Angie Jones on Goose, MCP, and the future of AI agents | Episode 9 | The GitHub Podcast

Key Concepts

• Goose: Block’s open-source AI agent, developed as a reference implementation of MCP (Multi-Party Computation).
• MCP (Multi-Party Computation): An open protocol enabling interoperability in the AI space, allowing agents to interact securely.
• AI Agents: Software entities capable of performing tasks autonomously, often leveraging Large Language Models (LLMs).
• Agent HQ (GitHub): GitHub’s recently released platform for building and deploying AI agents.
• DevRel (Developer Relations): The practice of building communities and fostering relationships with developers.
• Non-Deterministic AI: The inherent unpredictability of AI outputs, requiring careful control and context.
• AI Builder Fellowship (Block): A six-month program focused on training early-career developers in AI-assisted software development.
• Py Fluff: An open-source Python port of Bloflaft, enabling Furby Connect programming.

Introduction & Background

The podcast episode features Abby from GitHub’s Open Source Programs team interviewing Angie Jones, VP of Engineering at Block (Square and Cash App). The conversation centers around Block’s open-source AI agent, Goose, its relationship to the MCP protocol, and the broader implications of AI for software development and beyond. Angie emphasizes the importance of practical application and community involvement in shaping the future of AI tools.

Goose: Origins and Functionality

Goose was initially developed as the reference implementation for Anthropic’s MCP protocol in January 2025. Block decided to open-source Goose after realizing the internal benefits it provided, aiming to share those advantages with the wider community. While Block is known for financial technology (Square and Cash App), Goose represents their commitment to open-source development tools, alongside projects like okhttp. Goose is designed to be a highly permissive open-source project, actively accepting contributions from the community to drive its evolution.

MCP and the Future of AI Interoperability

Angie and Abby both express strong support for MCP, highlighting its potential to foster interoperability in the AI landscape, drawing parallels to successful open protocols of the past. They note the rapid adoption of MCP, with over 80,000 stars on GitHub in a short timeframe, demonstrating strong community interest. GitHub’s Agent HQ is positioned as building upon the foundation laid by Goose, acknowledging its pioneering role in multi-agent systems.

Real-World Applications of Goose

Goose is being utilized extensively within Block itself, across 15+ job functions and by all 12,000 employees. Examples discussed include:

• Resilient Coders Project (Leon): Goose monitors students’ physiological signals (blood pressure, heart rate) during coding exercises to detect frustration and provide assistance. It also tracks common sticking points to inform instructors.
• Sales Lead Segmentation: Goose processed 80,000 leads from a conference in an hour, a task that would have taken a week manually.
• Feature Development: A salesperson prototyped a customer-requested feature on a train ride using Goose, accelerating the development process.
• Internal Issue Triage: Goose is being used to triage issues and pull requests within the Goose repository itself.

The Role of Open Source and Community Input

Angie stresses the importance of open source in shaping the direction of AI development. Goose’s open-source nature allows the community to voice their needs and demand specific features, such as multi-modal support. The community’s contributions are driving the evolution of Goose, with patterns often appearing first in Goose before being adopted by larger clients. This collaborative approach ensures that AI tools are aligned with developer needs.

Transforming Workflows with AI Agents

Angie envisions AI agents as providing “extra hands” for workers, enabling multitasking and accelerating processes. She describes engineers using agents to automate tasks like test writing and documentation while focusing on core coding responsibilities. She cautions against over-reliance on AI-generated code, advocating for developers to retain ownership of the parts of their jobs they enjoy. The focus should be on leveraging AI to alleviate tedious tasks and enhance productivity.

Balancing Capability and Control

Addressing concerns about granting AI access to sensitive data, Angie acknowledges the trust issue and the non-deterministic nature of AI. She emphasizes the importance of control mechanisms like rules files and context provision to guide agents effectively. Gradual granting of permissions, based on observed performance, is recommended.

Block’s AI Builder Fellowship

Block is launching an AI Builder Fellowship to cultivate the next generation of AI developers. The program prioritizes practical skills and experience over traditional credentials, evaluating applicants based on projects they’ve built using AI tools. The six-month fellowship will embed participants in real-world projects, providing hands-on experience and potential full-time employment opportunities.

Open Source Shout-Outs

• Selenium: Angie highlights Selenium as an unsung hero of the open-source world, praising its 20-year longevity and volunteer-driven development.
• Py Fluff: Abby shares Martin Woodward’s open-source project, Py Fluff, which allows users to program their Furby Connect.

Conclusion

The conversation underscores the transformative potential of AI agents, particularly when coupled with open-source principles and community collaboration. Goose serves as a pioneering example of how AI can be integrated into various workflows, from software development to sales and education. Angie Jones advocates for a pragmatic approach to AI adoption, emphasizing the importance of control, context, and human oversight. The episode highlights Block’s commitment to fostering the next generation of AI builders through initiatives like the AI Builder Fellowship. The key takeaway is that the future of AI is not about replacing humans, but about augmenting their capabilities and empowering them to achieve more.