Teaching students to work with AI
AI assignment ideas for business courses
Business courses have an unusual problem: the deliverable students produce - a memo, a deck, a market analysis - is exactly what AI produces best. Here are ten assignments where that stops mattering.
A consulting-style recommendation memo used to take a student six hours and demonstrate that they understood the case. It now takes four minutes and demonstrates nothing. The artifact was never the learning; it was just the most convenient evidence of it. That convenience is gone.
What follows assumes students have AI, will use it, and still have to do the part that’s actually being taught: deciding under uncertainty and holding a position when someone with authority pushes back.
Negotiation and stakeholder pressure
1. Defend the margin
The student has to justify a price increase to a sceptical CFO over email. The CFO has a brief: they know the student’s cost structure, they’ve seen the competitor’s quote, and they will not accept “costs went up.” Every reply has to concede something real or hold the line with a number.
Grade: did the position survive, and did the student give ground on the right things? A student who caves on price in turn two and one who never moves at all have both failed differently.
2. The board that’s already decided
Students present a recommendation to a board that has privately committed to the opposite course. The AI characters have distinct motives - one protecting a budget, one protecting a relationship, one genuinely persuadable. The exercise is reading the room and finding the one person whose mind can change.
3. Bad news, early
The project will miss its deadline by six weeks. The student has to tell the client. Grade on what they disclose, when, and whether they arrive with options or just an apology. This is the single most common thing junior hires do badly and the rarest thing they’re taught.
Models and quantitative judgment
4. The model with the wrong assumption
Give students a working financial model that produces a plausible answer and contains one bad assumption - a growth rate borrowed from a different market, a cost that scales linearly when it shouldn’t. AI will happily extend the model without questioning it. The assignment is finding the assumption.
5. Build it, then break it
Students use AI to build a break-even or runway model. Then they have to state which single input, if wrong by 20%, changes the recommendation - and defend that choice. The formulas are free; the sensitivity judgment isn’t.
6. The number that doesn’t support the story
Hand students a dataset and a draft strategy memo where the recommendation isn’t actually supported by the data. Their job is to find the gap and decide what to do about it - revise the recommendation, or find what would need to be true for it to hold.
Case work that AI can’t shortcut
7. Your case, not a famous one
The moment you use a Harvard case, AI has read it and so has every student. Write a short case about a local company, a fictional firm with specific numbers, or a situation from your own industry experience. Two pages is enough. Specificity is the whole defence.
8. The incomplete brief
Give students 70% of what they need and no way to get the rest. The assignment is to state what’s missing, decide anyway, and specify what would change their mind. Models are poor at saying “I don’t have enough information” - they fill gaps confidently. Students who notice the gap are demonstrating exactly the thing.
9. Two defensible answers
Construct a case where expand-now and wait-a-quarter are both genuinely defensible. Assign each student a side at random. Grade the argument, not the conclusion. This removes the possibility of a right answer to look up.
Process over artifact
10. The annotated deck
Students submit their recommendation plus one slide they cut and why, and one AI suggestion they rejected and why. Two sentences each. It takes ninety seconds to grade and reveals more about their judgment than the deck does.
Where AI should be doing the work
Formatting, first-draft prose, formula syntax, competitor summaries, and the mechanics of building a model. None of that is the learning objective in most business courses, and pretending otherwise just taxes students who are bad at spreadsheets. Hand it over and move the assessment onto the assumptions and the defence.
Common questions
What about case competitions and group work?
They hold up better than individual written work, because the assessment is partly the live defence. Strengthen them by requiring each member to record a position before the group converges - it surfaces who actually did the thinking. Structures for this.
Can I still assign a written memo?
Yes, if the memo is about your specific case and the grade is weighted toward the recommendation’s reasoning rather than its polish. Add one follow-up question the student answers in class and the format is largely rescued.
How much of this works in a 200-person section?
Numbers 4, 5, 6 and 10 scale well - the assessment target is specific enough to grade fast. The negotiation and board exercises need a platform that runs the exchange and scores the transcript, or small sections.
See what this looks like in a real assignment
Kova turns a sentence into immersive, auto-graded coursework - AI conversations students have to argue their way through, models they build and defend, transcripts scored on the reasoning rather than the artifact.
The full guideHow Conjure builds itRequest a demo