AI cost per output hour

AI cost per output hour is the inverse of return on token spend: dollars of attributed AI tool spend per unit of engineering output, where one unit is the time an expert engineer would need to complete the merged pull request.

Like return on token spend, and unlike the cost and output series, it is computed only over engineers who have both attributed AI spend and merged code output in the same month, so unreported telemetry on either side does not read as zero. For that cohort, total AI spend is divided by total output. The cohort was 2,799 engineers in January 2026 and 7,020 in August 2026.

Ignored PRs are excluded from output. Spend comes from tool billing and usage telemetry, normalized into dollars; spend that cannot be attributed to a person, such as organization-level overage charges, is excluded. A rising line means AI spend is growing faster than the output it produces.

Sample and interpretation

The matched cohort included 2,799 engineers in January 2026 and 7,020 in August 2026. Monthly counts between those dates are not included in the public snapshot.

These are aggregate observations from Weave customer telemetry, not a controlled experiment or a representative sample of every engineering team. Changes in the observed cohort and connected telemetry can affect the series. The data alone does not establish that AI caused a change in output.

Estimated expert hours are a modeled measure of engineering output, not recorded working hours, hours saved, or a direct measure of business revenue.

For the wider measurement framework, read how to calculate AI coding ROI, including cost attribution, quality, and baseline comparisons.