Agentic AI
Bloomberg's Pomona automates code quality improvements through autonomous micro-pull requests
A lightweight AI agent deployed at Bloomberg continuously identifies and fixes code quality issues in small, manageable chunks that engineers trust and merge quickly.
Bloomberg researchers have developed Pomona, an agentic system that handles code quality improvements at scale by breaking work into tiny, low-risk contributions. Rather than attempting sweeping refactors, the tool follows a philosophy inspired by Kaizen—continuous, incremental improvement—to keep technical debt manageable while maintaining engineer confidence.
How Pomona works#
The system operates through two primary capabilities. A Scanning skill automatically discovers opportunities for improvement and builds a prioritized backlog of tasks. A Repair skill then generates small pull requests targeting individual issues, keeping diffs minimal and reviewable. This design keeps changes focused and easy to understand, reducing the friction that often prevents code quality work from merging.
Real-world results from deployment#
Bloomberg tested Pomona across a three-month team trial, documenting strong adoption outcomes. Of the 39 pull requests generated, 32 were merged—an 82% acceptance rate. Merges happened quickly: the median time from creation to merge was just over two hours, suggesting engineers found the work credible and straightforward to review.
The team also surveyed 12 senior engineers about the experience. Ten expressed enthusiasm about continuing or expanding Pomona adoption, citing appreciation for the small, focused changes and the tool's emphasis on sustainable code health. Since the formal evaluation concluded, another team at Bloomberg has already begun using the system.
Why incremental automation matters#
Traditional code quality initiatives often stall because large refactoring tasks require significant review overhead and carry higher risk of introducing bugs. By automating discovery and issuing dozens of tiny, targeted fixes instead, Pomona reduces the barrier to merge while still accumulating meaningful improvements. The approach appears to satisfy both the practical concerns of busy engineers and the long-term health goals of development teams.
