Your AI coding spend bought 25% more output. Duplication rose 81%.

Since AI tools arrived on the software scene, many companies have pinned their hopes on them. However, the tone has shifted recently as CFOs and CTOs have demanded visibility into AI spend. IBM's Gary Cohn recently said the ROI has "not been nearly as high as people might think."

GitClear analyzed 623 million code changes from 2023 to 2026 and found that heavy AI users gained only 25% on their prior velocity, far from the claimed 10x increases. These heavy users still out-produced non-AI users by 4 to 10x, but this improvement was present before AI tooling was adopted. Most teams will face harder usage limits and may not sustain the usage levels.

Refactoring, a key factor in maintainability, fell to 3.8% of changed lines in 2026 from 21% in 2022. Code duplication rose 81% over the same period, with multiple expressions of the same concept becoming more common. Productivity gains in terms of output are not guaranteed to translate into value for the organization or its customers.

Technical practices like test automation and refactoring have become less prioritized as developers aim for faster output. The data suggests that the promise of 10x boosts with AI coding assistants has not materialized. Instead, the software industry may be returning to a code-and-fix era. Without understanding the inherent properties of value streams, organizations risk paying a painful price when trying to track ROI.