MOSAIC Agentic 3M: Pull Requests from AI Agents and Humans

June through August 2025 saw 111,969 Pull Requests from AI coding agents and human contributors, released with a paper accepted at MSR 2026. The dataset includes activity metadata for analyzing agent contributions in software development. It was created by AISE-TUDelft and last updated on Hugging Face in April 2026.

Use Cases

• Compare code change patterns between human developers and AI agents based on pull request data.
• Analyze temporal activity patterns of different AI coding assistants mentioned in the description.
• Study the evolution of AI agent contributions over a three-month period based on the described time range.

Strengths

• Contains 111,969 Pull Requests, providing a substantial sample for analysis.
• Includes data from five named AI coding agents (Claude Code, OpenAI Codex, GitHub Copilot, Google Jules, and Devin) and human contributors.
• Associated with peer-reviewed research accepted at the MSR 2026 conference.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for specific agent types is unknown, which may limit granular analysis.
• Data freshness should be verified as the last update was on 2026-04-01.

Provenance

Source

AISE-TUDelft via Hugging Face.

Collection Method

Collected from software development platforms, likely GitHub, as indicated by the pull request focus.

Time Range

June through August 2025.

Freshness

Last updated 2026-04-01 12:45:43.