NYC and LA are restricting student AI. We used it to expose what school missed—and repair two years of gaps in six weeks. Part 1 of 3 in a series.
Between my husband and I, three of our parents were public-school teachers. Even the fourth one had worked, for a brief time, within the public school system. We had tremendous respect for the system itself.
I use past tense for a reason.
For twelve years we had assumed: we assumed that by being in a good school district, we would be safe from the decline of public education.
My son (“Bobby”) was an excellent student in elementary school, but had trouble adjusting to middle school. We decided to start touring the local private school, and was shocked at how far ahead they were of public schooling. The private school was mandating essays written in cursive by third grade, teaching algebra in sixth grade, and actually held students accountable from an early age on.
We realized we were failing our children as parents, and we had an emergency on our hands. We sent Bobby to be homeschooled by my parents for about six weeks, finishing out the remainder of his term learning cursive and sentence diagramming. Then we sent him to full time summer school at the private school, and hoped that would be enough.
One month before the year started, he took an admissions test. He was supposed to be entering seventh grade.
Bobby failed the admissions test.
He failed even the sixth grade placement, effectively putting him two years behind.
This was a blow to the entire family. Bobby cried and cried many tears over this - after all his hard work, and sacrificing a whole summer, he didn’t pass.
But the story has a happy ending. After six weeks, Bobby was able to pass the placement retest.
This is the story of how we used AI to fix two years’ worth of deficiencies in Saxon math and grammar.
We turned a deadline into an engineering problem
The subjects Bobby failed were in math and grammar. The private school had a Saxon math curriculum. We had six weeks to try again. We could not use a curriculum and start on page one. We need to find and identify missing pre-requisites and weaknesses, repair them, and get him to pass the test in six weeks.
I downloaded Saxon math textbooks. I had AI design a comprehensive diagnostic test covering up to sixth grade Saxon math.
Bobby scored 35 out of 60 on his initial assessment. Structural weaknesses were identified with long division and decimal multiplication. The test also revealed he was never taught percentages, order of operations, geometry, or statistics. A few correctable habits were identified: failure to simplify ratios and final answers.
Bobby’s diagnostic results.
Grammar needed a different repair plan. There, the gap was much more critical: he did not know the difference between a noun and a verb. (How and why this happened in Utah public schools is a topic for part 2). He had learned sentence diagramming, so he had an intuitive understanding of English structure, but he needed to learn proper labels for them.
Bobby Never Interacted with a Chatbot
Bobby did not spend these six weeks in front of a computer screen. All the work he did, was on pen and paper worksheets generated by AI.
The loop:
- Give Bobby a diagnostic without hints or review.
- Identify the reason behind each miss.
- Have AI generate a short paper lesson or test aimed at that problem.
- Have an adult in the family administer and score it, and share the results with AI.
- AI decides whether each miss came from a missing concept, a shaky procedure, or a simple slip.
- Change the next worksheet in response.
- Test the skill later without hints and mixed with later material.
How the AI curriculum worked.
Different errors called for different work. If Bobby did not know what a percent was, he needed instruction. If Bobby understood percent but subtracted tax instead of adding it, he needed a direction check. If he made mistakes in the arithmetic itself rather than the concept, rehashing the concept would simply waste time.
This customized-by-AI approach targeted exactly where Bobby needed in order to master Saxon math.
For example, Bobby did not know percentages at all at the beginning. After he learned the basic concept, he quickly became fluent in converting among fractions, decimals, and percentages. Then his errors stopped being about doing wrong calculations, and became more about how to correctly apply them. So AI made the judgment call to evolve his worksheets to emphasize word problems and real-life scenarios.
Mid-way through these six weeks, he scored 22 out of 26 on a checkpoint, proving that he still needed additional drilling and practice before he could claim mastery.
Worksheet sample.
AI was able to identify habits in a way that a parent might not. Bobby scored 24 out of 29 in order of operations, but every miss came from the same habit. When multiplication or division shared a line with addition or subtraction, he reverted to strict left-to-right calculation. A custom worksheet simply asked him to circle the right order. He did not have to do tedious equations in order to fix the habit.
Personalization made the work more fun
Mastering two years’ worth of Saxon math in six weeks is a lot of work for a child. Bobby likes machines, science fiction, rubber ducks, and board games, so the worksheets were customized to his interests.
The math track centered around the Terminator. Mastering each major concept would get him promoted as the next Terminator model.
Gamifying the worksheets.
The customization also worked with his thinking style. Structuring concepts as an assembly line helped him master them faster, for instance. Bobby was allowed to “advance” only after he passed an unassisted concept check.
AI changed the production economics
A family may know that a child needs more practice with percentages. But producing the right practice that enables mastery in the shortest amount of time is something else altogether.
Only a system that already knew Bobby inside and outside knows that a worksheet must explain percent as “per hundred” and can leverage Bobby’s existing knowledge of fractions to accelerate his mastery. And when the system detects that Bobby was confusing discounts with sale prices, the system must adapt and quickly correct that error overnight.
AI made that revision cheap enough to do it overnight. Altogether, we produced 60 worksheets over 6 weeks, totalling 398 pages.
We could produce a typeset lesson, a new answer key, and a fresh assessment whenever evidence changed. Without this automation, we would almost certainly have been stuck with giving standard Saxon drills without any clear path to mastery in such a short time window.
What we learned
AI is able to diagnose narrowly enough to act.
A broad label such as “behind” is difficult to remediate. AI has the ability to drill down on exact weaknesses, and leverage strengths (e.g., fraction mastery) to establish the quickest path to overall mastery.
AI is able to generate the smallest useful lesson.
Repeated drills of Saxon math may be useful when learning concepts over an entire year. They are not nearly as useful when trying to pass an admissions test in six weeks. AI was able to diagnose the minimal cognitive load to achieve mastery. When a bad habit was identified, Bobby was not required to work through tedious calculations over and over to correct it. He would fill in the blanks or draw circles instead.
AI does not remove humans from the equation.
An adult in the family was able to tell whether Bobby was rushing through the questions, was refusing to read instructions, and was able to correct with some prompting. These distinctions were often more useful for AI in generating the next sets of lessons than the actual score.
AI can be used to reward progress and create fun.
Hyper-localized scoring guides were able to reward (or penalize) Bobby when he skipped units, or when he missed reading instructions. Gamifying the worksheets according to Bobby’s interests enabled Bobby to have fun with them rather than finding them tedious.
Humans were still the scarce resource.
AI reduced the cost of curriculum production. AI did not print the worksheets, or sit with Bobby, or enforce the schedule, or score the work, or persuade an exhausted child to try again. This was a high intensity family intervention.
The final takeaway
The result resembled a small curriculum engineering team organized around one student. AI repeatedly re-wrote lessons, assessments, incentives, and cumulative review around one student.
Bobby still worked on paper and never consulted a screen. Adults still retained judgment. The six week pass was the family outcome.
But there is still a remaining question: how did a high-performer in a good public school district have so many gaps in math and grammar?
Part 2 explores why and how public school systems got to this state.
