# Coding Agents: Autonomy and trust

> Human review, time-to-merge, and test-inclusion for merged agent pull requests, plus settled merge rates. Machine-readable version of https://amplifying.ai/coding-agents/autonomy.

## Autonomy and trust

Across 230 merged agent PRs in significant repos (org-owned or 10+ stars, 2026-05-07 to 2026-06-05), 55.7% had a human GitHub reviewer and 44.3% merged with none.

| Agent | Human-reviewed (significant repos) | Median time to merge | Touches tests |
| ----- | ---------------------------------- | -------------------- | ------------- |
| Cursor | 87.5% | 4.3h | 35.8% |
| GitHub Copilot | 78.3% | 43 min | 28.3% |
| Devin | 41.7% | 17 min | 19.2% |
| OpenAI Codex | 38.7% | 11 min | 40.8% |
| Jules | 37.5% | 1.19h | 17.5% |
| Claude Code | 33.3% | 25 min | 42.5% |

Note: "reviewed" means a non-author, non-bot GitHub reviewer. A high merge rate with a low review rate means the agent self-merges, not that a human signed off. Detail: https://amplifying.ai/coding-agents/autonomy

## Merge rates (settled window)

| Agent | PRs merged |
| ----- | ---------- |
| Claude Code | 91.5% |
| Amazon Q Developer | 88.3% |
| OpenAI Codex | 87.1% |
| GitHub Copilot | 80.5% |
| Cursor | 66.4% |
| Devin | 61.3% |
| Jules | 43.3% |

## How to read this

Each agent's output carries a visibility class, and they are not equal:

- **Exact**: visible through the agent's own bot or workflow, counted exactly.
- **Estimate**: inferred from public signals, either attribution the agent declares (large counts can swing 30 to 40 percent between runs) or the pull requests a cloud service opens (well-covered for the cloud path, blind to local CLI work).
- **No public signature**: the agent works through the developer's own identity and leaves nothing public to attribute. Absence means invisible, not low usage.

Coverage limits: public repositories only; private-repo work is not counted. Squash-merges and disabled attribution further undercount. Treat every figure as a floor and a proxy for the ecosystem, not a census. Output snapshot as of 2026-06-09.