HubSpot's Sidekick: Multi-Model AI Code Review with 90% Faster Feedback and 80% Engineer Approval

"What we found might surprise you: our AI code reviewer catches real issues, understands HubSpot‑specific context, and maintains a high signal to noise ratio, often leaving no comments at all."

"Sidekick provides immediate pull request feedback, letting human reviewers focus on architecture and higher-level design, improving efficiency and reducing review bottlenecks."

"The first version of the system ran on an internal platform called Crucible. Large language model agents operated in Kubernetes environments and interacted with GitHub repositories via the command line. While this approach demonstrated that LLMs could provide useful feedback, it introduced operational complexity."

HubSpot engineers developed Sidekick, an AI-powered code review agent that analyzes pull request changes and provides automated feedback through GitHub. The system uses large language models to identify issues and post comments directly in repositories. Sidekick reduced time-to-first-feedback by approximately 90 percent, addressing delays caused by unavailable human reviewers. The tool catches real issues while maintaining high signal-to-noise ratio, often leaving no comments when unnecessary. Initially built on an internal Crucible platform using Kubernetes containers, the system introduced operational complexity and latency. The team subsequently migrated Sidekick to a Java-based agent framework called Aviator, which integrates with HubSpot's development platform for improved efficiency and reduced infrastructure overhead.

Read at InfoQ