🛰️ Daily AI Frontier
‹ back to 2026-07-16

Early Adoption of Agentic Coding Tools by GitHub Projects

Research Agents & Tool Use

Merged summary

TL;DR — An empirical study of 25,264 agent-generated pull requests across 2,361 popular GitHub repositories, examining how agentic coding tools are adopted and how humans oversee them at the project level. It matters because successful agent integration depends on human/organizational processes, not just agent capability.

  • Adoption is concentrated: the median repository produced only one to two agentic PRs over a three-month period, so intensive use is limited to a small subset of projects.
  • Small projects (1–5 contributors) showed higher participation ratios and average agentic PR activity than medium and large projects.
  • Productivity varied widely; a few projects exceeded an industry-reported benchmark of 36 PRs per participant per three months, but most fell below it.
  • Human-agent collaboration was dominated by a single-human oversight model (one developer reviews/modifies the agent's work); multi-human collaboration patterns remained uncommon.

Sources (1)

Early Adoption of Agentic Coding Tools by GitHub Projects

arXiv cs.SE Maliha Noushin Raida, Daqing Hou 2026-07-15 arXiv:2607.14037

TL;DR — An empirical study of 25,264 agent-generated pull requests across 2,361 popular GitHub repositories, examining how agentic coding tools are adopted and how humans oversee them at the project level. It matters because successful agent integration depends on human/organizational processes, not just agent capability.

  • Adoption is concentrated: the median repository produced only one to two agentic PRs over a three-month period, so intensive use is limited to a small subset of projects.
  • Small projects (1–5 contributors) showed higher participation ratios and average agentic PR activity than medium and large projects.
  • Productivity varied widely; a few projects exceeded an industry-reported benchmark of 36 PRs per participant per three months, but most fell below it.
  • Human-agent collaboration was dominated by a single-human oversight model (one developer reviews/modifies the agent's work); multi-human collaboration patterns remained uncommon.
item →