Redefining AI Agent Orchestration Inside the Development Workflow
AI agent orchestration at Block is the coordinated management of many AI coding agents that plan, write, and ship code across hundreds of services, all controlled through Slack and tightly integrated with issue tracking, continuous integration, and a monorepo architecture. Instead of treating AI coding agents as isolated tools inside a single repository, Block built an orchestration layer, called Goose, and an application framework, called Builderbot, to connect them directly to everyday engineering workflows. Developers tag @builderbot in a Slack thread, describe the change, and the agent takes over research, planning, coding, and CI monitoring while humans supervise in real time. This turns the conversation itself into the development environment, so teams avoid context switching between chat, IDEs, and dashboards. The result is software development automation that operates across hundreds of millions of lines of code, not just one repo at a time.
Goose and Builderbot: A Slack-Native Orchestration Stack
Goose is Block’s internal AI agent orchestration framework; Builderbot is the application built on top of it that coordinates AI coding agents from Slack. Engineers initiate work by tagging @builderbot in a channel, often linked to tickets from Linear or Jira. Builderbot then creates branches, opens pull requests, tracks CI, and iterates on feedback without leaving the thread. According to Block, the system 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. Goose keeps these AI coding agents aligned with company-wide conventions, service boundaries, and permissions, so a Cash App engineer can safely modify a Square service they have never seen before. The orchestration layer gives teams a single place to manage agents, inspect progress, and intervene, turning Slack into the front door of enterprise automation infrastructure.
Builderbot as a Cross-Service Development Engine
Builderbot acts as a central coordination layer for software development automation across Block’s complex backend. The framework understands every internal API, service contract, and engineering convention spread over hundreds of services and hundreds of millions of lines of code. Standard single-repository coding assistants could not process this scale; they broke down when tasks spanned multiple repositories and databases. Builderbot instead maps the entire codebase so it can plan multi-step changes, from small bug fixes to wide architectural refactors. Engineers issue natural-language instructions in Slack, then watch as Builderbot performs research, drafts changes, runs tests, and responds to CI results in the same thread. Human supervisors can redirect the plan mid-flight, keeping the AI agents aligned with business context. This orchestration model compresses timelines: work that used to take months can be delivered in days because the system removes much of the coordination and mechanical grunt work between teams.
Monorepo Architecture and 8,800 Weekly Builds as a Force Multiplier
To fully unlock AI agent orchestration, Block consolidated about 450 JVM repositories into a single monorepo for Cash App and Square backend systems. Under the old polyrepo model, shared libraries and services lived in separate repos, which led to dependency drift, duplicated upgrade work, and occasional runtime failures from version mismatches. In the monorepo architecture, shared dependencies are resolved from source, and cross-service changes land as a single atomic commit. The new repository supports around 8,800 builds per week, with p90 CI times of about 10 minutes on a reliably green main branch. This predictable CI foundation allows Goose and Builderbot to operate at scale, safely driving large numbers of automated builds and pull requests. Engineers benefit from faster feedback, more cohesive IDE workflows, and simpler coordination, while AI coding agents gain a consistent, global view of dependencies and build graphs across the entire codebase.
Enterprise Automation Infrastructure and Real-Time Agent Management
Block’s approach shows what enterprise automation infrastructure looks like when AI coding agents are first-class actors in the development process. Goose provides real-time agent management, routing tasks to specialized agents, tracking their progress, and coordinating responses in shared Slack threads across distributed teams. Builderbot is deeply integrated with issue trackers, CI pipelines, and source control, so automation is not a sidecar—it is embedded in how work flows from ticket to deployment. Engineers can observe, steer, or pause agents as they modify code across multiple services, which keeps humans in control while delegating routine mechanics to machines. With thousands of weekly builds, coordinated through a monorepo and an AI-aware orchestration layer, Block has turned its chat system into a central command console for software development automation. The result is a blueprint for organizations that want to scale AI agent orchestration beyond single repositories and into whole engineering ecosystems.
