Code review has become the bottleneck of modern software development. AI coding assistants generate code faster than ever, but the review process remains largely unchanged — senior engineers manually reading through massive pull requests, context-switching between reviews, and waiting for CI to pass before merging. Graphite attacks this problem from a fundamentally different angle than most AI code review tools. Instead of just adding smarter comments to an unchanged workflow, it reimagines the workflow itself through stacked pull requests combined with AI-assisted review.
Founded with $52 million in Series B funding, Graphite has quickly become one of the most widely adopted developer productivity platforms among high-velocity engineering teams. Shopify routes 75% of its pull requests through Graphite and reports 33% more PRs merged per developer. Asana engineers save 7 hours weekly, ship 21% more code, and have reduced their median PR size by 11%. These are not theoretical projections — they are reported production metrics from teams operating at scale, and they stem from the combination of workflow improvement and AI assistance rather than AI alone.
The stacked PR concept is the foundation of Graphite's approach. Instead of creating one large pull request with hundreds or thousands of changed lines, developers break their work into a series of small, dependent PRs that build on each other. Each PR in the stack focuses on a single logical change — adding a data model, implementing an endpoint, writing tests. Developers can continue working on PR number three while PR number one is still under review, and when earlier PRs merge, Graphite automatically rebases the rest of the stack. This eliminates the single biggest source of developer idle time: waiting for reviews before starting dependent work.
The AI component, Graphite Agent, is tightly integrated into the stacked PR workflow. When a developer opens a PR, Agent provides codebase-aware feedback directly on the PR page. The critical insight is that smaller PRs give AI reviewers dramatically better results. A 200-line PR with a clear scope allows the AI to provide genuinely useful feedback on type errors, race conditions, security vulnerabilities, and optimization opportunities. The same AI struggling with a 2,000-line monolithic PR produces generic advice or misses critical issues entirely. Graphite's workflow innovation makes its AI reviewer more effective by design.
Agent goes beyond static comments. It functions as an interactive collaborator that lives on the PR page. Developers can ask contextual questions about their changes, request explanations of how modifications affect other parts of the codebase, and get help resolving CI failures directly in the conversation. One-click fixes allow developers to apply Agent's suggestions and commit changes without leaving the PR page. This conversational model reduces the back-and-forth cycles that typically slow down code review, turning what would be multi-round review iterations into single-session resolutions.