A teacher asks what makes a child special. A customer asks why a replacement hasn’t arrived. Both questions might land on the same kitchen table at the end of a working day. An AI assistant can draft a fluent answer to either. Only one of those answers can be checked against a delivery record.
That distinction helps explain the reaction to Dazzle AI CEO Marissa Mayer’s account of using an assistant to help describe her daughter for school. On November 1, TechCrunch’s Anthony Ha reported on an Equity discussion about why executives’ favorite uses for AI don’t always sound appealing to the people they’re meant to persuade. Mayer’s example raises a more useful question than whether a parent should ever ask for writing help: which part of the task did the parent want help with?
Finding the words and finding the memory are different jobs
Imagine a parent who remembers exactly what to say: her daughter pulls up a chair for the new kid at lunch, even when her own table is full. The parent is writing in a second language and wants help making that sentence clear. The observation is hers. A tool can help the teacher hear it.
Now imagine starting with a blank box and asking for three flattering qualities. “Kind” might be accurate, but the teacher loses the lunch-table detail. In the first case, the assistant helps carry an observation from one person to another. In the second, it can make the form look finished before anyone has supplied the most valuable part. That’s a distinction about the task, not a verdict on Mayer’s parenting.
Where the evidence for saving time is stronger
Customer support offers a case where researchers could measure the outcome. In a 2025 Quarterly Journal of Economics study, Erik Brynjolfsson, Danielle Li and Lindsey Raymond examined the phased introduction of an AI assistant among 5,172 support agents. With access to the tool, workers resolved 15% more issues per hour on average.
The average hides the finding that matters most to someone buying software for a team. Less experienced workers improved both speed and quality. Among the most experienced, highest-skilled agents, the researchers found small speed gains alongside a small decline in quality. A manager tracking only the team’s hourly total could miss that trade-off entirely.
The study measured resolved support issues in one workplace. It didn’t measure how much free time workers took home, and it says nothing about whether a family wants an assistant involved in a personal conversation.
Real Talk
Consider the missing delivery. If a worker has already checked the order, an assistant can turn the confirmed facts into a clear reply. If nobody knows where the package is, a polished promise to deliver tomorrow creates work for the next person who answers the phone. The useful measure is whether the customer got an accurate answer without having to ask again. Minutes saved writing the first email tell only half the story.
💡 IskraCore Take
Before using AI for a recurring task, decide what a good result looks like to the person receiving it. For a support customer, it might be a reliable delivery date. For a teacher, it might be one detail only a parent would know. Hand off the wording when wording is the hard part. Keep the observation or the decision with the person who can actually make it.