Agentic AI
Block's BuilderBot: Managing AI Coding Agents at Scale via Slack
Block built BuilderBot on its open-source Goose framework to orchestrate AI coding agents across hundreds of services, running over 200,000 operations daily.
The New Stack
Block—the fintech company behind Square, Cash App, and Tidal—has developed an internal system for coordinating AI coding agents at scale that it's now sharing as a blueprint for other engineering organizations facing similar challenges.
How BuilderBot Works#
BuilderBot, built on Block's open-source Goose framework, allows engineers to spawn and manage coding agents directly from Slack. By tagging @builderbot in a thread, developers can delegate research, planning, and code generation while remaining in a live feedback loop. The system picks up tasks from Linear and Jira, creates branches, opens pull requests, monitors CI pipelines, and iterates on real-time feedback—eliminating context-switching overhead.
Key capability: BuilderBot understands Block's entire service catalog, APIs, and engineering conventions, letting developers modify services they've never touched before.
Scale and Impact#
The numbers are significant:
- 200,000+ operations daily
- ~1,500 pull requests merged weekly (roughly 15% of all production code changes)
- Timeline compression: Projects that once took months now take days
Brad Axen, Block's head of AI capabilities, describes it as "the missing layer between AI coding tools and how engineering actually works at scale." On the Square product side, engineers deployed features that had been backlogged for months in just days, with BuilderBot handling scaffolding and repetitive work while humans directed strategy.
Open-Source Foundations#
The system emerged from Block's two-year push to make AI "native" to its engineering practice. The company open-sourced Goose in January 2025 and later contributed it to the Agentic AI Foundation (formed by the Linux Foundation in December 2024), joining Anthropic's Model Context Protocol and OpenAI's AGENTS.md.
Integration challenges during Goose development led Block to co-develop MCP with Anthropic, now an industry standard adopted by OpenAI, Google, and others.
