AI Fluency at Work
Two people at the same company got Claude access in the same week. One typed "write a quarterly business review" into chat, got something generic, and concluded AI is overhyped. The other gave Claude last quarter's win-loss notes, the account's renewal history, and the one-page format her VP actually reads — and cut a four-hour prep job to forty minutes. Same tool, same week, very different outcome.
The difference is not talent, and it is not a secret prompt. It is fluency: the ability to choose a useful task, give Claude the right context, judge the result, and turn what worked into a method someone else can repeat. That skill is what this program builds. It is not measured by how often you open a chat window.
From scattered experiments to shared methods
Most companies already have early adopters quietly saving hours. The larger opportunity — the reason your company bought this program — is turning those private wins into shared working methods. That requires three shifts:
- Start from a business outcome, not a feature. "Use AI more" is not an outcome. "Cut account-prep time in half without losing source accuracy" is.
- Keep the human owner visible. Every workflow has a person who directs it, checks it, and answers for the result.
- Save what works. A good prompt that lives in one person's head is a trick. Written down, tested, and handed over, it becomes an asset.
Throughout the course we follow Northstar Components, a fictional 400-person industrial sensor manufacturer. Every concept lands on a concrete desk there — sales, support, finance, HR, engineering — because abstract advice does not survive contact with a Monday morning.
Placeholder: replace with a simple spectrum from one-off assistance to a controlled, repeatable workflow.
Three levels of reliance
Claude can contribute at three levels, and knowing which one you are on tells you how much control the work needs.
| Mode | Claude contributes | The person remains responsible for |
|---|---|---|
| Assist | Ideas, summaries, first drafts | Direction, fact-checking, final output |
| Analyze | Patterns, comparisons, options | Evidence quality, interpretation, decision |
| Act | Steps completed through approved tools | Boundaries, approvals, exceptions, monitoring |
At Northstar, brainstorming names for an internal tool is assist: if the output is bad, you lose five minutes. Summarizing churn patterns across 200 support tickets is analyze: a wrong pattern could steer a real decision, so the evidence gets checked. Updating a customer record through a connected tool is act: that changes company data, so it runs inside explicit boundaries with approval gates.
The rule underneath the table: the harder an action is to reverse, the more direct human control it needs. A draft costs nothing to discard. A sent email, a changed record, or a customer commitment does.
Start with work, not features
Pick a recurring task with a clear input and a recognizable good result. The strongest candidates are frequent, information-heavy, and slowed by searching, comparing, drafting, classifying, or reformatting — the connective tissue of office work. A weekly pipeline summary qualifies. "Transform our strategy with AI" does not, yet.
Practice
Write down one real task from your last seven days — you will improve this same task throughout the program:
- What outcome did the task support?
- Which step consumed the most time?
- What information was required to do it well?
- What could Claude assist with?
- What judgment must stay with a person?
Ready to move on
You can name a specific workflow, its owner, the result that matters, and the step where Claude could create value — and you can say which reliance level it needs and why.