Engineering output rose 78% in Q3 as AI spend grew 2.5×

From Weave Research, a publication of Weave.

Engineering output per employee rose 78% in Weave's platform-wide monthly averages, from 76 units in June to 136 in September. Average AI spend per employee rose 2.5× over those months, from $257 to $637. In September, output grew 16% while spend grew 9%, the first month this quarter when output growth outpaced spend.

The Q3 2026 AI Impact Report draws on system-measured analysis covering 22,623 engineers, rather than survey responses. Its platform-wide monthly averages show the steepest three-month output increase in 21 months of platform data. The August Weave Index figure below uses a separate matched cohort.

Weave connects complexity-weighted engineering output with AI-tool spend, code quality, review, and delivery, giving leaders a connected view of how engineering outcomes and AI investment move together.

Output and AI spend both rose sharply

Weave's Q3 report shows platform-wide average monthly output per employee rising from 76 units in June to 136 in September. July added 28%, August 20%, and September 16%. The report also tracks a constant cohort: holding the set of organizations fixed, its median rose 3.1× over four quarters, compared with 2.8× for the blended average.

Average AI spend per employee rose from $257 a month in June to $637 in September. Its month-over-month growth slowed from 69% in July to 34% in August and 9% in September.

Source: Q3 2026 AI Impact Report, platform-wide monthly averages.

September was the first month this quarter when output growth outpaced spend growth. Spend was still rising, while the gap between the two growth rates narrowed.

Pair platform trends with a matched-cohort return measure

The report's output-per-employee and spend-per-employee series are platform-wide averages. The Weave Index publishes a separate matched-cohort measure for readers who want an output-to-spend ratio: engineers must have both attributed AI spend and merged output in the same month. The August Index cohort contained 7,020 engineers, drawn using those matched-data requirements.

For August 2026, the Weave Index's return on token spend was 0.091625 estimated expert hours per attributed dollar. The ratio is total estimated output divided by total attributed AI spend for that cohort and month.

The Weave Index expresses modeled engineering output as estimated expert hours per attributed dollar. Its public methodology explains how the unit is modeled from pull-request analysis and what cost and repository coverage it includes. This output-to-spend ratio differs from a financial ROI percentage or recorded hours saved.

Put output, spend, and quality in one view

The advantage is seeing engineering output alongside AI adoption and cost. Weave connects supported AI-tool usage and spend with complexity-weighted output, code quality, review, and delivery signals. Leaders can use those views together to understand how outcomes move as teams scale AI investment and where to focus next.

For a company-level evaluation, compare the same teams and period, reconcile usage and subscription costs to billing data, and pair output with review, revert, quality, and delivery signals. That makes it possible to decide which tools and workflows to expand using the outcomes that matter to the business.

Read the Q3 2026 AI Impact Report, the return-on-token-spend series, and the AI ROI measurement methodology. To see these measures across your own engineering tools and workflows, request a Weave walkthrough.

How to read the numbers

What does the 78% represent?

Weave's platform-wide average output per employee rose from 76 units in June to 136 in September. The Q3 report's system-measured analysis covers 22,623 engineers. The 78% is separate from the Index's matched-cohort return measure. For a company-level view, compare the same teams and period using aligned output, spend, and quality measures.

What does return on token spend measure?

For August 2026, the Weave Index reported 0.091625 estimated expert hours per attributed dollar across a separate matched cohort of 7,020 engineers. It is total estimated output divided by total attributed AI spend for that cohort and month. It describes modeled output per dollar, rather than a financial ROI percentage or recorded hours saved.

How can a team calculate its own return?

Match output and attributed spend for the same group and period, then review quality and delivery alongside both. The Weave Index publishes an aggregate matched-cohort series using that structure.