The case for a timed workflow

Reflection without structure produces
intentions, not improvements.

Most teachers reflect on their lessons. Very few have a structured process for doing it — which means that reflection tends to produce either vague intentions (“I should explain that differently next time”) that evaporate by the following week, or unstructured planning sessions that take hours and still don't produce targeted improvements.

The 20-minute workflow exists because the time constraint is the point. Twenty minutes is short enough to fit into the gap between one lesson and the next. It is also short enough that it forces discipline: you cannot rewrite the whole lesson in 20 minutes, so you have to decide what actually needs changing and change only that.

The output is not a new lesson plan. It is a set of specific, targeted modifications to an existing one. The lesson that comes out of this workflow is not a different lesson. It is a better version of the same lesson. The sequence, the learning objective, and the core activities remain. What changes is the specific explanation that didn't land, the activity that produced the wrong data, the pacing that lost the class in the second half.

⏱️Why 20 minutes and not longer?
The 20-minute ceiling prevents over-engineering. A teacher who spends 90 minutes “improving” a lesson typically rebuilds it rather than improving it — they start from scratch because iterating on existing work feels more constrained than starting fresh. The constraint of 20 minutes forces the most important skill in agile teaching: identifying the smallest change that will produce the biggest improvement.
The workflow

Four steps. Twenty minutes.
A better lesson tomorrow.

1
Read your observation notes — decide what to change
4 minutes · The single most important decision in the workflow

Open your observation notes from today's lesson. Your job in this step is to answer one question: what is the one thing that, if I changed it, would most improve how this lesson produces understanding? The right answer is almost always one of three things: the specific explanation that produced the systematic wrong answer; the activity that generated the wrong type of work; the transition that lost half the class.

What to look for
You are looking for the one element where students diverged from correct understanding in a systematic, predictable way. Random errors indicate individual variation. Systematic errors indicate a lesson-level problem.
2
Decide: repair or replace?
2 minutes · Covered in detail in A2

Once you've identified what to change, decide whether to repair the element (adjust what's there) or replace it (remove and substitute something different). Repair is faster and appropriate when the element is fundamentally sound but poorly calibrated. Replace is necessary when the element failed because of its structure, not its execution.

Quick heuristic
If the same explanation with different wording would have worked, repair. If no version of that explanation was going to work for this class at this point, replace.
3
Make the change
12 minutes · The writing phase — kept short by being targeted

Write the revised explanation, activity, or transition. You are not writing new curriculum. You are rewriting one specific element of existing curriculum. Most elements can be revised in 8–12 minutes of focused writing. If you find yourself going over, you have either identified too many things to change or you are replacing when you should be repairing.

What targeted rewriting looks like
Replacing a worked example: 3 minutes. Rewriting an explanation using a different analogy: 5 minutes. Adding a check-for-understanding question between two activities: 2 minutes. Redesigning an activity from the same stimulus: 10–12 minutes.
4
Write the iteration note
2 minutes · The institutional memory that makes the change permanent

After making the change, write one sentence in your lesson notes: what you changed, why you changed it, and what you'll check to know whether the change worked. Without this note, the change exists only in this version of the document — and will be lost the next time you copy the lesson for a new cohort.

Example iteration note
"Changed osmosis explanation from arrow diagram to particle-ratio demonstration — 14/25 students were using pulling language. Check: same Q3 on next exit ticket. If still >4 students using pulling language, try concrete-materials demonstration instead."
What changes and what stays

The elements worth changing —
and the ones to leave alone.

Not every element of a lesson should be subject to iteration. Some elements are fixed for good reasons: the learning objective anchors everything downstream and should not change based on a single lesson's data. The lesson's position in the sequence and its relationship to surrounding lessons are usually structural and not iteration targets.

Element
Iterate when…
Leave alone when…
Explanations & analogies
Students produced a systematic wrong answer that reveals a faulty mental model
Leave alone when most students understood — one or two didn't is not a signal to change
Activities
Activity produced the wrong cognitive level (recall instead of application)
Leave alone when activity produced the right type of work but some students needed more time
Worked examples
Examples are all from the same context; students can't transfer
Leave alone when examples are varied; transfer failure is about the concept, not the examples
Pacing
Half the class lost the thread at a specific transition
Leave alone when a few students worked slower — pacing serves the majority
Learning objective
The lesson genuinely targeted the wrong thing for this class
Leave alone when students didn't reach the objective — that's a delivery issue, not an objective issue
Formative checks
Checks produced vague data (confidence ratings, content restate)
Leave alone when checks produced actionable data but intervention was too slow
The AI shortcut

How AI compresses the writing phase
from 12 minutes to 3.

Steps 1, 2, and 4 require your judgment — they cannot be automated because they require you to understand your class, read the data, and make the strategic decision about what to change. Step 3 — the writing — is the step AI handles well, because you've already made the decisions.

🤖The Step 3 AI prompt — for explanation replacement
Give the AI four things: (1) the original explanation that produced the wrong answer; (2) the specific wrong answer you observed; (3) the misconception that wrong answer reveals; (4) a constraint about the new approach.

Prompt template: “My original explanation of [concept] used [method]. [N] out of [total] students responded with [specific wrong answer], which tells me they have the misconception that [specific incorrect model]. Rewrite the explanation using a completely different approach — avoid [original method]. The explanation should be [word limit] words maximum, suitable for [year group], and should specifically address [misconception].”

The output is a draft replacement explanation in under 30 seconds. Total Step 3 time with AI: 2–3 minutes. The key is the specificity of the prompt — vague prompts produce generic replacements that may replicate the same problem in different words. A3 in this cluster covers the full AI regeneration workflow.
In practice

The workflow across a real
teaching week.

The following shows how the 20-minute workflow accumulates across a week of teaching Year 10 Chemistry — five lessons, five iteration cycles, and the compound effect on lesson quality by Friday.

🧪Year 10 Chemistry — Exothermic reactions, one week of iteration
Monday night — iteration cycle 1: Exit ticket shows 18/25 students described exothermic reactions as “releasing energy from the bond.” Correct: energy released when bonds form is greater than energy required to break bonds. Systematic misattribution of direction identified.

Tuesday morning: Opens with a worked example that explicitly shows the energy balance. New analogy: building blocks — constructing releases energy; breaking apart also does, but construction releases more. Time to iterate: 18 minutes.

Tuesday exit ticket: 11/25 still showing old language. Q2 correct for 22/25. Pattern: concept understood, vocabulary not yet detached from old model.

Wednesday iteration: Change is vocabulary-level only. Opening 3 minutes: explicit language replacement exercise — students rewrite their Monday answer using the correct framing. No new content. Total iteration time: 6 minutes.

By Friday: Q3 application correct for 23/25. Iteration log note: replace energy-released-from-bonds language with energy balance demonstration. Two-day lag between conceptual and linguistic correction — build language exercise into original lesson for next cohort.

This is the compounding effect of agile teaching at the lesson level. Not a dramatic transformation — a series of small, targeted improvements that accumulate into a lesson that works significantly better for every subsequent cohort.