The Coaching Note
Two weeks ago, I got a text from my friend Eric Porres about my latest newsletter.
The entire message was a list. A handful of phrases, quoted back to me, with one label attached: Claude. "That's it, that's the whole leadership move." "Three things worth watching for as you read." No commentary. Just the receipts.
I asked if the tells had ruined the piece for him.
"Not at all," he wrote back. "But when you see it, you can't unsee it."
If you've heard Eric on Beyond the Prompt, you can imagine exactly how this felt. The insight that makes him such a great podcast guest is every bit as brutal when it's pointed at you.
To state the obvious: Eric was not trying to discourage me from writing with AI. He is the chief AI officer of Logitech. He's one of the most sophisticated AI-assisted writers I know, one of the most sophisticated AI-assisted builders I know, and one of the most sophisticated AI-assisted leaders I know, helping unleash one of the most sophisticated AI organizations I've seen. He's even written publicly about why he doesn't hide that AI helps him write.
But that's not how it felt.
The artisanal retreat
My first instinct, I'll confess, was to go backward.
I guess I have to go back to handcrafted. Artisanal. Organic, free-range prose, typed by hand the way my grandmother intended.
I've started calling that reflex the artisanal retreat: responding to criticism of your collaboration by collaborating less.
It feels virtuous. It feels humble. And it moves in exactly the wrong direction, because it misreads what kind of thing collaboration is.
Working with AI is a practice, not a position.
Positions are binary. When a position gets criticized, you have two moves: defend it or abandon it. Practices have a skill gradient. When a practice gets criticized, you have a third move: train.
Nobody asks whether you're "for or against" delegation after one badly delegated project. They ask what you'll brief differently next time. Eric's text belongs in that category. My flinch put it in the other one.
Everybody argued about the wrong thing
You may have watched this exact confusion play out at national scale last week.
Stanley Druckenmiller published an op-ed in the Wall Street Journal criticizing Treasury Secretary Scott Bessent. Within hours, readers ran it through Pangram, the AI detector, which scored it 100% AI. Druckenmiller didn't flinch: "Of course I used AI... I write everything using AI now for the same reason I use a calculator when I do math problems. My name is on the piece. It's my message."
The discourse machine did its thing, and nowhere more perfectly than among the co-hosts of the All-In podcast. Jason Calacanis called publishing AI-written work without disclosure "unforgivable." Chamath Palihapitiya called disclosure demands "virtue signaling." David Sacks asked why it matters, if the idea was Druckenmiller's and he approved every word. Meanwhile the Journal's own opinion editor called AI "a fact of modern life."
Four smart people, arguing about whether he should have.
Not one asking how well he did.
Because there was a legitimate critique available, and it went almost entirely unmade: the piece had tells. Commenters flagged the "it's not X, but Y" constructions; one called them "Claude-isms." (I know that tell intimately. It's item one on my own kill list.) The collaboration was shallow enough to leave fingerprints. That was the critique worth making, and almost nobody made it.
Eric said to me privately what nobody said to Druckenmiller publicly.
The vow is the floor
Reid Hoffman wrote the best defense of the position I've read, in an essay called The Pangloss of Pangram. His argument lands on what he calls a vow: "My name on a page is a vow. I invested energy in these thoughts. I am present in these words." The true transgression, he says, is hitting a button, not bothering to weigh the output, and publishing it as your own. Regardless of the tools, a human must mean what is said.
He's right. And the vow is the floor.
Druckenmiller kept the vow perfectly. His argument, his name. He still shipped the tells. Which means sincerity can't be the whole standard.
Everything between "I meant it" and "it's excellent" is the practice. Reid defends the right to collaborate. The practice is where the craft lives. And the practice is where the slop lives, too.
What deeper actually looked like
Here's my n-of-1 experiment, one week long.
The piece Eric flagged and the piece I published last week, The Future's Losses Have Names, were made by the same human and the same AI. The model didn't improve in seven days. What changed was the management.
On the second piece, I checked in constantly instead of reviewing once at the end. I rejected suggestions, with reasons, and watched the reasons improve the next round. And when Eric's critique arrived, I didn't just fix a draft. I turned his list of tells into a standing quality-control system my AI now applies to everything we make together.
Then that system failed me. The final draft went out for review with twenty-three em-dashes in fourteen hundred words. The system had protected em-dashes because I love them, and never checked the density. My first instinct wasn't to scrap it. I updated it: authenticity protects a habit, never its frequency.
Checking in. Bringing judgment. Giving feedback that outlives the draft. Fixing your systems when they fail. In my book I argue these are exceptional management behaviors, and this is the part nobody advertises: they take real effort. The promise of effortless excellence was always wrong. What AI changes is what the effort is for: less production, more leadership.
Same human. Same AI. One week apart. The only variable was how deeply I engaged.
Put the feedback in the system
So here's the move, and it takes five minutes.
The next time you get a critique of AI-assisted work — from a colleague, a reader, your own wince on re-reading — don't just fix the draft. Install the standard. Tell your AI teammate:
"Here's a piece of feedback I received on our work: [paste it]. Turn it into a standing instruction you apply to everything we make together, starting now. Then show me three places in our recent work where it would have applied."
Then run the audit on your last collaboration:
1. Did I brief, or just prompt?
2. Did I review mid-flight, or only at the end?
3. Did my feedback update the system, or just the draft?
4. Did I bring judgment the AI can't have?
Any "no" is your next depth of engagement.
Criticism of your collaboration is a commentary on how you collaborated this time. Treat it like a performance review for a manager: the remedy for a bad one has never been to stop managing.
Eric sent me a coaching note. It took me a week to read it that way.
And Druckenmiller is right: his name is on the piece. But a name doesn't end the scrutiny. It assigns it. A byline answers exactly one question — who is accountable for the quality of this?
If your name is on the work, the AI-tells are yours too.
Go dePage_UpPage_UpPage_UpPage_UpPage_UpPage_UpPage_UpPage_Upeper.
(And yes: I wrote this one with Claude. Eric — tell me what you see.)
Related: The Future's Losses Have Names. Its Possibilities Don't.
Related: Be the Prompt
Related: The Bug Is the Brief
Join 54,147 creators and leaders who read Methods of the Masters each week.