From Fragmented Repos to AI-Orchestrated Engineering
Block’s infrastructure transformation combines a massive monorepo migration with AI coding agents orchestrated through Slack, creating a single, automated system for enterprise-scale software development automation. Instead of treating repository consolidation and AI adoption as separate projects, Block tied them together: a unified codebase plus a fleet of coordinated agents that understand it end to end. The company behind Square and Cash App describes this as a new operational layer where issue tracking, code changes, and CI all converge. Standard coding assistants could not cope with hundreds of services and hundreds of millions of lines of proprietary code, especially when tasks spanned many backends. Block responded by rethinking its core engineering stack so that monorepo migration, agent orchestration, and conversational workflows in Slack form one continuous path from ticket to production.
Monorepo Migration: 450 JVM Repositories into One
Block consolidated about 450 JVM-based repositories for Cash App and Square into a single monorepo to cut coordination overhead and dependency drift. Under the earlier polyrepo model, teams owned separate services and shared libraries, which gave autonomy but led to diamond dependency problems, duplicated upgrade work, and runtime failures when versions diverged. In the monorepo, all services share one codebase and dependency graph, allowing atomic, cross-service updates in a single commit instead of multi-team rollouts. According to InfoQ, the monorepo “now supports approximately 8,800 builds per week, with p90 CI times of around 10 minutes on a reliably green main branch.” To keep this sustainable, Block built custom IntelliJ tooling to load only needed projects, adopted merge queues to protect main under heavy commit volume, and tuned Gradle builds and git performance to handle the growing repository.
Builderbot: AI Coding Agents as a Central Orchestration Layer
With the monorepo in place, Block deployed the Builderbot framework across its codebase to automate complex cross-service work that traditional tools could not handle. Standard single-repository assistants broke down when faced with hundreds of interconnected services, so Builderbot was designed as an orchestration layer that coordinates many AI coding agents across the entire codebase. Engineers trigger it by tagging @builderbot in Slack and describing the task, from fixing bugs to large architectural migrations across databases or services. Builderbot then performs research, plans changes, edits code, and keeps working as humans steer in the same Slack thread. Block says the system now runs more than 200,000 operations per day and merges about 1,500 pull requests per week, representing roughly 15% of production code changes. Work that once took months can now be completed in days, transforming day-to-day software development automation.
Goose and Slack: Managing a Fleet of AI Agents
Builderbot is built on Block’s open-source Goose framework, which handles low-level agent orchestration while Slack acts as the main control plane. Goose coordinates multiple agents that each specialize in parts of the workflow: reading context, planning changes, editing code, and reacting to CI feedback. When a developer tags @builderbot, the system can pick up tickets from Linear or Jira, create a branch, open a pull request, monitor CI pipelines, and iterate on review comments—all inside one Slack thread. The conversation becomes the development environment and reduces context switching between chat, IDE, and dashboards. Because Builderbot understands every service, API, and engineering convention, a Cash App engineer can safely modify a Square backend they have never touched. At the same time, Builderbot only operates on source code and configurations, not customer data, which keeps the agent orchestration layer focused on code-centric tasks.
From Pilot to Production-Grade Enterprise AI Infrastructure
What makes Block’s approach stand out is that AI coding agents are wired into core engineering infrastructure rather than left in experimental pilots. The monorepo’s shared dependency graph, merge queues, and IDE tooling give Builderbot a stable, consistent view of the system. In turn, Builderbot’s continuous operation at a scale of 1,500 merged pull requests per week and 8,800 weekly builds shows production-grade reliability for AI coding agents, not just point-in-time demos. According to Block’s head of AI capabilities Brad Axen, Builderbot is “the missing layer between AI coding tools and how engineering actually works at scale.” By unifying monorepo migration with agent orchestration over Goose and Slack, Block has built an enterprise AI infrastructure where issue trackers, CI, and conversational agents form a single, automated loop from idea to deployed change.
