Skip to content
Kirk Drake
Back to the blog

All Three of My Kids Told Me They Don't Need to Write

8 min readAI in Education

Artificial Summer School series banner in navy and orange. Part 3 of 11: All Three of My Kids Told Me They Don't Need to Write.

They’re half right, and the half they got wrong is our fault

I asked each of them the same question, separately, on tape. Do you think you’ll need to write well long-term?

Three kids. Three no’s.

Violet wants to go premed and then into surgery. She said basic writing through about ninth grade matters and past that it doesn’t much. She’ll write patient reports. Maybe publish something eventually. What matters more, she said, is knowing what you’re talking about, and if the prose isn’t beautiful, she can run it through AI and get it polished. Then she said something sharper: she has three teachers who grade writing three different ways, in history and English and science, and she doesn’t see much value in learning to write in one particular teacher’s preferred style.

Warren was more specific. The job he wants isn’t author. He’ll send emails and write analytically now and then. But he doesn’t believe knowing MLA format or Chicago-style citations will matter to him unless he goes into academia.

He also pointed out, on camera, to my face, that his father has written multiple books and had the first one ghostwritten.

Jasper is fifteen and the most economical speaker in the family. I asked if he’d need to write well. He said no. I asked why.

“The AI can do it.”

That’s the whole answer. Four words. Jasper’s is the one that should worry us, though not for the reason it looks like.

Let’s deal with the ghostwriter

My son is right, and I want to be precise about it, because the precision turns out to be the entire point.

My first book was ghostwritten in the sense that I didn’t produce the first draft. I had the ideas, I did the interviews, I gave the direction, and a professional turned it into prose.

Then I materially rewrote it.

I fought with that manuscript for the better part of a year. And somewhere in there I hit a section where my own argument didn’t survive being written down in order. I’d been saying the thing out loud at conferences for two years and it had always landed fine, because talking lets you move fast past the soft spots. Nobody interrupts a keynote to say “wait, that doesn’t follow.” On the page, in sequence, the soft spot just sat there staring at me. I had to go find out what I actually thought, and what I actually thought turned out to be different from what I’d been saying.

Here’s what I want you to notice. That didn’t happen while I was generating prose. It happened while I was reading someone else’s version of my ideas closely enough to catch where it was wrong.

I’ve since done a second book, and I’m in the middle of a third that also involves a co-writer and a lot of session work. I’m not going to pretend I’ve sworn off help. What I’ve learned is that the help isn’t the variable. The variable is whether anything in the loop is capable of telling me I’m wrong, and whether I’m reading closely enough to hear it.

A ghostwriter who just cleans up my ideas can’t do that. A ghostwriter I argue with can. AI that I accept on the first pass can’t. AI I argue with can.

It was never about who types.

Where the kids are right, and where schools should be embarrassed

Violet’s complaint about three teachers grading three different ways isn’t a teenager dodging work. It’s an accurate description of what’s being measured. When the same essay earns different marks in history and English and science based on stylistic preferences nobody ever wrote down, the skill being trained isn’t writing. It’s reading the grader.

That’s a real skill, by the way. It’s most of corporate life. But it isn’t what anybody claims to be teaching, and the kids can tell.

Warren’s point about citation formats is just correct. MLA and Chicago are conventions for a profession he isn’t entering, taught with a moral weight wildly out of proportion to their function. All three of my kids can now produce a flawless bibliography in about four seconds. Defending that as a skill in 2026 is defending a corpse.

The five-paragraph essay has the same problem. It was a scaffold. Scaffolds are supposed to come down. Defending the scaffold after the constraint changes is not the same thing as defending writing.

If the case for writing instruction is that students should produce correctly formatted prose in an assigned house style, my kids have won the argument and it isn’t close. The machine does that better, faster, and without complaining.

Ask AI to write something without much direction and you get the statistical middle: competent, conventional, average-ish prose. We enjoy mocking that as “AI writing.” There’s an awkward math problem with mocking average writing, which is that roughly half of us are below the median. (Yes, I switched from average to median. Somewhere a math teacher just unclenched.) And almost everyone reading this has already placed themselves in the good half, which suggests either an extraordinary demographic anomaly among my readers or that humans remain delightfully human.

The economic point is simple. If AI can produce competent prose instantly and for free, being an average writer just lost most of its market value. If you are exceptional, with original ideas, a voice, humor, taste, AI may make you more powerful. If your professional moat was producing a grammatically correct 900-word memo by Thursday, I have less encouraging news.

What those three actually did all summer

But none of that is the case for writing. And the reason my kids don’t know that is on us.

Warren built a t-shirt company that scans the internet for trends, generates designs, publishes them, and is close to running its own ad campaigns and selling autonomously. To get there he had to specify what he wanted, precisely, hundreds of times, in language exact enough that a system with no context and no ability to read his mind could act on it.

Jasper built a study platform with a cost-optimization layer underneath it, a mobile app on top, and an email agent that answers questions sent to it. Same requirement. Say exactly what you mean, in writing, or receive something you didn’t want.

Violet started the summer not knowing what a terminal was. She got herself unstuck by writing instructions like explain this to someone who doesn’t know what they’re doing. That isn’t a trick. That’s a person accurately diagnosing her own state of understanding and articulating it clearly enough to get help. Half the executives I work with can’t do that in a meeting.

And here’s Violet again, describing what she’d tell her teachers about how to use AI on a paper: ask it for edits, tell it not to change your wording or your sentences, get its read on what could be improved, then go back and make the changes yourself.

That is a description of an editorial loop. She invented it independently, at fifteen, because it works. It’s approximately what I did with my first manuscript.

All three of them wrote constantly for eight weeks. Specs. Prompts. Instructions. Corrections. Precise descriptions of what was wrong with the thing they got back.

Not one of them counts any of it as writing.

The actual finding

The OECD has started drawing almost exactly this distinction. Its work on AI and the writing curriculum separates writing as artifact production from writing as a way to summarize, structure, self-explain and critically appraise ideas. That is the argument I care about. I am not trying to preserve manual prose production as a heritage craft. I am trying to preserve the cognitive friction that tells you when your idea is nonsense.

Ten years of school taught my children that writing means an essay, in a format, for a grade, judged by a teacher’s taste. So when the machine took over the format, they concluded writing was over, while spending an entire summer doing the harder half of it.

The harder half is the only half that ever mattered. Writing is how you find out whether you actually know something. It’s a thinking instrument that happens to leave a paper trail.

That function has not been automated. It’s more valuable now, and more urgent, because we have a machine that will cheerfully produce five hundred fluent words about anything at all, including things you don’t understand, and hand them over with total confidence.

Fluency just went to zero. Fluency was never the product. The product is knowing whether what you just produced is true, whether the argument holds, and whether you’d bet money on it. And the cheaper fluency gets, the harder it is to notice when you’ve skipped that part.

Violet said it herself without seeing where it pointed: it’s more important to know what you’re talking about than to write beautifully. She’s completely right. And the way you find out whether you know what you’re talking about is usually by trying to write it down and watching it fall apart.

Jasper’s four words are the pure form of the confusion. The AI can do it. Do what? It can do the prose. It can’t do the knowing. And nobody has ever told him those were two different jobs, because for his entire education they’ve been graded as one.

So what do we do

Stop teaching the half that’s over. Kill the citation-style exams, the five-paragraph mold, the unspoken house preferences of individual graders. My kids are right that all of it is finished.

Then put the other half at the center of the whole enterprise. Argument. Precision. The specific discomfort of a claim that won’t hold once it’s written down in order. Make students edit: their own drafts, each other’s drafts, and yes, the machine’s drafts, with the reasoning made visible. Then have them defend it out loud for ten minutes. An oral defense is not magic, but it will tell you considerably more about ownership and understanding than another probability score.

And tell them the part nobody has told them, because it’s the whole thing: you don’t write in order to produce a document. Anyone can produce a document now.

You write because it’s the cheapest way ever invented to catch yourself being wrong.

I paid a professional and spent a year rewriting her work to figure that out. My kids can have it free.

Next: Violet asked the AI to build something she was sure was impossible, and it did. Then it couldn’t change the color of a heading. What that inversion tells us about which skills are actually scarce now.

Sources and further reading

  • OECD, “Evolving AI capabilities and the school curriculum: Emerging implications and a case study on writing,” 2025. Source
  • OECD, Digital Education Outlook 2026, 2026. Source

Want this energy at your event?