The PB&J Problem: Why Students Need to Learn to Give Directions Again
I'm sure you all remember the exercise. Somewhere around third or fourth grade, a teacher stood in front of the class holding a loaf of bread, a jar of peanut butter, and a knife, and asked you to write instructions for making a sandwich. And you weren't just supposed to think about it...you actually had to write it down, step by step, and precise enough that a stranger could follow it with no assumptions filled in.
And then the teacher followed your instructions exactly as written.
"Put the peanut butter on the bread" because the whole jar dumped on top of a single slice, because that's what the sentence actually said. Some kids groaned while others laughed, but everyone in that room learned something they didn't get from a worksheet: that an instruction/direction is only as good as the thinking behind it. Directions that were vague didn't produce a vague sandwich...they produced a wrong one that was made with total obedience to a total lack of clarity.
I think we owe students that same lesson again, right now, for a completely different tool.
The New Jar of Peanut Butter
AI will do exactly what you tell it to do. And in this context, I see that as both the whole problem AND the whole opportunity. When you type something vague into a chatbot you'll get something vague back that's delivered with total confidence and formatted so nicely/cleanly that it can look right even when it isn't.
A student who types "write me an essay about the causes of World War I" is doing the sandwich exercise all over again, except this time nobody is stopping them mid-process to show them the mess. The tool just produced something that reads fine and might even be graded fine, depending on the assignment and the teacher. But the student never learns the part that matters, which is that getting something useful out of AI is a process, not a request.
We spent decades teaching kids to write clear directions for a sandwich, and yet we haven't spent any real time teaching them to write clear directions for a thinking partner that can generate a plausible-sounding wrong answer just as easily as a right one. And I think that gap is where a lot of the current anxiety about AI in classrooms lives right now. It's not really about cheating, but about students outsourcing a task to a tool they never learned to direct, and mistaking the output for their own thinking because it appeared instantly and looked finished.
Learning from the Sandwich Exercise
Let's go back to what made that exercise work. It wasn't just "be specific." It was a whole sequence of skills (most of which have nothing to do with sandwiches):
Noticing what you're assuming the other party already knows. We know a slice of bread had two sides and that 'spread' means a thin layer and not a pile. The exercise forces you to surface assumptions you didn't know you were making.
Sequencing. It showed us that order matters! So, you gotta open the jar before you dip the knife in it.
Checking the output against the intent w/ feedback. You didn't just write the instructions once and call it a day. You were able to watch them get followed, and almost immediately get feedback to see where they broke and how you should revise them.
Accepting that failure was informative and not embarrassing. The kid whose sandwich was a disaster learned more than the one who just happened to write good instructions by accident.
And I think that last point is the connector for me with AI. A vague prompt that gets a mediocre response is a teaching moment, but only if the student is asked to look at the gap between what they wanted and what they got. If they're not asked to look at that gap, then it's just a shortcut that returns lower-quality thinking that's dressed up in polished sentences.
Moving Forward: Treat Prompting as a Taught Process, Not an Assumed Skill
So how do we move forward in the quest to truly provide students with the best way to utilize AI? It's definitely not another one-off "how to prompt [insert LLM]" handout. It needs to be a recurring, visible process that gets applied across subjects, the same way we taught the writing process itself: draft, revise, get feedback, revise again.
I'd build it around four moves that would be taught explicitly and practiced repeatedly:
State the goal before you state the request. Before typing anything into the tool, students write one sentence that helps them understand what this is actually for and how they will know if it worked. A paragraph for an audience of sixth graders is a different request than a paragraph for a science journal, even if the topic is identical. Skipping this step would be like skipping the "open the jar" in the sandwich instructions. It seems obvious until it isn't there.
Draft the prompt, then audit it for the assumptions you didn't write down. Have students trade prompts the way they'd trade sandwich instructions, and ask: what did you assume I already knew? What's missing that you took for granted? This is the direct descendant of the PB&J exercise, just aimed at a keyboard instead of a kitchen.
Read the output like a first draft (not a final answer). This is the step that's most skipped entirely because the output looks finished. Teach students to ask specific questions of it: What's being claimed here? Where would I push back if a classmate wrote this? What's missing that I know should be here? Treating AI output as a rough draft is the single highest-leverage thing we can instill in utilizing these tools.
Revise the prompt based on what the output revealed beyond the topic. If the response missed the mark, the fix usually isn't just "try again with different words." It requires going back to step one and asking what wasn't clear about the goal. Doing this closes the loop the way watching your sandwich instructions get followed closed the loop. You don't just retry, but you learn something about your own thinking and apply it.
None of this requires banning AI or pretending it doesn't exist in classrooms, which was never a realistic strategy anyway. It requires treating "how do you direct a tool that will do exactly, and only, what you tell it to do" as a skill with the same seriousness we once gave to writing directions for a sandwich. We knew (decades ago) that clarity of thought shows up in clarity of instruction, and that the fastest way to teach it was to let students watch their own vague thinking get carried out literally.
AI just raised the stakes on the same lesson. The sandwich was messy and a little funny. A student's understanding of history, or science, or their own writing voice, hollowed out because they never learned to ask better questions of a tool that answers everything with confidence, is not funny at all.
We already know how to teach this. We've been teaching it since third grade. We just need to point it at a new jar of peanut butter.
Comments