What this is
50,000 GitHub stars and over 6,000 forks — goose, the AI agent open-sourced by Block (Square's parent), is making a play for Agent-era infrastructure. Open-sourced by Block in January 2025, goose writes code, drafts text, runs scripts, and gathers information. It's now a community-governed project under the Agentic AI Foundation at the Linux Foundation.
Two protocols are key to understanding goose. MCP (Model Context Protocol, an interface standard pushed by Anthropic that lets agents call external tools and data) acts like a universal socket for agents, letting them plug into databases, GitHub, browsers, and other external systems; goose currently integrates over 70 official extensions. On the other side, goose isn't locked to a single model — Anthropic, OpenAI, Google, or a local Ollama can all serve as its brain, and the companion ACP protocol lets editors like VS Code and JetBrains call goose directly.
Industry view
On the bullish side, we note that Block CTO Dhanji Prasanna has publicly said open-sourcing goose is meant to take invention and growth to a new level — and it's unusual for a payments company to give away a core efficiency tool for free. 50,000 stars is a hard metric in engineering circles, proof that real demand exists for this direction.
But reservations are worth airing. First, the real hard part of Agents isn't execution — it's task planning and error recovery; goose solves the "general execution" layer, and engineering it for production still has plenty of pitfalls. Second, 50K stars isn't top-tier in open source (CNCF flagships like Kubernetes sit above 100K), so the real usage density in the developer community is still to be observed. Third, MCP is pushed by Anthropic; OpenAI and Google each have their own agent access schemes, and the standards war is far from settled — goose betting its fortunes on MCP means dependence on a single ecosystem path.
Impact on regular people
For enterprise IT: If your team is already evaluating or procuring Agent tools, whether a product supports the MCP protocol may matter more than which vendor you pick — it's an integration-cost issue; pick wrong once and you'll redo it later.
For individual careers: Tools like goose don't lower the bar of "whether you can code" — they lower the bar of "whether you're willing to spell out a task completely." People who can write clear instructions and break work into executable steps will capture the upside first.
For the consumer market: It's still far from regular users in the short term. But once Agent infrastructure is standardized, work that used to require outsourcing — batch-processing Excel files, auto-organizing invoices, "book my trip" — will get cheaper and more on-demand.