Coding Agent Radar

Adoption of AI coding agents across public GitHub repositories

Loading...
GitHub Star
⚠ Draft methodology — work in progress. The detection rules are a starting point, not a definitive standard. A config file in a repo is a weak signal: it may be a template, a fork, an experiment, or simply outdated. Treat all numbers as rough approximations, not ground truth.
Methodology

Detection strategy

A repository is counted as using a coding agent when it contains one or more agent-specific config files. No file contents are read and no LLM is involved.

Data collection

For each signal file the GitHub Code Search API is queried (up to 1000 results per signal). Results are returned in best-match order. Each matched repository is enriched with full metadata via a separate API call, cached to avoid redundant fetches across signals.

A repository is counted once per agent regardless of how many of its signal files match. The scanner runs weekly and preserves original discovery dates across re-runs.

Limitations

  • Repos using an agent without committing a config file are not counted.
  • Only public GitHub repositories are covered.
  • The timeline reflects scan date, not the original commit date.
  • All GitHub Code Search API limitations apply — including the 1000 result cap per query, index delays, and potential gaps in coverage for recently pushed or low-activity repositories.
  • GitHub forces results to be sorted by relevance ("best match") — sorting by stars is not supported by the API. High-starred repos are therefore not guaranteed to appear and may be missing from the data.
Total detections
Agents tracked
Repos last scan
New last scan

Agent Radar

Adoption Over Time

Repository Popularity by Agent (stars)

Agent Share by Language (%)

Language Share by Agent (%)