/slashscore

agent report.

How much of the merged code at 20 well-known engineering orgs names an AI agent as its author or co-author. Read from public GitHub pull requests over the 90 days to September 28, 2026.

29.8%

of merged pull requests disclosed agent work.

  • Claude Code
  • Cursor
  • GitHub Copilot
  • Codex
  • Devin
  • other agents
  • inferred
  • people

32.0% counting inferred agents as well: bots whose pull requests read as agent work but never say so. They are drawn hatched and kept out of the headline.

20
orgs, up to 10 active repos each
85,047
merged pull requests
72,326
by people or agents, the base
1,621
people steering agents

by org

declared share · with inferred

open any org for its repositories, agents, models and bus factor

  1. mastra-ai
    63.3%3,396 PRsview →
  2. microsoft
    52.6%7,589 PRsview →
  3. langchain-ai
    51.4%52.4%3,140 PRsview →
  4. PostHog
    46.9%48.3%18,447 PRsview →
  5. getsentry
    32.7%35.4%7,332 PRsview →
  6. calcom
    23.1%78 PRs · small sampleview →
  7. grafana
    21.7%5,288 PRsview →
  8. facebook
    17.5%423 PRs · small sampleview →
  9. huggingface
    13.8%15.2%2,508 PRsview →
  10. anthropics
    13.7%380 PRs · small sampleview →
  11. supabase
    13.3%2,825 PRsview →
  12. cloudflare
    9.3%1,412 PRsview →
  13. denoland
    7.9%11.2%694 PRsview →
  14. vercel
    6.5%18.6%5,294 PRsview →
  15. hashicorp
    4.1%1,307 PRsview →
  16. openai
    2.3%2,075 PRsview →
  17. kubernetes
    1.3%2,822 PRsview →
  18. astral-sh
    0.8%17.2%2,632 PRsview →
  19. google
    0.4%1,137 PRsview →
  20. rust-lang
    0.4%3,547 PRsview →

which agents

declared PRs · orgs using it
  1. Claude Code7,90020 orgs
  2. Cursor58215 orgs
  3. GitHub Copilot3,99312 orgs
  4. Codex3826 orgs
  5. Devin1053 orgs
  6. Personal agent account92 orgs
  7. PostHog Desktop5,1621 org
  8. Mastra Code1,9361 org
  9. Open SWE1,4431 org
  10. Junior41 org
  11. LangSmith31 org
  12. Inkeep31 org
  13. Serge31 org
  14. ai-sdk-factory11 org
  15. astral-automations-bot11 org

models named

where a trailer names one
  • Claude Opus 5 · 2,177
  • Claude Opus 4.8 · 1,836
  • Claude Fable 5 · 1,435
  • anthropic/claude-fable-5 · 466
  • Claude Fable 5.1 · 465
  • openai/gpt-5.6-sol · 366
  • Claude Opus 5.5 · 276
  • Claude Sonnet 5 · 263

how this is measured

  • A pull request counts as agent work when it carries an agent's co-author trailer, an agent's footer (a session or task link), or was opened by an agent's own account. Each one is credited to a single agent, so the parts of a bar add up.
  • Dependency bots, release bots, mirrors and AI code reviewers are left out of the base. A bot we could not classify is left out too, so it can shrink the base but never raise the share.
  • These figures count disclosed agent work only. An org that turns attribution off reads close to zero here whatever its real usage.
  • The 20 orgs were picked by hand for this demo and lean toward teams that use agents heavily. The pooled figure describes them, and open source as a whole will differ.
  • No person is named anywhere in this report.

This report reads public repositories. slashscore for Teams draws the same map for your private ones: the people, the agents working beside them, and the repos they share.