GitHub - powerset-co/github-coding-agent-tracker: Tracking the use of AI coding agents in public GitHub commits

GitHub Coding Agent Monitor

A public, auditable log of AI coding agent commit counts on public GitHub repos over time. The following chart and table are updated automatically by a GitHub Action running on a daily schedule.

10-day rolling average, as a % of all public commits on GitHub of the top 3 coding agents (by detected commit count).

Caveats

Given the methodology described below, there are some implicit limits to this data:

1. Only public GitHub activity is monitored, private repos are not accessible by the queries used.
2. Only commits where a coding agent has left a "signature" can be detected (e.g. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>).
3. Only commits on default branches are indexed by GitHub's search.

Some coding agents may be more prevalent in private repos vs public repos. Some may not show up much in this data because they don't leave a "signature." As such, be wary of what you conclude from this data.

In general, the data is meant to highlight broad trends around coding agent adoption overall.

We'd welcome more companies and developers to leave signatures in their commits so we can track this for the community. Additionally, we'd welcome any contributions to this repo to improve how data is collected, additional analysis, or other improvements.

How It Works

A daily GitHub Action uses the GitHub Search API to count new public commits matching each coding agent's signature. Total public commits are counted in 24x 1-hour windows and summed.

Results are stored as flat CSVs in data/YYYY-MM-DD.csv and committed back to this repo, along with an updated chart.

Specific coding agents are detected using the following search queries:

Query the Data with DuckDB

Since the data lives in this repo in CSV files, you can use DuckDB to query it.

SELECT * FROM read_csv('data/*.csv');

-- Daily agent percentages
SELECT
  date,
  query AS agent,
  count,
  count * 100.0 / SUM(count) FILTER (WHERE query = 'total') OVER (PARTITION BY date) AS pct
FROM read_csv('data/*.csv')
WHERE query NOT LIKE 'total%'
ORDER BY date, count DESC;

Run Locally

# Fetch a single day
GITHUB_TOKEN=ghp_... bun run src/fetch.ts 2026-02-14

# Fetch a date range (inclusive)
GITHUB_TOKEN=ghp_... bun run src/fetch.ts 2025-02-17 2026-02-15

# Generate chart from existing data
bun run src/chart.ts

Backfill

At ~30 requests/min (GitHub search API rate limit), each day requires 34 queries (24 hourly windows + 10 agents), so backfilling runs at ~1 day/minute (~6 hours for a full year).

GITHUB_TOKEN=ghp_... bun run src/fetch.ts 2025-02-17 2026-02-15