Position Paper No. 2

The Egg in the
Cake Mix

Why the lesson planning debate is asking the wrong question — and what actually determines whether a teacher improves.

4,163
lesson plans generated by teachers using Rumi's AI assistant since launch
98%
of teachers on AI-assisted platforms actively use the lesson planning feature1
89.7%
of teachers observed not engaged with or only partially using a scripted plan2

The Debate

Two sides. Both partly right.

The EdTech community has been arguing about lesson planning for years. On one side: standardized, scripted plans that guarantee minimum quality, reduce cognitive load, and protect the 80% of teachers who have never received training on how to plan. On the other: open-ended AI tools that give teachers autonomy — the chance to make something their own, adapt to their classroom, build professional skill.

Both sides have evidence. Both sides have failures. The debate has produced heat. It has not produced answers.

We think that's because both sides are answering the wrong question. The question is not "scripted or generative?" The question is: what gives a teacher enough of a stake in a lesson that they actually deliver it?

"The lesson plan is not the intervention. The teacher is the intervention. The plan only matters insofar as it changes what the teacher does in the room."

A Lesson from Consumer Psychology

The Betty Crocker Paradox

In the 1950s, General Mills launched Betty Crocker instant cake mix. Everything was included. You added water, stirred, and baked. Sales were terrible.

Psychologists diagnosed the problem: the process was too complete. Mothers felt they hadn't actually made the cake. The product required no contribution from them — and so they felt no ownership over the outcome. General Mills' fix was counterintuitive. They removed the powdered egg from the mix. Now you had to add a real egg. Sales jumped.

The egg did nothing nutritionally significant. But it gave the maker one genuine act of contribution — and that changed the psychological relationship to the product. It was no longer a factory cake. It was her cake.

This is not a metaphor we are reaching for. It is a precise description of what we have observed in 1,101 schools across Pakistan.

What the Data Shows

Three sources. One story.

We are drawing on three distinct data sources: an independent evaluation by the EdTech Hub, a teacher survey conducted by GSMA/Radicle Development across 211 teachers on our AI coaching platform in Islamabad, and our own Rumi platform database covering 4,335+ lesson plans generated by teachers across Pakistan.

Source 1 — EdTech Hub Independent Evaluation, 2024

The Ownership Problem

The EdTech Hub evaluation of scripted lesson plans across Pakistan government schools found four failure modes that recur regardless of plan quality:

Plans arrive without context

Teachers receive a plan but don't know why it was structured this way — no pedagogical rationale, no discussion. It lands as an instruction, not a tool.

No space for local adaptation

A plan designed for Grade 4 Islamabad schools may not fit a mixed-grade classroom in Quetta. Scripted plans don't know your classroom. Teachers do.

Assessment and follow-through disconnect

69% of teachers report greater satisfaction with AI-assisted plans over government-issued scripts — but satisfaction doesn't automatically translate into better delivery.

No feedback loop

The plan is delivered (or not). No one asks what worked. No one helps the teacher close the gap between what they planned and what happened in the room.

89.7% of teachers observed were not engaged or only partially engaged with their scripted lesson plans. The plans were technically present. The teachers had moved on.

Source 2 — GSMA / Radicle Development Teacher Survey, March 2026

High adoption. High ownership. Mixed delivery.

211 teachers. 6 focus groups. Islamabad schools using Taleemabad's AI coaching platform. The numbers are striking — but the nuance matters more.

79%
use the platform daily. 94% use it at least 3 times per week.
91%
report improved teaching effectiveness since using AI lesson planning
73%
say their lesson preparation time has decreased — more time for reflection

But buried in the focus group data is the contradiction: teachers who rated the platform highly also wanted more control. More flexibility. The ability to adjust timing, sequence, and emphasis. They were not rejecting the tool. They were claiming it.

The survey's own conclusion: "Overall impact depends on the educator." The tool is not the variable. The teacher's relationship to the tool is the variable.

This is the egg. Not the mix.

Source 3 — Rumi Platform Data, Pakistan

What teachers actually choose

Rumi's lesson planning data tells a story about teacher preference when given real choice. We have two plan formats: pre-generated chapter plans (curriculum-linked, fully scripted, launched March 2026) and user-generated plans (teacher specifies topic, grade, and parameters; AI builds from their inputs).

User-Generated
4,163 plans
Pre-Generated
172 plans

The ratio is 24:1. Teachers, when given the choice, overwhelmingly generate their own plans. This holds even when the pre-generated plans are higher quality by most objective pedagogical measures.

But here is the more interesting observation: pre-generated plans are growing fast. 10 in March. 162 in the first ten days of April. Teachers who started with user-generated plans are migrating toward structured pre-generated formats — now that those formats exist and feel complete rather than imposed.

The lesson: when the pre-generated plan is designed to accommodate agency rather than eliminate it, teachers welcome it. The issue was never scripted vs generative. It was whose hand is on the plan.

One more signal from the data: teachers generate lesson plans at 1am Pakistan time. Nobody asks them to do this. Tuesdays and Wednesdays account for 49% of all weekly generation. This is not compliance. This is profession.

What We Built and Why

The National Institute of Excellence
in Teacher Education (NIETE)

In December 2023, Taleemabad — in partnership with the Ministry of Federal Education and Professional Training — built and launched the National Institute of Excellence in Teacher Education. This is not a partnership with a pre-existing institution. We designed it, built the curriculum and technology backbone, and operate the programme. NIETE is a government programme: MoFEPT launched it officially, the Federal Directorate of Education and Federal College of Education are co-partners, and LUMS School of Education contributed to curriculum design.

NIETE's design reflects exactly the lesson described in this paper. We did not build a platform that delivers scripted plans to teachers. We built a coaching environment in which teachers participate in their own development — submitting audio recordings, receiving AI-generated feedback against the OECD coaching framework, and generating their own lesson plans on Rumi.

The 69% satisfaction advantage of AI-assisted plans over government scripts — cited above from the EdTech Hub evaluation — was documented in NIETE's context. It is the egg in action.

Our Framework

A three-layer model for adaptive planning

1
Pre-Generated Chapter Plans — for capacity-constrained teachers

Curriculum-linked, pedagogically structured plans ready to use. Designed for teachers in high-load contexts who lack time or confidence to generate from scratch. The floor, not the ceiling.

2
AI-Assisted User Generation — for teachers who want agency

Teacher specifies topic, grade, classroom size, and parameters. AI builds the plan. The teacher made the choices. The plan is theirs. This is the egg. This is what 96% of Rumi's lesson plans are.

3
Coaching Feedback Loop — for closing the gap between plan and classroom

Teacher submits audio from their class. AI analyzes against OECD coaching framework. Feedback identifies the gap between what was planned and what happened. This is where real improvement occurs — not in planning, but in reflection.

The Argument

What actually changes a teacher

The EdTech field has a theory of change problem. Most lesson planning interventions assume that if you improve the plan, you improve the lesson. The evidence — including 89.7% non-engagement with scripted plans — suggests this is mostly false.

Plans don't change classrooms. Teachers change classrooms. And teachers change when three things happen in sequence:

1

They feel ownership

They made a choice about this lesson. It's not something that happened to them — it's something they built, even partially.

2

They receive feedback on the gap

What they planned versus what actually happened in the room. Without this, the plan is orphaned from reality.

3

They iterate

The next lesson is better because of what they learned from the last one. Not because the platform updated. Because the teacher did.

The loop closes

This is not a one-time event. It is a professional identity. Teaching becomes something you improve at — not something that happens to you.

The GSMA survey confirmed this directly: "Overall impact depends on the educator." Not the tool. The educator. Every teacher who reported improved outcomes had something in common — they were using the tool actively, adapting it, treating the AI as a collaborator rather than an authority.

The 7% who showed no improvement were using the platform passively. They received the output. They did not engage with it. Same tool. Different relationship. Wildly different outcomes.

Recommendations

For policymakers and programme designers

The cake mix was never the problem. Neither was the instruction to add water. The problem was that there was nothing left for her to do. The moment she added her own egg — she owned the outcome. That is the entire lesson.
1GSMA / Radicle Development. Orenda Platform Validation Study. March 2026. N=211 teachers, Islamabad. 6 focus groups.
2EdTech Hub. AI in the Classroom: Independent Evaluation of Taleemabad. 2024. Government school observation data, Islamabad.
3Rumi platform data. Lesson plan type breakdown: user-generated = 4,163; pre-generated chapter = 172. April 2026. Pakistan only, non-test users.
4National Institute of Excellence in Teacher Education (NIETE). Launched December 2023. Partners: Ministry of Federal Education and Professional Training (MoFEPT), Federal Directorate of Education (FDE), Federal College of Education (FCE), LUMS School of Education.