AI Tool Adoption Report

Last updated: October 28, 2026

The Tool utilization report shows whether your developers use the AI coding tools you pay for: how many use them, how often, and how much AI output they accept, by tool, team, and person.

Why it matters

Seat counts tell you what you bought. Tool utilization tells you what people use. Use it to find teams that haven't adopted a tool, to compare tools side by side, and to track whether usage is growing. For how much of your shipped code AI wrote, use the Comparison report instead; see Understanding differences between AI Impact (Comparison), Tool Utilization, and Agent Traces.

How it works

Open AI Effectiveness → AI insights → Tool utilization. Pick All tools or one tool in the tool selector, and switch between Weekly and Monthly.

Who counts. Active users and AI days / week count only active contributors. A person is an active contributor on a given day when all of these hold:

They are marked as a code contributor on their Span person record.
They had coding activity (a commit, merged PR, PR review, or PR comment) that day or in the 30 days before.
They were not out of office that day.
They are active in Span and their tool account matches their Span person.

Accepted lines count every code contributor whose tool account matches their Span person, without the coding-activity and out-of-office checks.

What counts as AI use on a day depends on the tool:

Tool	Active on a day when
Cursor	Cursor's own daily active flag is set
GitHub Copilot	The person appears in GitHub's per-user usage report
Claude Code (API connection)	Any Claude Code session, line, commit, or pull request
Claude Code (Enterprise connection)	Any Claude Code activity, or any token use in Claude Code or Claude.ai chat
Claude Cowork	Any Cowork token use (Enterprise connection)
ChatGPT (ChatGPT & Codex Workspace)	At least one chat, message, or any spend in ChatGPT web, desktop, or mobile
Codex CLI (ChatGPT & Codex Workspace)	At least one chat, message, or any spend in any other client, such as the Codex CLI and Codex IDE extensions. Codex lines added and removed count as accepted lines.
OpenAI API (OpenAI Platform)	At least one model request or any cost. Adds no accepted lines, because OpenAI doesn't report lines of code.
Augment Code	Any usage row for the day
Devin, Codegen, Codex Cloud, Async Copilot	The person requested a pull request the agent opened that day
Tools with agent traces (Claude Code, Cursor, Codex CLI, Copilot CLI, VS Code)	Trace activity that day. Trace data replaces the vendor's data for that person, tool, and day.

The metrics

Metric	Calculation
Weekly active users	Active contributors who used an AI tool on any of the last 7 days
Monthly active users	Active contributors who used an AI tool on any of the last 30 days
Active users ratio	Weekly (or monthly) active users ÷ active contributors
AI days / week	`(days with AI use ÷ active contributor days) × 7`
Total Active AI Days	Count of person-days with AI use
Accepted lines of code	Lines the tool reports as accepted: added + deleted, with each Cursor tab completion counted as 2 lines
Accepted AI Lines vs. All Merged Lines	Accepted lines ÷ lines in merged pull requests. Can exceed 100%.

The headline number on the active users card is a daily ratio averaged over your date range, weighted by headcount: Span adds up each day's active AI users and divides by the sum of each day's active contributors.

AI days / week distribution. Span computes each person's AI days per week over the range and groups people into 0-1 (under 1.5), 2-3 (1.5 to under 3.5), 4-5 (3.5 to under 5.5), and 6-7 (5.5 and up), alongside the organization average.

Per-tool and all-tools views. A person who uses two tools counts once in the all-tools number and once in each tool's own number, so per-tool counts can add up to more than the all-tools count. The Adoption breakdown and Breakdown grids show the same metrics by team and by person; accepted-line columns appear when you select one tool.

How it relates to other views

View	Relationship
Comparison (AI Impact)	Attributes the merged diff to AI (AI code ratio), from agent traces. It doesn't use vendor APIs, so an active Copilot user can still have a low AI code ratio on merged work.
Agent traces	A session-level record of what the agent did, joined to pull requests. Where traces cover a person and day, Tool utilization uses them.
PR throughput and cycle time	Outcome metrics. Decide which AI source you're comparing against (AI code ratio or vendor accepted lines) before you correlate.

Supported tools

Connected through an API: GitHub Copilot (Business and Enterprise), Cursor (team plans), Claude Code (API plan through Claude Code API, Enterprise through Claude Code Enterprise, which also covers Claude Cowork), Augment Code, ChatGPT Enterprise through ChatGPT & Codex Workspace, which reports Codex usage under Codex CLI, and the OpenAI API through OpenAI Platform, which reports all API usage, including Codex on API keys, under OpenAI API.
Detected from pull requests, no setup: Devin, Codegen, Codex Cloud, and Async Copilot (Copilot's coding agent).
Through agent traces: Claude Code, Cursor, Codex CLI, Copilot CLI, and VS Code Copilot Chat, including users on plans the API connections don't cover.

Setup

Connect a version control integration. The report needs it.
Connect each AI tool under Settings → Integrations. The report shows only tools you've connected or that Span recognizes from pull requests. See the GitHub Copilot Integration Guide and the Cursor Integration Guide.
Mark who writes code: set Code Contribution on each person record so managers and non-developers don't count.
Check matching: click Integrations near the top of the report and select a tool. The Missing and Unknown tabs show people and accounts that didn't match. OpenAI Platform isn't listed here; check it in the Integrations status on the AI Spend page.

Things to know

Span counts use, not seats. An assigned seat nobody uses doesn't show here. Span has no seat-utilization or idle-seat metric.
Span doesn't measure acceptance rate. Span stores accepted lines, not accepts divided by suggestions shown.
Any use counts as a day. One autocomplete counts the same as a full day of agent work.
One day of use counts for the whole window. A person who uses a tool once stays a weekly active user for 7 days and a monthly active user for 30. Small teams swing more as a result.
Weekends count. AI days / week divides by every eligible calendar day, so someone who uses AI every weekday and never on weekends scores 5.0, and someone who uses it every eligible day scores 7.0.
High AI days with few accepted lines is normal. Chat, research, and planning count as use but add few accepted lines, and tools differ in what they count as a line.
Adoption can rise while AI code ratio stays flat. People can use a tool without much AI-written code reaching merged pull requests. Check trace coverage and PR qualification on the Comparison report.
Vendor dashboards count differently. Vendors count every account. Span drops unmatched accounts, non-developers, people without recent coding activity, and out-of-office days, so Span's numbers usually run lower.
Accepted lines include unshipped code. They count code accepted on branches that never merge and code later rewritten. See Understanding Accepted AI Lines vs. All Merged Lines.
Copilot data starts October 10, 2025. GitHub's per-user usage report has no earlier data. A banner warns you when your date range reaches past a tool's data.
Gemini CLI isn't supported. Neither are Claude Team or Max plans through the API connections; use agent traces for those.
Worked example. Your team has 20 active contributors this week. 15 of them used an AI tool on at least one of the last 7 days, so the weekly active users ratio is 15 ÷ 20 = 75%. One developer had 20 eligible days in the range and used AI on 15 of them: AI days / week = (15 ÷ 20) × 7 = 5.25, which lands in the 4-5 bucket.