What a Kindergarten Sandwich Lesson Teaches Us About AI Prompting

A Kindergarten Teacher Just Taught the Best AI Prompting Lesson on the Internet, and It's Not What You Think

TLDR: A kindergarten teacher followed her students' sandwich-making instructions literally, with hilarious and messy results. The lesson maps perfectly onto why most people's AI prompts fail, and why the most experienced people in a room are often the worst at writing them.

Summary: There's a video going around of a kindergarten teacher who promised her five-year-olds she would follow their written instructions for making a peanut butter and jelly sandwich exactly as written, no assumptions, no common sense, nothing added. The results are predictably chaotic. She picks up the bread. Picks up the jars. Holds them. "Did I make it?" The kids lose it. Someone forgot to mention a knife, so she spreads jelly with her fingers. Jelly gets stacked directly on bare bread. No plate. No knife. No flip. Just a mess, and a room full of delighted kindergarteners who are suddenly very aware of what they left out.

What makes this clip remarkable is not the chaos itself but the revision process. The class works together to fix the instructions step by step. "Put bread on the plate. Grab the knife. Scoop peanut butter. Spread it. Get another piece. Put jelly on it. Flip it onto the first one." Sandwich. Kids cheer. The teacher grins. Nobody in that room needed to say the quiet part out loud, but here it is: every adult watching that video has the exact same problem with AI.

The author Kamil makes a point that I think cuts deeper than most "write better prompts" advice: the more senior you are, the worse you tend to be at this. A junior analyst writing a procedure has to think through every step because nothing is automatic yet. A 20-year veteran skips half the steps in their head. Their expertise has gone underground. Ask a senior salesperson how they close deals and you get vibes, not steps. Ask a marketing director what makes a campaign land and you get taste, not a recipe. That tacit knowledge is invisible to them, and it's invisible to the model.

Some roles have lived in checklists their whole careers: software engineers, operations people, pilots, surgeons, lawyers, QA teams, anyone who has ever written a runbook. AI feels native to those people because they were already translating expertise into explicit instructions every day. Other roles built their value on ambiguity: executives selling vision, marketers chasing taste, strategists pointing at the horizon. They've rarely needed a written procedure, and now they're being asked to brief a machine that needs one. The professional hierarchy quietly inverts.

The practical exercise Kamil suggests is blunt and effective. Pick one task you want AI to handle this week. Write the instructions like you're handing them to a five-year-old who has never made a sandwich. Plate. Knife. Order. Flip. Then read your prompt out loud. Every time you catch yourself thinking "obviously" or "you know what I mean," that's a missing step. Add it. Run that test once and you'll see what your AI has been seeing for months.

Key takeaways:

• AI has no common sense and will execute your instructions exactly as written, no more, no less
• Senior professionals often give worse AI prompts than junior ones because their expertise is tacit and hard to articulate
• Roles that already work with checklists and procedures adapt to AI more naturally than those who operate on instinct and judgment
• Improving AI output starts with writing instructions at the level of explicitness you'd need for a complete newcomer

Why do I care: This hits close to home for me because I've watched the same pattern play out in software teams repeatedly. Architects and senior engineers have deep mental models but they've internalized so much that they can't always surface what they know. Writing prompts forces the same discipline as writing good documentation or a clear code review comment: you have to externalize assumptions. The teams I've seen succeed with AI fastest are the ones who already had strong runbooks and clear processes. It's not about the prompting tricks. It's about whether your team can write down what it actually does.

A Kindergarten Teacher Just Taught the Best AI Prompting Lesson on the Internet