AI Fluency Across Roles Best Practices
Ten practices for applying the 4D Framework, Delegation, Description, Discernment, Diligence, in a way that actually fits the role you're in, rather than a generic version of the framework that ignores your specific context.
How to Use This List
- Treat these as habits to build, not a one-time checklist to complete.
- Each practice names the stage of the framework it strengthens, so you can see which of the four you tend to under-apply.
- Revisit this list whenever your role or responsibilities shift, since the practices that matter most change with context.
A - Know Your Context Before You Delegate
- Name your role for the specific task at hand. A single person can be a student, a builder, and a small-organization volunteer in the same week - identify which role applies before deciding how much to hand to Claude.
- Match your delegation level to what's actually at stake. Low-stakes, easily reversible tasks can be delegated freely; graded work, public statements, and production code need a lighter delegation touch and a heavier review step.
- Delegate the parts that aren't the point of the task. Formatting, summarizing, and first-draft generation are usually safe to hand off; the reasoning or decision that the task exists to test or protect should stay closer to you.
B - Describe With the Context Claude Can't Guess
- Give Claude the constraints that matter for your role, not just the topic. A nonprofit's donor tone, a classroom's grade level, or a codebase's deployment topology are all context Claude cannot infer - state them explicitly.
- Keep a reusable context note for repeat tasks. A short standing reference (audience, tone, things to avoid, technical constraints) saves you from rebuilding Description from scratch every time, which is especially valuable under time pressure.
C - Apply Discernment at the Level Your Role Requires
- Check whether an answer reflects real understanding, not just a plausible tone. This matters most for students and educators evaluating whether learning actually happened, but it applies to anyone reviewing AI-assisted reasoning.
- Review technical output the way you'd review a teammate's pull request. For builders, this means checking assumptions against the real system, not just whether the code runs without errors.
- Read the explanation alongside the output, not just the output itself. Assumptions and caveats are often stated in prose before you ever look at the actual content, code, or draft - reading them first surfaces gaps faster.
D - Follow Through With Diligence
- Ask who is affected before anything goes out. A grader, a student, a donor, a customer, or a teammate reviewing a merge all carry different stakes - let that answer set how careful your final check needs to be.
- Own the outcome after you hit send, not just the moment of production. Diligence includes noticing later if something needs correcting, not only reviewing once before publishing, submitting, or merging.
FAQs
Do I need to apply all ten practices to every task?
No - match the practices to the stakes of the specific task.
A low-stakes personal note doesn't need the same level of Discernment and Diligence as a public-facing statement or graded submission.
Which practice matters most if I only have time for one?
Naming your role for the task at hand (practice A1), since it determines which of the other nine practices deserves the most attention in that specific context.
How is this list different from the pitfalls list in this section?
This list states the positive version of good habits; the pitfalls list names the specific mistakes that happen when these habits are skipped.
Reading both together gives you the practice and the failure mode it prevents.
Is keeping a reusable context note really worth the setup time?
Yes, for any task you repeat regularly - it turns a Description step that would otherwise be rebuilt from scratch each time into a quick paste, which matters most under time pressure when Description is most likely to get skipped.
Does "delegate the parts that aren't the point of the task" apply outside of school?
Yes - the same principle applies to a builder delegating boilerplate but keeping architecture decisions close, or a small-business owner delegating a first draft but keeping the final sign-off.
The point of the task, whatever that is in your context, is what should stay closest to you.
What does "review technical output like a pull request" mean for non-developers?
It's specific to builders and technical output, but the underlying idea (checking assumptions against your real situation before accepting an answer) applies to any role reviewing AI-assisted work.
How do I know if I'm under-applying Diligence?
A common sign is sending or publishing AI-assisted work with only one quick read, or considering the task done the moment something is sent rather than staying responsible for what happens after.
Can these practices change as my role changes?
Yes, and they should.
Revisiting this list when your responsibilities shift, a new job, a new class, a new project, keeps the practices matched to your actual current context rather than an old one.
Is under-delegating ever a problem this list addresses?
Yes, indirectly - practice A3 and B1 both push toward delegating what's safe to delegate, since under-delegating wastes the efficiency Claude offers on tasks that don't require your full personal effort.
Where should I go next after this list?
The role-specific articles in this section (students, educators, small businesses and nonprofits, builders) apply these ten practices in much greater depth for each context.
Related
- How AI Fluency Looks Different Across Roles - the concept these practices are built on
- Applying AI Fluency Basics - a walkthrough for putting these practices into action
- Common AI Fluency Pitfalls by Role - the mistakes these practices are designed to prevent
- Choosing the Right AI Fluency Variant for Your Context - matches these practices to a specific role variant
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.