The July cohort of Supermanagers will be fully async; if you have been interested in the course but scheduling has been an issue, now’s your chance. It will also be the last time I run it until October. All students get lifetime access to the materials + any future class updates.
I recently ran a survey of 100 leaders about their biggest AI-related pain points. One of the most common answers I heard was a frustration with how badly everyone around them was using AI: “Everyone is generating a million garbage AI documents and no one reads them.”
This struck me as a solvable problem: hold people to a higher standard. A manager should be able to say, clearly and directly, that someone’s work is not meeting the bar.
So I posted a poll on LinkedIn: when you receive an obviously AI-generated document from a work colleague, do you say anything to them about it being obviously AI-generated?
Nearly 600 people voted, and only 17% said yes.
The most common answer was “it depends on who they are,” which makes sense to me. As frustrating as your AI-brained boss may be, you are wise not to tell them off. But the second most common answer was “it depends on something else,” which surprised me. What else?
This comment from Ken Romano was illuminating:
To me it’s less about ‘was it AI generated’ It’s “was it AI generated and high quality/accurate” or “was it AI generated and low quality/inaccurate” Another framing is “did it take more time for me to process/correct/fact check/understand it than it did for you to create it?” In other words, are you passing off your time savings to me?
Anna Bergevin added,
A good document takes work no matter what path it took to get there.
The question is not: did a robot touch this?
The question is: what work got skipped, and who pays the cost?
Sometimes the cost is paid by the reader. You saved yourself twenty minutes by producing something bloated, vague, and inaccurate, and now someone else has to spend forty minutes figuring out what you meant.
Sometimes the cost is paid by you. You saved yourself the discomfort of wrestling with an idea, and in doing so diminished your own cognitive ability.
Sometimes the cost is negligible.
This got me wondering: is there a good way to write with AI?
When in doubt, ask yourself three questions:
What is the writing supposed to do for the reader?
What part of the writing process am I asking AI to do?
Is this a one-time shortcut, or a repeated process?
Not all writing is equal.
If you are writing a status update to your manager, the goal may simply be to communicate information clearly. In that case, all that really matters is that the update is accurate, clear, useful, and timely, so using AI is probably fine.
If you are writing a wedding toast, the goal is different. You are not merely trying to communicate information about the bride and groom. You want them to feel loved and seen, which comes from putting time and thought into the difficult act of writing a toast. If everyone learns later that the toast was mostly generated by AI, you have undermined that goal.
Performance reviews are similar. Writing them takes forever. They often get squeezed into the margins of already overfull weeks. They do not usually advance your own work. So, of course, managers are tempted to use AI to write them. But a performance review is a signal of care, attention, and judgment. When I ask people about moments when they lost faith in a manager, “They totally phoned in my performance review” comes up more often than you might expect. So I advise managers to tread carefully here.
Original essays, reporting, and criticism are their own beasts. The reader is showing up because they expect a mind at work. So realizing that the work is AI-generated, even partially so, makes them feel misled, like finding out the ingredients on the package are not telling the truth.
So the first question is: what promise is this piece of writing making to the reader? Is it promising information? Care? Attention? Original thought? The more the writing depends on anything besides information transfer, the more cautious I would be about outsourcing the parts that create those things.
The second question is more granular: what exactly are you asking AI to do?
Before you can make decisions about where AI might belong, you have to break writing down into its component parts.
What we call “writing” is a long process exercising a wide variety of skills. In the course of writing an essay, you might:
Have an idea
Capture the idea
Research the topic
Synthesize a variety of inputs
Talk it out with a friend
Externalize thoughts onto the page
Structure your thoughts
Form a clear point of view
Write a draft
Rewrite with a different framing
Despair (without quitting)
Poke apart your arguments
Identify gaps and research again
Rewrite
Edit for clarity
Edit for narrative
Send it to someone else for feedback
Copyedit
AI can help in different ways at each one of these points. But its help comes at a cost, which is that in outsourcing a step, your ability to do that thing will atrophy. Sometimes this doesn’t matter; the speed or efficiency gains more than make up for the lost muscle, or maybe you weren’t even doing that step to being with. Sometimes this matters a lot, and if you are not careful this will hollow out a skill that matters an awful lot to you.
Of course, AI can do more than enable shortcuts. At any one of these steps, it can help you reach a quality bar you otherwise would not have been able to reach yourself, sometimes by sheer force of quantity. For example, you may have in your head a few different possible framings of an essay. You may only have time to sketch out one or two of these versions. But you could ask AI to write you 10 or 20 different framings, essentially prototyping out the essay, so you can get a quick read on which frame is most compelling, then throw all the prototypes out and sit down to write it yourself.
If you have never thought about writing in such a granular way — if you, perhaps, skip straight from “an idea popped into my head” to “I asked AI to turn it into an essay” — the best advice I can give you is to try to break the process down more. (My Custom GPT Wally the Writing Partner helps with this; the prompt is behind the paywall if you want to turn it into a Skill).
Taking shortcuts here and there will probably not have any serious consequences. It only becomes a problem if you take that shortcut every time.
This is why I can't give a simple answer to how I use AI in my writing: my process bends to the needs of the subject at hand. I'll reach for it at various steps, but I'm careful not to let my brain get comfortable skipping steps for too long. If I notice I'm letting an important skill get soft, I'll cut it out of that part of the process for a while. Some essays, I get stuck on a framing or a conclusion and ask for a little help getting over the wall. And I go through whole stretches where I don't touch it at all.
So if you want to write well, here’s my advice: whatever process you land on, keep changing it.
The version of this essay that you are reading is my fourth rewrite. It is quite a different essay than the one I set out to write. Halfway through the second draft, stuck, I asked my husband for help. After the third rewrite, I fed it to Claude and Codex, asked both for notes, and did another pass using the ones I agreed with (about half). When I finished that version, I sent it back to my husband for one more read. Then I went through it again and rewrote swaths of it.
It's a very different process from how I write for social media (I like putting a lot of time into my newsletter writing, but want to spend as little time writing social media posts as possible) or how I make my course content (most of which comes from videos and demos I've already recorded).
Sometimes, taking a shortcut is a problem; other times, it is smart. When writing a product requirements document, there are some details you should spend much longer working through than might feel natural, and others where additional thought is a waste of time. Knowing which details are which is not always straightforward, which is why developing the judgement is so important.
The one step I’m happy to hand off to AI completely is capturing and organizing my ideas. My brain is a leaky sieve, and my scattered notes docs were not cutting it. Unlike the rest of my creative process, this part benefits from consistency — plus, there’s no insight lost in sorting a notes doc the same way every time.
It's not lost on me that much of the enterprise AI story runs on the opposite bet: that you can codify a process and run it the same way, every time. That works when the output is a commodity; when, as with organizing my notes, no insight is at stake. But when the value of the work depends on a person improving along with the system, a fixed, repeatable process is, unless very carefully designed, unlikely to produce good work over time.
I gave a talk in business school about a hero of mine, The Cat in the Hat. I love the book as a model for how you make new futures out of fixed constraints.
the sun did not shine.
it was too wet to play.
so we sat in the house
all that cold, cold, wet day.
In college I used to write things like this all the time, for fun. But sitting down to write that talk, I realized something jarring. In the years between studying literature and studying business, I had fallen completely out of practice. I hadn’t just stopped writing the things I used to work hard on; I’d picked up a stack of bad habits from powering through “business” assignments where speed mattered more than quality. Worst of all, I hadn’t noticed it happening. I only found out when I tried to write about the Cat in the Hat and could not stick the landing. Framing the piece, setting up the argument, weaving a story — all of it evaded me.
This was troubling, because I had always considered “being a good writer” a core element of my identity. But here I was faced with irrefutable evidence that writing was just another muscle that would atrophy without focused attention.
And, in fact, it had.
xoxo,
hils
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.