AI pull requests
Are coding agents shipping code in public?
Contributes to: Is it being used to build?. Signal score: 5 / 100. Relative weight within this question: 1.
This undercounts AI work. Most agent-written code is opened under a person's account, so it looks human here. Only accounts and branch names that identify an agent are counted.
It also counts opening, not merging or quality. Codex is matched by its default branch name, so some of those may be people, and a pull request can be automated without any AI.
Observed trend
Public GitHub pull requests from identifiable coding agents, per month.
Public GitHub pull requests from identifiable coding agents, per month. · 2024-12 – 2026-10
View data table
| Period / observation | Public GitHub pull requests from identifiable coding agents, per month. |
|---|---|
| 2024-12 | 1,895 |
| 2025-01 | 2,363 |
| 2025-02 | 2,861 |
| 2025-03 | 3,322 |
| 2025-04 | 4,905 |
| 2025-05 | 70,099 |
| 2025-06 | 400,429 |
| 2025-07 | 300,846 |
| 2025-08 | 434,829 |
| 2025-09 | 450,995 |
| 2025-10 | 559,400 |
| 2025-11 | 473,850 |
| 2025-12 | 474,276 |
| 2026-01 | 537,322 |
| 2026-02 | 676,716 |
| 2026-03 | 817,335 |
| 2026-04 | 735,445 |
| 2026-05 | 779,567 |
| 2026-06 | 646,052 |
| 2026-07 | 798,827 |
| 2026-08 | 906,062 |
| 2026-09 | 1,092,496 |
| 2026-10 | 1,549,090 |
What it measures
- Metric
- Public GitHub pull requests clearly labelled as coming from coding agents, latest full month
- Solved if
- Coding agents open most pull requests
- Signal score
- 5 / 100. 4.7% of public pull requests clearly come from agents, and most agent work is hidden. How it is scored
- Status rule
- More agents when more agent pull requests are opened than in the same month a year before.
Which agents
These are 4.7% of all public pull requests, against 6.8% a year earlier.
- Codex: 1,381,684, against 418,960 a year earlier.
- Copilot: 61,241, against 120,409 a year earlier.
- Devin: 53,084, against 4,102 a year earlier.
- Cursor: 29,333, against 35 a year earlier.
- Claude: 12,280, against 148 a year earlier.
- Jules: 11,468, against 15,746 a year earlier.
Update history
The reading on each date we checked, separate from the historical observations above.
| Date | Value | Year over year | Status |
|---|---|---|---|
| 2026-11-02 | 1.5M | +177% | More agents |
| 2026-11-01 | 1.5M | +177% | More agents |
Source
| Source | Role | Measures | Latest data |
|---|---|---|---|
| GitHub pull request search | Scored measurement | Public pull requests by coding agent, by month | 2026-10 |