Research

Weave Research

Long-form reports on what AI is measurably doing to engineering output. Every figure is computed from production telemetry across Weave customers, never from surveys.

Q2 2026

1,470 organizations · 21,409 engineers

System-measured AI and engineering benchmarks from 1,470 organizations and 21,409 engineers. No surveys.

  • AI-assisted PRs crossed half of June's shipped output value
  • The human review gate got at least 2.5x faster quarter on quarter
  • 14x more tokens consumed per unit of output

The next edition lands at the end of the year and revisits the findings above against a further two quarters of telemetry.