It is not just knowing which buttons to click.
It is having the judgment to use AI in the right way, at the right time, for the right kind of task.
If you want AI to help with real work, AI fluency is the skill to build.
The 4D Framework for AI Fluency
The 4D Framework for AI Fluency comes from a research collaboration between Professor Rick Dakan (Ringling College of Art and Design) and Professor Joseph Feller (University College Cork).
It breaks AI fluency into four skills that work together.
1) Delegation
Delegation is deciding what should be done by a person, what should be done with AI, and how to combine the two.
It includes:
- Clarifying the goal
- Knowing what AI is good at, and where it often fails
- Making smart choices about who does what in the workflow
2) Description
Description is communicating clearly with an AI system.
It includes:
- Stating the output you want
- Sharing useful context, examples, and constraints
- Setting expectations for tone, format, and how the tool should behave
If you have used a prompt pattern like “set the stage, define the task, specify rules,” you have already been practicing Description.
3) Discernment
Discernment is judging AI output with a critical eye.
It includes:
- Checking quality, accuracy, and fit for purpose
- Catching subtle mistakes or missing context
- Noticing what to improve in your prompt or process
4) Diligence
Diligence is using AI responsibly.
It includes:
- Choosing tools and data carefully
- Being transparent about AI help when it matters
- Taking accountability for the final result
Evaluating AI tools for your workflows
Once you start using AI in more places, a practical question shows up fast:
How do I know if an AI tool is actually good at this task for my work?
That is where Discernment becomes a daily habit.
One simple way to build it is to run small, lightweight evals (short for evaluations).
Evals are a structured way to test how well an AI tool performs on the tasks you care about.
Why evals matter
Your work is specific.
An AI tool might be great at drafting marketing copy, but need more guidance for technical documentation in your domain.
Simple evals help you:
- See where AI adds the most value in your workflow
- Spot tasks where you need more context, examples, or constraints
- Build confidence in results for repeatable work
A simple eval you can run this week
You do not need complex infrastructure.
You can get useful signal with a small set of examples and a consistent way to compare.
- Gather examples. Collect 5 to 10 real examples of a task you do often.
- Emails you have written
- Reports you have created
- Analyses you have done
- Create test prompts. Write prompts that should produce similar outputs.
- Include the context you normally have
- Add constraints like format, length, tone, and audience
- Compare outputs. Run your prompts and compare the AI responses to your originals.
- Does it include the key information?
- Is the tone right?
- What is missing?
- What is incorrect?
- Refine your approach. Update prompts and process based on what you learn.
- Add examples that show what “good” looks like
- Tighten constraints
- Decide where human review is always required
If you work with data
This approach is especially helpful for analysis work.
To test an AI tool with your data:
- Pick a dataset you have already analyzed manually
- Ask the AI to do the same analysis
- Compare the results to your original work
- Track patterns and adjust the prompt
- For example, the numbers might be right, but the tool may miss the larger pattern or the business implication
The goal is not to prove an AI tool is “perfect.”
The goal is to build a clear sense of where it is strong, where it needs help, and where you should slow down and review carefully.