AI Fluency: Framework & Foundations
AI fluency is the ability to collaborate with AI in ways that are effective, efficient, ethical, and safe. Those four adverbs are the whole definition. This course teaches collaboration as a durable skill, so what you learn outlasts any model generation.
- Define AI fluency and its four adverbs: effective, efficient, ethical, safe
- Apply the 4D framework: Delegation, Description, Discernment, Diligence
- Run the Description-Discernment loop on real work
- Disclose AI's role in your work with a diligence statement
Introduction to AI fluency
Most advice about AI tells you what to type. This course teaches something sturdier: how to think about working with AI at all. Models keep changing; a prompt trick that works today may be pointless next year. The skill of collaborating well, deciding what to hand over, communicating what you need, judging what you get, and standing behind what you use, does not expire.
AI fluency is the ability to collaborate with AI in ways that are effective (it achieves your goal), efficient (it is worth the time), ethical (it aligns with your values and obligations), and safe (it protects you and others from harm).
By the end you will have four things: a framework for thoughtful interaction, confidence about when and how to use AI, collaboration skill you have actually exercised, and the habits to responsibly evaluate and share AI-assisted work.
- AI fluency = effective, efficient, ethical, safe collaboration
- A durable framework, not a bag of prompt tricks
- Four competencies: Delegation, Description, Discernment, Diligence
Up next: the three ways people engage with AI, and why mixing them up causes most confusion about "using AI well".
Why we need AI fluency
Fluency combines practical skills, knowledge, insights, and values. To see why it matters, start with the three fundamentally different ways people engage with AI:
AI completes specific, well-defined tasks on your instructions. Reformatting a document, translating a paragraph, generating fifty product descriptions.
You and the AI collaborate as thinking partners. Ideas move in both directions: a strategy drafted together, a hard analysis worked through.
You configure AI to work independently on your behalf, shaping its knowledge and behavior rather than handing it single tasks.
Each mode demands different things from you. Automation needs precise instructions and spot checks. Augmentation needs genuine back-and-forth. Agency needs careful configuration up front and accountability for everything the system does while you are not watching.
Label your last three AI uses: automation, augmentation, or agency. Did you bring the right kind of attention to each? The common failure is treating augmentation like automation: one message sent, whatever comes back accepted.
- Automation executes, augmentation thinks with you, agency acts for you
- The mode changes what fluency requires
Up next: the framework itself, four competencies in four sentences.
The 4D framework
Everything else in this course unpacks these four definitions:
Thoughtfully deciding what work to do with AI and what to do yourself.
Communicating clearly with AI systems.
Evaluating AI outputs and behavior with a critical eye.
Ensuring you interact with AI responsibly.
The framework's central claim: these competencies survive model evolution. Models get better at vague requests, but deciding what belongs with AI never stops being your call. Outputs get more polished, but judging whether polished equals right never stops being your job.
The four questions
Each D becomes a question you can ask about any collaboration:
- What would I keep for myself, and why? (Delegation)
- What context and instructions does the AI need? (Description)
- How will I verify what comes back? (Discernment)
- What do I owe the people who receive this work? (Diligence)
A colleague says a report "looks finished" and sends it out unreviewed. Which competency did they skip?
Discernment is the evaluating competency, and it is the one research shows people skip most, especially when output looks polished. (Diligence fails too, but the skipped step was the evaluation.)
- Delegation decides, Description communicates, Discernment evaluates, Diligence owns
- The competencies outlast any specific model
Up next: the machine you are collaborating with, in two short lessons.
Generative AI fundamentals
You do not need to be an engineer to be fluent, but you need a working model of the machine. Generative AI creates new content rather than only analyzing existing data. Three ingredients made modern language models possible: the transformer architecture, vast training data, and large-scale compute. Learning happens in two stages: pre-training (the model reads enormous amounts of text and learns the patterns of language and the world) and fine-tuning (that raw capability is shaped into assistant behavior).
Everything the model is currently paying attention to lives in a fixed-size working memory. Anything outside it has to reach the model another way: through what its weights already hold, or through search, memory features and tools.
Abilities nobody explicitly programmed appeared as models scaled: translation, reasoning steps, style imitation. This is why capability maps date quickly and testing beats assuming.
- Transformers + data + compute; pre-training then fine-tuning
- Hold two concepts: the context window and emergence
Capabilities and limitations
Versatility across language tasks, conversational awareness, switching between tasks without retraining, connecting to external tools.
Knowledge cutoffs, hallucination, context window limits, genuinely complex multi-step reasoning.
The best applications combine strengths: AI supplies speed, breadth, and drafting power; the human supplies critical thinking, judgment, creativity, and ethical oversight. Neither alone matches the pair.
The field moves fast, so treat any capability map, including this one, as dated the day it is written. The durable move is knowing how to probe: give it a task you can verify, and look.
Up next: Delegation, the deciding competency, in depth.
A closer look at Delegation
Delegation is deciding what to do yourself, what to do with AI, what AI handles alone, and how to distribute the work. Three components:
Understand your goal and the work before involving AI. If you cannot say what success looks like, no tool can help you reach it.
Know what different AI systems can and cannot do. "AI" is not one thing.
Strategically divide the work so each side does what it is genuinely better at.
Good delegation needs both domain expertise and AI-capability knowledge. One without the other fails: the domain expert who does not know the tool hands over the wrong things; the tool expert who does not know the domain cannot tell good output from plausible output.
The goal is not to automate everything. It is the most effective partnership per task. Some work should stay entirely yours, and recognizing that is fluency, not failure.
Take a real upcoming task and discuss the delegation plan with an AI as a conversation, not a list. Tell it the goal, ask where it would struggle, and push back. You each may see something the other does not.
- Delegation = problem awareness + platform awareness + task delegation
- It requires knowing both your domain and your tools
Practice: project planning with Delegation
This course threads one real project through everything that follows. Pick it now: something multi-step you could complete in about an hour. An event to plan, a purchase to research, a proposal to draft, a study plan to build.
- Define the vision with the AI. Describe your project and invite the AI to ask questions until the success criteria are genuinely clear. Its questions will expose what you had not decided.
- Break the project into tasks. Every step, however small.
- Run each task through the delegation questions, one at a time. Do I do this myself? With AI? Does AI do it alone? Why? Challenge its suggestions and let it challenge your assumptions.
- Save the plan. Later lessons return to this project.
- Define success first, with the AI asking questions until it is clear
- Delegate task by task, with reasons, not wholesale
Up next: Description, and why it is more than prompting.
A closer look at Description
Description is more than prompting: it is creating a collaborative environment. AI systems are interactive partners, not databases or vending machines, and they cannot read your mind. Three parts:
What you want: the output, its format, its audience, its style.
How the AI should approach the request. The path can matter as much as the destination.
How the AI should behave in the collaboration: concise or detailed, challenging or supportive.
Most people only ever describe the product. The other two thirds are where collaborations turn from adequate to excellent.
Write the laziest version of a request you actually need ("write me a newsletter"). Improve it yourself using all three descriptions. Then give the AI the lazy version and ask it to improve the prompt. Compare what information it added and how it organized it. Run this both ways a few times and you will internalize what good description contains.
- Describe the product, the process, and the performance
- A partner, not a vending machine, and it cannot read your mind
Effective prompting techniques
Prompt engineering, plainly: designing effective instructions, using ordinary communication principles plus a few AI-specific considerations. Six techniques cover most of the craft:
What this is, why you need it, the background a new colleague would require.
One real example of good output beats three paragraphs of adjectives.
Format, length, structure, hard requirements.
Rather than asking for everything at once.
Give it room to reason before answering, especially on anything analytical.
Who it should be, and how it should sound.
Ask the AI itself to improve your prompt. It knows what information it is missing better than you do, and it will tell you if you ask.
Your draft request reads: "Make a good presentation about our project." What is the biggest missing element?
"Good" is undefined and the AI knows nothing about the project, the audience, or the goal. Context is technique one for a reason; without it, every other instruction shapes a guess.
The first response is a starting point. Refinement based on results is the actual work, and research shows every other fluency behavior strengthens with it.
- Context, examples, constraints, steps, thinking room, role
- Iterate: the first response is a draft of the conversation, not the end
Up next: Discernment, the flip side of Description.
A closer look at Discernment
Discernment evaluates what the AI produces, how it produced it, and how it behaves. It mirrors Description exactly:
Quality of the output: accuracy, appropriateness, coherence, relevance.
How the AI got there: logical errors, attention gaps, unsupported leaps.
How it behaved: is the style, length, and manner working for you?
Description and Discernment form a continuous feedback loop: what you notice while evaluating becomes what you describe better next time. And a plain truth: even the most advanced AI systems benefit from human judgment. Research behind this framework found discernment behaviors are the least practiced of all, and they drop further exactly when outputs look most polished.
Pick something you know deeply. Ask the AI for three different explanations of it. Grade them with your expertise: what is right, what is subtly off, what is missing. Feed back specifics and co-create an improved version. The lesson underneath: domain knowledge powers discernment, and novices struggle to evaluate outputs precisely where they most need to.
- Evaluate product, process, and performance, the same three lenses
- The most skipped competency, most valuable when output looks finished
The Description-Discernment loop
This four-beat loop is the operating rhythm of augmentation, the engine you run for every real task:
Product, process, performance.
Evaluate through the same three lenses.
Feedback, adjusted description, iterate.
Add your expertise, make final decisions, take responsibility.
Return to your project from lesson 7. Pick its most substantial task and run the full loop until you have something you would genuinely use. Then reflect: which side cost more effort, describing or discerning? What description patterns produced the best outcomes? Those answers are your personal fluency profile.
- Describe, discern, refine, integrate, per task, until done
- Integration means the final call, and the responsibility, are yours
Up next: Diligence, the ethical and safety half of the definition.
A closer look at Diligence
The first three Ds cover effectiveness and efficiency. Diligence covers the ethical and safety half. Three components:
Thoughtful choice of AI systems and engagement: what data you share, and the privacy, security, and ethics of the choice itself.
Honesty about AI's role with everyone who needs to know: colleagues, clients, readers, examiners.
Ownership: you verify, and you vouch for, the outputs you share.
Diligence is context-sensitive: personal, academic, and professional settings carry different disclosure and verification expectations, and your job is to know and meet the ones that apply to you.
"In creating this document, I collaborated with [AI tool] to assist with [specific tasks]. I affirm all AI-generated and co-created content underwent thorough review. The final output accurately reflects my understanding, expertise, and intended meaning. While AI assistance was instrumental, I maintain full responsibility for the content, its accuracy, and its presentation."
Place a statement like this in the footer, appendix, or metadata of finished work. Adapt the wording; keep all its parts. The reference Writing an AI diligence statement goes deeper.
You used AI to draft a report for a client, reviewed everything, and rewrote half of it. What does transparency diligence require?
Transparency diligence is honesty with the people who receive your work, calibrated to the setting. A short statement covers it; hiding it fails the competency, and dumping raw logs is not disclosure either.
- Creation, transparency, deployment: choose responsibly, disclose honestly, own what you ship
- Disclosure norms differ by context; meeting them is part of the competency
Conclusion, and keeping the practice alive
The whole frame in one paragraph: four competencies, exercised across three engagement modes. Fluency develops through practice, not overnight mastery. AI systems are powerful but not magical; they are only as useful and safe as we enable them to be through thoughtful engagement.
Three ways to keep practicing
Rate yourself novice, developing, or confident per D. Pick one or two priorities. Plan practice with a timeline and success markers.
Turn your best past conversations into template prompts for your five to ten recurring tasks. Your own history is the best textbook you own.
Puzzles are description and discernment gyms: exact rules, deliberately tricky. Swap riddles, solve crosswords cooperatively, play twenty questions.
Pick one recurring task this week and run the full loop on it: describe, discern, refine, integrate. One change actually made beats ten understood.
Course quiz
Attribution. Adapted from the AI Fluency courseware by Prof. Rick Dakan (Ringling College of Art and Design) and Prof. Joseph Feller (University College Cork), developed in collaboration with Anthropic. CC BY-NC-SA 4.0. This adaptation © 2026 AI Literacy Foundation, shared under the same license.