What Parenting Taught Me About Prompting AI Agents
A confident "yes" and a bone-dry toothbrush taught me more about working with AI than any tutorial. On hallucination, verification, and why setting context is a deeply human skill.
Every summer our house runs on Summer Rules: a short list of things the kids have to finish before any screen turns on. Make the bed. Read. Chores. Brush your teeth.
That last one has taught me more about working with AI than any tutorial.
“Did you brush your teeth?”
“Yes.”
“…when did you brush them?”
“This morning.” Confident. Specific. Completely wrong.
I know it’s wrong because I’ve learned to check the toothbrush. Bone dry. And here’s the part that gets me every time: only when I hold up the dry toothbrush does the kid actually realize they didn’t do it. They weren’t lying, exactly. They’d half-convinced themselves it was true.
The confident “yes”
I see that exact pattern working with large language models almost every day. Ask an agent whether it ran the tests, updated the config, or handled the edge case, and you’ll often get a confident yes. Ask for more detail and it will happily produce detail: a plausible account of work that didn’t happen. It isn’t lying. It’s doing the same thing my kids do, reporting the answer it expects is true instead of the one it verified.
The industry calls this hallucination. Any parent would recognize it immediately.
Check the toothbrush, not the answer
The fix is the same in both cases, and it’s the single most important habit I bring to agent work: verify against ground truth, not against the self-report.
I don’t ask my kids whether they brushed. I check the toothbrush. I don’t ask an agent whether the tests pass. I look at the actual test run, the actual file, the actual state of the world. The claim is not the evidence. The dry toothbrush is the evidence.
This is the thread that runs through everything I write about AI: the machine can do the work, but the human keeps the judgment and the accountability. You delegate the task. You do not delegate the checking.
“Clean your room” is a bad prompt
The other half of parenting is upstream of the toothbrush. It’s getting the instruction right in the first place.
Ask a kid to “clean your room” and you’ll get a wide range of interpretations, most technically compliant and none of them what you meant. Everything shoved under the bed? Room is clean. The instruction was ambiguous, so the outcome was a gamble.
Agents are the same. Vague in, vague out. Parenting cured me of assuming shared context. I learned to say “clothes in the hamper, books on the shelf, floor clear enough to vacuum,” because specificity is the difference between the result you wanted and the one you have to redo.
Say the why, not just the what
“Because I said so” doesn’t travel well, either. When my kids understand why (we tidy before dinner so we can find things tomorrow), they make better calls in the gray areas I didn’t spell out. Give an agent the intent behind a task and it handles the cases you didn’t anticipate; give it a bare command and it follows the letter and misses the point. Good context isn’t just instructions. It’s enough of the “why” that the agent can fill gaps the way you would have.
Managing more than one at a time
Any parent of several kids has learned to keep multiple, very different operations running at once: one who needs a lot of direction, one who just needs to be pointed at the goal, one who will test every boundary you set. Orchestrating a few agents with different strengths feels startlingly familiar. So does the patience it takes to calmly rephrase when you’ve been misunderstood, instead of just getting louder.
Why this is the point
I run Xyberal Solutions, and a version of this is the actual work: helping teams get real leverage from AI by getting the human part right. The tooling changes every few months. The durable skills do not: stating intent clearly, giving enough context, and above all verifying against ground truth instead of a confident claim. And they’re profoundly human.
That’s reassuring, honestly. What makes you good at directing agents isn’t a new technical trick. It’s clear thinking, clear communication, and the discipline to check the toothbrush: the same things that make you a good manager, a good teacher, or a parent trying to get everyone to the car with both shoes on.
The tools are new. The skills are ancient. I just happened to get a lot of practice before breakfast.
Have a project in mind?
Let's talk through the path from idea to production.