How Block manages its fleet of AI coding agents from Slack
Most AI coding tools work well inside a single repository. But few of them can operate across hundreds of services and hundreds of millions of lines of code. To make these coding agents for its needs, Block (the company behind Square, Cash App, and — for some reason — Tidal) developed BuilderBot, a tool on top of its open-source Goose framework, that lets engineers manage a fleet of AI coding agents from a single Slack thread by tagging @builderbot.
The company is now talking about this internal tool, it says, to provide a blueprint to other developers who likely face similar challenges.
When Block’s developers tag @builderbot, it gets to work in the thread and handles the research, planning, and coding, all while the developers steer in real time. Block says BuilderBot picks up tickets from Linear and Jira, creates a branch, opens a pull request, watches CI, and iterates on the feedback. This ensures that nobody has to switch context, because, as Block puts it, the conversation becomes the development environment.
BuilderBot also understands every service, API, and convention across the company, so an engineer working on Cash App can make a change in a Square service they have never touched. That’s what sets it apart from an ordinary coding assistant, Block argues.
As the company stresses, the tool operates solely on source code and system configurations, never on customer data or payment information.
Currently, 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. Work that used to take months now takes days, the company says.
Brad Axen, Block’s head of AI capabilities, says BuilderBot is “the missing layer between AI coding tools and how engineering actually works at scale.” On the Square side, he says, engineers pulled a list of features sellers had been waiting on for months and shipped them in days, with BuilderBot handling the scaffolding and the repetitive work while people made the calls that shaped the product.
All of Block’s engineers now regularly use AI, the company says, which is the result of a two-year push to make AI “native” to how Block builds.
That push included open-sourcing Goose in January 2025. Block later contributed Goose to the Agentic AI Foundation, which the Linux Foundation formed in December 2025 and which is also home to Anthropic’s Model Context Protocoland OpenAI’s AGENTS.md.
The integration problems Block ran into building Goose are what led it to co-develop MCP with Anthropic, now a standard for connecting agents to tools that OpenAI, Google, and others have adopted.
Block is doing all of this with a far smaller team than it had a year ago. In February, it laid off more than 4,000 people, over 40% of its workforce, in a restructuring CEO Jack Dorsey tied to the company’s bet on AI and agents. He told shareholders he expected most companies to reach the same conclusion within a year.
BuilderBot itself is staying inside Block; it is not a product anyone else can buy. The company is instead putting out the argument behind it: that the shift from AI-assisted coding to AI-native engineering is one of the bigger changes in how software gets built right now. The numbers behind that argument, for now, are Block’s own.
