The agent commit index

Almost all AI commits still have a human in the loop.

Coding agents sign their own commits. We counted those signatures across 123 open source repositories — 182,354 commits, all measured over the same window (February–August 2026). 93.7% turn out to be an agent committing alongside a developer, not running on its own. And a headline percentage still tells you almost nothing without knowing how many people are behind it.

Try

Figure 1 The shape of agent adoption across 123 repositories

123 repositories, sorted by agent share — hover any bar

median 3.9%

individualteamorg-widenonebar height = agent share · colour = who is driving it

There is no smooth middle. A handful of projects have gone a long way, then the curve collapses — half of everything measured sits below 3.9%, and 11 repositories are at exactly zero. Colour shows who is behind each bar, and the tallest ones are often a single person.

Figure 2 Two repositories. Same story, until you look.

pandas-dev/pandas

54.8%

54.8% agent · top contributor 97.4%12 contributors

Nearly three quarters of recent commits are agent-authored — but one maintainer accounts for 97.4% of them. This is a single person’s workflow, not a project-wide shift.

calcom/cal.com

41.8%

41.8% agent · top contributor 12.5%41 contributors

A lower number, spread across 41 contributors with no one above 12.5%. This is what an actual organisational rollout looks like.

0.0%

of all commits measured
are agent-authored

0.0%

median repository —
the typical project has barely started

0.0%

of agent commits are assisted —
a human authored them

repositories with
broad team adoption

What the numbers hide

Assisted, not autonomous

Only 0.7% of all commits come from a bot identity acting alone. The rest carry a developer’s name with the agent as co-author — the agent ran the commit, a person was there.

Systems languages lag

Go sits at 0.5%, Rust at 1.5%, Python at 1.4% — against 9–11% for web tooling and AI projects. Measured over the same window, so this is adoption, not commit velocity.

Concentration still decides meaning

29 of 123 repositories have agent commits coming from essentially one person — median top contributor 88.7%. The same percentage means opposite things depending on who is behind it.

The ecosystem divide

Figure 3

Median agent share by ecosystem. The gap between the least and most adopted is roughly 55x — systems-language projects have effectively not started, while product companies and AI tooling have.

Data/viz

10.9%

AI/agent tools

9.8%

Product companies

9.2%

UI libraries

JS/TS frameworks

5.2%

Build tools

4.4%

Backend/ORM

3.5%

Editors/desktop

2.7%

Rust

1.5%

Python

1.4%

0.5%

Run it on your own code

The detection engine is open source and does the work locally. Nothing is uploaded, so it runs on private and internal repositories that no hosted tool can reach.

qmmit-cli ↗

Reads git history directly with a blobless clone — commit metadata only, no file contents, no GitHub API and therefore no rate limits. A 9,000-commit repository takes about six seconds.

Works offline on any local clone, including private repos
Deterministic signature matching — the rules are readable in one file
JSON output for scripting and CI
MIT licensed, zero dependencies

Source on GitHub npm Read the detection rules

# this repository, right here
npx qmmit-cli

# a private repo you have cloned
npx qmmit-cli ~/work/internal-api

# same window as this index
npx qmmit-cli --since 2026-02-01

# machine readable
npx qmmit-cli owner/repo --json

It reports counts, never names. You will see how many contributors commit agent-signed work and how concentrated that is — never who they are. A per-person breakdown of AI usage is a surveillance artifact, and this tool will not produce one, locally or otherwise.

The index

Every repository measured, sortable. “Top” is the largest single contributor’s share of that repository’s agent commits — the higher it is, the more the headline number reflects one person.

#		Ecosystem				Pattern
1	All-Hands-AI/OpenHands	AI/agent tools	62.5%	54	19.2%	org-wide
2	framer/motion	UI libraries	61.2%	2	95.8%	individual
3	pandas-dev/pandas	Python	54.8%	12	97.4%	individual
4	denoland/fresh	JS/TS frameworks	54.1%	12	71.7%	team
5	cypress-io/cypress	Product companies	52%	15	43.2%	org-wide
6	calcom/cal.com	Product companies	41.8%	41	12.5%	org-wide
7	continuedev/continue	AI/agent tools	37.7%	13	45.7%	org-wide
8	microsoft/vscode	Editors/desktop	30.8%	155	9.7%	org-wide
9	babylonjs/Babylon.js	Data/viz	28%	23	26.6%	org-wide
10	sanity-io/sanity	Product companies	25%	27	28.2%	org-wide
11	mendableai/firecrawl	AI/agent tools	22.4%	22	18.6%	org-wide
12	BerriAI/litellm	AI/agent tools	21.6%	122	21.1%	org-wide
13	appwrite/appwrite	Product companies	21.1%	20	41.1%	org-wide
14	mermaid-js/mermaid	Data/viz	20.5%	20	23%	org-wide
15	ant-design/ant-design	UI libraries	20.2%	35	27.7%	org-wide
16	docker/compose	Go	19.5%	8	43.2%	team
17	preactjs/preact	JS/TS frameworks	18.3%	2	90%	individual
18	denoland/deno	Build tools	17.2%	31	79.6%	team
19	vitest-dev/vitest	Build tools	17.2%	15	78%	team
20	pulumi/pulumi	Product companies	16.8%	34	29.4%	org-wide
21	storybookjs/storybook	UI libraries	16.7%	35	22.7%	org-wide
22	mastra-ai/mastra	AI/agent tools	16.6%	124	33.2%	org-wide
23	langchain-ai/langgraph	AI/agent tools	15.6%	5	29.2%	team
24	vercel/ai	AI/agent tools	15%	35	57.7%	team
25	electron/electron	Editors/desktop	13.4%	14	31.4%	org-wide
26	tldraw/tldraw	UI libraries	13.3%	14	32.3%	org-wide
27	grafana/grafana	Go	12.9%	110	26%	org-wide
28	vllm-project/vllm	AI/agent tools	11.3%	358	3.8%	org-wide
29	python/cpython	Python	11.1%	17	90.4%	individual
30	strapi/strapi	Product companies	11%	23	19.8%	org-wide
31	mrdoob/three.js	Data/viz	10.9%	12	59.2%	team
32	crewAIInc/crewAI	AI/agent tools	10.2%	16	39%	org-wide
33	unclecode/crawl4ai	AI/agent tools	10%	4	84.2%	individual
34	block/goose	AI/agent tools	9.8%	74	13.8%	org-wide
35	huggingface/diffusers	AI/agent tools	9.7%	19	47.4%	org-wide
36	colinhacks/zod	Backend/ORM	9.4%	9	47.1%	org-wide
37	prisma/prisma	Product companies	9.3%	15	31.6%	org-wide
38	twentyhq/twenty	Product companies	9.3%	54	57.6%	team
39	cloudflare/workers-sdk	Product companies	9.2%	45	20%	org-wide
40	pnpm/pnpm	Build tools	9.2%	33	38.9%	org-wide
41	supabase/supabase	Product companies	9.2%	56	27.9%	org-wide
42	vercel/turborepo	Build tools	8.1%	19	71.6%	team
43	directus/directus	Product companies	8%	16	20%	org-wide
44	browser-use/browser-use	AI/agent tools	7.8%	14	64%	team
45	nrwl/nx	Build tools	7.5%	21	19.6%	org-wide
46	vercel/next.js	JS/TS frameworks	7.1%	25	51.7%	team
47	shadcn-ui/ui	UI libraries	7%	28	42.3%	org-wide
48	mui/material-ui	UI libraries	6.9%	12	25%	org-wide
49	oven-sh/bun	Build tools	6.9%	16	67%	team
50	astral-sh/uv	Rust	6.3%	12	81.3%	individual
51	huggingface/transformers	AI/agent tools	6.1%	52	22.2%	org-wide
52	payloadcms/payload	Product companies	6.1%	12	78.7%	team
53	facebook/react	JS/TS frameworks	5.3%	4	64.7%	team
54	prometheus/prometheus	Go	5.2%	15	34.8%	org-wide
55	sveltejs/svelte	JS/TS frameworks	5.2%	5	66.7%	team
56	npm/cli	Build tools	4.8%	5	66.7%	team
57	biomejs/biome	Build tools	4.4%	12	54%	team
58	vitejs/vite	Build tools	4.4%	13	38.5%	org-wide
59	remix-run/react-router	JS/TS frameworks	4.3%	6	70%	team
60	sharkdp/bat	Rust	4.2%	3	50%	individual
61	excalidraw/excalidraw	UI libraries	4.1%	4	33.3%	team
62	cline/cline	AI/agent tools	3.9%	17	19.1%	org-wide
63	fastify/fastify	Backend/ORM	3.9%	5	37.5%	team
64	microsoft/playwright	Product companies	3.7%	9	60%	team
65	TanStack/table	Backend/ORM	3.5%	4	72.7%	team
66	open-webui/open-webui	AI/agent tools	3.5%	9	88.7%	individual
67	swc-project/swc	Rust	3.5%	6	61.1%	team
68	medusajs/medusa	Product companies	3.4%	11	63.3%	team
69	run-llama/llama_index	AI/agent tools	3.2%	10	25%	org-wide
70	apache/arrow	Python	3.1%	15	25%	org-wide
71	prettier/prettier	Build tools	3.1%	3	81.8%	individual
72	nestjs/nest	Other	3%	6	40%	team
73	flutter/flutter	Mobile	2.9%	45	10.5%	org-wide
74	tailwindlabs/tailwindcss	UI libraries	2.8%	4	50%	team
75	danny-avila/LibreChat	AI/agent tools	2.7%	23	20.6%	org-wide
76	neovim/neovim	Editors/desktop	2.7%	5	89.8%	individual
77	astral-sh/ruff	Rust	2.3%	17	33.9%	org-wide
78	expo/expo	Mobile	2.3%	47	17%	org-wide
79	typeorm/typeorm	Backend/ORM	1.9%	3	66.7%	individual
80	eslint/eslint	Build tools	1.8%	2	83.3%	individual
81	langchain-ai/langchain	AI/agent tools	1.8%	7	33.3%	team
82	zed-industries/zed	Editors/desktop	1.8%	50	5.8%	org-wide
83	chakra-ui/chakra-ui	UI libraries	1.7%	3	33.3%	individual
84	plotly/plotly.js	Data/viz	1.7%	3	50%	individual
85	pydantic/pydantic	Python	1.7%	4	25%	team
86	honojs/hono	Backend/ORM	1.6%	4	25%	team
87	clap-rs/clap	Rust	1.5%	1	100%	individual
88	numpy/numpy	Python	1.4%	6	30%	team
89	duckdb/duckdb	Python	1.3%	10	43.4%	org-wide
90	nocodb/nocodb	Product companies	1.3%	7	67.3%	team
91	withastro/astro	JS/TS frameworks	1.3%	14	17.6%	org-wide
92	TanStack/query	Backend/ORM	1.2%	4	28.6%	team
93	webpack/webpack	Build tools	1.1%	4	37.5%	team
94	ggml-org/llama.cpp	AI/agent tools	1%	19	12%	org-wide
95	changesets/changesets	Build tools	0.9%	2	50%	individual
96	documenso/documenso	Product companies	0.9%	1	100%	individual
97	nuxt/nuxt	JS/TS frameworks	0.9%	7	14.3%	team
98	ionic-team/ionic-framework	Mobile	0.8%	1	100%	individual
99	pola-rs/polars	Rust	0.8%	8	22.2%	team
100	nodejs/node	Build tools	0.7%	8	41.7%	team
101	fastapi/fastapi	Python	0.6%	3	33.3%	individual
102	comfyanonymous/ComfyUI	AI/agent tools	0.5%	6	16.7%	team
103	ollama/ollama	Go	0.5%	3	33.3%	individual
104	scikit-learn/scikit-learn	Python	0.5%	2	66.7%	individual
105	zellij-org/zellij	Rust	0.5%	1	100%	individual
106	tauri-apps/tauri	Editors/desktop	0.4%	1	100%	individual
107	tokio-rs/tokio	Rust	0.4%	1	100%	individual
108	golang/go	Go	0.3%	2	50%	individual
109	kubernetes/kubernetes	Go	0.3%	3	66.7%	individual
110	hashicorp/terraform	Go	0.2%	1	100%	individual
111	helix-editor/helix	Rust	0.2%	1	100%	individual
112	bevyengine/bevy	Rust	0.1%	1	100%	individual
113	angular/angular	JS/TS frameworks	0%	—	—	none
114	apache/echarts	Data/viz	0%	—	—	none
115	django/django	Python	0%	—	—	none
116	evanw/esbuild	Build tools	0%	—	—	none
117	godotengine/godot	Data/viz	0%	—	—	none
118	jestjs/jest	Build tools	0%	—	—	none
119	pocketbase/pocketbase	Product companies	0%	—	—	none
120	puppeteer/puppeteer	Product companies	0%	—	—	none
121	radix-ui/primitives	UI libraries	0%	—	—	none
122	teableio/teable	Product companies	0%	—	—	none
123	vuejs/core	JS/TS frameworks	0%	—	—	none

Method, and what this does not measure

This counts commits authored by agents, not code written with AI help. A developer who writes code with Claude Code and then commits it by hand leaves no trace in git and correctly reads 0% here. Git cannot see who typed the code. What it can see is who ran the commit.
Detection is deterministic string matching against signatures agents write about themselves — Co-Authored-By: Claude, cursor[bot], devin-ai-integration[bot], Codex task footers, Aider’s (aider) suffix. No heuristics, no style analysis, no inference.
Every figure is a floor, not an estimate. Squash merges and rebases strip Co-Authored-By trailers, so real usage is higher than what is reported here.
CI, dependency and release bots are excluded from the denominator by the [bot] naming convention. This matters more than it sounds: LangChain’s recent history is 46% bot commits. Counting those as “AI-written code” would inflate every number on this page.
Last 500 commits per repository. Lifetime figures differ substantially — projects with long pre-agent histories look far lower over their full life.
Repositories with fewer than 200 considered commits are excluded as too noisy to publish.
Commit counts, not lines of code. A one-line fix and a 2,000-line refactor each count once.
Contributor figures are aggregate only. No individual is named anywhere in this index, the dataset, or the tool. This is a deliberate limit, not an oversight — a per-person breakdown of AI usage is a surveillance tool, and this is not one.