Block's Goose agent handles 90% of code

Goose signals a shift from "AI chat" to "AI doing" by putting a persistent, tool-enabled agent directly on the developer's machine.

• –Local execution ensures data privacy and gives the agent direct filesystem and terminal access, enabling it to solve end-to-end engineering problems autonomously.
• –Native support for the Model Context Protocol (MCP) allows Goose to connect with an expansive ecosystem of third-party tools and data sources.
• –Its model-agnostic architecture lets users choose between premium cloud models like Claude 3.5 Sonnet or fully local models via Ollama to balance cost and performance.
• –Transitioning the project to Linux Foundation stewardship demonstrates Block's commitment to building a standard, community-driven piece of open-source infrastructure.
• –Already deployed to over 4,000 engineers at Block, the tool has proven its value by automating up to 90% of code for certain routine engineering tasks.