> ## Documentation Index > Fetch the complete documentation index at: https://docs.cubic.dev/llms.txt > Use this file to discover all available pages before exploring further. # AI coding > See how AI tools contribute to your codebase: authorship, agents and models, review burden, and team adoption. The AI coding tab (Beta) shows how much of your merged code was written with AI tools, which tools and models wrote it, and how that code fares in review. > [!NOTE] > > Each member needs the [cubic CLI](https://docs.cubic.dev/ide/cli-review.md), which installs [git-ai](https://usegitai.com) > to record AI authorship on each commit. Until cubic finds git-ai data for the period, the tab > shows only setup steps and an **Install CLI** link. See [AI attribution > tracking](#ai-attribution-tracking). Organization accounts can also filter this tab by member, which no other tab offers. The repository, team, and period [filters](https://docs.cubic.dev/analytics/overview.md#filters-and-time-periods) apply too. ## AI coding adoption The **AI coding adoption** chart shows lines of code authored by AI and by humans over time. | Category | Meaning | | - | - | | **Human** | Written without AI assistance. | | **AI + human** | Generated by AI, then edited by a developer. | | **AI** | Generated by AI and accepted without edits. | The header shows **% of AI code**: the share of attributed lines written with AI help (**AI** plus **AI + human**). Lines from commits without attribution are left out. A **Total lines with AI** line runs over the bars. Two controls change the chart: * **Normalize:** switches from line counts to percentages. * **Day** and **Week:** change the interval each point covers. The subtitle shows the share of commits that carry attribution, for example **from 62% of commits tracked**. A low share means the chart covers only part of your code. See [AI attribution tracking](#ai-attribution-tracking). ## Team members The team members table shows adoption per developer or per team. Switch between **Members** and **Teams**, and filter by name. | Column | Meaning | | - | - | | **% AI code** | Share of lines authored with AI help (**AI** plus **AI + human**). | | **Lines by AI** | AI-attributed lines, including **AI + human**. | | **Lines by human** | Human-authored lines. | | **Top coding agent** | Most-used AI coding agent in the selected period. | | **Top model** | Most-used model in the selected period. | Members who merged PRs in the period but have no attribution data show **No CLI installed** in the **Top model** column, with a link to set it up. Select a column header to sort, except **Top coding agent** and **Top model**. The table starts sorted by **% AI code**. ### CSV export Select the download button to export the current view as CSV. The export includes every row that matches your name filter, across all pages, in your sort order. It leaves out rows marked **No CLI installed**. ## AI coding agents and models The **AI coding agents** section charts lines of code written per agent. Two tables list each coding agent and each model: | Column | Meaning | | - | - | | **Additions** and **Deletions** | Lines added and deleted by the agent or model. | | **Accepted** | AI-generated lines accepted without changes. | | **Overridden** | AI-generated lines that a human modified. | | **Override rate** | **Overridden** divided by **Accepted** plus **Overridden**. | | **Usage count** | How often the agent or model was used, counted per commit and model. | Use them to see which agents and models your team uses most, and whether their code is accepted as-is or edited before merge. ## Quality and review burden **Quality and review burden by AI contribution** compares merged PRs by how much of their code AI wrote. A PR is mostly AI-authored when AI-attributed lines are more than 50% of attributed lines. Otherwise it's mostly human-authored, and so are PRs with no attribution data. With low attribution, the human-authored group includes AI code that wasn't tracked. Four cards show two measures for each group: issues cubic flagged per 1,000 lines added, and average review rounds before merge. These are comparisons and don't show cause. AI-heavy PRs can differ from others in size and type. The section doesn't appear when the period has no data. ## Issues flagged by coding agent and model **Issues flagged by coding agent and model** shows, per agent and per model: | Column | Meaning | | - | - | | **Issues / 1k lines** | Issues cubic flagged in code attributed to the agent or model, per 1,000 lines it authored. | | **Lines authored** | Lines of code the agent or model authored. | | **Issues flagged** | Issues cubic flagged in code attributed to the agent or model. | | **Merged PRs** | Merged PRs that used the agent or model. | | **AI edited by human** | Share of AI-generated lines edited by a human before merge. | By default the tables show only agents and models with at least 5,000 lines authored, plus rows without line counts. Turn on **Show all** to include the rest. Some agents are detected only from commit metadata, such as a `Co-authored-by` line, not from git-ai. Their rows have no line data, so **Lines authored**, **Issues / 1k lines**, and **AI edited by human** don't describe them. Install the CLI for every member to get line-level attribution. ## AI attribution tracking **% of commits with AI attribution** is the share of merged-PR commits that carry AI attribution metadata, the same number as the chart's **commits tracked** subtitle. The card shows the commit counts and the change from the previous period. Low attribution usually has one of four causes: * Some members haven't installed the [cubic CLI](https://docs.cubic.dev/ide/cli-review.md). * Their coding agent, or its version, doesn't record AI authorship. * Code was pasted in from a chat app. * A squash merge on GitHub created new commits without the authorship data. cubic analyzes commits from May 9, 2025 or the day you installed cubic, whichever is later. ### Setup with the cubic CLI The [cubic CLI](https://docs.cubic.dev/ide/cli-review.md) installs [git-ai](https://usegitai.com) during setup. git-ai adds AI authorship notes to commits, so cubic records AI code even when your coding tool doesn't. Opt out at install time with `CUBIC_DISABLE_GIT_AI=true`, or turn tracking off and on later with `cubic stats disable` and `cubic stats enable`. See [Advanced settings](https://docs.cubic.dev/ide/cli-review.md#advanced-settings) in the CLI docs. ## Supported tools cubic detects contributions from more than 30 coding agents, including Claude Code, OpenAI Codex, Cursor, GitHub Copilot, Devin, Google Jules, Windsurf, Aider, Gemini, JetBrains Junie, Amp, and Replit. The list expands over time. If your tool is missing, [contact us](mailto:contact@cubic.dev?subject=AI%20tool%20detection%20request). This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.