AI code percentage

The AI code percentage report measures the share of eligible code changes in daemon-captured commits that DX attributes to supported AI coding agents. Use it to validate attribution, establish a captured-code baseline, and find teams or repositories worth investigating.

Note: AI code percentage is not an organization-wide adoption measure until daemon, repository, identity, and supported-tool coverage are credible. Code with no supported agent signal can appear as human-attributed.

When to use AI code percentage

AI code percentage answers the question: What share of the captured code changes in this cohort has an AI-attribution signal?

This report helps teams:

  • Validate instrumentation — Inspect known commits before relying on aggregated trends.
  • Establish exposure — See where AI-attributed code is landing across time, teams, groups, contributors, repositories, or attributes.
  • Compare captured and merged code — Check whether the attribution mix changes when you include only commits associated with merged PRs.
  • Choose the next investigation — Move from a difference in exposure to the PRs or sessions behind it.

Do not use the report alone to prove productivity, quality, developer adoption, or ROI.

How AI code percentage is calculated

For current daemon versions, DX calculates the weighted percentage as:

Sum of AI-attributed additions and deletions ÷ sum of all eligible additions and deletions × 100

The calculation sums code changes across the selected cohort before dividing. It does not average the percentages of individual commits.

Daemon versions before 0.5.63 use additions only because those versions did not track deletions reliably:

Sum of AI-attributed additions ÷ sum of all eligible additions × 100

DX returns no value when the denominator is zero. Time-series intervals with no eligible data are omitted rather than shown as 0%.

What is included

The default All code view includes daemon-reported commits in the selected date range after report filters are applied. The report supports date, team or group, attribute, repository, aggregation, interval, and comparison filters.

Select Merged code to include only captured commits associated with merged PRs. This option appears when pull_request_commits data is available.

The daemon excludes or misses changes in these cases:

  • The daemon was not running on the machine where the work occurred.
  • The repository was not imported into DX or its local remote could not be matched.
  • The coding agent was unsupported, its integration was unavailable, or the code was pasted from a browser.
  • The file was binary, vendored, generated, or matched an administrator-defined exclusion.
  • The change occurred before installation.

The report also excludes recognized merge-branch commits and source-control commit rows marked as deleted.

Understand 0%, missing data, and coverage

0% AI means DX recorded no AI-attributed changes in the eligible captured denominator. It does not prove that every change was written without AI. Unsupported tools, pasted code, missed hook events, and work outside monitored scope can all lack an AI signal.

A missing value means DX could not calculate a percentage for that interval or cohort, usually because no eligible code changes were present.

Use Verify deployment to check a known commit and document the following coverage before sharing an aggregate result:

  • Developers and machines in scope
  • Monitored repositories and remote-URL matching
  • Supported coding agents and installed integrations
  • Daemon versions and identity linking
  • Date when collection began

Inspect individual commits

Open a chart point or breakdown row to review the commits behind it. The drilldown includes commit message, author, total changes, AI code percentage, timestamp, commit SHA, and source-control link when available.

For a known commit, compare the reported percentage with the developer’s understanding of the accepted AI and manual changes. Submit attribution feedback from the commit row when the difference is material.

Turn the signal into action

Use the report as the first step in an investigation:

  1. Validate two or three known commits.
  2. Find a team, repository, or period with a meaningful difference and enough captured data.
  3. Open AI pull request overview to compare flow, review, revert, and work-allocation signals.
  4. Inspect the PRs behind the difference.
  5. Change one rollout, enablement, review, testing, or PR-sizing practice and compare the same cohort in the next period.