Block implements AI agents: neural networks now write 15% of the code
Payment and blockchain company Block, founded by Jack Dorsey, has taken a significant step in automating development. The implemented AI tool Builderbot already generates about 15% of the company's entire software code. This is not just an auxiliary utility, but a full-fledged system that is changing the approach to software creation.
Builderbot works as an orchestrator of AI agents integrated into the corporate messenger Slack. A developer only needs to tag the bot and describe the task — the artificial intelligence independently finds errors, suggests fixes, or creates new features. The key difference from analogs is access to Block's entire codebase. This allows, for example, an engineer from the Cash App team to make changes to Square services they have not worked with before.
The process is fully automated: the bot takes tasks from Jira, creates branches in the repository, writes code, and submits a Pull Request. The scale is impressive: the system performs over 200,000 operations per day and closes about 1,500 merge requests weekly.
Block's Head of AI Capabilities Brad Axen emphasizes: "What used to take months now takes days. The bot handles the routine and environment setup, allowing engineers to focus on solving complex problems." Importantly, the tool only works with source code and system configurations, having no access to customer data or payment information.
Technically, Builderbot is built on the open-source framework goose, which Block transferred to the Agentic AI Foundation. Anthropic also participated in the development — the partnership involved creating the Model Context Protocol (MCP).
The company is confident that the transition from simple AI-assisted code writing to "native" engineering processes based on neural networks will become the main trend in the IT industry.
My opinion: Block's initiative is not just another AI tool, but a demonstration of how deeply neural networks can integrate into production processes. If 15% of code is already being created automatically, then in the coming years we will see this share grow to 50% and beyond. The question is not whether AI will replace programmers, but how quickly companies will adapt their processes to the new reality. Block is setting a trend here that many will follow.