AutonomyPreviewThis index is in active development: we are still adding agents, refining attribution signatures, and verifying data against sources. The date in the header is when key figures were last re-checked. Numbers can move as coverage improves.
Public GitHub can show whether merged agent pull requests had a human reviewer, how long they took to merge, and whether they touched tests. Significant repos are org-owned or have at least 10 stars. Agent-level rates are withheld when the sample is too concentrated. Sparse exact-bot agents can use an expanded window; those rows are not included in the market rate.
Human reviewed
54.9%
Merged agent PRs in significant repos.
No human reviewer
45.1%
The same significant-repo sample.
Sample
244
Significant-repo PRs behind the market rate.
Agent-Level Autonomy
Window: 2026-05-21 to 2026-06-19. Published rates require at least 40 capped PRs across 15 repos. Significant-repo review rates require at least 10 significant PRs across 8 significant repos.
| Agent | Significant review | All-repo review | Time to merge | Tests touched | Sample | Status |
|---|---|---|---|---|---|---|
| GitHub Copilot | 83.3% | 67.5% | 1.23h | 20.8% | 120 PRs, 120 repos30-day window120 of 120 enriched | Published |
| Cursor | 74.5% | 46.7% | 1.95h | 45.8% | 120 PRs, 120 repos30-day window120 of 120 enriched | Published |
| Devin | 50% | 15% | 11 min | 22.5% | 120 PRs, 120 repos30-day window120 of 120 enriched | Published |
| OpenAI Codex | 35.1% | 34.2% | 19 min | 41.7% | 120 PRs, 120 repos30-day window120 of 120 enriched | Published |
| Claude Code | 33.3% | 24.2% | 11 min | 42.5% | 120 PRs, 120 repos30-day window120 of 120 enriched | Published |
| Amazon Q Developer | 32.3% | 38.3% | 22 min | 31.7% | 120 PRs, 120 reposexpanded 365-day window120 of 120 enriched | Published from 365-day sample. |
| Jules | 21.4% | 19.2% | 1.8h | 22.5% | 120 PRs, 120 repos30-day window120 of 120 enriched | Published |
Read This With the Ranking
Output volume counts what an agent leaves visible. Autonomy adds the lifecycle read: whether merged work had a human reviewer, how quickly it merged, and whether tests changed. For capture details and visibility limits, see the methodology.