Sixteen drafts. That is what I had done to the thing before another human being touched it.
The piece was for MillerKnoll’s Better World Report, our annual accounting of what we did with the year, and I had been asked to write about the AI work. I took that seriously in the way a person takes something seriously when they are slightly overcompensating for not being sure they belong on the list of contributors. Version one was long and pleased with itself. Version sixteen was 1,050 words, down from 1,586, and I remember sitting back from it with the specific glow of a man who believes he has just cut the fat.
Reader, I had cut nothing. I had rearranged the fat into a nicer shape.
Then Mimi wrote back. Mimi is the report’s outside editor, a professional, and her note was so gentle that it took me a full read to understand I had been mugged:
“It’s very enthusiastically written, but I’m afraid I had to massage it a bit to fit our needs. The piece came in much longer than we have space for. Ideally these articles should be between 500 and 800 words. You’ll see I’ve made some cuts.”
Very enthusiastically written. I have thought about that phrase more than is healthy. It is the editorial equivalent of your mother-in-law calling your casserole “interesting.” It is not an insult. It is not not an insult. It sits in a small warm room between the two and hums.
She also wanted a forty-word bio intro. She wanted more technical specificity, and her actual question was: “are you orienting MK teams to chat bots, LLMs, machine learning?” Which is a fair question, and one I had spent 1,050 words gracefully declining to answer. And she wanted two more numbered lessons, so the piece could end in a bulleted list and, in her words, “appease the AI/SEO gods.”
I loved that line. Somewhere in a spreadsheet, the gods are counting.
I turned it around in under twenty-four hours, partly because deadlines are the only reliable form of discipline I have ever encountered and partly because if I had slept on it I would have argued. Final count: 784 words. Sixteen under the ceiling. I did not need to check that number twice but I did.
What I wrote back to her: the cuts were fair and the piece is better for the discipline, even if I winced once or twice.
Both halves of that sentence are true and they do not get along.
The casualty list, briefly.
Gone: a small confession about being tired of proving I’m human, which I still think was the best paragraph in the draft and which I now understand was best in the way a solo is best, meaning it was about me and not about the reader. Gone: a whole run about writing something on a mirror, which I had been carrying around for months waiting for a place to put it. Gone: an entire passage about tuning the intuition, the part where you stop asking whether the tool is right and start noticing when your own gut goes quiet. That one I still mourn on the anniversary.
Gone, also, as a closing line: “How can I help?” Mimi did not kill it outright, and I ended up keeping a version of it in the middle, folded into a smaller sentence: one four-word question we should probably ask more, how can we help. It works there. It is a better piece for having it there. I would still rather have had it last.
The bio came in at forty-six words against her forty, which I would like the record to show is a rounding error and not defiance. I wrote into it the line “He is as far from an engineer as they get,” because if I am going to be the AI guy in a furniture company’s sustainability report, somebody should say the quiet part in print.
The best trade in the whole revision was accidental. I had written “toolkit, prompt library, and curriculum from scratch,” which is accurate and heavy. It became “capabilities, curriculum, and confidence.” Fewer nouns, more meaning, and a little alliteration I did not earn but will be accepting anyway.
There was a headline fight, briefly. One candidate had “not AI” in it, and I killed it for reading too negative, which in retrospect was me protecting a program more than a sentence. We landed on “A Year of Learning Out Loud.” The three lessons the piece ends on: learning compounds, curiosity is a permission structure, and opting out is a legitimate answer. I stand by all three, including and especially the last one, which is the only one anybody ever pushes back on.
Seventeen passes. Two humans, one of them a professional. And the published piece contains the phrase “what keep won’t be the tools,” which is not a sentence, it is a ransom note. There is a missing period after “no-brainer.” And my favorite: a passage got moved late, and in the move, “trading tips and tricks” became “trading tips and training.” Which sounds like something a consultant would say on purpose.
I love these errors. I do not want them fixed. They are the physical evidence that the process is not a machine, that two careful people can hold a document up to the light seventeen times and still miss the thing sitting in the middle of it, wearing a hat.
I have every model in the world available to me at all hours. I drafted with them. They were genuinely useful, in the way a very fast, very agreeable colleague is useful. And none of that made the piece good. What made the piece good was a woman with a word budget and a deadline telling me no. Generation is free now. Constraint is the scarce input. I teach iteration and verification for a living, and I evidently believed iteration meant me, alone, sixteen times, which is not iteration. That is pacing.
Sixteen drafts got me a shorter version of the same piece. One editor got me a better one, and she did it by not letting me decide what stayed. I would like to report that I have made peace with this. What I have actually done is send her the next thing.