents, skills, and connectors in an open GitHub repo. The pitch is simple — don’t start from a blank prompt if you want AI to help with pitchbooks, models, KYC, reconciliations, or research. Start from a workflow that already knows the job. ### What actually got released? Two things at once. First, Anthropic announced 10 new agent templates for financial services. Second, it pointed developers to a public repository called `financial-services`, which packages those workflows as installable plugins for Claude Cowork and Claude Code, plus cookbooks for Claude Managed Agents. So this is not just a blog post with examples — it is meant to be deployed. ### Why are people calling them “open-source agents”? (anthropic.com) Because the repo is public, Apache-2.0 licensed, and structured like reusable production scaffolding rather than a toy demo. Anthropic describes it as “reference agents, skills, and data connectors” for investment banking, equity research, private equity, and wealth management. In plain English, the company is open-sourcing the workflow logic around the model — prompts, task decomposition, tool hooks, and deployment templates — even though the Claude models themselves are still proprietary. ### What do these agents do? The announced templates cover a very bank-shaped list of chores. On the front-office side, there are agents for pitch building, meeting prep, earnings review, model building, and market research. On the finance and operations side, there are agents for valuation review, general-ledger reconciliation, month-end close, statement auditing, and KYC screening. Anthropic frames each one as a starting point that firms can adapt to their own approval rules and modeling conventions. (github.com) ### Why does that matter more than another chatbot? Because finance work is mostly process, not just prose. A useful banking or research agent has to pull the right data, follow a repeatable checklist, hand work from one step to another, and leave something a human can review. Anthropic’s templates bundle three pieces for that: skills, connectors, and subagents. Basically, the company is productizing the boring but crucial part — how an agent actually moves through a workflow without losing context. (anthropic.com) ### What’s the Microsoft 365 angle? Anthropic also tied the release to new Microsoft 365 add-ins for Excel, PowerPoint, Word, and eventually Outlook. That matters because a lot of finance work lives inside those apps. The company’s pitch is that context can move from spreadsheet to deck to memo without the analyst re-explaining everything every time. If that works well, the agent stops feeling like a sidecar chatbot and starts looking more like a junior operator living inside the actual work surface. (anthropic.com) ### Are these fully autonomous? No — and Anthropic is explicit about that. The repo says the agents draft work product for review by qualified professionals. They do not make investment recommendations, execute transactions, post to a ledger, or approve onboarding. That’s a big tell. Anthropic wants the upside of automation in regulated workflows, but without pretending the model can own the final decision. ### So why is this a bigger deal than the raw number? (anthropic.com) Because the count is almost beside the point. Whether you call it 10 templates, 12 agents, or a larger stack of agents plus plugins, the real shift is that Anthropic is publishing reusable workflow assets that other teams can fork, test, and port. You can already see that happening — a separate GitHub project has started adapting the same skill files to Microsoft Foundry and Azure OpenAI. That is how an internal product pattern turns into an ecosystem pattern. (github.com) ### Bottom line? This is Anthropic trying to make agents feel less magical and more operational. The models still matter, but the moat is creeping upward into workflows, connectors, and reviewable outputs. For finance teams, that is the difference between “interesting demo” and “maybe we can actually use this on Monday.” (anthropic.com) (github.com)