Every life stage has work to do and a distance that can be calculated.
Hand over the parent’s clerical load; keep the child’s practice, errors and choices with the child.
The test is not “can an agent do it.” The test is whether delegating removes your clerical repetition or removes your child’s struggle — and those look identical from the outside.
Verification signals (two or more and the need is real):
· The task is repetition of something your child already did by hand
· You can check the output against the original notebook in minutes
· Nothing in the output explains, judges or decides on the child’s behalf
Task decomposability + accountability
A task can be split only where each step has a named input, a named output and an acceptance check. Re-typing a notebook splits cleanly: the input is the page, the output is a typed list, the check is a visual comparison. Explaining why an answer is wrong does not split, because the output is a change in the child, not on a page.
This is why the same tool can be useful to the parent and harmful to the child in the same evening.
Basis taken from the cell framework. All three must pass; a failed test is a stop, not a warning.
Can I hand the sorting and re-typing of my child's wrong-answer notebook to an agent?
Basis: Verification cost economics
If checking takes longer than re-typing, keep the paper notebook and skip the agent.
Can I hand the search for extracurricular activity information (classes, camps, open days) to an agent?
Basis: Principal-agent cost
If sources are missing or stale, use the agent only to draft search keywords, not to recommend.
Can I hand the daily schedule and routine reminders (homework start, bedtime, class times) to an agent?
Basis: Accountability
If the agent starts enforcing or judging, switch to a plain alarm the child controls.
My evenings stop being data entry
The child does the correction, not the machine
I know which activity is worth a visit before we pay
What to look at
Hours per week on school admin · wrong answers actually reviewed by the child · reminders missed
How to judge
After the agent’s output, did the child still do the correction by hand?
Healthy state
≥80% of admin automated · 100% of corrections done by the child · zero reminders missed
Let the agent explain the wrong answer and hand it over as-is —— the correction is the learning
Automate the child’s practice —— the struggle is the point, not the friction to be removed
Let a comparison table pick the camp —— the visit is the evidence, the table is only the shortlist
Trust a summary of what the child studied —— summaries hide what was skipped
List what you did for school this week and mark each item: repetition or judgement
Automate only the repetition, with your child able to check the output
Keep every explanation of an error with the child, in person
Let the agent tabulate options with links; decide after one visit
Re-test monthly: anything the child can now do alone stops being delegated
This page does not collect a child’s data and asks for none. It is about where the boundary sits, not about building a profile.