Baz, a software development team, implemented an AI-driven code review system to address the limitations of traditional manual reviews. Developers often struggled to validate whether code met functional and design requirements, leading to inconsistencies and delays. By integrating Amazon Bedrock and Amazon Bedrock AgentCore, Baz automated the verification process, bridging the gap between code implementation and product intent. This solution enables teams to catch discrepancies early in the development lifecycle, reducing bugs and accelerating the merge process.
The Baz Spec Review agent orchestrates a multi-stage validation pipeline, combining static code analysis with dynamic runtime validation. When a GitHub pull request is submitted, the system queries Figma and Jira to gather design and functional specifications. It then spawns isolated sub-agents to evaluate each requirement, using Amazon Bedrock AgentCore to simulate user interactions and visually confirm that the implementation aligns with the intended design. This approach ensures that both the code and the user experience meet the specified requirements.
The implementation leverages Amazon Bedrock to interpret requirements and assess design intent, while AgentCore provides secure, isolated browser automation for real-time validation. This combination allows Baz to scale validation processes across multiple projects without requiring custom infrastructure. Teams report faster reviews, fewer regressions, and higher confidence that changes meet requirements before merging. The solution also reduces manual validation work by up to 70% and cuts time-to-merge by 30–70%.
Source: awsml