Teach it

1. Your commitment card, in silence (2 minutes). Below.
2. Workshop feedback (up to 3 minutes). Open the feedback form, or scan the code on the closing screen. Five short questions: your level, usefulness, what you finished, one takeaway and one improvement. Honest answers, including unfinished builds, help improve the next workshop. Permission to quote your takeaway is optional and starts private.

Use the last 1½ minutes of the gallery harvest (card 15) for "One build I'd steal": four or five people name another attendee's build and say, in one sentence, what they'd take from it. Signpost the next-step links (card 18) as the feedback form opens; explore them after the session.

By [date, two weeks from today], I will test [tool] with [one real user], and check [one success check].

Two weeks from today: ____

This checks one thing: does it work for the person it is for?

• The date: write the real date, two weeks from today, on the line under the sentence.
• One real user is one person the tool is for: a student in the class it is for, or the colleague it is for. They use the finished page, never the AI tool, and nothing about them goes into the tool.
• One success check is one of the three on your canvas (box 4), run while they use it.

Write it now. Keep it; photograph it. Afterwards, note one thing that broke, and put it in the spec for version 2 (card 01): ______________________________________

Then, with one class, ask the same question before and after. That checks a second thing: did they learn? (below).

• Quiz: after teaching, low stakes. Use it again next week and a few weeks later, mixed with other topics.
• Simulation or model: after a short explanation. Use it before explaining only with a prediction step and a debrief. Give students an open challenge question, not a click-by-click worksheet.
• Explainer: for students new to the idea: a worked example, then a partly worked one, then one to try.
• Never instead of your explanation. The tool leads into your teaching; it does not replace it.

Why: discovery with guidance did better than other teaching, and discovery with no help did worse than explicit teaching (Alfieri et al. 2011); predicting before a demonstration helped where watching alone did not (Crouch et al. 2004); meeting it again, spaced out, helps it stick (Cepeda et al. 2006).

For simulations, models and explainers.

1. Predict: "What do you think will happen when [ ]? Write it down."
2. Challenge: "[Your open question]." Students use the tool.
3. Explain: "What happened? Were you right? What explains the difference?"
4. The rule: "Say the rule in your own words." Then give the correct version.
5. The limits: "Where is this model different from the real thing?"
6. Retrieve: "Close the screen. Answer from memory: [ ]."

In one study, university physics students in active classes learned more but felt they had learned less (Deslauriers et al. 2019). "They enjoyed it" is not evidence.

1. Choose one question on the misconception you targeted: your outcome statement's evidence (canvas box 2). Make it explain, not choose. For example: "In one sentence, explain why 2/3 ≠ 3/4."
2. Before: ask it on paper, from memory, before the class uses the tool. Count how many students get it right.
3. After: ask the same question the same way, a week after they use it. Count again.

A rise is a good sign, not proof: your teaching happened in between too. Record results where you keep other class records, never in the tool or an AI chat. A class-level count is enough.

Adapted from DEN Lab Build 05, Design the Human Handoff.

• You decide, not the tool: marks or grades that count, sets and groups, report comments, support plans, and anything about behaviour or safeguarding. The Lab rule: "Never ask AI to make a safeguarding, disciplinary, legal, admissions, grading or employment decision."
• Check the stop. Your spec has the stop rule (card 03): after two wrong tries, the tool sends the student to you or to a worked example. Check it works before students use it.
• Plan the fallback. If the file, the link or the Wi-Fi fails, have the same questions on paper or on a slide.
• Don't rely on the tool to alert you. You see only what is on a student's screen. Your school's usual safeguarding routes apply.
• Check your school's rules on AI, and on sharing links and web tools with students, before you use it (card 08, safe building).

"AI can assist the judgement; it shouldn't replace the person responsible for it."

Alex Gray, Which Half of Marking Should AI Do?

• Skip "One build I'd steal". Instead, write down one thing you would change in your own build, and add it to the spec for version 2.
• Write your commitment card anyway, and put the date in your calendar.
• Tell one colleague the date, so someone asks how it went.