Most teachers use AI to create.
Agile teachers use it to improve.
Most teacher AI use follows the same pattern: “Write me a lesson on [topic].” The output is a generic lesson that may or may not suit the class, and which requires heavy modification before it can be used. It is generation from scratch — useful for new content, poor for iteration.
The agile AI workflow is different. You are not asking AI to create a lesson — you are asking it to modify a specific lesson you already taught, in a specific way, to address a specific failure. The inputs are more constrained, the output is more targeted, and the editing pass is shorter because you're reviewing a modification rather than generating from nothing.
Three inputs.
One targeted lesson modification.
The AI lesson regeneration prompt has three required inputs and one optional constraint. Each input has a specific purpose — removing any one of them degrades the output quality significantly.
Paste the specific lesson element you are modifying: the explanation, the activity, the worked example, or the assessment task. More specific is always better — if only the explanation needs changing, paste only the explanation rather than the full lesson. Without the original, AI generates from scratch — producing something that may contradict your other lesson elements, reuse examples you've already used, or use vocabulary your class hasn't encountered.
Paste the observation notes you wrote during or after the lesson. As covered in C4/A3, these need four components: the specific wrong answer, the mechanism behind it, the element to change, and the constraint on the new approach. If your notes are vague, upgrade them before pasting — the quality of this input determines everything.
Tell the AI exactly what the output should be: an explanation replacement, a new worked example, a redesigned activity. Specify format (word count, structure, level), audience (year group, prior knowledge), and purpose (to address the specific misconception named in Input 2).
What to check before using
the AI output tomorrow.
Run these three checks on every AI-generated element before use. No exceptions — AI is accurate often enough to develop false confidence, and wrong often enough that unchecked output occasionally contains significant errors.
The 3-minute iteration as
a teachable habit.
Teachers who use this workflow daily report a consistent experience: the first few uses feel mechanical and produce output that needs significant editing. By the third or fourth use, the observation notes are sharper, the prompt is more precise, and the output requires minimal adjustment. By the second week, the full cycle — scan, diagnose, prompt, edit — takes under 4 minutes and produces output that is usually better than a manually written first draft.
The habit formation depends on one condition: the observation notes must be written before the lesson ends, not reconstructed from memory at 9pm. A 2-minute note written at the classroom door while students are filing out is worth more than a 20-minute reconstruction attempt after dinner.
(1) Most common wrong answer: ___
(2) Why they got it wrong (my hypothesis): ___
(3) Element to change: ___
(4) What the new version must avoid: ___
Fill it before leaving the room. Use it verbatim as Input 2 in tonight's prompt.
Now do the same thing
for exit quizzes.
A2 covers AI exit quiz generation — the other side of the agile loop. If A1 is how to improve yesterday's lesson using tonight's AI time, A2 is how to generate tomorrow's data collection instrument using the same time. Together they create a daily AI-powered agile cycle that takes under 7 minutes total.