What AI agent orchestration means at Block
AI agent orchestration at Block is the coordinated control of multiple coding agents, integrated into developer chat and build systems, to plan, modify, test, and ship code across hundreds of services and repositories in a unified workflow. Block’s approach turns Slack into a command center for software development automation, connecting engineers to a fleet of agents that understand the company’s entire codebase. When developers tag @builderbot in a Slack thread, the system kicks off research, planning, and coding within that conversation, instead of inside a traditional IDE. Builderbot builds on the open-source Goose framework, which provides a shared foundation for these coding agents. This orchestration layer bridges AI tools and real engineering practices, allowing work to span many repositories and services without losing context, while still keeping sensitive customer and payment data outside the agents’ reach.
Goose, Builderbot, and Slack as the new IDE
Block’s Goose framework underpins Builderbot, a central AI agent that coordinates specialized coding agents across the company’s infrastructure. Engineers initiate tasks by describing the problem in Slack and tagging @builderbot, which then pulls tickets from systems like Linear and Jira, creates branches, opens pull requests, and watches CI. The conversation becomes the development environment: multiple teammates can follow the agent’s reasoning, steer decisions, and inspect changes without leaving chat. According to Block, the system runs more than 200,000 operations per day and merges about 1,500 pull requests a week, accounting for roughly 15% of all production code changes. Builderbot understands internal APIs, services, and conventions, so a Cash App engineer can safely make changes in a Square backend service they have never seen before, while the orchestration layer manages permissions and standards.
From single-repo assistants to cross-service software development automation
Traditional coding assistants work inside one repository and struggle with the complexity of hundreds of interconnected services and hundreds of millions of lines of code. Block’s Builderbot framework treats this as a systems problem, acting as a central orchestration layer for AI-powered software development automation. When a developer requests a change, Builderbot decomposes the work into steps across services, coordinates multiple agents, and performs operations that range from bug fixes to large architectural migrations touching several databases. Its context model maps every service, API, and engineering convention, giving it the authority and understanding to modify any repository in the estate. This is how Block turned AI from a code-completion helper into an operational engine that can execute real cross-system changes end to end, at scale, while still keeping humans in the loop through Slack-based guidance and review.
Why migrating 450 JVM repositories into a monorepo mattered
Block’s AI agents sit atop a major architectural shift: consolidating about 450 JVM-based repositories into a single monorepo for Cash App and Square backend systems. Under the previous polyrepo model, dependency drift, duplicated upgrade efforts, and diamond dependencies created frequent breaking changes and runtime failures. The monorepo allows atomic updates across services in a single commit and resolves shared dependencies directly from source. According to Block’s engineering team, the new monorepo supports around 8,800 builds per week with p90 CI times near 10 minutes on a reliably green main branch. To keep this scale workable, Block built a custom IntelliJ plugin that loads only relevant projects, adopted merge queues, and invested in shared Gradle plugins and dependency-graph-based build scoping. This cleaner foundation makes it far easier for AI coding agents to reason about the codebase and perform cross-service work.
Production-scale AI agent orchestration and what comes next
Builderbot now operates as a production system rather than an experiment, touching a significant share of Block’s code changes and integrating tightly with CI, issue trackers, and the new monorepo architecture. The orchestration stack supports more than 200,000 operations per day and around 1,500 weekly pull requests, underlining that AI coding agents can handle real, multi-service development work when given consistent structure and authority. For developers, this means less time spent on mechanical tasks like scaffolding, boilerplate, and dependency upgrades, and more time guiding higher-level design choices inside Slack threads. Block presents this stack—Goose, Builderbot, Slack workflows, and monorepo architecture—as a reference pattern for organizations that need AI agent orchestration at scale, where the challenge is not only writing code, but coordinating many agents and humans across thousands of builds and repositories.
