Can AI Really Replace Human Code Review? Inside Graphite’s Diamond Tool

Graphite, founded in 2020 by former Facebook, Google, and market‑research staff, builds code‑review tools.

The company’s core problem is large pull requests that are hard to review; their solution is “stacking” – splitting code into multiple PRs, which can cause upstream merge issues.

Graphite addresses this with the Graphite CLI and a VS Code extension.

AI‑assisted code review is now common in tools like GitHub Copilot, Amazon Q Developer, and Google Gemini. In October last year Graphite launched Graphite Reviewer to help developers spot issues before sending PRs for review.

Graphite has now turned Graphite Reviewer into a standalone product called Diamond , backed by fresh venture funding. Diamond can detect bugs, style inconsistencies, security vulnerabilities, performance problems, documentation issues, apply custom rules, provide context‑aware suggestions across the whole codebase, and flag accidentally committed code such as hard‑coded “return true” functions, offering fix recommendations.

Diamond’s current limitations are that it only integrates with GitHub organizations and its IDE integration is limited to VS Code.

Despite the AI push, co‑founder Greg Foster stresses that AI will never fully replace human code review, noting that AI lacks the domain expertise and accountability needed for critical decisions.

Foster also points out that AI may miss business context, team preferences, and cannot be held responsible for security‑critical bugs it introduces.

Anthropic CEO Dario Amodei recently predicted that AI could write 90 % of code within 3‑6 months, but Foster warns that unread, AI‑generated code poses huge risks if humans never truly understand it.

The broader challenge for enterprises is balancing the productivity gains of tools like Diamond with ensuring skilled humans remain involved, lest developers’ own abilities erode.