Which repositories AI-assisted commits actually land in

A leaderboard of people answers “who”. The more useful question for anyone deciding how to work is “where”: which codebases are actually absorbing AI-assisted commits, and what kind of work is going into them.

Every figure below comes from the snapshot named at the top of this page. Later snapshots will differ, and the current ranking is always live.

The shape of the list matters more than its top

Two things are worth noticing in the chart below, neither of which is the name in first place.

The distribution is steep. A small number of repositories absorb a large share of the detected commits, and the tail falls away quickly. That is the signature of concentrated agent work: a handful of projects where someone has wired an agent into their normal loop, rather than broad shallow adoption across many codebases.

Most of them are infrastructure, not products. Tooling, pipelines, configuration and test suites dominate. This matches what the tools are good at today: bounded, well-specified changes with fast feedback, where a wrong answer is caught by a test rather than a customer.

Counting caveat, stated once

These totals are built from the repositories named by ranked developers, so a repository whose contributors are not yet ranked is undercounted, and one with many ranked contributors is favoured. It is a ranking of visible activity, not a census. Coverage has the current scan figures.

When the commits happen

The weekday pattern is the other half of the picture. AI-assisted commits are not evenly spread, and the shape tells you something about who is producing them: weekday-heavy suggests professional work inside a normal week, while a flatter distribution points to side projects and personal repositories.

What this does not show

Nothing here measures value. A repository with many AI-assisted commits is not better maintained than one with none, and the tools are heavily represented in exactly the kind of high-volume, low-risk change that inflates commit counts without moving a product forward. Read this as a map of where the tools are being used, and nothing more.