# State of the Coding Agent Market

> The business of coding agents, measured from public evidence: attributed output, review patterns, segment cuts, and source confidence. Machine-readable companion to https://amplifying.ai/research/state-of-coding-agents (the full written analysis lives there).

## 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.

## Key findings

- Visible agent output is top-heavy: of 112.5M attributed commits and PRs, the top three (Claude Code, Replit Agent, Cursor) account for 90%, and Claude Code alone is 75%. Units differ by agent, so read this as concentration, not a like-for-like ranking.
- Most merged agent work still passes a human, but it splits by agent: across 244 merged PRs in significant repos, 54.9% had a human GitHub reviewer.
- Weekend usage is an enterprise-vs-consumer tell: tracked weekend volume runs 8.9% below weekdays, with Copilot agent down 25.3% while Jules barely moves.
- Inside Claude Code, Opus is 70.7% of observed runs.

## Rankings (attributed public output)

Total tracked attributed output: 112.5M commits and pull requests across 22 agents.

| # | Agent | Company | Attributed output | Signal | npm demand/wk | Channel |
| - | ----- | ------- | ----------------- | ------ | ------------- | ------- |
| 1 | Claude Code | Anthropic | 83.9M | Estimate | 8.2M | Native preferred |
| 2 | Replit Agent | Replit | 8.6M | Estimate | n/a |  |
| 3 | Cursor | Anysphere | 8.5M | Estimate | n/a |  |
| 4 | OpenAI Codex | OpenAI | 5.1M PRs | Estimate | 10.9M | npm primary |
| 5 | GitHub Copilot | GitHub / Microsoft | 3.9M | Exact | 1.5M | CLI/package channel |
| 6 | Jules | Google | 1.1M | Exact | n/a |  |
| 7 | Aider | Aider AI | 613K | Estimate | n/a |  |
| 8 | Devin | Cognition | 259K | Exact | n/a |  |
| 9 | Gemini CLI / Antigravity CLI | Google | 227K | Estimate | 721K | npm primary |
| 10 | Qwen Code | Alibaba / Qwen | 99K | Estimate | 83,956 | npm primary |
| 11 | opencode | Anomaly | 85,915 | Estimate | 1.5M | npm primary |
| 12 | OpenHands | All Hands AI | 58.5K | Estimate | n/a |  |
| 13 | Amp | Sourcegraph | 49.2K | Estimate | 36,970 | npm primary |
| 14 | Warp | Warp | 29.9K | Estimate | n/a |  |
| 15 | Roo Code | Roo Code | 19.1K | Estimate | n/a |  |
| 16 | JetBrains Junie | JetBrains | 9,602 | Estimate | n/a |  |
| 17 | Cline | Cline | 5,907 | Estimate | 65,236 | npm primary |
| 18 | Google Antigravity | Google | 4,863 | Estimate | 664K | Legacy npm / Antigravity native |
| 19 | Crush | Charm | 2,844 | Estimate | 3,013 | npm primary |
| 20 | Kiro | AWS | 2,369 | Estimate | n/a |  |
| 21 | Amazon Q Developer | AWS | 2,218 PRs | Exact | n/a |  |
| 22 | Codebuff | CodebuffAI | 1,205 | Estimate | 9,365 | npm primary |

Full table with filters: https://amplifying.ai/coding-agents

## Trends: weekly agent-marked PRs

Latest complete week per charted agent.

| Agent | PRs, latest week | Week over week |
| ----- | ---------------- | -------------- |
| Claude Code | 589K | -5% |
| Copilot agent | 15,721 | -8% |
| Cursor agents | 56,694 | +16% |
| Devin | 7,430 | +10% |
| Jules | 29,326 | -37% |
| Codex | 184K | +32% |

Live flow sample: of the 6,998 most recent public PRs, 19.7% carry an agent signal (Claude Code alone 16.3%), dependency bots another 9.6%.

Detail and daily view: https://amplifying.ai/coding-agents/trends

## Autonomy and trust

Across 244 merged agent PRs in significant repos (org-owned or 10+ stars, 2026-05-21 to 2026-06-19), 54.9% had a human GitHub reviewer and 45.1% merged with none.

| Agent | Human-reviewed (significant repos) | Median time to merge | Touches tests |
| ----- | ---------------------------------- | -------------------- | ------------- |
| GitHub Copilot | 83.3% | 1.23h | 20.8% |
| Cursor | 74.5% | 1.95h | 45.8% |
| Devin | 50% | 11 min | 22.5% |
| OpenAI Codex | 35.1% | 19 min | 41.7% |
| Claude Code | 33.3% | 11 min | 42.5% |
| Jules | 21.4% | 1.8h | 22.5% |

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

## Weekday vs weekend

Across tracked agents, weekend PR volume runs 8.9% below weekdays. The size of the dip is an enterprise-versus-consumer tell.

| Agent | Weekday/day | Weekend/day | Weekend change |
| ----- | ----------- | ----------- | -------------- |
| Copilot agent | 3,384 | 2,527 | -25.3% |
| Devin | 1,294 | 1,001 | -22.6% |
| Cursor agents | 5,170 | 4,300 | -16.8% |
| Codex | 21,484 | 19,386 | -9.8% |
| Claude Code | 67,662 | 62,411 | -7.8% |
| Jules | 7,315 | 7,231 | -1.1% |

Detail: https://amplifying.ai/coding-agents/trends#weekday-weekend

## Claude Code model mix

Which Claude model appears in public Claude Code runs (2026-05-18 to 2026-07-19):

- Opus: 70.7%
- Sonnet: 28.4%
- Haiku: 1%

## npm install demand (CLI agents)

Weekly npm downloads as an install-demand floor, not an active-user count. CLI agents only; IDE and web agents do not appear here.

| Agent | Demand/wk | Channel |
| ----- | --------- | ------- |
| OpenAI Codex | 10.9M | npm primary |
| Claude Code | 8.2M | Native preferred |
| GitHub Copilot | 1.5M | CLI/package channel |
| opencode | 1.5M | npm primary |
| Pi | 1.1M | npm primary |
| Gemini CLI / Antigravity CLI | 721K | npm primary |
| Google Antigravity | 664K | Legacy npm / Antigravity native |
| Qwen Code | 83,956 | npm primary |
| Cline | 65,236 | npm primary |
| Kilo Code | 50,329 | npm primary |
| Amp | 36,970 | npm primary |
| Codebuff | 9,365 | npm primary |
| Crush | 3,013 | npm primary |

Detail: https://amplifying.ai/coding-agents/npm

## Languages (agent penetration)

Share of all public PRs in each language that carry an agent mark, 2026-06-01 to 2026-06-30.

| Language | Agent-marked share | Leader |
| -------- | ------------------ | ------ |
| Swift | 37.6% | Claude Code |
| TypeScript | 24.7% | Claude Code |
| Rust | 24.2% | Claude Code |
| Python | 23.8% | Claude Code |
| Dart | 22.6% | Claude Code |
| JavaScript | 20.1% | Claude Code |
| HTML | 18.2% | Claude Code |
| PHP | 15.9% | Claude Code |
| Shell | 15.8% | Claude Code |
| Kotlin | 15.6% | Claude Code |
| Go | 14.7% | Claude Code |
| C# | 13.3% | Claude Code |
| C++ | 11.7% | Claude Code |
| C | 11.1% | Claude Code |
| Vue | 10.2% | Claude Code |
| Java | 7.9% | Claude Code |

Detail: https://amplifying.ai/coding-agents/languages

## Industries (where sampled agent work lands)

Sampled agent PRs by industry, 2026-06-29 to 2026-07-19.

| Industry | Share of classified | Leader |
| -------- | ------------------- | ------ |
| AI/ML | 40.4% | OpenAI Codex |
| Devtools | 9.9% | OpenAI Codex |
| Finance | 7% | Devin |
| Gaming | 6% | Cursor |
| Media | 4.9% | Jules |
| Data/Analytics | 3.9% | Devin |
| Security | 3.7% | Devin |
| Crypto/Web3 | 3.6% | Devin |
| Robotics | 3.6% | Jules |
| Productivity | 2.5% | Cursor |

Detail: https://amplifying.ai/coding-agents/industries

## More

- Full written analysis: https://amplifying.ai/research/state-of-coding-agents
- Live index: https://amplifying.ai/coding-agents
- Methodology and coverage: https://amplifying.ai/coding-agents/methodology
- Contact: team@amplifying.ai