Applying AI Fluency Basics

8 examples to get you started with AI Fluency Across Roles - 5 basic and 3 intermediate.

Prerequisites

• No setup is required beyond an active Claude.ai account.
• You should already be familiar with the four stages of the AI Fluency Framework: Delegation, Description, Discernment, Diligence.
• Know which role you are approaching this from (student, educator, small business or nonprofit staffer, builder, or some mix of these) before you start.

Basic Examples

1. Name Your Role Before You Start

Every task sits inside a role, and the role changes what "doing it well" means.

Before opening a chat with Claude, name the role you are acting in for this specific task: are you a student turning in work, an educator preparing a lesson, a staffer writing donor communications, or a developer reviewing generated code.

• The same prompt can be low-stakes in one role and high-stakes in another.
• Naming the role up front tells you which of the four stages deserves the most attention.
• A person can hold more than one role in a single day, so this step is worth repeating per task, not done once.

Related: How AI Fluency Looks Different Across Roles - the concept this walkthrough puts into practice

2. Decide What to Delegate

Delegation is the first decision point: what should Claude do, and what should you keep doing yourself.

Ask a simple question before delegating any task: "If Claude does this instead of me, what do I lose - time saved, or something I actually needed to do myself?"

• A repetitive, low-learning-value task (formatting a table, summarizing a long document) is usually safe to delegate fully.
• A task that is the actual point of the exercise (working through a proof, drafting your own argument) should stay partly or fully in your hands.
• When in doubt, delegate a first draft but keep the final judgment call yourself.

3. Describe the Task With Real Context

Description is how clearly you tell Claude what you need, including the constraints Claude cannot guess.

A vague request like "write a project update" gets a generic answer.

A described request like "write a two-paragraph project update for a nonprofit board that expects plain language, no jargon, and a clear ask for volunteer hours" gets something usable.

• Include the audience, the format, the length, and any constraints that matter to your role.
• If the task connects to earlier work, say so - Claude cannot infer history it was not given.
• Under-describing is the most common cause of a disappointing first answer, not a weak model.

4. Check the Output Before You Trust It

Discernment is the habit of evaluating what Claude gives back instead of accepting it on sight.

Read the response and ask: does this reflect real understanding of the problem, or does it just sound confident?

• For factual claims, check anything you could not have verified yourself before asking.
• For written work, check whether the reasoning actually holds up, not just whether the tone sounds right.
• If you cannot explain why an answer is correct, you are not ready to use it yet.

5. Take Responsibility for What Happens Next

Diligence is what you do after you accept Claude's output - how it gets used, shared, or built on.

Before sending, publishing, or submitting anything Claude helped produce, ask who is affected if it is wrong: a grader, a student, a customer, a donor, a teammate reviewing your code.

• The stakes of Diligence scale with who sees the output, not with how much effort went into producing it.
• Diligence does not end at "send" - it includes noticing later if something needs correcting.
• Skipping this step is the step most people skip when they are in a hurry, which is exactly when it matters most.

Intermediate Examples

6. Walk a Task Through All Four Stages

Take one real task from your role and walk it through Delegation, Description, Discernment, and Diligence in order, writing one line for each stage.

For example, a small business owner drafting a customer email: Delegation - "draft the first version, I'll edit the closing"; Description - "friendly tone, mention the delay, offer a discount code"; Discernment - "check the discount code and dates are actually correct"; Diligence - "this goes to 400 people, so I re-read it once more before sending."

• Writing the four lines out forces you to notice which stage you were about to skip.
• This exercise takes under a minute once it becomes a habit.
• Different roles will naturally spend more words on different stages - that is expected, not a mistake.

7. Adjust Emphasis When the Stakes Change

The same type of task can call for different emphasis depending on who it affects.

Drafting a personal note to a friend and drafting an official statement from your organization both start with "ask Claude to draft it," but the second needs far more Discernment and Diligence before it goes out.

• Ask "who sees this, and what happens if it's wrong" every time the audience changes.
• Higher stakes usually mean slower down at the Discernment and Diligence stages, not skipping Delegation.
• Treat every external-facing or graded piece of work as higher stakes by default.

8. Combine Roles Without Losing the Thread

Many people are more than one role at once - a graduate student who also teaches, a nonprofit staffer who also codes internal tools.

When a task straddles two roles, apply both variants deliberately instead of defaulting to whichever one feels more familiar.

For a teaching assistant grading work, that means applying the student variant's Discernment habit to their own understanding of the material, and the educator variant's assessment-design thinking to how they evaluate submissions.

• Name both roles explicitly rather than picking one and ignoring the other.
• Notice if you are consistently under-applying one role's emphasis - that is usually the one you are less experienced in.
• Revisit this combination periodically as your responsibilities shift.

Related: Choosing the Right AI Fluency Variant for Your Context - compares every variant side by side | Common AI Fluency Pitfalls by Role - what happens when a stage gets skipped

Stack versions: Written against the Claude model lineup current as of ~June 2026 - Claude Fable 5, Claude Opus 4.8, Claude Sonnet 5 (the default), and Claude Haiku 4.5. Model names, pricing, and product features move quickly - verify current specifics at platform.claude.com/docs before relying on them.