What an AI-Era Class Could Look Like in Practice: The Classroom Reimagined

1. Where Part 1 left us

The principle from Part 1 was structural. The class becomes the AI-free cognitive gym, specialising in live, unaided, observable reasoning. Homework becomes AI-augmented intake, retrieval, and spaced practice, with the at-home AI deliberately prompted to behave like a Socratic tutor. The teacher's job moves from explainer to coach, curator, diagnostician, and judge.

This week we get practical. Maybe there's something you could try on Monday.

2. The 7-stage trajectory and where each stage now lives

The 7-stage trajectory is a working model for how one concept gets acquired.

1. Encounter. The student meets the concept for the first time and asks why it exists.
2. Schema formation. The mental model takes shape.
3. Encoding. Retrieval, spacing, and interleaving lock it into long-term memory.
4. Familiar application. Standard problems get worked.
5. Novel transfer. The concept gets applied to unfamiliar contexts.
6. Evaluation. The student weighs trade-offs and takes a defensible position.
7. Integration. The concept gets used fluently alongside others.

The redesign inverts where these stages happen. Stages 1 to 3 move home, where AI is the on-demand tutor. Stages 4 to 7 move into class, where the only AI-free observation hour of the student's day still exists.

3. The weekly homework budget

Students have around two hours a week of homework capacity per subject, as decided by my school. Here what homework could look like:

Forward prep, the intake thread. Roughly forty to sixty minutes a week, split across two or three concepts. Each concept gets fifteen to twenty minutes of either a concept video or an AI tutor session. The at-home AI is prompted to behave like a Socratic tutor: one step at a time, no full solutions, prompts that force generation.

Backward retrieval, the encoding thread. Roughly thirty to forty minutes a week. The structural move is the cold-attempt MCQ protocol. Students attempt every question unaided, without notes, under time pressure, and write a one-sentence justification for each answer. Only after they lock in their answers does AI become a tutor for the wrong ones.

Ongoing maintenance. Roughly twenty-five to thirty-five minutes a week. Three minutes a day of spaced-repetition flashcards. Ten to fifteen minutes of a metacognitive journal and a current-events find on alternating weeks.

What is deliberately out of homework should also be of note. Essay outlines move into class as a live timed activity. Full timed essays move into class as the only honest mastery signal. Past papers move into class for the same reason. The home is for intake, retrieval, and maintenance.

4. The weekly in-class budget

Most lesson in-class should have one or some or all of the tasks below.

• An entry-slip retrieval and prep diagnostic takes five to seven minutes. It is a cold retrieval check on yesterday's content plus a spot-check on the prep that was due. I mark it live, and the patterns I see decide what the next chunk of the lesson needs.
• Cold-attempt structured practice runs for fifteen to twenty minutes. Exam-style questions, unaided, with verbal justification under cold call. Sprinkle these throughout the unit and add full essay practices near the end of a unit.
• Exit retrieval and metacognition takes three to five minutes. Students write a one-sentence summary of what they learned and a question they still have.

Other ideas I came up with include: reteaching when the entry slip shows a lack of understanding, data response sprints, current events critique, evaluation debate, timed exam-style writing.

One piece of evidence is also informing my in-class tech policy. Mueller and Oppenheimer's 2014 study compared laptop and longhand note-taking and found laptop note-takers transcribed verbatim, while longhand forced selective rewording. Conceptual understanding was stronger in the longhand group on a delayed test. The cognitive gym defaults to pen and paper. AI typing-up of notes is fine after class as a retrieval task. The encoding work happens by hand.

5. Planning gets flipped upside down

The planning horizon shifts. It doesn't make sense to plan by how long it takes to "go through" the content anymore. Rather, planning should not be around how long it takes to do the practice for each topic.

For example, here is what a lesson on elasticity could look like. The night before day one, students watch twenty minutes of a concept video plus do a five-question self-quiz. In class, we look at real life examples, and students come up with more, identifying the common characteristics of elasticity determinants together. After day one, fifteen minutes of a Socratic AI tutor session on determinants plus five MCQ cold-attempts. In the second class, students write an essay response. After day two, twenty minutes of pre-reading a stimulus and bringing one real-world example. In the third class, students give feedback to each other's essay responses. You'll notice that most of the content acquisition moved outside the class, and class was used for consolidation and authentic practice and feedback.

6. Cons honestly named

This is more planning, not less. Designing the topic arcs in advance, configuring the at-home AI tutor's prompt pattern, guesstimating how long learning outside of the classroom takes for students, and reading the room each lesson in real time all add work to a teacher who already has all the powerpoints built. Thankfully, students aren't the only ones equipped with AI.

A note on method: this issue was produced through the co-creation workflow I'm advocating. The idea, the angle, the practitioner observations, the curated sources, and the final wording are mine. An AI assistant calibrated to my voice (through a guide of phrases I've approved and rejected) did the research legwork on sources I selected and drafted from an outline we agreed on.

References

1. Bastani, H., Bastani, O., Sungu, A., Ge, H., Kabakcı, Ö., & Mariman, R. Generative AI without guardrails can harm learning: Evidence from high school mathematics. Proceedings of the National Academy of Sciences. 2025.
2. Roediger, H. L., & Karpicke, J. D. Test-enhanced learning: Taking memory tests improves long-term retention. Psychological Science 17(3): 249–255. 2006.
3. Mueller, P. A., & Oppenheimer, D. M. The pen is mightier than the keyboard: Advantages of longhand over laptop note taking. Psychological Science 25(6): 1159–1168. 2014.
4. Bjork, E. L., & Bjork, R. A. Making things hard on yourself, but in a good way: Creating desirable difficulties to enhance learning. In Gernsbacher et al. (Eds.), Psychology and the real world. Worth Publishers. 2011. (Previously cited; see the earlier AI in Schools issue on student thinking and AI.)
5. Anderson, L. W., & Krathwohl, D. R. (Eds.). A taxonomy for learning, teaching, and assessing: A revision of Bloom's taxonomy of educational objectives. Longman. 2001.