Baz automates code review with Amazon Bedrock AgentCore browser validation
AI agent checks code against Figma designs and Jira specs in real-time
Baz, a platform for accelerating development workflows, identified a critical gap in traditional code review: while syntax and compilation were checked, alignment with product requirements—like Figma designs and Jira acceptance criteria—was left to manual, error-prone QA. To solve this, Baz built a Spec Review agent powered by Amazon Bedrock and Amazon Bedrock AgentCore. The agent triggers on GitHub pull requests, then concurrently pulls visual specs from Figma (via MCP) and functional requirements from Jira. It decomposes these into discrete visual and behavioral requirements, spawning sub-agents that validate each one.
Each sub-agent combines deep code analysis with dynamic runtime validation using Amazon Bedrock AgentCore's Browser Tool. Instead of only checking static code, the sub-agents spin up temporary preview environments, perform DOM inspection, event simulation, and visual testing against the actual rendered UI. This end-to-end pipeline runs on Amazon EKS behind an ALB/NLB, with Amazon Bedrock powering the LLM inference across all stages. The result: Baz's agent catches discrepancies early, reduces manual QA effort, and ensures every pull request conforms to product intent before deployment.
- Baz's agent uses Amazon Bedrock AgentCore Browser Tool to visually validate live UI against Figma designs in real-time
- Multi-agent pipeline concurrently pulls requirements from Figma and Jira, then spawns isolated sub-agents per requirement
- Runs on Amazon EKS with Bedrock LLMs, reducing manual QA and catching behavioral mismatches before deployment
Why It Matters
Automates behavioral validation of code against product specs, cutting QA cycles and preventing design regressions.