Adoption is broad. Many contributors are committing agent-authored work, which is what a real organisational rollout looks like.
Assisted or autonomous
Assisted — a developer’s identity authored the commit and the agent added itself as co-author. The agent ran git commit during a working session, with a person there. Autonomous — a bot identity is the author, no human in the loop. Across the whole index, 93.7% of agent commits are assisted.
Agent share by month
Peak month: 18.4%. Bars are scaled to that peak, not to 100%.
Which agents
| Claude Code | Anthropic | 53 | 5.4% |
| Gemini CLI | 47 | 4.8% | |
| Cursor Agent | Anysphere | 15 | 1.5% |
| OpenAI Codex | OpenAI | 6 | 0.6% |
| GitHub Copilot | GitHub | 4 | 0.4% |
Verify it yourself
Sample commits that matched. Every one links to GitHub — check the trailers.
0e91ffdf[rollout] feat: continuous_token support Assistant Role in Merge and EncodClaude Code5b79827b[distillation] fix: Disable detokenization for teacher prompt_logprobs (#7Claude Codefcb4f7f7[sglang] fix: make DeepSeek-V4 full weight sync work with SGLang + MegatroClaude Code0e91ffdf[rollout] feat: continuous_token support Assistant Role in Merge and EncodGemini CLIdf9e2243[BREAKING][rollout] feat: Add Multimodal Continuous Token (#6804)Gemini CLI61f96993[ckpt] fix: added cuda stream synchronization in NCCL broadcast wait for cGemini CLIcf1649b2[hardware] test: include AMD in platform auto-detection expectations (#779Cursor Agent0e91ffdf[rollout] feat: continuous_token support Assistant Role in Merge and EncodCursor Agentb356f430[vllm] fix: raise max_num_batched_tokens to max_model_len when chunked preCursor Agent5cfb74fa[megatron] feat: bucket packed sequence lengths (#7358)OpenAI Codex7898ab49[tool, megatron] fix: resolve PrecisionDebugger model chunks (#7104)OpenAI Codexe0031631[rollout] fix: handle malformed Qwen3 XML tool calls (#7044)OpenAI Codex4493b868[training_utils, env, doc] feat: use Liger fused linear PPO kernel (#7461)GitHub Copilotfeff37ae[2/n][rollout] feat: flowgrpo - add diffusion agent loop support (#5716)GitHub Copilotf245e2dd[fsdp] fix: Support trust_remote_code during FSDP HugginFace checkpoint saGitHub Copilot
What was excluded
| Commits in window | 982 |
| Merge commits removed | −0 |
| CI / dependency bots removed | −0 |
| Considered | 982 |
Bots are excluded from the denominator entirely. Counting dependabot as “AI-written code” is the fastest way to produce a meaningless number.
How this was measured
Measured over February–August 2026, cloned directly. Detection is deterministic string matching against signatures agents write about themselves — no heuristics, no style analysis. Merge commits and CI bots are excluded from the denominator. The figure is a floor: squash merges strip trailers, so real usage is higher. Contributor counts are aggregate — no individual is named.
Run it yourself: npx qmmit-cli verl-project/verl