Artificial intelligence is already changing how software gets built, but Block just shared a glimpse into how far that shift has gone inside its own walls.
The fintech company behind Cash App and Square has unveiled Builderbot, an internal AI system that it says is now responsible for roughly 15 percent of all production code changes across the company. According to Block, Builderbot merges around 1,500 pull requests each week and performs more than 200,000 operations every day.
While AI coding assistants have become common tools for developers, Builderbot is designed to operate on a much larger scale. Rather than helping with code in a single project, Block says the system understands the company’s entire codebase, including hundreds of services, APIs, and internal conventions.
Engineers interact with Builderbot through Slack by tagging it in a conversation and describing a task. The AI can then research the issue, create a branch, write code, open a pull request, monitor CI results, and make revisions based on feedback. The company says employees can collaborate with the system in real time while maintaining oversight of the work being performed.
Block argues that the biggest difference between Builderbot and traditional coding assistants is scope. An engineer working on one product can use the system to make changes to services they have never worked with before because Builderbot already understands how those systems are connected.
The company claims the tool has dramatically reduced development timelines. What previously took months can now be completed in days, according to Block. Brad Axen, Head of AI Capabilities at Block, described Builderbot as the missing layer between AI coding tools and the realities of large-scale software engineering.
Notably, Builderbot is built on Goose, an open source AI agent framework developed by Block and later contributed to the Agentic AI Foundation under the Linux Foundation. The company also notes that its work on Goose helped inspire the creation of the Model Context Protocol, or MCP, which it co-developed with Anthropic and which has since become one of the most widely discussed standards in the AI industry.
Of course, all of the performance claims come directly from Block, and the company did not provide data on code quality, bug rates, or how often human engineers reject Builderbot’s work. Still, the numbers are hard to ignore. If AI is already responsible for 15 percent of production code at a company the size of Block, it raises an obvious question: how much software will humans be writing five years from now?
Support independent tech journalism
NERDS.xyz is independently owned and operated. If you enjoy my coverage of Linux, AI, hardware, cybersecurity, and tech culture, consider supporting the site on Ko-fi.
Support NERDS.xyz