Defining Block’s AI Agent Orchestration Stack
Block’s AI agent orchestration stack is an internal infrastructure that coordinates multiple AI coding agents, connects them to development tools, and automates software delivery steps across many services from within Slack. It combines Goose, an open-source framework for orchestrating agents, with Builderbot, a higher-level system that turns chat requests into branches, pull requests, and CI runs. Together, they target automated software development in a codebase spanning hundreds of services and hundreds of millions of lines of code. According to Block, the system now runs more than 200,000 operations a day and merges about 1,500 pull requests a week, accounting for roughly 15% of all production code changes. By focusing the agents on source code and configuration instead of customer data, Block has built an AI-assisted pipeline that is powerful but still constrained to engineering artifacts.
Goose and Builderbot: From Coding Assistant to Orchestration Layer
Standard AI coding assistants tend to work on a single repository and inside an IDE. Goose and Builderbot were designed to go further by turning AI coding agents into a coordinated fleet that understands the whole estate of services, APIs, and engineering conventions. Goose provides the underlying AI agent orchestration, while Builderbot acts as the application layer that maps human requests onto complex development workflows. Builderbot does the research, planning, and code changes, then carries work through CI and review. Block describes Builderbot as “the missing layer between AI coding tools and how engineering actually works at scale,” because it can modify any repository the company manages. That means an engineer focused on Cash App can trigger safe changes in a Square backend service they have never worked on, with the orchestration system supplying the context.
Slack as the Control Plane for Developer Workflow Automation
In Block’s model, Slack is the control plane for developer workflow automation. Engineers summon Builderbot by tagging @builderbot in a thread and describing the desired work: fixing a bug, wiring a new feature across services, or executing a multi-database migration. From there, the AI agents stay fully within the same conversation, turning the thread into the development environment. Builderbot picks up tickets from Linear and Jira, creates branches, opens pull requests, watches CI, and iterates on feedback in that channel. Multiple people can watch and steer the process without switching tools, which cuts the context-switching that usually comes from juggling chat, issue trackers, and IDEs. The agents’ understanding of company-wide code and conventions makes this Slack-driven loop more than chatops; it is a live orchestration layer that keeps developers in one place while the system handles repetitive delivery steps.
Monorepo Foundations: 450 JVM Repos and 8,800 Weekly Builds
The scale of Block’s AI agent orchestration depends on a foundation that makes cross-service changes predictable: the migration of about 450 JVM repositories into a single monorepo for Cash App and Square. The previous polyrepo approach allowed teams to move fast but led to dependency drift, duplicated upgrades, and runtime failures when shared libraries changed. In the monorepo, AI coding agents can apply atomic updates across many services in one commit and resolve dependencies directly from source. Block reports that the monorepo now supports around 8,800 builds per week, with p90 CI times of roughly 10 minutes on a reliably green main branch. According to engineering leadership, what began as a painful migration ended up as “a modern, cohesive codebase with an optimized IDE workflow” and a base for long-term velocity, which in turn makes automated software development through Goose and Builderbot much safer.
