Why differentiation gets skipped

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.

💡The single most important thing to understand about AI differentiation
“Easier” and “harder” are not useful specifications. AI cannot produce a good foundation-level activity from “make it easier” because it doesn't know which dimension to adjust. Cognitive demand, scaffolding level, task type, and expected output — these are the dimensions that need to be specified. The prompt template below encodes these dimensions explicitly.
What makes each level different

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.

Dimension
Foundation level
Extension level
Cognitive demand
Recognition, recall, and guided application using provided examples
Analysis, evaluation, and unprompted transfer to novel scenarios
Scaffolding
Sentence starters, partially completed models, step-by-step instructions, worked examples to follow
Minimal structure; student determines approach; success criteria provided without process guidance
Expected output
Shorter response with explicit format requirements; key terms provided
Open-form response; student selects format; expected to justify choices
Context
Familiar contexts from the lesson or closely related examples
New contexts requiring concept transfer; may combine multiple lesson concepts
The three-level prompt

The differentiation prompt that actually
specifies all three levels.

🤖The three-level differentiation prompt — copy and fill
Generate three differentiated versions of this activity:

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).
Connecting to the agile loop

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.

Stage
What happened
What it produced
Monday exit ticket data
Q1 correct: 24/25. Q2 correct: 18/25. Q3 correct: 9/25. Pattern: Q2 wrong answers cluster around "energy released FROM bonds" language.
Three distinct groups identified: 7 students need foundation level (Q2 wrong), 9 need core (Q2 right, Q3 wrong), 9 ready for extension (all correct).
Monday night AI differentiation
Foundation (7 students): guided application with scaffold addressing the "from bonds" misconception explicitly.
Core (9 students): application to new context with support. Extension (9 students): analysis of an edge case requiring evaluation of competing explanations.
Tuesday lesson
Students self-select or are assigned based on Monday's exit ticket. Foundation group works with teacher for 8 minutes.
Core group works independently. Extension group works on open analysis task. Teacher attention is targeted where it produces the most gain.
Tuesday exit ticket result
Foundation group improvement: 5/7 now correct. 2 still using "from bonds" language — individual intervention.
Core group: 7/9 correct. Extension group: all complete; quality of justification is the variable. Two more days has closed the gap that 20 minutes of general reteaching would not have.
The three-level choice architecture

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.

You've finished C8

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.

← Back to P6 pillar hub← Back to A2