Chapter 01 · Section II · 17 min read
The teaching workflow, mapped to where AI helps
A stage-by-stage walk through a real Nepali teaching week showing which parts of the job AI shifts, which stay exactly where they were, and why the difference matters.
The brochures will tell you that AI changes everything about teaching. It does not. It changes some of the hours, leaves others untouched, and the difference is the whole story. To see it clearly, you have to stop thinking about teaching in the abstract and walk through a real week — a real teacher, a real classroom, a real set of obligations between Sunday morning assembly and Friday afternoon staff meeting — and mark, stage by stage, where the technology actually moves the work.
A real teacher, a real week
Meet Ms. Sapana Karki. She teaches Class 8 science at a community school in Chautara, Sindhupalchok. Fifty-three students in the register, on a good day forty-five in the room. Her class contains children whose mother tongue is Nepali, three Tamang speakers from the upper wards, and two Newar children from the bazaar. Reading levels span Class 4 to Class 9. She has a CDC textbook, a chalkboard, intermittent grid electricity, a basic Android phone, and on Wednesdays the staff room laptop. Her week is the unit on the digestive system.
We will walk through her week stage by stage and mark each one high, medium, or low leverage from AI. By “leverage” we mean: how much of this stage’s time can the technology compress without making the teaching worse?
Sunday and Monday — lesson preparation (high leverage)
Sapana sits down on Sunday evening with the CDC unit and her diary. She has six 40-minute periods to teach the digestive system. Without AI, she would draft six period plans from memory and the textbook — perhaps two hours of work after the children are asleep.
With AI, she opens a chatbot on her phone, types the unit objectives from the CDC page, tells the model her class size, the spread of reading levels, and what materials she has — chalkboard, the textbook diagrams, no projector — and asks for six period plans. In two minutes she has drafts. She edits them. She knows the model does not know that Bishal in the third row freezes when called on cold, or that the morning assembly often runs long on Mondays. She rewrites the hook activities, drops the model’s suggested group work for Period 3 because her classroom benches are bolted down, and adds a question about the bel fruit because half her students will have eaten one for breakfast. Time saved: about ninety minutes a week. This is the highest-leverage stage in her week.
She uses the same chatbot to produce a tiered worksheet — foundational, on-level, stretch — for the Wednesday practice period, and a short Nepali-language explainer paragraph she will read aloud for the two students whose English is weakest. Both drafts take minutes; both need her eyes before they reach the class.
Tuesday through Friday — in-class teaching (low leverage)
Periods 1 and 2 on Tuesday. Sapana stands in front of her class. She watches Sushmita’s face go blank halfway through the explanation of peristalsis, slows down, asks a different question, lets a student demonstrate with hand gestures, switches briefly into Nepali for a definition, switches back. She notices that Prakash, who has been quiet for two weeks, makes a small joke under his breath, and she rewards him with a question he can answer. She manages the noise from the next classroom, the dust from the open window, the child who needs to leave for the toilet.
None of this is AI work. The minute the children are in the room, the leverage drops to nearly zero. The teaching is the relationship — judgement, attention, presence, the thousand small decisions that respond to faces she has known since Class 6. A model cannot make any of these decisions, and no useful AI tool tries to. The hours in front of the class are exactly the hours they were before.
Wednesday afternoon — worksheet and homework design (high leverage)
After lunch on Wednesday, Sapana has a free period. Without AI, she would spend it composing the practice questions for Thursday — slow work, especially generating variations that catch the children at the back without boring the children at the front. With AI, she pulls up the staff room laptop, gives the model the topic and the three ability tiers, and asks for fifteen questions across the tiers, with answer keys and brief hints. Six minutes. She reads through, fixes two questions where the model has used a vocabulary word her class has not met, and prints. The free period is now a free period. She uses it to call the mother of a student who has been absent.
Thursday evening — marking (medium leverage)
Forty-five exercise books from the Tuesday practice. Most of the work is short-answer, some is a short paragraph. Without AI, this is two hours of red pen. With AI, Sapana cannot — and should not — feed each child’s handwritten work to a model on her basic phone. But she uses the model differently. She drafts a rubric with the chatbot in five minutes — five criteria, three levels — and uses it consistently as she marks by hand. She types two or three of the longer paragraphs into the model and asks for feedback notes she can adapt. She drafts a one-paragraph comment template for the strong, middling, and struggling responses, and adapts each to the specific child.
The marking still takes most of the evening. But the feedback her students get is more useful than before, because the model freed her from composing each comment from scratch. Medium leverage. The grade is still hers. The relationship the comment carries is still hers.
Friday — parent communication and admin (medium-high leverage)
Two parents to write to. One a routine update; one a careful note about a child who has been struggling. Sapana drafts both in Nepali with the model, reads them critically, edits the careful note heavily — the model’s first draft was too formal and a touch cold — and sends them. Ten minutes for what used to take forty. The end-of-week attendance summary for the headmaster: the model produces a clean draft from her notes in three minutes.
Saturday — professional development (medium leverage)
Sapana is preparing for an NEB curriculum workshop on the new English textbook. Without AI, she would read the framing chapter and take notes. With AI, she also asks the model to summarise the key shifts from the previous edition, lists the pedagogical assumptions, and gives her three questions to raise with colleagues at the workshop. The model is not a substitute for reading the source — she still reads it — but it sharpens her engagement. Medium leverage, and it compounds over a year.
What the map shows
Across Sapana’s week, AI saves perhaps three to five hours. None of that time is taken from the children. All of it is taken from the paperwork, the drafting, and the administrative debris that the system has piled on top of teaching. She uses the recovered time to plan more carefully, mark more thoughtfully, talk to two parents she would otherwise not have reached, and sleep an hour longer on Thursday. Her teaching does not become “AI-driven.” Her week becomes survivable, and the parts of it that matter most get more of her attention.
A +2 mathematics lecturer in Pokhara, a private-school English teacher in Lalitpur, a tuition centre instructor in Birgunj — the specifics differ, but the shape of the map does not. Preparation, worksheets, marking feedback, parent and admin writing, and CPD reading: medium-to-high leverage. Time with students: low. Judgement calls about individual children: zero, and that is non-negotiable.
The verification-cost question, applied
When Sapana decides whether to use the model for a specific task, she does not consult a guru. She runs the same two-question test from the previous section: if the model’s output is wrong in a way I miss, what does it cost — and how fast can I catch it? A lesson plan with a wrong question: caught in seconds, no cost. A parent letter with a wrong tone in honorifics: caught in a careful re-read, embarrassing if missed. A grade on the SEE pre-board: not a job for the model. The map and the test, together, decide where the technology earns its place in her week and where it stays out.
Check your understanding
Quick check
—A Class 8 government-school teacher is mapping her week to decide where to use AI. Which stage gives her the highest leverage from current AI tools?
Quick check
—A +2 lecturer is deciding whether to use AI to produce the final marks for her students' internal assessment, which will count toward their board grade. What does the verification-cost and downside-cost test tell her?
What comes next
The workflow map answers the teacher’s question about her own time. There is another question waiting that the map cannot avoid — the one your students are already deciding for themselves. The next section is about the AI your students are using, with or without your permission, and what an honest teaching response actually looks like.