AI as a Craft Teaching Guide

Teaching is a craft

You already know teaching is a craft. You have spent years building a feel for it: when to pause, when to push, which analogy lands with which room. This guide is about bringing AI into that craft without hollowing it out.

Working with AI as a craft, as introduced in the AI as a Craft training, means one thing above all: the thinking stays yours and yours is the hand that guides AI. AI can hand you a finished-looking lecture in four seconds with nothing of you in it, and from the back of the room nobody can tell. For a week or two. Then students start asking questions the slides can't answer, and everybody can tell.

So this guide is not a set of rules for permitted use, and it is not a form to complete. It is a way of working where you can tell the good from the rubbish, your judgement is in everything you use, and you can stand behind whatever carries your name into a classroom.

The Five Stages: Define, Decide, Design, Evaluate, Reflect

Every time you reach for AI in your teaching, you'll follow the same short journey. At first you will walk it deliberately then in time you will stop thinking about it as it becomes natural, which is when you've achieved fluency.

1. Define: Why are you using AI?

Not which tool. Not which prompt. Why. What does this task actually need: your presence, your knowledge, or just your time? Sometimes the honest answer is that the task is drudgery and your time is better spent with students. Sometimes the honest answer is that you are avoiding thinking you should be doing yourself. Only one of those is a good reason to continue.

2. Decide: Choose your door

Four doors, and matching the door to the job is most of the skill.

3. Design: Who you are, what you want, and why

Whichever door you take (except the fourth), the way in is the same. Tell it who you are, what you want, and why, before any of the particulars. Don't open with "Create a lecture on thermodynamics". Build context: who your students are, where they struggle, what you have tried, what good looks like in your discipline. For agentic work, this becomes designing a full character using the six boxes: Purpose, Nature, Subject, Setting, Communication, Mechanisms.

4. Evaluate: Is it right, and is it you?

Two checks, not one. Is it right? Has it invented anything, dropped anything, flattened your discipline's nuance into confident mush? Is it you? Is your judgement actually in this, or is it the average of everyone and no one? Material that could have come from any lecturer anywhere teaches students that your module could be taught by anyone. It can't, and your materials should prove it.

5. Reflect: Are you answerable for it?

You stand behind whatever you put in front of students. Was this a job you should have handed over at all? Was the use fair to your students, your colleagues, and your subject? Would you be comfortable telling your class exactly how this was made? If not, go back a stage or two.

The Two Tests

Two rules of thumb, one protecting your students and one protecting you. Carry them everywhere.

The Park Bench Test: protects your students

If you're comfortable leaving this information on a park bench to be read by strangers, it's okay to share with AI.

What you CAN share: Anonymised small segments, your own observations and notes, general descriptions of student struggles, your own teaching materials.

What you CANNOT share: Complete student assignments, student names or IDs, confidential institutional documents, sensitive personal information.

The Corridor Test: protects you

Before you use AI-assisted material in your teaching, imagine a student stops you in the corridor afterwards and asks you to go deeper. No slides, no notes, just you. Can you?

If you can, AI has amplified your knowledge and bought you time for the parts of teaching only you can do. If you can't, you have not delegated a task, you have delegated your understanding, and eventually that debt will get called in, usually in front of your students or peers.

Students compare notes, and they learn quickly to tell the difference between the teacher whose AI-assisted materials sits superficially on top of a real depth of understanding and the teacher whose materials compliment their own depth of understanding. The superficial approach is quantitative, the complimentary approach is qualitative. The first earns a trust that survives any amount of AI use. The second loses a trust that no amount of polish wins back.

The Skeleton Underneath

The five stages are how you work. Underneath them sit four competencies that grow as you do, drawn from the AI Fluency Framework (Dakan & Feller, 2025). You don't need to memorise them, but it helps to know what's developing while you practise.

• Delegation grows at Define and Decide: judging whether, when, and through which door to use AI.
• Description grows at Design: communicating who you are, what you want, and why, well enough that the output is worth having.
• Discernment grows at Evaluate: catching what's wrong, missing, or invented.
• Diligence grows at Reflect: being transparent, ethical, and answerable for the result.

One question has no ancestor in any framework: Is it you? That question is what develops AI from a skill into a craft.

Before Semester: Preparation

Before you start building content, work through three foundational questions:

• Context: What analogies, metaphors, or frameworks make your subject genuinely accessible?
• Engagement: What makes this subject compelling? What hooks attention on day one, and what sustains momentum when novelty wears off around week six?
• Meaning: How do students connect what they're learning to their own sense of identity or purpose?

Creating Lecture Content

You're under pressure, with multiple modules, committee work, and administrative demands, but you know your teaching needs improvement.

Define and Decide: Before touching any AI system, ask: What exactly needs changing? Which door suits this work? What should remain human versus what could AI help with?

Design and Evaluate: Don't start with "Create a lecture on [topic]". Build context over multiple exchanges. Share your students' backgrounds, conceptual struggles, and what you've tried before. Evaluate each output critically: is it right, and is it you?

Reflect: Verify accuracy, check cultural appropriateness, ensure examples don't reinforce stereotypes, be transparent about AI involvement, and apply the corridor test before anything reaches a classroom.

Developing Course Materials

Define and Decide: What exactly needs updating? What requires your disciplinary judgement versus what AI can help with?

Design and Evaluate: Instead of "Summarise this article", build context: explain what students struggle with and what kind of scaffolding would help them access the original material without replacing it.

Creating Accessible Materials

You need materials that work for diverse learners: different processing styles, disabilities, English as an additional language.

During Semester: Teaching

This is where preparation meets reality. Students arrive with different needs than anticipated. Energy dips around Week 6. Your carefully designed plans need adaptation.

Facilitating Active Learning

Education follows a natural rhythm: present theory through story, students discuss to build understanding, they practise applying it, then reflect on learning.

For each cycle phase, build specific context with AI:

• Theory (Story): Build from everyday experience to principles. Make abstract principles feel concrete and inevitable.
• Discussion: Generate prompts addressing common misconceptions. Create productive conflict between intuitions and actual principles.
• Practice: Create problem sequences scaffolding from simple to complex.
• Reflection: Design questions helping students articulate how their thinking changed.

Load a Scenario, Not a Slide

The theory is taught. Now let them fly the plane. Design an agent that plays a part: an anxious patient, a difficult client, a select committee, a rival negotiator, a dig site with contradictory evidence. Load in a scenario, set the constraints, and let students apply the week's theory to something that talks back.

This is a flight simulator for your discipline: practice with consequences but without casualties. Use the six boxes to build the character, and use the Setting and Mechanisms boxes to load each week's scenario. The same simulation, reloaded with new situations, can carry a whole semester of applied practice. The Assessment Guide covers how to assess within one.

Supporting Struggling Students

Define and Decide: What specific struggles need addressing? What requires human connection versus what AI can provide?

Providing Feedback on Student Work

Critical principles:

• You cannot delegate judgement about quality or achievement to AI. Ever.
• You cannot upload student work to AI. It's their intellectual property and potentially identifiable information. Use the park bench test.

The correct process: Read the work yourself, make rough notes, then use AI to help you articulate your observations clearly (without sharing student work). Add your personal response.

Build Yourself a Feedback Companion

If you follow Robert's process every marking season, you are re-explaining the same standards, the same tone, and the same module outcomes every time. That repetition is the signal to go agentic. Design a feedback companion once, using the six boxes:

• Purpose: Turns my rough marking notes into constructive, developmental feedback for second-year students in my discipline.
• Nature: Encouraging but honest. Never inflates praise. Treats every piece of work as a draft of something better.
• Subject: My students: who they are, what they typically struggle with, what the module is building towards.
• Setting: Marking season, a stack of scripts, my notes scribbled in the margins.
• Communication: Feedback in my voice: specific, warm, forward-looking, always ending with what to try next.
• Mechanisms: My discipline's standards and the module outcomes, baked in. Works only from my notes, never from student work itself.

The judgement stays yours, the reading stays yours, the personal response stays yours. The agent only does the articulation, and it does it knowing your subject rather than needing the whole story again each time. The park bench test still applies absolutely: your notes go in, the student's work never does.

Adapting When Things Aren't Working

Don't ask: "My course isn't working, help me fix it." Build diagnostic understanding: share what you're observing, specific behaviours, assignment patterns, feedback themes.

Teaching Practical Application Without Industry Experience

Critical principle: AI cannot replace genuine practitioner insight. Be transparent about this. The corridor test applies doubly here: borrowed experience collapses under the first follow-up question.

After Semester: Reflection & Improvement

The semester ends. This is when the Reflect stage is most prominent: the moment when small investments create exponential improvements for next time.

Evaluating What Worked

Don't ask: "Summarise this feedback." Build analytical understanding: identify patterns across multiple data sources. Not just surface complaints or praise, but deeper themes about what enabled or blocked learning.

Planning Improvements

You can't redesign everything. Focus on targeted improvements with genuine impact.

Building Your Practice Over Time

Build a personal teaching playbook documenting: approaches that work consistently, common misconceptions and how to address them, effective analogies, activity structures that generate genuine engagement. After each iteration, add what worked unexpectedly well, what didn't work despite planning, new insights, and refinements. This is how craft accumulates: not in the tools, but in you.

Putting It All Together

The five stages aren't a rigid process to follow, but a way of thinking that keeps you grounded in what matters: teaching that serves student learning, made by a teacher whose thinking is in it.

In Practice

DEFINE → Why are you using AI at all?
DECIDE → Automate, Collaborate, Agentic, or No AI
DESIGN → Who you are, what you want, why
EVALUATE → Is it right? Is it you?
REFLECT → Are you answerable for it?

Moving Forward

Start small. Pick one teaching challenge that resonates with your current situation and walk it through the five stages deliberately. Run the two tests before anything reaches students. As you practise, the stages become instinct, the way a potter stops thinking about centring the clay.

But you'll always remain the expert. The human. The teacher. AI is material in your hands, and the work is only worth what you put of yourself into it.