The Questions AI Shouldn't Answer
Last week I sat down to write a difficult email.
You know the kind, the one you draft in your head for days before your fingers ever touch the keyboard. I had let someone down. Not catastrophically, but enough that it needed a real acknowledgment, not a breezy "so sorry, crazy week!"
So I opened a blank email. And then, almost on reflex, I opened a second tab to ask AI to help me find the words.
And then I stopped.
Because here is what I realized: if AI wrote my apology, I wouldn't really have apologized. I would have outsourced the one part that actually mattered - the wince. The sitting-with-it. The slow, uncomfortable work of finding my own words for something I wish I hadn't done.
In my last post, I wrote about the questions AI can't answer — the ones with no data-able answer, because the answer comes from inside you. This is the flip side: the questions AI could take a decent swing at, but shouldn't. Not because it would fail, but because succeeding would cost you something.
The hard stuff is hard for a reason. The difficulty isn't a flaw to be optimized away. It's the point.
When you struggle to write the condolence note, that struggle is you honoring the person who is hurting. When you sit in the discomfort of a hard conversation with someone on your team, that discomfort is you taking your leadership seriously. When you wrestle with whether something is the right thing to do, the wrestling is your character getting its reps in. Hand those to a machine and you get a tidy result and an atrophied muscle. The growth was never in the outcome. It was in the doing.
Some tasks are transactional, and some are relational. AI is a gift for the transactional — the scheduling, the summarizing, the first draft of the boring memo. I use it every single day, and I'm not giving it back. But the relational ones — the apology, the thank-you that actually means something, the "I've been thinking about you," the hard truth delivered with care — those are where our humanity does its quiet, essential work. Automate them and the other person feels it, even if they can't name it. Something rings hollow. We are remarkably good at sensing when words came from somewhere other than the heart.
So I've started using a test. Before I ask AI to write something for me, I ask one question: Is the doing of this the point?
If the point is simply to arrive at the finished thing, delegate away. But if the doing is the thing, if the effort itself is what makes it matter to me or to someone I love, then it's mine to do. It's a small question, and it has quietly reshaped how I spend my days.
Here are a few I've decided to keep for myself, no matter how tempting the shortcut:
How do I apologize to someone I've hurt?
What do I say to a friend who is grieving?
Is this the right thing to do?
Is there a lesson to be learned in the writing of it?
Am I being the kind of person I want to be?
None of these are efficient. That's the point. Some things are supposed to require something.
So the next time you're tempted to let the machine handle the human part, pause and ask whether the doing is the point. If it is, close the tab. Sit down. Find your own words, clumsy as they may be. They will be worth more than anything a machine could write, precisely because they cost you something.