Chapter Lab

How is AI being used at work?

If you want a clearer picture of how you or your company use AI, this skill is for you. Admins drop a usage file. Everyone else drops their own chats. You leave with one file you can run in Claude, Cursor, or Codex.

Who: You manage seats, or you only have your own chats and you want to see what you repeat.
What: One skill. It turns a usage export, or five of your chats, into a picture and three next steps.
Why: Paid seats pile up. People keep starting the same blank chat. A picture beats a guess.
When: After the first month on a team plan, or the week you notice the same ask coming back.

Field notes

The sample below is the shape of the result. Open a note if you want where the file comes from, or which path is yours.

Who this is for

You paid for seats, or you keep opening the same kind of chat, and you cannot point at a picture of either one.

Company file (admin)

You can export usage from Claude, Cursor, or Codex. The skill turns that file into seats, products, skill names, and three next steps. It will not invent what people asked. That text is not in the file.

My own work (not an admin)

You only have your chats. Paste five of them, or a personal export. The skill themes your week and tells you what to save. It will not guess about teammates.

Where the file comes from

Claude admin

Owners: initials → Analytics. Export the spend report, and copy skill / product tiles if the CSV is thin.

Cursor or Codex admin

Team analytics in the Cursor dashboard. Codex / ChatGPT Enterprise: workspace analytics export.

Memory

Claude memory is on by default on Free, Pro, and Max. On Team and Enterprise it stays off until you or an admin turn it on. ChatGPT has saved memories. Cursor uses this project's transcripts. Memory is a head start, not the archive.

No export yet

Start with five of your own chats. That is enough to see what you repeat. Ask for the admin file later if you need the company view.

What you leave with

A company usage picture

Sample from a Claude Team export. Yours will follow the same shape after you drop the file into the skill.

Last 30 days

28of 32 seats active

Chat61%

Code22%

Other17%

Skills in the dashboard

Follow-up drafts142
Meeting notes88
Research brief21

From the file

4 people sent 55% of requests. Most of the rest still sits in private chats.

Not in the file

This file does not include what people asked. Topics are not here.

Recommendations

Put the two busy skills in one shared home
Follow-up drafts and meeting notes already have names and volume. People should start there, not in a blank chat.
Ask Sales and Support what they run every morning
The export cannot see the questions. Two short conversations will tell you if those skills match the real work.
Watch the four heavy users before you buy more seats
More than half the requests come from a small group. Learn their motion before you scale it.

What you give the skill

A usage export from Claude Analytics, Cursor team analytics, or Codex / ChatGPT Enterprise. That file has seats, products, volume, and often skill names. It does not have the text of anyone's questions.

Take it with you

One skill for your tool

This is the thing you keep. Same skill in Claude, Cursor, or Codex. Download it, then follow the steps for that tool.

How to install it

01Download the Claude skill.
02Open Claude. Start a Project for this job, or open one you already use.
03Paste the file into Project instructions, or add it as a skill / reference file in that Project.
04Start a new chat in that Project.
05Ask it to use the company-ai-usage skill, then drop the usage export in the chat.

What is in the file

# Company AI usage

Turn an admin usage export into a one-page picture: who is using which product, which skills get used, where requests pile up, and three things to change.

## What you need

One export or a pasted table from the admin dashboard. Accept:

- Claude: Settings → Analytics. The spend / usage CSV, or notes from the Analytics pages (products, skills, projects, seats).
- Cursor: team analytics export, or a careful copy of the Usage Analytics view (active users, models, skills, messages).
- Codex or ChatGPT Enterprise: the workspace analytics export.

If the person has no file, stop. Tell them where to get it in their tool. Do not interview them into a fake company dashboard.

## What the file actually contains

Typical columns and dashboard tiles:

- People: email or name, seats, active users
- Product: Chat, Claude Code, Cowork, Cursor Agent, Tab, Codex
- Volume: requests, tokens, messages, spend
- Skills: name and use count, when the dashboard or API includes it
- Projects: count of users and conversations, on Claude Analytics → Chat

It does not contain the text of anyone's questions or answers. Do not invent topics, "main requests," or what a team is working on. If a column is missing, write "not in this file."

## Method

1. Name the tool and the date range.
2. Report only metrics that are present.
3. Flag concentration (a few people = most usage), chat-heavy product mix with few shared skills, and skills with volume.
4. Write three recommendations that follow from those flags. Each one names the evidence.
5. If they want to know what people ask, tell them to talk to three people this week, or use Cursor Conversation Insights if they are on Enterprise. Do not fill that gap from the CSV.

## Output

Markdown dashboard:

- Headline numbers (seats, active, range)
- By product
- Skills, or "not in this file"
- Concentration
- A one-line note that topics are not in the export
- Three recommendations

## Voice

Plain language. Short sentences. No hype. Do not say the writing is simple. Just be simple.

When a job keeps coming back, decide where it should live. When should I use a chat, a Project, a skill, or an automation?