A unit is not ten lessons. It is one argument delivered across ten days, where each day assumes the ones before it. Almost every AI lesson tool treats it as ten independent requests — and everything that goes wrong downstream follows from that.
Failure one: the same lesson, four times
Ask for five days on the same topic in five separate prompts and you will get five well-formed lessons that are mostly the same lesson. The model has no way to know that Tuesday already did a gallery walk, because Tuesday was a different conversation. Teachers notice this in week one and usually conclude the tool is shallow. It is not shallow — it is amnesiac.
Failure two: the flat difficulty curve
Gradual release only means something across time. Within a single lesson you can move from modelling to independence; across a unit you should be able to assume that what was modelled on day two is available on day seven. A tool that cannot see day two has to re-scaffold on day seven, so every day begins near the bottom and the unit never accumulates.
The gradual release of responsibility model describes a shift over time from teacher modelling toward independent student practice — a progression, not a single lesson's structure.— Pearson & Gallagher, The Instruction of Reading Comprehension (1983)
Failure three: the assessment nobody taught toward
The unit test is usually generated last, from the unit's topic rather than from the lessons that were actually delivered. So it assesses a reasonable version of the unit — just not yours. This is the most expensive failure of the three, because you find out after teaching it.
What fixes it is context, not a better prompt
The fix is unglamorous: every generation has to receive the whole structure, not the current request. That means the course, the unit, the week, the lesson's target, the earlier lessons in the unit, the sections already written for this lesson, and the materials the teacher attached — all of it, on every call.
That is how TeacherHero is built, and it is the reason the product is a hierarchy rather than a prompt box. Some specifics:
- Earlier lessons travel forward. A lesson's context includes the earlier lessons in its unit, which is what makes a warm-up that revisits Monday possible rather than coincidental.
- Sections know their siblings. When the third section of a lesson is generated it receives the two already written, so it continues them instead of restarting the day.
- The assessment is anchored to the lessons. A connected unit assessment is built from every lesson's target and final product plus the unit's vocabulary and materials. You approve a coverage blueprint before any items are written.
- Editing a lesson stales the assessment. Approval pins which version of each lesson the assessment was built from. Edit, add, or delete a lesson afterward and generation is blocked until you re-approve — so the paper cannot quietly stop matching the unit.