Coding Horizon

GitHub Just Made The AI Coding War Impossible To Fake

Every figure, field name and capability the finished picture puts on screen, chased to a primary source. GitHub’s own documentation and changelog are the primary sources for what the metrics report; where a claim could not be found stated there, it is listed under Not checked rather than drawn.

Copilot usage metrics exist at organization and enterprise level

Copilot usage metrics reached general availability on 27 February 2026, after a public preview that opened in October 2025. The offering covers a usage dashboard (code completion activity, IDE usage, model and language breakdown) and a code generation dashboard (lines suggested, added and deleted), at enterprise and organization granularity, through both dashboards and APIs.

Active users

The dashboard reports IDE daily active users, “unique users who interacted with Copilot each day”, and IDE weekly active users, “unique users active over a 7-day rolling window”. It also reports a code completions acceptance rate, “percentage of suggestions accepted”.

Chat modes: ask, edit, plan and agent

Requests per chat mode breaks interactions down by mode: Ask, Edit, Plan or Agent. In the underlying data these appear in totals_by_model_feature as chat_panel_ask_mode, chat_panel_edit_mode, chat_panel_plan_mode and chat_panel_agent_mode.

Model usage, and model usage per chat mode

Model usage per day “shows which AI models power Copilot Chat activity”, and model usage per chat mode “breaks down model usage by chat mode (Ask, Edit, Plan, Agent)”. The model dimension carries specific model identifiers plus auto, unknown and others.

One limit, stated by GitHub itself: model usage charts “currently represent chat activity only. Completions data is not included in model breakdowns.” So model usage is not a complete picture of which model wrote the code that landed.

Code generation, accepted code, lines added and lines deleted

Lines of code metrics quantify “the lines it suggested, added, or deleted across completions, chat, and agent features”. The fields are:

Field What it holds
loc_suggested_to_add_sum Lines of code Copilot suggested to add
loc_suggested_to_delete_sum Lines of code Copilot suggested to delete
loc_added_sum Lines of code actually added to the editor
loc_deleted_sum Lines of code deleted from the editor

Agent activity is measured differently from completions

Agent work is not counted the same way, and GitHub says why: “Copilot agent does not follow a ‘suggest then accept’ flow.” Agents plan and execute multi-step tasks, editing multiple files without an explicit acceptance.

So agent file edits are counted as loc_added_sum and loc_deleted_sum under the agent_edit feature bucket, and are not included in the suggested metrics. On a multi-file operation, “each file edit contributes to total added and deleted lines, even if triggered by one prompt”.

The dashboard separately reports agent adoption, “percentage of active users who used Copilot agent”.

The surfaces: IDE chat, IDE agent, code review, CLI and cloud agent

Each surface is a distinct per-user field:

Field What it holds
used_chat Whether the user used IDE chat that day
used_agent Whether the user used agent mode in the IDE that day. Does not include Copilot code review activity
used_cli Whether the user used Copilot CLI that day
used_copilot_cloud_agent Whether the user used Copilot cloud agent that day (used_copilot_coding_agent is retained for backward compatibility)
used_copilot_code_review_active Whether the user actively engaged with Copilot code review that day
used_copilot_code_review_passive Whether the user had Copilot automatically assigned to review their pull request that day, without actively engaging

used_agent and used_copilot_cloud_agent being separate fields is what makes IDE agent mode distinguishable from cloud agent activity.

Activity attributed to individual users

User-level reporting is explicit: “User-level: Analyze individual Copilot usage for a specific day to support enablement and identify where teams may need training or better documentation.” Per-user data is available as NDJSON downloads and through REST API endpoints, and on 28 July 2026 individual Copilot app activity was also “attributed to users in the enterprise-user and organization-user reports”.

Surfaces are compared against each other

The July 2026 rollup expansion states that Copilot app coding activity “is now broken out in the feature, model, and language rollups alongside every other Copilot surface”, so a team can “compare the Copilot app against the IDE, chat, code review, and coding agent surfaces using the same fields you already consume”.

Code review and pull request activity

The code review surface covers pull request creation, review, merge and suggestion activity, including activity performed by Copilot cloud agent and Copilot code review. This is what connects Copilot activity to the delivery artifacts a team already uses: pull requests, reviews and checks.

Pull requests merged per user, as an outcome measure

GitHub does publish one measure that reaches past adoption toward impact: an adoption multiplier, which compares engaged against passive users on “pull requests merged per user per month” and on time to merge pull requests. It is the closest thing in the dashboard to an outcome rather than an activity count.

Not checked