Human+AI.
Practices for thinking, deciding and creating alongside intelligent systems — without quietly disappearing inside them.
Six chapters · ~ 8 minute read
A writer told us, recently and without drama, that she had not written a sentence from scratch in three months.
She still considered herself a writer. She was no longer sure what that meant.
This is the quiet phase of the AI transition. Not the science-fiction phase. Not the productivity-revolution phase. The phase where capable people sit down at their desks and slowly realize they are not entirely sure where their work ends and the model's begins.
The right response is not to recoil from these tools, and it is not to surrender to them. It is to develop a serious practice for using them — one that keeps you sharper, not softer, as their capabilities expand.
This essay is that practice, in six chapters.
Before any technique, there is a posture. A way of standing in relation to the machine. Most people are inheriting their posture by accident — from the marketing of the tools, from peer pressure, from whatever video autoplayed yesterday. That is the most expensive way to encounter a technology this powerful.
The useful posture is neither awe nor refusal. It is something older and more useful: curious authority. You assume the tool is remarkable. You also assume you are the one responsible for what gets made.
Awe gets you a workflow you do not understand. Refusal gets you left out of a conversation that is shaping the world. Curious authority lets you experiment without losing yourself, and adopt without losing your standards.
The tool is not your collaborator. It is not your replacement. It is a remarkably capable instrument — and instruments do not choose what gets played.
Every model on the market is exquisitely sensitive to the quality of the question. This is not a prompting tip. It is the entire game.
The interesting truth is that most of the difficulty of intelligent work has always been in the question, not the answer. We just used to be allowed to hide that fact behind the effort of producing the answer. When the answer becomes nearly free, the difficulty walks out into the open and stands at your desk.
A well-formed question contains a model of what you actually want, a model of why you want it, and a model of what would count as a wrong answer. The people who get unusual results from these tools are the people who have done that work — usually with a pen, away from the keyboard, before the first prompt is ever typed.
You are upstream of the answer. Whatever you put into the river is what comes back down.
A healthy human–AI workflow is not a single transaction. It is a loop with four moves, in this order, every time:
- 01Draft
Write the rough version yourself. Not for quality. For ownership of the idea.
- 02Reflect
Hand it to the model with a specific request. Read what comes back not as the answer, but as a mirror.
- 03Refuse
Throw out the parts that don't sound like you, don't match your standard, or are subtly wrong in ways only you can see.
- 04Revise
Rewrite the result by hand in the parts that matter. The final mile is always human.
Skip the first step and you outsource the thinking. Skip the third and you publish someone else's voice under your name. Skip the fourth and the work is, in a quiet but real sense, no longer yours.
The loop is small. The discipline of doing it every time is not.
The most underrated skill in the AI era is the ability to say, calmly and without explanation, this part I will do myself.
There is a kind of work that is valuable precisely because it required you. A note written to a friend in your own clumsy phrasing. A decision you sat with for three days. A piece of code you understood line by line. A conversation you did not rehearse with a model first.
These are not inefficiencies to be optimized away. They are the places where character is built, where intuition is trained, where a self is kept intact. Delegating them feels like a win in the moment and registers as a loss across a year.
A serious AI practice always contains a list of things the tool will never be allowed to touch. That list is not a limitation. It is the shape of who you intend to remain.
Knowing what to delegate is half the practice. Knowing what to refuse is the other half — and the more important one.
As capability becomes abundant, taste becomes the differentiator. When anyone can produce a competent draft, the question that remains is which draft is actually good — and only a human with taste can answer it.
Taste is not a personality trait. It is an accumulated record of what you have paid close attention to, what you have refused, what you have rewritten, what you have loved enough to study. People who have spent years caring deeply about something develop a kind of internal compass for it. The compass does not get cheaper as models improve. It gets rarer.
The strange consequence: in an era of infinite output, the people who matter most are not the fastest producers. They are the slowest, most opinionated readers — the ones who can look at a hundred plausible drafts and say, with quiet certainty, this one, and not the others, and here is the small reason why.
The last mile of any creative or strategic work is taste. Taste cannot be prompted. It can only be earned, slowly, by paying attention to things you actually care about.
Posture, question, loop, refusal, taste — these are stances. A stance becomes a life only through small daily acts. The following five are the ones we keep returning to, both in our own work and in conversations with the people we guide.
Think first, prompt second
Write the first version of any important thought in your own words, even badly, before letting a model touch it. The thinking is the point; the prose is downstream.
Name what you are outsourcing
Before delegating a task, say out loud what part of yourself you are handing over — speed, memory, phrasing, judgment. Some are fine to delegate. Some quietly atrophy.
Keep one craft analog
Choose one thing — writing, drawing, cooking, code, conversation — that you continue to do without assistance. Not out of nostalgia. Out of self-preservation.
End with a human pass
Whatever the model produces, the last edit is yours. Not for grammar. For voice, for taste, for whether you actually believe what it says.
Track your own surprise
If nothing in your week genuinely surprised you, the model is choosing your inputs. Reintroduce friction on purpose — a long book, a difficult person, an opposing view.
None of these are dramatic. None of them require a manifesto. They are simply the small, repeated movements that keep a human recognizably human as the surrounding intelligence keeps expanding.
"The goal is not to use AI well. The goal is to remain a person worth being, while using it."
What we are practicing for.
It is tempting to treat the rise of intelligent systems as a question about productivity. Most of the public conversation does. But the deeper question — the one that will quietly decide what the next twenty years feel like to live through — is a question about formation.
What kind of mind do you want to have in ten years? What kind of judgment? What kind of attention? What kind of voice?
Every interaction with one of these tools is a small vote on that answer. Used with curious authority, they sharpen you. Used without one, they slowly do your thinking for you, in ways that are very difficult to notice from the inside.
A practice is the difference between the two. Posture, question, loop, refusal, taste — repeated long enough to become a temperament, not a technique.
Intelligence is becoming abundant.
The humans using it well are not.
Where you might go from here.
Human Advantage Assessment
A 15-minute reflection that surfaces your unique human signal — the part no model can produce on your behalf.
Find your North Star 02North Star Blueprint + GPS
Your personal navigation system. A Blueprint of who you are, paired with a year of guided GPS to stay clear as the tools keep shifting.
Get your Blueprint + GPS 03Guided Programs
For individuals, teams and organizations learning to think, decide and create alongside intelligent systems — on purpose.
Start a conversation