GitHub hosts Angie Jones, VP of Engineering at Block, who shares insights on Goose, their open source AI agent, the MCP standard, and supporting the next generation of AI builders.

Angie Jones on Goose, MCP, and AI Agent Innovation

In this episode of The GitHub Podcast, Abigail Cabunoc Mayes interviews Angie Jones, the VP of Engineering at Block, live from GitHub Universe. The discussion centers on Goose, Block’s open source AI agent and reference implementation for the Model Context Protocol (MCP).

Key Topics Covered

  • Goose – from Internal Tool to Open Source: Angie details how Goose transitioned from solving internal challenges at Block to empowering a global open source community. Community contributions have shaped features such as multimodel support.
  • Model Context Protocol (MCP): The importance of MCP as a standard for AI agents and how Goose acts as its reference implementation.
  • AI Beyond Engineering: Goose is now used by more than 12,000 Block employees in over 15 job roles, from engineering to sales, expanding the scope and value of AI-powered agents.
  • Practical Use Cases: The discussion focuses on non-hype, real-world use cases like detecting student stress, segmenting over 80,000 sales leads, and triaging open source project issues.
  • Trust and Control: Angie explains challenges and solutions around trusting AI agents with access to sensitive codebases.
  • Developers and the AI Hype Curve: Reflections on why many developers are skeptical of flashy AI demos and how practical applications are winning trust.
  • ‘Vibe Coding’: The phenomenon of non-traditional contributors (like salespeople) using Goose to build features on the go.
  • AI Builder Fellowship: Angie talks about her new fellowship to empower and mentor emerging ‘native AI builders.’

Other tools and initiatives discussed:

About the Guests

  • Angie Jones: VP of Engineering at Block, leader in AI and automation, open source advocate
  • Host: Abigail Cabunoc Mayes, with production by Victoria Marin and the GitHub Podcast team

Show Chapters

  • 00:00 - Live from GitHub Universe
  • 01:34 - What is Goose?
  • 03:20 - The MCP reference implementation
  • 05:11 - How non-engineers use agents at Block
  • 06:01 - Detecting student stress
  • 10:59 - Trusting AI with your codebase
  • 12:59 - Why devs are tired of flashy demos
  • 17:24 - Vibe coding on the train
  • 19:40 - The AI Builder Fellowship
  • 21:59 - Open source picks: Selenium & PyFluff

For more episodes and updates, follow the GitHub Podcast.