DEV Community

Raj Narayanasamy
Raj Narayanasamy

Posted on

Employee AI usage metrics when the team uses more than one coding agent

A coding team rarely stays on one assistant, and the same person rarely stays on one surface. Someone lives in Cursor on the desktop, someone else in Claude Code or Codex in the terminal, a third person still has GitHub Copilot open in the editor, and Devin shows up both as Devin Desktop and as a cloud session. Each vendor can show its own usage. None of those consoles answer the question a manager actually asks: which employee used which agent, on which surface, and what did that prompt cost in tokens?

That gap is what employee AI usage metrics are for.

One row per finished prompt

AI Usage Metrics (https://aiusagemetrics.com/) records one row when an employee finishes a prompt. The row names the person, the agent, the surface, and three counts: input tokens, output tokens, and cached tokens. Prompt text stays on the employee’s machine. The report stores the counts.

The surface is one of three:

Desktop. Cursor, GitHub Copilot in the editor, and Devin Desktop.

CLI. Claude Code and Codex in the terminal.

Cloud. A Devin session that ran in the cloud rather than on the desktop app.

Devin Desktop and Devin in the cloud are the same agent and different surfaces. A company that only counts “Devin” will mix a person working at their machine with a session running in the cloud.

A skipped prompt never becomes a row. A later prompt does not overwrite an earlier one. Turns are the number of rows.

Cursor, Claude Code, Codex, and Copilot can send that row from the editor as a hook, and the hook can name the surface. Devin Desktop can send the same kind of row. The company’s Devin cloud usage comes from the enterprise account, once a day, and that pull is measured in ACUs. It does not, by itself, say which of those sessions were Devin Desktop. A row that names the surface is what separates Devin Desktop from Devin in the cloud.

The same report lists them together:

Claude metrics (https://aiusagemetrics.com/claude-metrics.html) for Claude Code

Cursor metrics (https://aiusagemetrics.com/cursor-metrics.html)

Codex metrics (https://aiusagemetrics.com/codex-metrics.html)

Copilot metrics (https://aiusagemetrics.com/copilot-metrics.html)

Devin metrics (https://aiusagemetrics.com/devin-metrics.html), from the enterprise key, in the same employee report, with Devin Desktop kept distinct from a cloud session when the row names the surface

The employee AI usage metrics page (https://aiusagemetrics.com/employee-ai-usage-metrics.html) is that report: who ran the agent, which agent it was, and how many tokens that prompt consumed.

What you can total

From those rows a company can total:

tokens per employee

tokens per agent, for Cursor, Claude Code, Codex, and Copilot

tokens per surface, so desktop, CLI, and cloud stay visible inside one agent

turns, as completed prompts

cache share, from cached input tokens next to raw input tokens

Devin cloud usage stays in ACUs. It is not added into the token columns used for Cursor, Claude Code, Codex, and Copilot.

One email for the combined total

Combined usage alerts (https://aiusagemetrics.com/combined-usage-alerts.html) add that person’s sessions across Cursor, Claude Code, Codex, Copilot, and Devin. The default cap is 250 USD over 7 days. Both numbers are set per company, along with the currency and the time zone.

A person at or over the cap gets one email on a local day. Someone with no email address is left out. The mail goes out during the hours when that company’s usage is lowest. Jobs keeps one row for that run: the quiet hours, how many people were mailed, and whether the run finished. Settings is where the cap is saved.

Configure the company

Sign in with a company email. Personal inboxes such as Gmail are declined. The company id is the email domain.

Devin and Copilot keys

Open Settings after sign-in. Choose Add.

Devin enterprise. Paste the enterprise key. Leave scope empty. This key covers the whole enterprise. Saving it turns the daily Devin pull on.

Copilot enterprise. Enter the GitHub enterprise name and the enterprise token. Saving it turns the daily Copilot pull on.

Repository access. A GitHub App id, an installation id, and a private key. This key can be stored with the others. It does not start the pull.

The key is entered once. The list then shows Configured and an id. The secret is not shown again.

The daily pull

Open Jobs. One ai_metrics_pull row is stored for the company each UTC day. Starting it again the same day does not run a second time.

When the keys are saved, Devin pulls sessions updated on the previous UTC day. Copilot pulls the report from two UTC days ago. A missing key does not stop the other pull. Devin usage for that day is measured in ACUs. Copilot usage is measured in AI credits.

Editor hooks

Settings also lists hook connections: Cursor, Claude, Codex, Copilot, Devin Desktop, Antigravity, OpenCode, and a direct feed. These are messages from the editor after a prompt. One plugin (https://aiusagemetrics.com/plugin.html) covers those editors, plus Gemini Code Assist once a Google Cloud key is saved. The Devin Desktop hook is that desktop app. The daily Devin pull remains the enterprise total and does not mark the desktop hook as received. Copilot’s daily pull is separate from the Copilot hook in the same way.

Jobs shows hook health for the same connections:

Not connected means no message has arrived.

Receiving means a message arrived in the last day.

Quiet means the last message is older than a day.

Failed means the last call failed and it was still within a day.

Several laptops in one company share one row per connection. The direct feed is the same list, for a company that posts usage itself.

What this is not

Vendor docs remain the source for a single product. Claude, Cursor, Codex, Copilot, and Devin each explain their own dashboards. This report does not replace those consoles. It lines the people up across them.

The public demo is sample data.

Try the sample

Open AI Usage Metrics (https://aiusagemetrics.com/) and use View the demo with a company email. Read Employee AI usage metrics (https://aiusagemetrics.com/employee-ai-usage-metrics.html) first if you want the row described before you sign in. After sign-in, add the Devin enterprise key on Settings, then check Jobs for the daily pull, the hook health, and the combined usage alert run.

Top comments (0)