What AI Agent Orchestration Means in Real Software Development
AI agent orchestration in software development is the practice of coordinating multiple specialised AI coding agents through a central control layer so they can understand large codebases, interact with developer tools, and execute end-to-end engineering workflows that span planning, coding, testing, and deployment while humans supervise the work in real time. Block’s internal Builderbot system is a concrete example of this idea. Built on its open-source Goose framework, Builderbot acts as a control tower for AI coding agents that operate across hundreds of services and hundreds of millions of lines of code. Instead of living inside a single repository or IDE plugin, the orchestration layer sits on top of existing tools and repositories, turning everyday conversations into programmable workflows. For enterprise teams, this shows that AI agent orchestration is less about one smart bot and more about a coordinated system that plugs into the existing development environment.
Inside Builderbot: A Central AI Orchestration Layer Running from Slack
Block’s engineers interact with their AI coding agents from a place they already live in: Slack. By tagging @builderbot in a thread and describing the task, they trigger a full software development automation sequence without leaving the chat. Builderbot then manages research, planning, and code generation directly in that thread, turning it into a shared execution space where multiple humans can review and steer the work. According to Block, this “conversation becomes the development environment,” reducing context switching between chat, IDE, and ticketing tools. Under the hood, Goose-powered agents coordinate a stream of actions: picking up issues from Linear and Jira, creating branches, opening pull requests, watching CI, and iterating on failures or review comments. This is AI agent orchestration as a living process: Slack becomes the command console, Goose provides the agent framework deployment mechanism, and Builderbot coordinates the workflow across the codebase.
From Single-Repo Helpers to Cross-Service Software Development Automation
Block’s experience highlights the limits of typical AI coding assistants that stay inside a single repository. In a mature environment with hundreds of interconnected services, those tools could not handle cross-system changes or reason over the entire architecture. Builderbot fills that gap by actively mapping the complete structural context of Block’s codebase, documenting every service, internal API, and engineering convention. It has the permissions and understanding to modify any repository, whether it belongs to Cash App or a Square backend. As Block describes it, a Cash App engineer can ask for a change in a Square service they have never touched, and the AI agent orchestration layer fills in the missing context. This turns high-friction, cross-service work—such as multi-database migrations or coordinated refactors—into tasks that agents can plan and execute under human direction, instead of threads of manual handoffs across teams.
Autonomous Workflow: Tickets, CI, and Pull Requests without Manual Glue
Where Builderbot stands out for enterprise AI agents is its end-to-end workflow ownership. Once triggered, Builderbot retrieves tickets from Linear or Jira, claims the work, creates a feature branch, and generates the needed source code. It then opens a pull request, monitors continuous integration, and keeps iterating on failures or human review comments until the change is production-ready. Humans stay in the loop for judgment, not for repetitive typing. Block also constrained the system’s reach: Builderbot works only with source code and system configurations, and cannot access customer data or payment information. One quotable statement from Block’s deployment underscores the impact: “The system runs more than 200,000 operations a day and merges about 1,500 pull requests a week, or roughly 15% of all production code changes across Block.” Those numbers show that agent framework deployment can become a significant share of real production output, not a side experiment.
What Enterprise Teams Can Learn from Block’s AI Agent Strategy
Block’s rollout offers a playbook for enterprises exploring AI agent orchestration. First, embed agent control where people already work—in this case Slack—so developers can summon AI coding agents inline, collaborate on decisions, and avoid tool fatigue. Second, treat orchestration as a central infrastructure layer, not a single assistant, with an agent framework deployment that understands services, APIs, and conventions across the entire organisation. Third, integrate the agents into existing systems of record, such as issue trackers and CI pipelines, so software development automation mirrors current processes rather than replacing them outright. Finally, adopt clear security boundaries: Builderbot manipulates code and configs while remaining isolated from production customer data. As Block’s head of AI capabilities Brad Axen puts it, Builderbot is “the missing layer between AI coding tools and how engineering actually works at scale,” a description many enterprise engineering leaders will recognise in their own environments.
