AI code attribution
Also known as AI contribution attribution
Definition
AI code attribution identifies or estimates the contribution of AI tools to a software change. It can rely on direct tool records, explicit metadata, or inference, and its confidence depends on the evidence available.
Attribution is a claim about evidence
A tool may record that it suggested code, that a developer accepted the suggestion, or that an agent produced a commit. These events are related, but they describe different things.
Accepted code can later be edited or removed. A human-authored change may incorporate an AI explanation without containing copied output. No single event necessarily captures every way assistance influenced the result.
Separate direct records from estimates
Consider an illustrative pull request with an AI co-author marker. The marker supports an attribution claim about the commit metadata. It does not reveal exactly which lines came from the tool or how much human review was required.
An estimate based on usage timing has different limitations. Two activities occurring during the same period do not prove that one caused the other. The estimate can still be useful if its method and uncertainty are clearly labeled.
Do not confuse contribution with value
A larger AI-attributed share is not automatically a better outcome. An agent can generate a substantial change that is rejected, or a short correction that resolves a costly problem.
Pair attribution with the fate of the work, its quality, and the effort required to complete it. Compare similar tasks and account for changes in the team or workflow. Attribution helps locate assistance; evaluating its return requires additional evidence about what that assistance achieved.
How this relates to Weave
Weave distinguishes direct integration signals, commit-metadata inference, and estimates based on activity windows. Those evidence types should remain visible when interpreting AI contribution. Token Intelligence connects usage and spending with engineering output, helping teams examine whether adoption is accompanied by useful changes in the work.
Explore Token intelligence