AI Fluency: The 4D Framework
Having access to an AI assistant is not the same as being fluent with it. The Framework for AI Fluency names four competencies that make the difference: Delegation, Description, Discernment and Diligence. Read together, they form two loops: a strategic loop that decides what AI should do and how to do it responsibly, and a tactical loop of describing, evaluating and refining, prompt by prompt.
Four competencies, two loops: what separates using AI from working well with it.
Most teams now have access to a capable AI assistant. Fewer get consistently good results from it. The answers miss the point, the request is hard to put into words, and nobody is quite sure which data may go in. That gap between access and fluency is what the Framework for AI Fluency addresses.
Three ways of working with AI
The framework starts by naming how people actually use AI [1][2]:
- Automation. The AI completes a defined task from your instructions, such as summarising a report or drafting an email.
- Augmentation. You and the AI work on a task together, as thinking partners, for example when you develop an argument or test a concept.
- Agency. You configure the AI to act on its own on your behalf, such as sorting incoming requests and drafting replies.
No mode is better than the others, and one project often uses all three. The four competencies apply to each of them.
The four Ds
Each competency has three parts. Knowing the parts makes the competency teachable and checkable [2].
| Competency | The question it answers | Its three parts |
|---|---|---|
| Delegation | What should AI do, and what stays with me? | Problem awareness, platform awareness, task delegation |
| Description | How do I tell the AI what I need? | Product, process and performance description |
| Discernment | Is what I got good enough? | Product, process and performance discernment |
| Diligence | Am I using AI responsibly? | Creation, transparency and deployment diligence |
Delegation begins before any AI is involved. Problem awareness means knowing your goal and what kind of work it needs. Platform awareness means knowing what the available systems can and cannot do. Task delegation is the deliberate split: what is automated, what is done together, what an agent may handle and what stays with people because it needs judgment [2]. The most effective AI users are experts in their field first and delegators second.
Description goes beyond writing prompts. Product description says what you want: context, task, format, audience. Process description says how the AI should approach it, for example the steps, the data to use or the method. Performance description says how the AI should behave with you: challenge your assumptions or follow your lead, answer briefly or in depth [2].
Discernment is your quality control. Product discernment asks whether the output is accurate, fitting and useful. Process discernment checks how the AI got there and watches for logical gaps, circular reasoning or a fixation on one detail. Performance discernment asks whether the interaction itself works for you [2]. Discernment depends on domain knowledge: you can only judge what you understand.
Diligence covers responsibility. Creation diligence means choosing systems and data deliberately: who owns the data you enter, and who can see it afterwards? Transparency diligence means being open about the role AI played with everyone who needs to know. Deployment diligence means checking facts, sources, bias and usage rights before you use or share a result, because the responsibility stays with you, not with the AI [2].
Two loops, not a checklist
The four Ds are easy to learn as a sequence. In practice they behave differently. The Teaching AI Fluency course describes a more useful view: two loops that constantly inform each other [3].
Strategic loop: the big picture
Delegation → Diligence: each decision about who does the work raises questions of responsibility.
Tactical loop: prompt by prompt
Describe, evaluate, refine, repeat.
Diligence → Delegation: each obligation reshapes who does the work.
The strategic loop: Delegation and Diligence
This loop holds the big decisions: what we are trying to achieve, how we divide the work, which systems fit our values and obligations, and how we keep the collaboration responsible [3]. It moves from asking “how do I use AI?” to “how do I make good decisions about AI?”. It runs in both directions.
Forward, from Delegation to Diligence. Suppose a quality team wants a summary of last year’s supplier audit reports. Problem awareness: the goal is an overview of recurring findings for the management review. Platform awareness: which approved assistant handles long documents well? Task delegation: the AI groups and summarises the findings, the team draws the conclusions. Each of these choices immediately raises diligence questions. May supplier names and audit details go into this system? Who needs to know that AI produced the first summary? How will we check that every finding in the summary exists in the source reports?
Backwards, from Diligence to Delegation. Often the obligation comes first and reshapes the plan [3]. If the rule is that confidential supplier data stays inside the company’s approved environment, the delegation changes: the AI works only in that environment, or only on anonymised excerpts, and the interpretation stays with people. If you commit to documenting every AI contribution, you know exactly what you are accountable for, and that makes it easier to hand over more work with confidence.
The point of this loop is that effective collaboration is not only about what AI can do, but about what it should do, given the context, the constraints and the consequences [3]. Teams that work this way treat rules as clarifying, not as limiting.
The tactical loop: Description and Discernment
This is the moment-to-moment work: how do I express this specific need, does the output meet my standard, how do I refine my request [3]? You describe, you evaluate, you refine, and you repeat. Most interactions with AI are small loops of exactly this kind [2].
The loop runs on the same three lenses on both sides [3]:
| Lens | Description: what you say | Discernment: what you check |
|---|---|---|
| Product | What you are creating together | Is the result good, and what does good look like now? |
| Process | How to think through the problem | Is the approach finding things we would miss alone? |
| Performance | How the AI should work with you | Does this way of working still serve the goal? |
A typical product loop starts broad. “Write something about supplier quality” returns a generic text. Discernment shows it is not what you need, so the next description is narrower: recurring findings in incoming goods inspection, for an audience of plant managers. Each round improves the output and sharpens your own idea of the goal [3].
A performance loop works the same way. You might ask the AI to act as a friendly but strict sceptic. Discernment then shows that the challenges come too early, so you refine the instruction: hold back the scepticism until I have developed my first ideas [3]. This is why the course moves away from the idea of the perfect prompt. The quality comes from a conversation that builds shared context over time, not from a single instruction [3].
How the two loops fit together
The strategic loop sets the direction and builds the frame: what we create, with which assistant and under which rules. The tactical loop fills that frame with the actual work [3]. Information also flows back. Discernment can show that the chosen system does not suit the task, which sends you back to Delegation. Diligence shapes Description, for example when data protection means a prompt must work without personal data. And if you cannot describe what you want, you have usually not yet finished the Delegation step of defining your goal [3].
What this means for your team
Fluency develops through practice, not through a single training day [2]. Four habits make a good start:
- Define the goal before opening a tool. Write down what success looks like. This is problem awareness, and everything else builds on it.
- Agree on quality criteria in advance. Discernment is much easier when you know what you are checking for before you read the output.
- Settle the diligence questions per use case. Which data may go in, who needs to know about AI involvement, who checks and signs off the result.
- Keep a short diligence statement. Note which AI you used, what it contributed and how you reviewed it. It makes your work transparent and your accountability clear [2].
Teams that work this way do not depend on a collection of copied prompts. They know why they use AI for a task, how to steer it and when to stop and check. From AI tools to productive use describes how to build that practice across a whole organisation.
Sources
- Dakan R, Feller J. Framework for AI Fluency: practical overview document. Version 1.1. Ringling College of Art and Design and University College Cork; 13 January 2025. ringling.libguides.com
- Anthropic. AI Fluency: Framework and Foundations. Anthropic Academy. anthropic.skilljar.com
- Anthropic. Teaching AI Fluency. Anthropic Academy. anthropic.skilljar.com
AI Enablement
We enable you and your people to use AI properly, tailored to your skills, your processes and the tools of your working day. Your team grows confident in putting AI to effective use in daily work.
Frequently asked questions
Is AI fluency the same as prompt engineering?
No. Prompting is one part of Description. AI fluency also covers deciding whether and how to use AI at all, judging what it produces, and taking responsibility for the result [1][2]. Prompt tricks also go out of date as models change, while the four competencies stay relevant.
Do I have to work through the four Ds in a fixed order?
No. A linear order from Delegation to Diligence is a good way to learn them. In real work the competencies influence each other constantly, which is why the Teaching AI Fluency course presents them as two connected loops [3].
Is the framework tied to Claude or to Anthropic?
No. The framework was developed by Prof. Rick Dakan (Ringling College of Art and Design) and Prof. Joseph Feller (University College Cork) and applies to any AI system [1]. Anthropic built two courses on it with the authors, and these use Claude for the exercises [2][3].
What is a diligence statement?
A short note that says which AI systems you used, what they contributed, how you reviewed the output and that you take responsibility for the final result [2]. It turns transparency and deployment diligence into a concrete habit.
Where do I start with a team?
With problem awareness: a clear goal and a clear picture of what good looks like, before anyone opens an AI tool. Teams that agree on quality criteria first find Description and Discernment much easier, because they know what to ask for and what to check.