How AWS CodeGuru Uses AI to Automate Code Reviews and Boost Performance

AWS (Amazon Web Services) has launched CodeGuru, an artificial‑intelligence‑driven tool that automatically reviews code.

CodeGuru employs a suite of machine‑learning models to examine developers’ code, propose possible optimizations—including performance assessments—and continuously improve its recommendations using collected data. The service consists of two components, Reviewer and Profiler, which were first previewed in December of the previous year.

Manual code review is challenging: developers must have sufficient experience and time, and even seasoned reviewers can miss bugs or performance issues that surface at runtime. Continuous code generation makes it hard to keep up with thorough reviews.

CodeGuru also addresses runtime monitoring, aiming to keep applications running efficiently by detecting performance problems early.

After deployment, developers continue using their usual workflows with GitHub or Bitbucket Cloud. CodeGuru Reviewer runs in the background, analyzes commits, creates a pull request, automatically adds comments, and provides information about errors or performance improvements.

Our customers run many applications with millions of lines of code. Ensuring code quality and efficiency is critical; even a few inefficient lines can be extremely costly. Manual identification of code quality issues is time‑consuming and error‑prone, especially at scale. CodeGuru combines Amazon’s decades of large‑scale development and deployment experience with deep machine‑learning expertise to deliver high‑quality software, better application performance, and reduced maintenance costs.

Several large companies have adopted CodeGuru, including Atlassian, EagleDream, and DevFactory.

Atlassian’s technical team lead Zak Lsam noted, “CodeGuru cannot always predict system behavior under high‑stress conditions, especially when managing complex data, and it sometimes falls short during frequent deployments.” He added, “After detecting anomalies in production, CodeGuru provides continuous analysis that can surface issues within hours or even minutes.”

Compiled by: Wang Chun Yong Source: 21CTO