The most frequently skipped part of lesson planning
is also the most valuable.
Differentiation is universally acknowledged as important and widely skipped in practice. The reason is not indifference — it is time. Writing three versions of an activity from scratch takes 45–60 minutes for a well-designed set. For a teacher who has 6 lessons to plan, differentiation is the first thing cut when time runs short.
AI changes this calculation. Not by generating differentiation automatically — the teacher still needs to define what makes each level different, which requires pedagogical judgment about the class. But once that judgment is made and articulated, AI can produce the three versions in under 2 minutes. The work shifts from generation to specification. Specification takes 2 minutes. Generation takes 90 seconds. The total time cost drops from 45–60 minutes to under 5 minutes.
Three dimensions.
Not one. Not just “easier” and “harder.”
Good differentiation adjusts three dimensions simultaneously: cognitive demand (what type of thinking the activity requires), scaffolding level (how much structure and support is built in), and expected output (what the student produces and at what length and depth). Adjusting only one dimension produces weak differentiation — a foundation activity that asks for a shorter answer to the same hard question is not differentiated, it is truncated.
The differentiation prompt that actually
specifies all three levels.
Original activity: [paste the core-level activity you've already designed or have from your lesson plan]
Learning objective: [what students should demonstrate by the end of the activity]
Foundation version requirements:
— Cognitive demand: recognition and guided application only
— Scaffolding: include [sentence starters / partially completed model / step-by-step instructions — choose one]
— Reduce the number of steps to [N]
— Provide [N] worked examples for students to follow
— Expected output: [shorter format — e.g. “3 sentences using the provided sentence starters”]
Extension version requirements:
— Cognitive demand: analysis/evaluation — requires comparing, justifying, or evaluating
— No scaffolding — state the outcome required but not the process
— Novel context: the scenario must NOT be from today's lesson
— Expected output: [open-form — e.g. “written argument of 150–200 words with explicit justification of position”]
Core version: keep the original activity exactly as written — do not modify it.
For each version, note which students it is designed for (not by name — by prior attainment signal from today's exit ticket).
Differentiation informed by
yesterday's exit ticket data.
Static differentiation — the same three levels for every lesson — is standard practice. Dynamic differentiation — three levels that are informed by what yesterday's exit ticket revealed about this specific class's distribution — is agile practice. The AI differentiation workflow enables the dynamic version by making the production time short enough that you can customise the levels based on last night's scan.
How to offer three levels without
making it a sorting exercise.
The three levels should be presented as different routes to the same destination, not as different destinations for different students. All three versions target the same learning objective. The difference is how much support is available on the route.
Present the three levels by what they offer, not by who they're for. “The foundation version includes worked examples and sentence starters.” “The extension version invites you to compare two competing explanations.” “The core version is the activity as designed — the standard challenge.”
Students self-select based on honest self-assessment. Pair this with the exit ticket: students who self-selected foundation level and produced a Q3 correct answer on Monday's exit ticket should be nudged to try core level on Tuesday. The exit ticket data is the objective signal that updates the student's self-assessment. This is the connection between C5 (student agency) and C8 (AI tools): student voice provides qualitative self-assessment, and exit ticket data provides objective calibration.
The complete AI agile toolkit:
7 minutes daily.
C8 has covered the three AI tools that compress the agile teaching cycle: lesson regeneration from observation notes (A1), exit quiz generation from learning objectives (A2), and three-level differentiation from one prompt (A3). Together they take under 7 minutes daily and cover the full agile loop — improve yesterday's lesson, generate today's data collection, differentiate today's activities.
You've now completed all 8 clusters of P6 — Agile teaching. The full pillar covers the theory (C1), the observation practice (C2), the data collection loop (C3), individual iteration (C4), student agency (C5), school-wide culture (C6), curriculum design (C7), and the AI toolkit (C8). The practices compound: each cluster builds on the previous one, and the complete system produces a school that improves continuously — one lesson, one week, one cohort at a time.