Anthropic dropped a repo on GitHub last week. No blog post, no tweet. Quiet as if it never happened.
But the numbers don't lie: anthropics/financial-services, 21,500 stars, 13,176 new this week. 2,900 forks.
The repo contains a set of Claude Agent templates for the financial industry — investment banking, equity research, private equity, wealth management. Four vertical scenarios, each with reference Agents, skill definitions, and data connectors. Installable as a Claude Cowork plugin, or deployable via the Claude Managed Agents API.
13k stars in a week. In Anthropic's official repos, this growth rate is second only to Claude Code itself.
This Isn't an Open Source Contribution — It's Seizing Industry Standards
If you think Anthropic is doing charity, you're being naive.
OpenAI has industry solutions too, but takes the enterprise customization route — one-on-one deals with large financial institutions. Anthropic chose a different approach: open-source the templates, let the community validate and spread them.
Why is this smart? When an investment bank builds its Agent workflow using these templates, switching to another model later isn't as simple as changing an API key. Your system prompts, skill orchestrations, and data connectors all need to be rewritten.
That's vendor lock-in, dressed up as open source.
The repo structure tells the story too: .claude-plugin, managed-agent-cookbooks, plugins, scripts — not a demo-level toy, but a proper engineering skeleton. 52 commits, most recent 4 days ago, actively maintained.
But there's a detail worth noting: 29 branches, 0 tags. No formal releases. For a template library claiming to be production-ready, that's a bit sloppy.
The New Battleground for Model Companies
Over the past year, model company competition has focused on two dimensions: capability and price. Now Anthropic opened a third: industry depth.
When you build Agent templates that an investment banking analyst can use directly, you're no longer selling API tokens. You're selling industry know-how. The margins on that are far higher than tokens, and once customers start using it, switching costs are enormous.
My Take
I'm bullish on this direction, not because the templates are that great, but because Anthropic figured something out: the gap in model capability is shrinking, but the gap in industry solution capability is widening.
When the differences between GPT-5, Claude Opus, and Gemini Ultra become harder to perceive, whoever embeds their model more deeply into specific industry workflows wins.
OpenAI will follow — probably. Google? Gemini's financial industry solutions are already in the works.
In the second half of 2026, we'll likely see a wave of "model company × industry template" launches.
When that happens, choosing a model won't just be about benchmarks. You'll need to ask — which model's template library fits your business best.
Main sources:
- GitHub: anthropics/financial-services (21.5k stars, 52 commits)
- Anthropic Claude Managed Agents API documentation
- GitHub Trending weekly data