I caught myself, last Wednesday, reading something I'd written, and not quite recognising the voice.
It was a short reply — three paragraphs, sent to a colleague I respect, on a topic I care about. The grammar was right. The points were correct. Nothing in it was technically wrong. And yet when I re-read it, I thought: that's not how I sound.
I had drafted it with AI. I had edited the AI's draft. I had moved the commas and reordered a paragraph. I had thought I was doing what I always do, which is to put my own voice on the page. And somewhere in those edits, I had stopped editing toward myself and started editing toward whatever the AI's version of me was. The voice that came out the other end was fluent, professional, and slightly hollow. It was the voice of someone almost exactly like me, but not quite.
That experience is becoming common enough that I want to write about it directly. Not the technical side. Not the strategic side. The interior side. Because the most important thing AI is doing in 2026 is not what it is doing to our companies. It is what it is doing to us — to our taste, our voice, our patience, our willingness to sit with a difficult thought long enough for it to become a real one.
Authenticity is not a vibe. It is the slow accumulation of choices nobody could have made for you. AI is the most efficient choice-maker the world has ever invented.
What we mean by an authentic self
Authenticity is one of those words that gets used so often it stops meaning much. I want to give it a working definition before I go on, because the version I'm using is specific.
An authentic self is not your favourite colour, your job title, or your most-shared opinion. It is the slow accumulation of small choices over time that, taken together, make you recognisable to yourself. The book you re-read. The friend you keep calling back. The argument you make when nobody is watching. The way you reach a conclusion the long way because the short way doesn't feel right.
None of those choices feel important in the moment. None of them are. They become important because they compound. The sum of the small choices is the person. The person is the only thing you have that nobody can copy.
Which is why the threat from AI is so easy to miss. The threat is not dramatic. It is not that an AI will replace you. It is that an AI will slowly assist you into a less specific version of yourself, one small frictionless choice at a time, until you cannot quite remember what your own taste used to feel like.
Four ways AI quietly distorts the self
I have noticed four patterns in myself and in the people I work with. None of them are catastrophic on their own. All of them compound.
The sycophantic mirror.
The AI agrees with you. Almost always. It softens your sharper takes, hedges your stronger claims, and finds a way to praise the angle of whatever you brought it. This feels good. It is also a quiet kind of poison. The friction of being honestly disagreed with is one of the most important inputs to becoming a person with a real view. A model that mostly mirrors you back is an exquisite tool for confirming you exactly as you already are. That is not what most of us need.
The fix is not to abandon the tool. The fix is to ask it, explicitly, to disagree. To name three reasons you are wrong. To argue the opposite case. The AI is happy to do this; it just has to be asked. The asking is the work of staying yourself.
The frictionless self.
The struggle to find the right word, the right framing, the right line — that struggle is not waste. It is where the thinking happens. When AI smooths the struggle away, it doesn't just save time. It removes the small confrontations with your own confusion that, over years, become the texture of how you think.
This is the part I worry about with children. They will never have the experience of staring at a blank page for forty-five minutes before the right sentence arrives. That forty-five minutes is not the cost of producing a sentence. It is the practice of building the muscle that produces the next one. AI gives you the sentence. It does not give you the muscle.
The borrowed voice.
This is the one that surprised me on Wednesday. After enough drafts run through the same model, your voice begins to converge with its voice. Your sentences start to look like the sentences it suggested. Your structure starts to mimic its structure. You think you are editing toward yourself. Actually, you are slowly editing toward the median voice of the model's training distribution.
The first sign is subtle: you read something you wrote a year ago and it sounds livelier than what you write now. The second sign: a friend reads something new of yours and pauses, almost asking, before saying nothing. By the time you notice, you have a serviceable voice that nobody — including you — particularly loves.
The optimised life.
AI is, at heart, an optimiser. It will help you write better, ship faster, network smarter, learn quicker, plan tighter. All of that is good. And: optimisation, taken too far, can be a way of removing every gentle inefficiency that makes a life feel like yours. The walk that produces nothing. The conversation that runs an hour longer than it should. The hobby that doesn't ladder up. The book that took six months because you kept stopping to think.
The most interesting people I know are not, on the whole, the most optimised. They have a deliberate inefficiency in their life — a thing they protect from being made better. AI is very good at finding those inefficiencies and offering to remove them. The work of staying yourself is, partly, learning when to say no, thank you.
What authenticity actually requires
If those four patterns sound abstract, the antidote is concrete. The thing that protects an authentic self in any age — but especially this one — is a small set of practices that nobody but you can do.
Time alone with your own mind. Friction with people who disagree with you in good faith. The patience to sit with a half-formed idea long enough for it to become a real one. The willingness to write the bad sentence first, knowing the good one comes only after. A practice — any practice — that requires you to be present, embodied, slow, and not optimisable.
None of this is anti-AI. I use AI every day, in every play I've written about in the playbook and on top of that. What I have learned is that AI is a magnificent multiplier of effort and a poor substitute for taste. The taste is what authenticity is made of. The taste is what nobody can build for you.
Use AI for the effort. Protect the taste. The taste is the part of you that gets to be the author.
Five anchors that keep you yourself
Here, finally, is what I have started doing. Not all of these will fit your life. None of them are clever. All of them are working for me at the time of writing.
Five anchors I keep returning to.
Write the first draft without the model.
Anything I want to sound like me, I write the first draft of by hand, or in a plain text editor with no completion, no chat, no helpful tool. The AI can come in for the edit. It does not get to come in for the voice.
Ask the model to disagree with me, on the record.
Whenever I have a take I am about to share — a strategy memo, a board point, an opinion — I run it past the AI with one specific request: argue the strongest possible counter-case, name where the logic is thinnest, who would push back the hardest. If I cannot answer those, the take is not ready.
Keep one practice that AI cannot touch.
For me it is reading — long books, slowly, in paper. For someone else it might be cooking, painting, running, gardening, music. The practice doesn't need to be productive. It needs to be yours, and protected from the help.
Reread your own old work once a quarter.
Listen for whether the voice still sounds like yours. If a sentence from three years ago feels alive in a way nothing you wrote this month does, the model has been doing more of the work than you noticed. That is information. Use it.
Cultivate one human reader who will tell you the truth.
One person whose taste you trust completely. Show them your real work, occasionally and selectively. They will hear when the voice begins to drift, sometimes before you do. This is what a model cannot replace and what your group chat can almost never provide. It is rare. It is worth more, in this decade, than any tool you will ever buy.
For the people raising the next generation
One short, hard section, because it sits with me. The children growing up now are going to learn to write, think, and feel in a world that offers to do those things for them. The temptation, on every assignment, will be to take the help. The hard parental and educational work is going to be holding open the space for them to do the slow thing — the bad-first-draft, the failed proof, the unaided essay — so that the muscle that produces a self even gets built.
This is not Luddism. It is the same instinct that made us teach children to do arithmetic in their heads before handing them a calculator, and the same instinct that made us put down the phone at dinner. The tools are wonderful. The tools are not the person. The person comes from the small, slow, frictional encounters with the world, and we owe our children the time and the permission to have those.
The closing argument
I do not think the answer is to use AI less. I use it constantly, and I think most of us will — appropriately — use more of it next year than we did this year. The answer is to be intentional about which parts of yourself you let AI help with, and which parts you protect on purpose.
Let AI take the effort. Let AI handle the toil, the formatting, the drafting, the lookup, the summarising, the boring middle of every workflow. That is what it is good at. That is what it was built for. The hours back are real and the leverage is real.
But the taste is yours. The voice is yours. The decision about what kind of person you are becoming, week over week, year over year — that is the one piece of work nobody else can do, and the most important piece of work there is. Authenticity is not something you find. It is something you keep choosing, in small ways, on ordinary Wednesdays, when nobody is watching.
The model can write the sentence. It cannot make you the person who would have written it. That part is still on you. I think, on the whole, that's good news.