Skip to main content

How it works

AI output % measures the proportion of code output (PR estimates) that is attributed to AI assistants. This metric uses Weave’s proprietary Output algorithm to quantify productivity rather than raw line counts. It answers the question: “What percentage of our engineering output is coming from AI?” Because Output measures the meaningful impact of code changes rather than raw line counts, this metric provides a more accurate picture of AI’s contribution to your team’s productivity.

Calculation

The AI output % is calculated as the total output attributed to AI assistants, divided by the total output overall, expressed as a percentage. Formula: AI output % = AI output / Total output Where:
  • AI output = sum of PR estimates for contributions attributed to AI assistants
  • Total output = sum of PR estimates for all merged, non-ignored PRs (both human and AI)
To understand how we distinguish between human and AI output, see AI output attribution.

Technical details

Weave’s code output algorithm analyzes the actual meaning and impact of code changes, providing a consistent measure that’s comparable across individuals, teams, languages, and organizations. This makes AI output % a more meaningful measure of AI’s contribution to your team.
Output is attributed to AI if Weave detects an AI assistant, such as Cursor or Claude Code, from connected AI tool data or from signals in git like co-author trailers. The attribution is determined at the PR level based on which contributors (human or AI) made changes to the pull request.
Output is split within each PR, not assigned to one side. A PR that is 90% AI contributes 90% of its output to AI output and 10% to human output.When you click a slice of an AI vs human chart, the drill-down lists every PR that contributed to that slice:
  • AI output lists PRs with any output attributed to AI.
  • Human output lists PRs with any output not attributed to AI.
A PR that mixes AI and human work appears in both lists. The AI output % and Code output columns show the whole PR, so in the human output list, only the remaining share of that PR’s output counts toward the slice.
Weave uses multiple strategies to attribute output between humans and AI tools, automatically selecting the most accurate method based on available data. See AI output attribution for a detailed explanation of how attribution works.
Last modified on October 17, 2026