Defining Block’s Unified AI Agent Orchestration Strategy
Block’s unified AI agent orchestration strategy is a software development automation approach that combines a centralized coding agent framework, a Slack-based workflow, and a large-scale monorepo architecture so AI agents can execute coordinated changes across hundreds of services and repositories as a single, production-grade system. Instead of treating AI coding assistants as isolated tools inside individual projects, Block built Goose and its internal Builderbot framework as an orchestration layer that understands company-wide code, APIs, and conventions. Builderbot coordinates multiple agents to research, plan, implement, and ship changes that span different products and backends. By pairing this framework with a consolidated JVM monorepo, Block gives its agents a single, coherent view of dependencies and build pipelines. The result is a platform where AI is embedded directly in day‑to‑day engineering practice, not bolted on as a point solution.
From Polyrepo Pain to a JVM Monorepo Backbone
Before adopting monorepo architecture, Block’s JVM ecosystem lived in roughly 450 separate repositories, with services and shared libraries evolving on their own timelines. Over time, dependency drift, duplicated upgrade work, and diamond dependency surprises made coordination harder and riskier. Engineering leaders report that rolling out a breaking library change across downstream services often required “heroic effort.” The migration to a single JVM monorepo created one shared codebase where dependencies are resolved from source and cross-service changes can ship as atomic commits. The new repository now supports about 8,800 builds per week, with p90 continuous integration times around 10 minutes on a consistently green main branch. To keep the experience practical at this scale, Block invested in an IntelliJ plugin that loads only relevant projects, merge queues to protect main, and dependency-graph-aware build scoping so engineers and AI agents do not rebuild the world for every change.
Goose and Builderbot: A Coding Agent Framework for the Entire Stack
On top of Goose, Block built Builderbot as a coding agent framework designed for enterprise AI infrastructure rather than a single-repo assistant. Builderbot acts as a central orchestration layer that coordinates multiple specialized agents across the company’s codebase. It maps every internal service and API and encodes engineering conventions so agents can work safely across products like Cash App and Square. According to Block, the system runs more than 200,000 operations a day and now merges about 1,500 pull requests a week, or roughly 15% of all production code changes. Builderbot does more than generate code: it picks up tickets from tools such as Linear and Jira, creates branches, opens pull requests, monitors CI, and iterates on feedback. By operating solely on source code and configuration, without touching customer or payment data, it aligns software development automation with strict production and compliance boundaries.
Slack as the Control Plane for AI Agent Orchestration
Block’s most visible shift is turning Slack into the control plane for AI agent orchestration. Engineers trigger Builderbot by tagging @builderbot inside a channel or thread and describing the desired change in plain language. The agent replies in the same thread, carries out research, drafts plans, edits code, and responds to feedback while multiple teammates observe and guide the work. The conversation becomes the development environment, replacing much of the back‑and‑forth between chat, IDE, and CI dashboards. Builderbot can handle tasks that range from small bug fixes to large, cross-database migrations, coordinating multiple agents under the hood. Because it has the necessary permissions and context, a Cash App engineer can request changes in a Square backend without knowing its internals. This Slack-first workflow makes AI agents feel like standard collaborators embedded directly in daily engineering practice.
What Block’s Approach Signals for Enterprise AI Infrastructure
Block’s combination of monorepo architecture and AI agent orchestration hints at a new pattern for enterprise AI infrastructure. Instead of scattering small assistants across tools, it treats AI as a first-class execution layer that depends on clean dependency graphs, fast CI, and shared conventions. Builderbot’s deployment shows that AI agents can operate at production scale, with total adoption by engineers and a material share of weekly pull requests, when they fit into existing workflows and issue trackers. Engineering leaders described the monorepo migration as a “step-change in developer experience,” but it also gave Builderbot a coherent substrate for cross-service work. For organizations wrestling with hundreds of services, Block’s model suggests that successful AI coding systems will be less about a single powerful model and more about the surrounding frameworks, automation, and repository design that let agents act across the entire software estate.
