What Block’s AI Coding Agents and Monorepo Strategy Solve
Block’s approach to AI coding agents combines a Slack-controlled orchestration layer with a large-scale monorepo migration so that automated software development can work reliably across hundreds of interconnected services and millions of lines of code. Instead of limiting agents to a single repository, Block built Goose-based agents that understand company-wide services, APIs, and conventions, and then removed coordination bottlenecks by consolidating about 450 JVM repositories into one shared codebase. This dual strategy targets the chronic problems of dependency management, cross-team coordination, and context switching that plague complex distributed systems. Engineers interact with AI coding agents inside Slack while the monorepo structure keeps dependencies aligned and builds predictable. Together, they show how agent orchestration and monorepo migration can reinforce each other and turn AI-generated changes into dependable production code at scale.
Goose and Builderbot: Agent Orchestration from a Slack Thread
Block’s Goose framework underpins Builderbot, a system that lets engineers run a fleet of AI coding agents from a single Slack thread. Tagging @builderbot kicks off agent orchestration: the system researches, plans, writes code, creates a branch, opens a pull request, watches CI, and iterates on feedback without leaving the conversation. The thread becomes the development environment, so there is no context switching between chat, IDE, and ticketing systems. Builderbot integrates with Linear and Jira to pull tickets and understands every internal service and API, so a Cash App engineer can change a Square service they have never touched. According to Block, the system runs more than 200,000 operations a day and merges about 1,500 pull requests a week, which is roughly 15% of all production code changes across the company.
Builderbot as a Central Layer for Automated Software Development
Builderbot is more than a code assistant; it is a central agent orchestration layer for automated software development across Block’s full codebase. Standard single-repository tools could not manage the complexity of hundreds of interconnected services and hundreds of millions of lines of proprietary code. In response, Block built Builderbot to coordinate multiple AI coding agents that handle tasks ranging from bug fixes to multi-database migrations. Engineers launch operations with a short Slack message, then guide the agents as they research, plan, and apply changes in real time. Builderbot has the permissions and context to modify any repository, mapping every network service, internal API endpoint, and engineering convention so that cross-service work becomes routine. Brad Axen, Block’s head of AI capabilities, described Builderbot as “the missing layer between AI coding tools and how engineering actually works at scale,” highlighting its role in shipping long-awaited features faster.
Monorepo Migration: 450 JVM Repositories, One Codebase
To make AI-driven changes safe and consistent, Block also restructured its backend into a single JVM monorepo. The company migrated about 450 repositories supporting Cash App and Square into one codebase to fix mounting dependency drift and coordination overhead. Under the old polyrepo setup, separate repositories for services and shared libraries created duplicated upgrade work, diamond dependency problems, and runtime incompatibilities that made dependency management “unmanageable,” according to engineering leadership. In the monorepo, shared dependencies are resolved directly from source, and cross-service updates happen atomically in a single commit instead of many coordinated releases. The new repository now supports roughly 8,800 builds per week, with p90 CI times around 10 minutes and a reliably green main branch. Senior Engineering Manager Gabor Pap called the migration “a step-change in developer experience,” citing a modern, cohesive codebase and faster CI as key outcomes.
End-to-End Efficiency: Agents on Top of a Stable Monorepo
The most interesting part of Block’s story is how AI coding agents and monorepo migration reinforce each other. Builderbot’s agent orchestration from Slack cuts human coordination cost: tickets feed straight into agents that plan, code, and respond to CI feedback, with human steering kept inside one thread. Meanwhile, the monorepo removes many structural hazards that would otherwise trip up large-scale automation, such as dependency version drift and cross-repository deployment choreography. Build scoping based on a dependency graph, shared Gradle plugins, merge queues, and a custom IntelliJ plugin keep the large codebase usable while agents drive a growing share of changes. In this environment, AI coding agents can operate across services with fewer surprises, and automated software development becomes a practical default rather than a fragile experiment, turning structural consolidation and intelligent agent orchestration into a single continuous delivery system.
