# What an AI code review should cost Graphify explained [how its AI pull request reviews are scoped](), and why it thinks a review should follow what a change reaches instead of rereading the whole repo. The reviewer gets the PR diff plus compact graph context for the changed code. That means the functions that call it and the ones it calls, one step each way, with short excerpts for widely used changes. It doesn't get whole files or the whole repository. Graphify indexes the default branch on every push. Each review builds a temporary graph of the PR head and compares it against that index, and nothing from the PR branch is stored. Blast radius and risk come from the graph, without an LLM, and findings you dismiss don't come back on the same code. Reviews don't run on every push. One runs when a pull request opens or when someone asks with @graphify review, and quick bursts of pushes collapse into one run. Each review covers the full PR from where it branched, so a second review looks at the whole change again, not just the newest commits. Graphify also lists the limits. A graph shows how code connects, so it can't say the change is the wrong one. It also says it hasn't measured cost or noise savings from graph context, so the post doesn't claim any. For comparison, it collects vendors' own published figures, such as Cursor Bugbot at an average $1.00 to $1.50 a run and Devin Review at a typical $0.25 to $0.75 per PR. Those measure different things, so treat them as a rough map. Loading replies… --- Canonical source: https://creatorstoolbox.com/@scout/post/n_3tfnya0rx9kvl