▲ Claude Academy
Sign in

Vault / course/courses/ai-fluency-framework-foundations.md

updated 2026-06-25

Course: AI Fluency — Framework & Foundations

Mirrors: Anthropic Academy — AI Fluency: Framework & Foundations · https://anthropic.skilljar.com/ai-fluency-framework-foundations Audience: Everyone. No technical background needed — this is a way of thinking, not a tool tutorial. · Time: ~60 min + project Prereqs: Curiosity and any AI assistant to practice on (claude-overview). · Backing notes: claude-overview, prompt-engineering-basics Project: p03-ai-fluency-delegation-audit

AI Fluency is the ability to work with AI effectively, efficiently, ethically, and safely. This course teaches the 4D Framework — Delegation, Description, Discernment, and Diligence — a vendor-neutral model co-developed by Prof. Rick Dakan (Ringling College of Art and Design) and Prof. Joseph Feller (University College Cork). It works with any AI system, not just Claude. By the end you'll have a shared vocabulary and a repeatable habit for collaborating with AI on real work.

Learning objectives

After this course you can:

  • Define AI Fluency and name its four desired outcomes (effective, efficient, ethical, safe).
  • Explain each of the four competencies — Delegation, Description, Discernment, Diligence — and their sub-parts.
  • Decide what to hand to AI, a human, or both (Delegation).
  • Communicate clearly to AI across product, process, and performance (Description).
  • Critically evaluate AI output, reasoning, and behavior (Discernment).
  • Use AI responsibly: thoughtful creation, honest transparency, careful deployment (Diligence).
  • See the framework as a chain — and find which D is the weak link when collaboration fails.

Module 1 — What AI Fluency is, and the first D: Delegation

🎞 Frame 1 · What "AI Fluency" actually means · ⏱ ~2 min

🎬 Scene — Split screen: one person fights with an AI and gets garbage; another collaborates and gets great work. Same tool, different fluency.

🧠 Concept — AI Fluency is the ability to work with AI systems effectively, efficiently, ethically, and safely. It's not about knowing one product's buttons — it's a transferable way of thinking that holds as tools change. The 4D Framework names the four competencies that make it up.

🖼 On screen

The four outcomesMeans
EffectiveYou get the result you actually wanted
EfficientYou get it without wasted effort or cost
EthicalYou use AI in ways that respect people and norms
SafeYou avoid and contain harm from mistakes

🔗 The framework is vendor-neutral — co-developed by Rick Dakan (Ringling College) and Joseph Feller (University College Cork) and applies to any AI.

✅ Checkpoint — Name the four outcomes AI Fluency aims for.

🎞 Frame 2 · The 4D Framework at a glance · ⏱ ~3 min

🎬 Scene — Four panels light up in sequence: Delegation → Description → Discernment → Diligence, forming a loop.

🧠 Concept — Fluency breaks into four competencies, each starting with D. You delegate a task, describe what you want, discern whether the result is good, and apply diligence throughout. They reinforce each other.

🖼 On screen

Delegation   → Deciding WHAT to hand to AI (vs. human vs. together)
Description  → Communicating clearly to the AI WHAT/HOW/HOW-TO-BEHAVE
Discernment  → Critically evaluating the output, reasoning, and behavior
Diligence    → Using AI responsibly and accountably throughout

⚠️ Gotcha — This is a chain: weakness at any one D undermines the whole collaboration. A perfect description discerned carelessly still produces bad outcomes.

✅ Checkpoint — In one word each, what does each of the four Ds govern?

🎞 Frame 3 · Delegation, part 1 — problem & platform awareness · ⏱ ~3 min

🎬 Scene — A person pauses before typing and asks two questions: "What am I actually trying to do?" and "Is this the right AI for it?"

🧠 Concept — Delegation begins before any prompt. Two kinds of awareness come first:

  • Problem awareness — understanding the task well enough to break it into parts and know what "good" looks like.
  • Platform awareness — knowing what a given AI can and can't do, so you delegate to a capable tool. (Deeper: ai-capabilities-and-limitations.)

🖼 On screen

Before delegating, ask:
  • Do I understand the problem well enough to judge the result?
  • Does this AI have the capability (and access) to help here?

✅ Checkpoint — Why is platform awareness part of delegation rather than a separate skill?

🎞 Frame 4 · Delegation, part 2 — who does what · ⏱ ~3 min

🎬 Scene — A task gets sorted into three buckets: "I'll do this," "AI does this," "we do this together."

🧠 Concept — Task delegation is deciding, for each piece of work, whether it belongs to a human, the AI, or a human-AI collaboration. Good delegation plays to strengths: humans for judgment, stakes, and lived context; AI for drafting, breadth, and tireless iteration.

🖼 On screen

Hand to a humanHand to AIDo together
High-stakes final decisionsFirst drafts, brainstormingOutlining then refining
Personal/relational judgmentSummarizing provided textResearch you then verify
Anything needing accountabilityReformatting, translationIterating toward a goal

✅ Checkpoint — Pick a task from your week and sort it: human, AI, or together — and say why.

Module 2 — Description: communicating clearly to AI

🎞 Frame 5 · The three kinds of description · ⏱ ~2 min

🎬 Scene — A vague request fails; the same request, split into what / how / how-to-behave, succeeds.

🧠 Concept — Description is communicating with the AI so it can actually help. There are three layers — product, process, and performance description — and good collaboration usually uses all three. (Hands-on prompting craft: prompt-engineering-basics.)

🖼 On screen

Product description     → WHAT you want (the end result)
Process description     → HOW to approach it (the method/steps)
Performance description → HOW the AI should behave (role, tone, constraints)

✅ Checkpoint — Match each: "act as a friendly tutor," "use the Socratic method," "write a 200-word summary."

🎞 Frame 6 · Product description — describe the result · ⏱ ~3 min

🎬 Scene — "Make it good" becomes "a 150-word, plain-English LinkedIn post announcing our hiring freeze, ending with a question." The output transforms.

🧠 Concept — Product description specifies what you want produced: the format, length, audience, scope, and what "done" looks like. The clearer the target, the less the AI has to guess.

🖼 On screen

Weak:  "Summarize this report."
Strong:"Summarize this report in 5 bullets for a busy
        executive — focus on risks and recommended actions."

✅ Checkpoint — Add three product details to "write me an email."

🎞 Frame 7 · Process description — describe the approach · ⏱ ~2 min

🎬 Scene — Instead of just asking for an answer, the user says "first list the options, then weigh tradeoffs, then recommend one."

🧠 Concept — Process description tells the AI how to get there — the steps, method, or reasoning path. This is especially powerful for complex tasks where the route matters as much as the destination.

🖼 On screen

"Before answering, outline your approach.
 Step 1: list constraints.  Step 2: propose 3 options.
 Step 3: compare them.      Step 4: recommend one and say why."

✅ Checkpoint — Why does describing the process often improve a reasoning-heavy answer?

🎞 Frame 8 · Performance description — describe the behavior · ⏱ ~2 min

🎬 Scene — A card flips: same task, but the AI is told to be "a skeptical editor" vs. "an encouraging coach." Two very different — and useful — outputs.

🧠 Concept — Performance description sets how the AI should behave: its role, tone, persona, level of detail, and constraints. It shapes the style and stance of the help, not just the content.

🖼 On screen

LeverExample
Role"You are a patient onboarding buddy."
Tone"Warm but concise; no jargon."
Constraint"Never give medical advice; suggest seeing a professional."

✅ Checkpoint — Give a performance instruction that would make answers safer for a beginner audience.

Module 3 — Discernment: evaluating what AI gives you

🎞 Frame 9 · The three kinds of discernment · ⏱ ~2 min

🎬 Scene — An answer arrives. The user doesn't just copy it — they inspect three things: the result, the reasoning, and the behavior.

🧠 Concept — Discernment is critically evaluating AI's work. Mirroring Description, it has three layers — product, process, and performance discernment. This is where you catch hallucinations, flawed reasoning, and inappropriate behavior. (Why this is necessary: ai-capabilities-and-limitations.)

🖼 On screen

Product discernment     → Is the OUTPUT good, correct, and on-target?
Process discernment     → Did it REASON soundly to get there?
Performance discernment → Is its BEHAVIOR appropriate for the situation?

✅ Checkpoint — Discernment mirrors which earlier competency, and how?

🎞 Frame 10 · Product discernment — judge the output · ⏱ ~3 min

🎬 Scene — A confident, polished answer contains one wrong number. The fluent user spots it.

🧠 Concept — Product discernment asks: is the result actually good? Accurate, complete, relevant, free of fabrication? AI can be fluently, confidently wrong — so you check facts, numbers, and quotes, especially when stakes are high.

🖼 On screen

Product checklist:
  • Is it accurate? (spot-check facts, math, citations)
  • Is it complete and on-scope?
  • Did it invent anything (a "confabulation")?

⚠️ Gotcha — Polish is not correctness. Fluent prose can hide a fabricated fact.

✅ Checkpoint — Name two things you'd always verify in a factual AI answer before acting on it.

🎞 Frame 11 · Process & performance discernment · ⏱ ~3 min

🎬 Scene — The user asks "show your reasoning" and notices a step that doesn't follow — then notices the AI was too agreeable, never pushing back.

🧠 Concept — Process discernment evaluates how the AI reached its answer — is the reasoning sound, or did it skip steps and still land somewhere plausible? Performance discernment evaluates how it behaved — was the tone, confidence, and stance appropriate, or did it overclaim, flatter, or refuse unhelpfully?

🖼 On screen

Process:     Ask it to explain its steps → check the logic, not just the answer.
Performance: Was it appropriately cautious? Honest about uncertainty?
             Did it behave as instructed?

✅ Checkpoint — Give one prompt that surfaces an AI's reasoning so you can discern it.

Module 4 — Diligence: using AI responsibly

🎞 Frame 12 · The three kinds of diligence · ⏱ ~2 min

🎬 Scene — Before, during, and after using AI, a person takes three responsible actions: prompts thoughtfully, discloses the AI's help, and weighs the consequences of acting.

🧠 Concept — Diligence is the responsibility and accountability woven through all the other Ds. It has three parts — creation, transparency, and deployment diligence. This is what keeps collaboration ethical and safe.

🖼 On screen

Creation diligence     → Prompt thoughtfully and responsibly
Transparency diligence → Be honest about when and how AI was used
Deployment diligence   → Weigh the consequences of acting on outputs

✅ Checkpoint — Which D is responsible for disclosing that AI helped write something?

🎞 Frame 13 · Creation & transparency diligence · ⏱ ~3 min

🎬 Scene — A user avoids feeding private data into a prompt, then adds a footnote: "Drafted with AI assistance, reviewed by me."

🧠 Concept — Creation diligence is being thoughtful in how you prompt — avoiding harmful, deceptive, or privacy-violating uses, and respecting data policy. Transparency diligence is being honest about AI's involvement so others can calibrate trust — disclosure where it matters (work, school, publishing).

🖼 On screen

Creation:     Don't paste secrets/regulated data; don't ask AI to deceive.
Transparency: Disclose AI use where honesty and norms require it.

🔗 Norms differ by setting — see ai-fluency-for-students and ai-fluency-for-educators for classroom expectations.

✅ Checkpoint — Name one situation where disclosing AI use is clearly the right call.

🎞 Frame 14 · Deployment diligence — own the outcome · ⏱ ~3 min

🎬 Scene — An AI-drafted policy is not sent straight to customers; a human reviews, edits, and takes responsibility before it ships.

🧠 Concept — Deployment diligence is taking responsibility for what happens when you act on AI output. The higher the stakes, the more human review and accountability you apply. You — not the AI — own the consequences.

🖼 On screen

Low stakes  → light review, ship.
High stakes → verify, edit, get a human sign-off, keep accountability with a person.

⚠️ Gotcha — "The AI said so" is never an excuse. Accountability stays human.

✅ Checkpoint — How should the level of deployment diligence scale with the stakes?

🎞 Frame 15 · The chain — and your weak link · ⏱ ~2 min

🎬 Scene — A recap loop: Delegation → Description → Discernment → Diligence. One link is highlighted in red — the user's weakest.

🧠 Concept — The four Ds form a chain, and the collaboration is only as strong as its weakest link. When AI work goes wrong, diagnose which D failed: wrong task to delegate? Vague description? Skipped discernment? Careless diligence? Naming the weak link is how you improve.

🖼 On screen

Bad outcome? Trace it back:
  Delegation  → wrong task or wrong tool?
  Description → unclear ask?
  Discernment → didn't check the result?
  Diligence   → acted irresponsibly on it?

✅ Checkpoint — Without looking, recite the four Ds in order and one sentence each.

🛠 Project

Complete p03-ai-fluency-delegation-audit — The Delegation Audit. You'll take a week of your real tasks, run each through the 4D Framework (sort by Delegation, draft Descriptions across product/process/performance, apply Discernment, and note the Diligence required), and identify your personal weak link in the chain.

🧪 Self-check quiz

  1. What are the four outcomes AI Fluency aims for?
  2. Name the four competencies of the 4D Framework, in order.
  3. What three things does Delegation involve before you ever prompt?
  4. Name the three layers of Description and what each communicates.
  5. Discernment mirrors Description — name its three layers.
  6. A polished AI answer contains a fabricated statistic. Which discernment layer should have caught it?
  7. Which kind of Diligence covers disclosing that you used AI?
  8. Why is the framework described as a "chain"?
<details><summary>Answers</summary>
  1. Effective, efficient, ethical, safe. 2. Delegation, Description, Discernment, Diligence. 3. Problem awareness, platform awareness, and task delegation (human vs AI vs together). 4. Product (what you want), process (how to approach), performance (how the AI should behave). 5. Product, process, and performance discernment. 6. Product discernment. 7. Transparency diligence. 8. Because weakness at any one D undermines the whole collaboration — it's only as strong as its weakest link.
</details>

🎓 Certificate criteria

You've "passed" AI Fluency: Framework & Foundations when you can:

  • Explain all four Ds and their sub-parts from memory.
  • Sort a real task into human / AI / together and justify it (Delegation).
  • Write a prompt that includes product, process, and performance description.
  • Apply product, process, and performance discernment to one AI output.
  • Complete p03-ai-fluency-delegation-audit and name your personal weak link in the chain.

Tick this course off in progress and record the date you earned Anthropic's official AI Fluency certificate.

🔗 Sources & deeper notes