AI Fluency: Framework & Foundations: teaching notes

"Tips and tricks date fast; the models change every few months. This is the thing underneath that doesn't: how to actually collaborate with AI. Four habits. Once you have them, they keep working whatever the tool does next."

"Here's the shape. We set the foundations, look at how AI actually works under the hood, learn the four habits, then practise them on something real."

"Fluency isn't knowing facts about AI. It's working with it well, on four counts: effective, efficient, ethical, safe. The mental shift that unlocks it is to stop treating it like a vending machine you put a question into, and start treating it like a colleague you build a working relationship with."

"Three modes. Automation: you define it, it does it. Augmentation: you think together, the most common one. Agency: you set it up to act on its own. These aren't levels. You don't graduate from one to the next. You pick the mode that fits the task."

"Four habits. Delegation: decide what's even worth doing together. Description: say what you want, and how. Discernment: judge what comes back. Diligence: own the final result. The middle two are a loop you'll run over and over. Delegation sets it up, Diligence keeps it honest. That's the whole framework."

Pick the route by audience (framings.md: deterministic vs probabilistic). For analysts and regulated rooms: "All of this is predicting the next word. 'The cat sat on the…': most likely 'mat'. It's picking the most plausible next thing. Brilliant for drafting, wrong for anything that has to be identical every time." For generalists and designers: the Excel-formula contrast. For technical peers: one sentence and move on.

"It's strong exactly where most of us are weak: speed, breadth, tireless drafting. And weak where you're strong: your judgement, your context, knowing what 'right' looks like here. The limits aren't temporary. They're the reason you stay in the loop. Best results pair its breadth with your oversight."

"Delegation happens before you type a word. Two things meet: what you know about the problem, and what you know the tool can do. Where they overlap, you split the work. Not 'automate everything', the right split. The parts that are yours stay yours."

"Delegation becomes a plan in four steps: name what good looks like, break the work into tasks, mark each as yours, the AI's, or shared, then carry it through. The first step is the one people skip and the one that matters most: if you don't name success up front, you've got nothing to judge the output against later."

"Three layers to a good description. The product: what you want made. The process: how it should get there, the method. The performance: how it should behave with you. Almost everyone gives the first and forgets the second. The process layer is where the quality hides."

"Six practical moves. Context, examples, constraints, steps, ask it to think before answering, give it a role. You won't use all six every time. And the shortcut most people never try: paste your own prompt back to it and ask it to make it better. It's good at that."

"Discernment is Description in reverse. You judge the output, the reasoning, and the behaviour. Here's the uncomfortable part: you can only catch what you'd notice. Your expertise isn't replaced here; it's exactly what makes you able to judge. This is where the human earns their place."

"Describe, look at the output, discern, then describe it better and go again. That loop is where the actual work lives. Not the first answer, the third or the fourth. And at the end of the loop, the decision is yours. You own the final call."

"Diligence is owning it. Three parts: which tool you chose and whether it was right for sensitive work; being open about AI's role where that matters; and verifying what you put your name to. You're accountable for the output, whether or not AI helped make it. A diligence statement just names AI's role and your responsibility for the result."

"You don't get fluent by finishing a course, any more than you get fluent in a language that way. You get there by doing it, badly at first, then better. The four habits are the thing you practise. AI is powerful, not magical. It's only ever as good as what you bring to it."

"Four habits, three modes, one practice. If the tools change next year, and they will, this still holds. Delegate well, describe and discern in a loop, stay diligent. That's fluency."